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>
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", 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 vma_stop(struct proc_maps_private *priv)
128 struct mm_struct *mm = priv->mm;
130 release_task_mempolicy(priv);
131 up_read(&mm->mmap_sem);
135 static struct vm_area_struct *
136 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
138 if (vma == priv->tail_vma)
140 return vma->vm_next ?: priv->tail_vma;
143 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
145 if (m->count < m->size) /* vma is copied successfully */
146 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
149 static void *m_start(struct seq_file *m, loff_t *ppos)
151 struct proc_maps_private *priv = m->private;
152 unsigned long last_addr = m->version;
153 struct mm_struct *mm;
154 struct vm_area_struct *vma;
155 unsigned int pos = *ppos;
157 /* See m_cache_vma(). Zero at the start or after lseek. */
158 if (last_addr == -1UL)
161 priv->task = get_proc_task(priv->inode);
163 return ERR_PTR(-ESRCH);
166 if (!mm || !mmget_not_zero(mm))
169 down_read(&mm->mmap_sem);
170 hold_task_mempolicy(priv);
171 priv->tail_vma = get_gate_vma(mm);
174 vma = find_vma(mm, last_addr - 1);
175 if (vma && vma->vm_start <= last_addr)
176 vma = m_next_vma(priv, vma);
182 if (pos < mm->map_count) {
183 for (vma = mm->mmap; pos; pos--) {
184 m->version = vma->vm_start;
190 /* we do not bother to update m->version in this case */
191 if (pos == mm->map_count && priv->tail_vma)
192 return priv->tail_vma;
198 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
200 struct proc_maps_private *priv = m->private;
201 struct vm_area_struct *next;
204 next = m_next_vma(priv, v);
210 static void m_stop(struct seq_file *m, void *v)
212 struct proc_maps_private *priv = m->private;
214 if (!IS_ERR_OR_NULL(v))
217 put_task_struct(priv->task);
222 static int proc_maps_open(struct inode *inode, struct file *file,
223 const struct seq_operations *ops, int psize)
225 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
231 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
232 if (IS_ERR(priv->mm)) {
233 int err = PTR_ERR(priv->mm);
235 seq_release_private(inode, file);
242 static int proc_map_release(struct inode *inode, struct file *file)
244 struct seq_file *seq = file->private_data;
245 struct proc_maps_private *priv = seq->private;
251 return seq_release_private(inode, file);
254 static int do_maps_open(struct inode *inode, struct file *file,
255 const struct seq_operations *ops)
257 return proc_maps_open(inode, file, ops,
258 sizeof(struct proc_maps_private));
262 * Indicate if the VMA is a stack for the given task; for
263 * /proc/PID/maps that is the stack of the main task.
265 static int is_stack(struct vm_area_struct *vma)
268 * We make no effort to guess what a given thread considers to be
269 * its "stack". It's not even well-defined for programs written
272 return vma->vm_start <= vma->vm_mm->start_stack &&
273 vma->vm_end >= vma->vm_mm->start_stack;
276 static void show_vma_header_prefix(struct seq_file *m,
277 unsigned long start, unsigned long end,
278 vm_flags_t flags, unsigned long long pgoff,
279 dev_t dev, unsigned long ino)
281 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
282 seq_put_hex_ll(m, NULL, start, 8);
283 seq_put_hex_ll(m, "-", end, 8);
285 seq_putc(m, flags & VM_READ ? 'r' : '-');
286 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
287 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
288 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
289 seq_put_hex_ll(m, " ", pgoff, 8);
290 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
291 seq_put_hex_ll(m, ":", MINOR(dev), 2);
292 seq_put_decimal_ull(m, " ", ino);
297 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
299 struct mm_struct *mm = vma->vm_mm;
300 struct file *file = vma->vm_file;
301 vm_flags_t flags = vma->vm_flags;
302 unsigned long ino = 0;
303 unsigned long long pgoff = 0;
304 unsigned long start, end;
306 const char *name = NULL;
309 struct inode *inode = file_inode(vma->vm_file);
310 dev = inode->i_sb->s_dev;
312 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
315 start = vma->vm_start;
317 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
320 * Print the dentry name for named mappings, and a
321 * special [heap] marker for the heap:
325 seq_file_path(m, file, "\n");
329 if (vma->vm_ops && vma->vm_ops->name) {
330 name = vma->vm_ops->name(vma);
335 name = arch_vma_name(vma);
342 if (vma->vm_start <= mm->brk &&
343 vma->vm_end >= mm->start_brk) {
360 static int show_map(struct seq_file *m, void *v, int is_pid)
362 show_map_vma(m, v, is_pid);
367 static int show_pid_map(struct seq_file *m, void *v)
369 return show_map(m, v, 1);
372 static int show_tid_map(struct seq_file *m, void *v)
374 return show_map(m, v, 0);
377 static const struct seq_operations proc_pid_maps_op = {
384 static const struct seq_operations proc_tid_maps_op = {
391 static int pid_maps_open(struct inode *inode, struct file *file)
393 return do_maps_open(inode, file, &proc_pid_maps_op);
396 static int tid_maps_open(struct inode *inode, struct file *file)
398 return do_maps_open(inode, file, &proc_tid_maps_op);
401 const struct file_operations proc_pid_maps_operations = {
402 .open = pid_maps_open,
405 .release = proc_map_release,
408 const struct file_operations proc_tid_maps_operations = {
409 .open = tid_maps_open,
412 .release = proc_map_release,
416 * Proportional Set Size(PSS): my share of RSS.
