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 #define SEQ_PUT_DEC(str, val) \
28 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
29 void task_mem(struct seq_file *m, struct mm_struct *mm)
31 unsigned long text, lib, swap, anon, file, shmem;
32 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
34 anon = get_mm_counter(mm, MM_ANONPAGES);
35 file = get_mm_counter(mm, MM_FILEPAGES);
36 shmem = get_mm_counter(mm, MM_SHMEMPAGES);
39 * Note: to minimize their overhead, mm maintains hiwater_vm and
40 * hiwater_rss only when about to *lower* total_vm or rss. Any
41 * collector of these hiwater stats must therefore get total_vm
42 * and rss too, which will usually be the higher. Barriers? not
43 * worth the effort, such snapshots can always be inconsistent.
45 hiwater_vm = total_vm = mm->total_vm;
46 if (hiwater_vm < mm->hiwater_vm)
47 hiwater_vm = mm->hiwater_vm;
48 hiwater_rss = total_rss = anon + file + shmem;
49 if (hiwater_rss < mm->hiwater_rss)
50 hiwater_rss = mm->hiwater_rss;
52 /* split executable areas between text and lib */
53 text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
54 text = min(text, mm->exec_vm << PAGE_SHIFT);
55 lib = (mm->exec_vm << PAGE_SHIFT) - text;
57 swap = get_mm_counter(mm, MM_SWAPENTS);
58 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
59 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
60 SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
61 SEQ_PUT_DEC(" kB\nVmPin:\t", mm->pinned_vm);
62 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
63 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
64 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
65 SEQ_PUT_DEC(" kB\nRssFile:\t", file);
66 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
67 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
68 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
69 seq_put_decimal_ull_width(m,
70 " kB\nVmExe:\t", text >> 10, 8);
71 seq_put_decimal_ull_width(m,
72 " kB\nVmLib:\t", lib >> 10, 8);
73 seq_put_decimal_ull_width(m,
74 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
75 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
77 hugetlb_report_usage(m, mm);
81 unsigned long task_vsize(struct mm_struct *mm)
83 return PAGE_SIZE * mm->total_vm;
86 unsigned long task_statm(struct mm_struct *mm,
87 unsigned long *shared, unsigned long *text,
88 unsigned long *data, unsigned long *resident)
90 *shared = get_mm_counter(mm, MM_FILEPAGES) +
91 get_mm_counter(mm, MM_SHMEMPAGES);
92 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
94 *data = mm->data_vm + mm->stack_vm;
95 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
101 * Save get_task_policy() for show_numa_map().
103 static void hold_task_mempolicy(struct proc_maps_private *priv)
105 struct task_struct *task = priv->task;
108 priv->task_mempolicy = get_task_policy(task);
109 mpol_get(priv->task_mempolicy);
112 static void release_task_mempolicy(struct proc_maps_private *priv)
114 mpol_put(priv->task_mempolicy);
117 static void hold_task_mempolicy(struct proc_maps_private *priv)
120 static void release_task_mempolicy(struct proc_maps_private *priv)
125 static void vma_stop(struct proc_maps_private *priv)
127 struct mm_struct *mm = priv->mm;
129 release_task_mempolicy(priv);
130 up_read(&mm->mmap_sem);
134 static struct vm_area_struct *
135 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
137 if (vma == priv->tail_vma)
139 return vma->vm_next ?: priv->tail_vma;
142 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
144 if (m->count < m->size) /* vma is copied successfully */
145 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
148 static void *m_start(struct seq_file *m, loff_t *ppos)
150 struct proc_maps_private *priv = m->private;
151 unsigned long last_addr = m->version;
152 struct mm_struct *mm;
153 struct vm_area_struct *vma;
154 unsigned int pos = *ppos;
156 /* See m_cache_vma(). Zero at the start or after lseek. */
157 if (last_addr == -1UL)
160 priv->task = get_proc_task(priv->inode);
162 return ERR_PTR(-ESRCH);
165 if (!mm || !mmget_not_zero(mm))
168 down_read(&mm->mmap_sem);
169 hold_task_mempolicy(priv);
170 priv->tail_vma = get_gate_vma(mm);
173 vma = find_vma(mm, last_addr - 1);
174 if (vma && vma->vm_start <= last_addr)