418 * PSS of a process is the count of pages it has in memory, where each
419 * page is divided by the number of processes sharing it. So if a
420 * process has 1000 pages all to itself, and 1000 shared with one other
421 * process, its PSS will be 1500.
423 * To keep (accumulated) division errors low, we adopt a 64bit
424 * fixed-point pss counter to minimize division errors. So (pss >>
425 * PSS_SHIFT) would be the real byte count.
427 * A shift of 12 before division means (assuming 4K page size):
428 * - 1M 3-user-pages add up to 8KB errors;
429 * - supports mapcount up to 2^24, or 16M;
430 * - supports PSS up to 2^52 bytes, or 4PB.
434 #ifdef CONFIG_PROC_PAGE_MONITOR
435 struct mem_size_stats {
437 unsigned long resident;
438 unsigned long shared_clean;
439 unsigned long shared_dirty;
440 unsigned long private_clean;
441 unsigned long private_dirty;
442 unsigned long referenced;
443 unsigned long anonymous;
444 unsigned long lazyfree;
445 unsigned long anonymous_thp;
446 unsigned long shmem_thp;
448 unsigned long shared_hugetlb;
449 unsigned long private_hugetlb;
450 unsigned long first_vma_start;
454 bool check_shmem_swap;
457 static void smaps_account(struct mem_size_stats *mss, struct page *page,
458 bool compound, bool young, bool dirty)
460 int i, nr = compound ? 1 << compound_order(page) : 1;
461 unsigned long size = nr * PAGE_SIZE;
463 if (PageAnon(page)) {
464 mss->anonymous += size;
465 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
466 mss->lazyfree += size;
469 mss->resident += size;
470 /* Accumulate the size in pages that have been accessed. */
471 if (young || page_is_young(page) || PageReferenced(page))
472 mss->referenced += size;
475 * page_count(page) == 1 guarantees the page is mapped exactly once.
476 * If any subpage of the compound page mapped with PTE it would elevate
479 if (page_count(page) == 1) {
480 if (dirty || PageDirty(page))
481 mss->private_dirty += size;
483 mss->private_clean += size;
484 mss->pss += (u64)size << PSS_SHIFT;
488 for (i = 0; i < nr; i++, page++) {
489 int mapcount = page_mapcount(page);
492 if (dirty || PageDirty(page))
493 mss->shared_dirty += PAGE_SIZE;
495 mss->shared_clean += PAGE_SIZE;
496 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
498 if (dirty || PageDirty(page))
499 mss->private_dirty += PAGE_SIZE;
501 mss->private_clean += PAGE_SIZE;
502 mss->pss += PAGE_SIZE << PSS_SHIFT;
508 static int smaps_pte_hole(unsigned long addr, unsigned long end,
509 struct mm_walk *walk)
511 struct mem_size_stats *mss = walk->private;
513 mss->swap += shmem_partial_swap_usage(
514 walk->vma->vm_file->f_mapping, addr, end);
520 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
521 struct mm_walk *walk)
523 struct mem_size_stats *mss = walk->private;
524 struct vm_area_struct *vma = walk->vma;
525 struct page *page = NULL;
527 if (pte_present(*pte)) {
528 page = vm_normal_page(vma, addr, *pte);
529 } else if (is_swap_pte(*pte)) {
530 swp_entry_t swpent = pte_to_swp_entry(*pte);
532 if (!non_swap_entry(swpent)) {
535 mss->swap += PAGE_SIZE;
536 mapcount = swp_swapcount(swpent);
538 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
540 do_div(pss_delta, mapcount);
541 mss->swap_pss += pss_delta;
543 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
545 } else if (is_migration_entry(swpent))
546 page = migration_entry_to_page(swpent);
547 else if (is_device_private_entry(swpent))
548 page = device_private_entry_to_page(swpent);
549 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
550 && pte_none(*pte))) {
551 page = find_get_entry(vma->vm_file->f_mapping,
552 linear_page_index(vma, addr));
556 if (radix_tree_exceptional_entry(page))
557 mss->swap += PAGE_SIZE;
567 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
570 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
571 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
572 struct mm_walk *walk)
574 struct mem_size_stats *mss = walk->private;
575 struct vm_area_struct *vma = walk->vma;
578 /* FOLL_DUMP will return -EFAULT on huge zero page */
579 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
580 if (IS_ERR_OR_NULL(page))
583 mss->anonymous_thp += HPAGE_PMD_SIZE;
584 else if (PageSwapBacked(page))
585 mss->shmem_thp += HPAGE_PMD_SIZE;
586 else if (is_zone_device_page(page))
589 VM_BUG_ON_PAGE(1, page);
590 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
593 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
594 struct mm_walk *walk)
599 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
600 struct mm_walk *walk)
602 struct vm_area_struct *vma = walk->vma;
606 ptl = pmd_trans_huge_lock(pmd, vma);
608 if (pmd_present(*pmd))
609 smaps_pmd_entry(pmd, addr, walk);
614 if (pmd_trans_unstable(pmd))
617 * The mmap_sem held all the way back in m_start() is what
618 * keeps khugepaged out of here and from collapsing things
621 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
622 for (; addr != end; pte++, addr += PAGE_SIZE)
623 smaps_pte_entry(pte, addr, walk);
624 pte_unmap_unlock(pte - 1, ptl);
630 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
633 * Don't forget to update Documentation/ on changes.
635 static const char mnemonics[BITS_PER_LONG][2] = {
637 * In case if we meet a flag we don't know about.
639 [0 ... (BITS_PER_LONG-1)] = "??",
641 [ilog2(VM_READ)] = "rd",
642 [ilog2(VM_WRITE)] = "wr",
643 [ilog2(VM_EXEC)] = "ex",
644 [ilog2(VM_SHARED)] = "sh",
645 [ilog2(VM_MAYREAD)] = "mr",
646 [ilog2(VM_MAYWRITE)] = "mw",
647 [ilog2(VM_MAYEXEC)] = "me",
648 [ilog2(VM_MAYSHARE)] = "ms",
649 [ilog2(VM_GROWSDOWN)] = "gd",
650 [ilog2(VM_PFNMAP)] = "pf",
651 [ilog2(VM_DENYWRITE)] = "dw",
652 #ifdef CONFIG_X86_INTEL_MPX
653 [ilog2(VM_MPX)] = "mp",
655 [ilog2(VM_LOCKED)] = "lo",
656 [ilog2(VM_IO)] = "io",
657 [ilog2(VM_SEQ_READ)] = "sr",
658 [ilog2(VM_RAND_READ)] = "rr",
659 [ilog2(VM_DONTCOPY)] = "dc",
660 [ilog2(VM_DONTEXPAND)] = "de",
661 [ilog2(VM_ACCOUNT)] = "ac",
662 [ilog2(VM_NORESERVE)] = "nr",
663 [ilog2(VM_HUGETLB)] = "ht",
664 [ilog2(VM_SYNC)] = "sf",
665 [ilog2(VM_ARCH_1)] = "ar",
666 [ilog2(VM_WIPEONFORK)] = "wf",
667 [ilog2(VM_DONTDUMP)] = "dd",
668 #ifdef CONFIG_MEM_SOFT_DIRTY
669 [ilog2(VM_SOFTDIRTY)] = "sd",
671 [ilog2(VM_MIXEDMAP)] = "mm",
672 [ilog2(VM_HUGEPAGE)] = "hg",
673 [ilog2(VM_NOHUGEPAGE)] = "nh",
674 [ilog2(VM_MERGEABLE)] = "mg",
675 [ilog2(VM_UFFD_MISSING)]= "um",
676 [ilog2(VM_UFFD_WP)] = "uw",
677 #ifdef CONFIG_ARCH_HAS_PKEYS
678 /* These come out via ProtectionKey: */
679 [ilog2(VM_PKEY_BIT0)] = "",
680 [ilog2(VM_PKEY_BIT1)] = "",
681 [ilog2(VM_PKEY_BIT2)] = "",
682 [ilog2(VM_PKEY_BIT3)] = "",
684 [ilog2(VM_PKEY_BIT4)] = "",
686 #endif /* CONFIG_ARCH_HAS_PKEYS */
690 seq_puts(m, "VmFlags: ");
691 for (i = 0; i < BITS_PER_LONG; i++) {
692 if (!mnemonics[i][0])
694 if (vma->vm_flags & (1UL << i)) {
695 seq_putc(m, mnemonics[i][0]);
696 seq_putc(m, mnemonics[i][1]);
703 #ifdef CONFIG_HUGETLB_PAGE
704 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
705 unsigned long addr, unsigned long end,
706 struct mm_walk *walk)
708 struct mem_size_stats *mss = walk->private;
709 struct vm_area_struct *vma = walk->vma;
710 struct page *page = NULL;
712 if (pte_present(*pte)) {
713 page = vm_normal_page(vma, addr, *pte);
714 } else if (is_swap_pte(*pte)) {
715 swp_entry_t swpent = pte_to_swp_entry(*pte);
717 if (is_migration_entry(swpent))
718 page = migration_entry_to_page(swpent);
719 else if (is_device_private_entry(swpent))
720 page = device_private_entry_to_page(swpent);
723 int mapcount = page_mapcount(page);
726 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
728 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
732 #endif /* HUGETLB_PAGE */
734 #define SEQ_PUT_DEC(str, val) \
735 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
736 static int show_smap(struct seq_file *m, void *v, int is_pid)
738 struct proc_maps_private *priv = m->private;
739 struct vm_area_struct *vma = v;
740 struct mem_size_stats mss_stack;
741 struct mem_size_stats *mss;
742 struct mm_walk smaps_walk = {
743 .pmd_entry = smaps_pte_range,
744 #ifdef CONFIG_HUGETLB_PAGE
745 .hugetlb_entry = smaps_hugetlb_range,
757 mss->first_vma_start = vma->vm_start;
760 last_vma = !m_next_vma(priv, vma);
763 memset(&mss_stack, 0, sizeof(mss_stack));
767 smaps_walk.private = mss;
770 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
772 * For shared or readonly shmem mappings we know that all
773 * swapped out pages belong to the shmem object, and we can
774 * obtain the swap value much more efficiently. For private
775 * writable mappings, we might have COW pages that are
776 * not affected by the parent swapped out pages of the shmem
777 * object, so we have to distinguish them during the page walk.