175 vma = m_next_vma(priv, vma);
181 if (pos < mm->map_count) {
182 for (vma = mm->mmap; pos; pos--) {
183 m->version = vma->vm_start;
189 /* we do not bother to update m->version in this case */
190 if (pos == mm->map_count && priv->tail_vma)
191 return priv->tail_vma;
197 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
199 struct proc_maps_private *priv = m->private;
200 struct vm_area_struct *next;
203 next = m_next_vma(priv, v);
209 static void m_stop(struct seq_file *m, void *v)
211 struct proc_maps_private *priv = m->private;
213 if (!IS_ERR_OR_NULL(v))
216 put_task_struct(priv->task);
221 static int proc_maps_open(struct inode *inode, struct file *file,
222 const struct seq_operations *ops, int psize)
224 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
230 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
231 if (IS_ERR(priv->mm)) {
232 int err = PTR_ERR(priv->mm);
234 seq_release_private(inode, file);
241 static int proc_map_release(struct inode *inode, struct file *file)
243 struct seq_file *seq = file->private_data;
244 struct proc_maps_private *priv = seq->private;
250 return seq_release_private(inode, file);
253 static int do_maps_open(struct inode *inode, struct file *file,
254 const struct seq_operations *ops)
256 return proc_maps_open(inode, file, ops,
257 sizeof(struct proc_maps_private));
261 * Indicate if the VMA is a stack for the given task; for
262 * /proc/PID/maps that is the stack of the main task.
264 static int is_stack(struct vm_area_struct *vma)
267 * We make no effort to guess what a given thread considers to be
268 * its "stack". It's not even well-defined for programs written
271 return vma->vm_start <= vma->vm_mm->start_stack &&
272 vma->vm_end >= vma->vm_mm->start_stack;
275 static void show_vma_header_prefix(struct seq_file *m,
276 unsigned long start, unsigned long end,
277 vm_flags_t flags, unsigned long long pgoff,
278 dev_t dev, unsigned long ino)
280 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
281 seq_put_hex_ll(m, NULL, start, 8);
282 seq_put_hex_ll(m, "-", end, 8);
284 seq_putc(m, flags & VM_READ ? 'r' : '-');
285 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
286 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
287 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
288 seq_put_hex_ll(m, " ", pgoff, 8);
289 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
290 seq_put_hex_ll(m, ":", MINOR(dev), 2);
291 seq_put_decimal_ull(m, " ", ino);
296 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
298 struct mm_struct *mm = vma->vm_mm;
299 struct file *file = vma->vm_file;
300 vm_flags_t flags = vma->vm_flags;
301 unsigned long ino = 0;
302 unsigned long long pgoff = 0;
303 unsigned long start, end;
305 const char *name = NULL;
308 struct inode *inode = file_inode(vma->vm_file);
309 dev = inode->i_sb->s_dev;
311 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
314 start = vma->vm_start;
316 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
319 * Print the dentry name for named mappings, and a
320 * special [heap] marker for the heap:
324 seq_file_path(m, file, "\n");
328 if (vma->vm_ops && vma->vm_ops->name) {
329 name = vma->vm_ops->name(vma);
334 name = arch_vma_name(vma);
341 if (vma->vm_start <= mm->brk &&
342 vma->vm_end >= mm->start_brk) {
359 static int show_map(struct seq_file *m, void *v, int is_pid)
361 show_map_vma(m, v, is_pid);
366 static int show_pid_map(struct seq_file *m, void *v)
368 return show_map(m, v, 1);
371 static int show_tid_map(struct seq_file *m, void *v)
373 return show_map(m, v, 0);
376 static const struct seq_operations proc_pid_maps_op = {
383 static const struct seq_operations proc_tid_maps_op = {
390 static int pid_maps_open(struct inode *inode, struct file *file)
392 return do_maps_open(inode, file, &proc_pid_maps_op);
395 static int tid_maps_open(struct inode *inode, struct file *file)
397 return do_maps_open(inode, file, &proc_tid_maps_op);
400 const struct file_operations proc_pid_maps_operations = {
401 .open = pid_maps_open,
404 .release = proc_map_release,
407 const struct file_operations proc_tid_maps_operations = {
408 .open = tid_maps_open,
411 .release = proc_map_release,
415 * Proportional Set Size(PSS): my share of RSS.