778 * Unless we know that the shmem object (or the part mapped by
779 * our VMA) has no swapped out pages at all.
781 unsigned long shmem_swapped = shmem_swap_usage(vma);
783 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
784 !(vma->vm_flags & VM_WRITE)) {
785 mss->swap = shmem_swapped;
787 mss->check_shmem_swap = true;
788 smaps_walk.pte_hole = smaps_pte_hole;
793 /* mmap_sem is held in m_start */
794 walk_page_vma(vma, &smaps_walk);
795 if (vma->vm_flags & VM_LOCKED)
796 mss->pss_locked += mss->pss;
799 show_map_vma(m, vma, is_pid);
800 } else if (last_vma) {
801 show_vma_header_prefix(
802 m, mss->first_vma_start, vma->vm_end, 0, 0, 0, 0);
804 seq_puts(m, "[rollup]\n");
810 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
811 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
812 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
813 seq_puts(m, " kB\n");
816 if (!rollup_mode || last_vma) {
817 SEQ_PUT_DEC("Rss: ", mss->resident);
818 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
819 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
820 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
821 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
822 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
823 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
824 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
825 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
826 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
827 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
828 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
829 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
830 mss->private_hugetlb >> 10, 7);
831 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
832 SEQ_PUT_DEC(" kB\nSwapPss: ",
833 mss->swap_pss >> PSS_SHIFT);
834 SEQ_PUT_DEC(" kB\nLocked: ", mss->pss >> PSS_SHIFT);
835 seq_puts(m, " kB\n");
838 if (arch_pkeys_enabled())
839 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
840 show_smap_vma_flags(m, vma);
847 static int show_pid_smap(struct seq_file *m, void *v)
849 return show_smap(m, v, 1);
852 static int show_tid_smap(struct seq_file *m, void *v)
854 return show_smap(m, v, 0);
857 static const struct seq_operations proc_pid_smaps_op = {
861 .show = show_pid_smap
864 static const struct seq_operations proc_tid_smaps_op = {
868 .show = show_tid_smap
871 static int pid_smaps_open(struct inode *inode, struct file *file)
873 return do_maps_open(inode, file, &proc_pid_smaps_op);
876 static int pid_smaps_rollup_open(struct inode *inode, struct file *file)
878 struct seq_file *seq;
879 struct proc_maps_private *priv;
880 int ret = do_maps_open(inode, file, &proc_pid_smaps_op);
884 seq = file->private_data;
886 priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL);
888 proc_map_release(inode, file);
891 priv->rollup->first = true;
895 static int tid_smaps_open(struct inode *inode, struct file *file)
897 return do_maps_open(inode, file, &proc_tid_smaps_op);
900 const struct file_operations proc_pid_smaps_operations = {
901 .open = pid_smaps_open,
904 .release = proc_map_release,
907 const struct file_operations proc_pid_smaps_rollup_operations = {
908 .open = pid_smaps_rollup_open,
911 .release = proc_map_release,
914 const struct file_operations proc_tid_smaps_operations = {
915 .open = tid_smaps_open,
918 .release = proc_map_release,
921 enum clear_refs_types {
925 CLEAR_REFS_SOFT_DIRTY,
926 CLEAR_REFS_MM_HIWATER_RSS,
930 struct clear_refs_private {
931 enum clear_refs_types type;
934 #ifdef CONFIG_MEM_SOFT_DIRTY
935 static inline void clear_soft_dirty(struct vm_area_struct *vma,
936 unsigned long addr, pte_t *pte)
939 * The soft-dirty tracker uses #PF-s to catch writes
940 * to pages, so write-protect the pte as well. See the
941 * Documentation/admin-guide/mm/soft-dirty.rst for full description
942 * of how soft-dirty works.
946 if (pte_present(ptent)) {
947 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
948 ptent = pte_wrprotect(ptent);
949 ptent = pte_clear_soft_dirty(ptent);
950 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
951 } else if (is_swap_pte(ptent)) {
952 ptent = pte_swp_clear_soft_dirty(ptent);
953 set_pte_at(vma->vm_mm, addr, pte, ptent);
957 static inline void clear_soft_dirty(struct vm_area_struct *vma,
958 unsigned long addr, pte_t *pte)
963 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
964 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
965 unsigned long addr, pmd_t *pmdp)
967 pmd_t old, pmd = *pmdp;
969 if (pmd_present(pmd)) {
970 /* See comment in change_huge_pmd() */
971 old = pmdp_invalidate(vma, addr, pmdp);
973 pmd = pmd_mkdirty(pmd);
975 pmd = pmd_mkyoung(pmd);
977 pmd = pmd_wrprotect(pmd);
978 pmd = pmd_clear_soft_dirty(pmd);
980 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
981 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
982 pmd = pmd_swp_clear_soft_dirty(pmd);
983 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
987 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
988 unsigned long addr, pmd_t *pmdp)
993 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
994 unsigned long end, struct mm_walk *walk)
996 struct clear_refs_private *cp = walk->private;
997 struct vm_area_struct *vma = walk->vma;
1002 ptl = pmd_trans_huge_lock(pmd, vma);
1004 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1005 clear_soft_dirty_pmd(vma, addr, pmd);
1009 if (!pmd_present(*pmd))
1012 page = pmd_page(*pmd);
1014 /* Clear accessed and referenced bits. */
1015 pmdp_test_and_clear_young(vma, addr, pmd);
1016 test_and_clear_page_young(page);
1017 ClearPageReferenced(page);
1023 if (pmd_trans_unstable(pmd))
1026 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1027 for (; addr != end; pte++, addr += PAGE_SIZE) {
1030 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1031 clear_soft_dirty(vma, addr, pte);
1035 if (!pte_present(ptent))
1038 page = vm_normal_page(vma, addr, ptent);
1042 /* Clear accessed and referenced bits. */
1043 ptep_test_and_clear_young(vma, addr, pte);
1044 test_and_clear_page_young(page);
1045 ClearPageReferenced(page);
1047 pte_unmap_unlock(pte - 1, ptl);
1052 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1053 struct mm_walk *walk)
1055 struct clear_refs_private *cp = walk->private;
1056 struct vm_area_struct *vma = walk->vma;
1058 if (vma->vm_flags & VM_PFNMAP)
1062 * Writing 1 to /proc/pid/clear_refs affects all pages.
1063 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1064 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1065 * Writing 4 to /proc/pid/clear_refs affects all pages.
1067 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1069 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1074 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1075 size_t count, loff_t *ppos)
1077 struct task_struct *task;
1078 char buffer[PROC_NUMBUF];
1079 struct mm_struct *mm;
1080 struct vm_area_struct *vma;
1081 enum clear_refs_types type;
1082 struct mmu_gather tlb;
1086 memset(buffer, 0, sizeof(buffer));
1087 if (count > sizeof(buffer) - 1)
1088 count = sizeof(buffer) - 1;
1089 if (copy_from_user(buffer, buf, count))
1091 rv = kstrtoint(strstrip(buffer), 10, &itype);
1094 type = (enum clear_refs_types)itype;
1095 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1098 task = get_proc_task(file_inode(file));
1101 mm = get_task_mm(task);
1103 struct clear_refs_private cp = {
1106 struct mm_walk clear_refs_walk = {
1107 .pmd_entry = clear_refs_pte_range,
1108 .test_walk = clear_refs_test_walk,
1113 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1114 if (down_write_killable(&mm->mmap_sem)) {
1120 * Writing 5 to /proc/pid/clear_refs resets the peak
1121 * resident set size to this mm's current rss value.
1123 reset_mm_hiwater_rss(mm);
1124 up_write(&mm->mmap_sem);
1128 down_read(&mm->mmap_sem);
1129 tlb_gather_mmu(&tlb, mm, 0, -1);
1130 if (type == CLEAR_REFS_SOFT_DIRTY) {
1131 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1132 if (!(vma->vm_flags & VM_SOFTDIRTY))
1134 up_read(&mm->mmap_sem);
1135 if (down_write_killable(&mm->mmap_sem)) {
1139 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1140 vma->vm_flags &= ~VM_SOFTDIRTY;
1141 vma_set_page_prot(vma);
1143 downgrade_write(&mm->mmap_sem);
1146 mmu_notifier_invalidate_range_start(mm, 0, -1);
1148 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1149 if (type == CLEAR_REFS_SOFT_DIRTY)
1150 mmu_notifier_invalidate_range_end(mm, 0, -1);
1151 tlb_finish_mmu(&tlb, 0, -1);
1152 up_read(&mm->mmap_sem);
1156 put_task_struct(task);
1161 const struct file_operations proc_clear_refs_operations = {
1162 .write = clear_refs_write,
1163 .llseek = noop_llseek,
1170 struct pagemapread {
1171 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1172 pagemap_entry_t *buffer;
1176 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1177 #define PAGEMAP_WALK_MASK (PMD_MASK)
1179 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1180 #define PM_PFRAME_BITS 55
1181 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1182 #define PM_SOFT_DIRTY BIT_ULL(55)
1183 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1184 #define PM_FILE BIT_ULL(61)
1185 #define PM_SWAP BIT_ULL(62)
1186 #define PM_PRESENT BIT_ULL(63)
1188 #define PM_END_OF_BUFFER 1
1190 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1192 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1195 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1196 struct pagemapread *pm)
1198 pm->buffer[pm->pos++] = *pme;
1199 if (pm->pos >= pm->len)
1200 return PM_END_OF_BUFFER;
1204 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1205 struct mm_walk *walk)
1207 struct pagemapread *pm = walk->private;
1208 unsigned long addr = start;
1211 while (addr < end) {
1212 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1213 pagemap_entry_t pme = make_pme(0, 0);
1214 /* End of address space hole, which we mark as non-present. */
1215 unsigned long hole_end;
1218 hole_end = min(end, vma->vm_start);
1222 for (; addr < hole_end; addr += PAGE_SIZE) {
1223 err = add_to_pagemap(addr, &pme, pm);
1231 /* Addresses in the VMA. */
1232 if (vma->vm_flags & VM_SOFTDIRTY)
1233 pme = make_pme(0, PM_SOFT_DIRTY);
1234 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1235 err = add_to_pagemap(addr, &pme, pm);
1244 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1245 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1247 u64 frame = 0, flags = 0;
1248 struct page *page = NULL;
1250 if (pte_present(pte)) {
1252 frame = pte_pfn(pte);
1253 flags |= PM_PRESENT;
1254 page = _vm_normal_page(vma, addr, pte, true);
1255 if (pte_soft_dirty(pte))
1256 flags |= PM_SOFT_DIRTY;
1257 } else if (is_swap_pte(pte)) {
1259 if (pte_swp_soft_dirty(pte))
1260 flags |= PM_SOFT_DIRTY;
1261 entry = pte_to_swp_entry(pte);
1263 frame = swp_type(entry) |
1264 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1266 if (is_migration_entry(entry))
1267 page = migration_entry_to_page(entry);
1269 if (is_device_private_entry(entry))
1270 page = device_private_entry_to_page(entry);
1273 if (page && !PageAnon(page))
1275 if (page && page_mapcount(page) == 1)
1276 flags |= PM_MMAP_EXCLUSIVE;
1277 if (vma->vm_flags & VM_SOFTDIRTY)
1278 flags |= PM_SOFT_DIRTY;
1280 return make_pme(frame, flags);
1283 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1284 struct mm_walk *walk)
1286 struct vm_area_struct *vma = walk->vma;
1287 struct pagemapread *pm = walk->private;
1289 pte_t *pte, *orig_pte;
1292 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1293 ptl = pmd_trans_huge_lock(pmdp, vma);
1295 u64 flags = 0, frame = 0;
1297 struct page *page = NULL;
1299 if (vma->vm_flags & VM_SOFTDIRTY)
1300 flags |= PM_SOFT_DIRTY;
1302 if (pmd_present(pmd)) {
1303 page = pmd_page(pmd);
1305 flags |= PM_PRESENT;
1306 if (pmd_soft_dirty(pmd))
1307 flags |= PM_SOFT_DIRTY;
1309 frame = pmd_pfn(pmd) +
1310 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1312 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1313 else if (is_swap_pmd(pmd)) {
1314 swp_entry_t entry = pmd_to_swp_entry(pmd);
1315 unsigned long offset;
1318 offset = swp_offset(entry) +
1319 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1320 frame = swp_type(entry) |
1321 (offset << MAX_SWAPFILES_SHIFT);
1324 if (pmd_swp_soft_dirty(pmd))
1325 flags |= PM_SOFT_DIRTY;
1326 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1327 page = migration_entry_to_page(entry);
1331 if (page && page_mapcount(page) == 1)
1332 flags |= PM_MMAP_EXCLUSIVE;
1334 for (; addr != end; addr += PAGE_SIZE) {
1335 pagemap_entry_t pme = make_pme(frame, flags);
1337 err = add_to_pagemap(addr, &pme, pm);
1341 if (flags & PM_PRESENT)
1343 else if (flags & PM_SWAP)
1344 frame += (1 << MAX_SWAPFILES_SHIFT);
1351 if (pmd_trans_unstable(pmdp))
1353 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1356 * We can assume that @vma always points to a valid one and @end never
1357 * goes beyond vma->vm_end.
1359 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1360 for (; addr < end; pte++, addr += PAGE_SIZE) {
1361 pagemap_entry_t pme;
1363 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1364 err = add_to_pagemap(addr, &pme, pm);
1368 pte_unmap_unlock(orig_pte, ptl);
1375 #ifdef CONFIG_HUGETLB_PAGE
1376 /* This function walks within one hugetlb entry in the single call */
1377 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1378 unsigned long addr, unsigned long end,
1379 struct mm_walk *walk)
1381 struct pagemapread *pm = walk->private;
1382 struct vm_area_struct *vma = walk->vma;
1383 u64 flags = 0, frame = 0;
1387 if (vma->vm_flags & VM_SOFTDIRTY)
1388 flags |= PM_SOFT_DIRTY;
1390 pte = huge_ptep_get(ptep);
1391 if (pte_present(pte)) {
1392 struct page *page = pte_page(pte);
1394 if (!PageAnon(page))
1397 if (page_mapcount(page) == 1)
1398 flags |= PM_MMAP_EXCLUSIVE;
1400 flags |= PM_PRESENT;
1402 frame = pte_pfn(pte) +
1403 ((addr & ~hmask) >> PAGE_SHIFT);
1406 for (; addr != end; addr += PAGE_SIZE) {
1407 pagemap_entry_t pme = make_pme(frame, flags);
1409 err = add_to_pagemap(addr, &pme, pm);
1412 if (pm->show_pfn && (flags & PM_PRESENT))
1420 #endif /* HUGETLB_PAGE */
1423 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1425 * For each page in the address space, this file contains one 64-bit entry
1426 * consisting of the following:
1428 * Bits 0-54 page frame number (PFN) if present
1429 * Bits 0-4 swap type if swapped
1430 * Bits 5-54 swap offset if swapped
1431 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1432 * Bit 56 page exclusively mapped
1434 * Bit 61 page is file-page or shared-anon
1435 * Bit 62 page swapped
1436 * Bit 63 page present
1438 * If the page is not present but in swap, then the PFN contains an
1439 * encoding of the swap file number and the page's offset into the
1440 * swap. Unmapped pages return a null PFN. This allows determining
1441 * precisely which pages are mapped (or in swap) and comparing mapped
1442 * pages between processes.
1444 * Efficient users of this interface will use /proc/pid/maps to
1445 * determine which areas of memory are actually mapped and llseek to
1446 * skip over unmapped regions.
1448 static ssize_t pagemap_read(struct file *file, char __user *buf,
1449 size_t count, loff_t *ppos)
1451 struct mm_struct *mm = file->private_data;
1452 struct pagemapread pm;
1453 struct mm_walk pagemap_walk = {};
1455 unsigned long svpfn;
1456 unsigned long start_vaddr;
1457 unsigned long end_vaddr;
1458 int ret = 0, copied = 0;
1460 if (!mm || !mmget_not_zero(mm))
1464 /* file position must be aligned */
1465 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1472 /* do not disclose physical addresses: attack vector */
1473 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1475 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1476 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1481 pagemap_walk.pmd_entry = pagemap_pmd_range;
1482 pagemap_walk.pte_hole = pagemap_pte_hole;
1483 #ifdef CONFIG_HUGETLB_PAGE
1484 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1486 pagemap_walk.mm = mm;
1487 pagemap_walk.private = ±
1490 svpfn = src / PM_ENTRY_BYTES;
1491 start_vaddr = svpfn << PAGE_SHIFT;
1492 end_vaddr = mm->task_size;
1494 /* watch out for wraparound */
1495 if (svpfn > mm->task_size >> PAGE_SHIFT)
1496 start_vaddr = end_vaddr;
1499 * The odds are that this will stop walking way
1500 * before end_vaddr, because the length of the
1501 * user buffer is tracked in "pm", and the walk
1502 * will stop when we hit the end of the buffer.
1505 while (count && (start_vaddr < end_vaddr)) {
1510 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1512 if (end < start_vaddr || end > end_vaddr)
1514 down_read(&mm->mmap_sem);
1515 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1516 up_read(&mm->mmap_sem);
1519 len = min(count, PM_ENTRY_BYTES * pm.pos);
1520 if (copy_to_user(buf, pm.buffer, len)) {
1529 if (!ret || ret == PM_END_OF_BUFFER)
1540 static int pagemap_open(struct inode *inode, struct file *file)
1542 struct mm_struct *mm;
1544 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1547 file->private_data = mm;
1551 static int pagemap_release(struct inode *inode, struct file *file)
1553 struct mm_struct *mm = file->private_data;
1560 const struct file_operations proc_pagemap_operations = {
1561 .llseek = mem_lseek, /* borrow this */
1562 .read = pagemap_read,
1563 .open = pagemap_open,
1564 .release = pagemap_release,
1566 #endif /* CONFIG_PROC_PAGE_MONITOR */
1571 unsigned long pages;
1573 unsigned long active;
1574 unsigned long writeback;
1575 unsigned long mapcount_max;
1576 unsigned long dirty;
1577 unsigned long swapcache;
1578 unsigned long node[MAX_NUMNODES];
1581 struct numa_maps_private {
1582 struct proc_maps_private proc_maps;
1583 struct numa_maps md;
1586 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1587 unsigned long nr_pages)
1589 int count = page_mapcount(page);
1591 md->pages += nr_pages;
1592 if (pte_dirty || PageDirty(page))
1593 md->dirty += nr_pages;
1595 if (PageSwapCache(page))
1596 md->swapcache += nr_pages;
1598 if (PageActive(page) || PageUnevictable(page))
1599 md->active += nr_pages;
1601 if (PageWriteback(page))
1602 md->writeback += nr_pages;
1605 md->anon += nr_pages;
1607 if (count > md->mapcount_max)
1608 md->mapcount_max = count;
1610 md->node[page_to_nid(page)] += nr_pages;
1613 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1619 if (!pte_present(pte))
1622 page = vm_normal_page(vma, addr, pte);
1626 if (PageReserved(page))
1629 nid = page_to_nid(page);
1630 if (!node_isset(nid, node_states[N_MEMORY]))
1636 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1637 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1638 struct vm_area_struct *vma,
1644 if (!pmd_present(pmd))
1647 page = vm_normal_page_pmd(vma, addr, pmd);
1651 if (PageReserved(page))
1654 nid = page_to_nid(page);
1655 if (!node_isset(nid, node_states[N_MEMORY]))
1662 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1663 unsigned long end, struct mm_walk *walk)
1665 struct numa_maps *md = walk->private;
1666 struct vm_area_struct *vma = walk->vma;
1671 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1672 ptl = pmd_trans_huge_lock(pmd, vma);
1676 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1678 gather_stats(page, md, pmd_dirty(*pmd),
1679 HPAGE_PMD_SIZE/PAGE_SIZE);
1684 if (pmd_trans_unstable(pmd))
1687 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1689 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1692 gather_stats(page, md, pte_dirty(*pte), 1);
1694 } while (pte++, addr += PAGE_SIZE, addr != end);
1695 pte_unmap_unlock(orig_pte, ptl);
1699 #ifdef CONFIG_HUGETLB_PAGE
1700 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1701 unsigned long addr, unsigned long end, struct mm_walk *walk)
1703 pte_t huge_pte = huge_ptep_get(pte);
1704 struct numa_maps *md;
1707 if (!pte_present(huge_pte))
1710 page = pte_page(huge_pte);
1715 gather_stats(page, md, pte_dirty(huge_pte), 1);
1720 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1721 unsigned long addr, unsigned long end, struct mm_walk *walk)
1728 * Display pages allocated per node and memory policy via /proc.
1730 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1732 struct numa_maps_private *numa_priv = m->private;
1733 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1734 struct vm_area_struct *vma = v;
1735 struct numa_maps *md = &numa_priv->md;
1736 struct file *file = vma->vm_file;
1737 struct mm_struct *mm = vma->vm_mm;
1738 struct mm_walk walk = {
1739 .hugetlb_entry = gather_hugetlb_stats,
1740 .pmd_entry = gather_pte_stats,
1744 struct mempolicy *pol;
1751 /* Ensure we start with an empty set of numa_maps statistics. */
1752 memset(md, 0, sizeof(*md));
1754 pol = __get_vma_policy(vma, vma->vm_start);
1756 mpol_to_str(buffer, sizeof(buffer), pol);
1759 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1762 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1765 seq_puts(m, " file=");
1766 seq_file_path(m, file, "\n\t= ");
1767 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1768 seq_puts(m, " heap");
1769 } else if (is_stack(vma)) {
1770 seq_puts(m, " stack");
1773 if (is_vm_hugetlb_page(vma))
1774 seq_puts(m, " huge");
1776 /* mmap_sem is held by m_start */
1777 walk_page_vma(vma, &walk);
1783 seq_printf(m, " anon=%lu", md->anon);
1786 seq_printf(m, " dirty=%lu", md->dirty);
1788 if (md->pages != md->anon && md->pages != md->dirty)
1789 seq_printf(m, " mapped=%lu", md->pages);
1791 if (md->mapcount_max > 1)
1792 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1795 seq_printf(m, " swapcache=%lu", md->swapcache);
1797 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1798 seq_printf(m, " active=%lu", md->active);
1801 seq_printf(m, " writeback=%lu", md->writeback);
1803 for_each_node_state(nid, N_MEMORY)
1805 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1807 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1810 m_cache_vma(m, vma);
1814 static int show_pid_numa_map(struct seq_file *m, void *v)
1816 return show_numa_map(m, v, 1);
1819 static int show_tid_numa_map(struct seq_file *m, void *v)
1821 return show_numa_map(m, v, 0);
1824 static const struct seq_operations proc_pid_numa_maps_op = {
1828 .show = show_pid_numa_map,
1831 static const struct seq_operations proc_tid_numa_maps_op = {
1835 .show = show_tid_numa_map,
1838 static int numa_maps_open(struct inode *inode, struct file *file,
1839 const struct seq_operations *ops)
1841 return proc_maps_open(inode, file, ops,
1842 sizeof(struct numa_maps_private));
1845 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1847 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1850 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1852 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1855 const struct file_operations proc_pid_numa_maps_operations = {
1856 .open = pid_numa_maps_open,
1858 .llseek = seq_lseek,
1859 .release = proc_map_release,
1862 const struct file_operations proc_tid_numa_maps_operations = {
1863 .open = tid_numa_maps_open,
1865 .llseek = seq_lseek,
1866 .release = proc_map_release,
1868 #endif /* CONFIG_NUMA */