417 * PSS of a process is the count of pages it has in memory, where each
418 * page is divided by the number of processes sharing it. So if a
419 * process has 1000 pages all to itself, and 1000 shared with one other
420 * process, its PSS will be 1500.
422 * To keep (accumulated) division errors low, we adopt a 64bit
423 * fixed-point pss counter to minimize division errors. So (pss >>
424 * PSS_SHIFT) would be the real byte count.
426 * A shift of 12 before division means (assuming 4K page size):
427 * - 1M 3-user-pages add up to 8KB errors;
428 * - supports mapcount up to 2^24, or 16M;
429 * - supports PSS up to 2^52 bytes, or 4PB.
433 #ifdef CONFIG_PROC_PAGE_MONITOR
434 struct mem_size_stats {
436 unsigned long resident;
437 unsigned long shared_clean;
438 unsigned long shared_dirty;
439 unsigned long private_clean;
440 unsigned long private_dirty;
441 unsigned long referenced;
442 unsigned long anonymous;
443 unsigned long lazyfree;
444 unsigned long anonymous_thp;
445 unsigned long shmem_thp;
447 unsigned long shared_hugetlb;
448 unsigned long private_hugetlb;
449 unsigned long first_vma_start;
453 bool check_shmem_swap;
456 static void smaps_account(struct mem_size_stats *mss, struct page *page,
457 bool compound, bool young, bool dirty)
459 int i, nr = compound ? 1 << compound_order(page) : 1;
460 unsigned long size = nr * PAGE_SIZE;
462 if (PageAnon(page)) {
463 mss->anonymous += size;
464 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
465 mss->lazyfree += size;
468 mss->resident += size;
469 /* Accumulate the size in pages that have been accessed. */
470 if (young || page_is_young(page) || PageReferenced(page))
471 mss->referenced += size;
474 * page_count(page) == 1 guarantees the page is mapped exactly once.
475 * If any subpage of the compound page mapped with PTE it would elevate
478 if (page_count(page) == 1) {
479 if (dirty || PageDirty(page))
480 mss->private_dirty += size;
482 mss->private_clean += size;
483 mss->pss += (u64)size << PSS_SHIFT;
487 for (i = 0; i < nr; i++, page++) {
488 int mapcount = page_mapcount(page);
491 if (dirty || PageDirty(page))
492 mss->shared_dirty += PAGE_SIZE;
494 mss->shared_clean += PAGE_SIZE;
495 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
497 if (dirty || PageDirty(page))
498 mss->private_dirty += PAGE_SIZE;
500 mss->private_clean += PAGE_SIZE;
501 mss->pss += PAGE_SIZE << PSS_SHIFT;
507 static int smaps_pte_hole(unsigned long addr, unsigned long end,
508 struct mm_walk *walk)
510 struct mem_size_stats *mss = walk->private;
512 mss->swap += shmem_partial_swap_usage(
513 walk->vma->vm_file->f_mapping, addr, end);
519 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
520 struct mm_walk *walk)
522 struct mem_size_stats *mss = walk->private;
523 struct vm_area_struct *vma = walk->vma;
524 struct page *page = NULL;
526 if (pte_present(*pte)) {
527 page = vm_normal_page(vma, addr, *pte);
528 } else if (is_swap_pte(*pte)) {
529 swp_entry_t swpent = pte_to_swp_entry(*pte);
531 if (!non_swap_entry(swpent)) {
534 mss->swap += PAGE_SIZE;
535 mapcount = swp_swapcount(swpent);
537 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
539 do_div(pss_delta, mapcount);
540 mss->swap_pss += pss_delta;
542 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
544 } else if (is_migration_entry(swpent))
545 page = migration_entry_to_page(swpent);
546 else if (is_device_private_entry(swpent))
547 page = device_private_entry_to_page(swpent);
548 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
549 && pte_none(*pte))) {
550 page = find_get_entry(vma->vm_file->f_mapping,
551 linear_page_index(vma, addr));
555 if (radix_tree_exceptional_entry(page))
556 mss->swap += PAGE_SIZE;
566 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
569 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
570 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
571 struct mm_walk *walk)
573 struct mem_size_stats *mss = walk->private;
574 struct vm_area_struct *vma = walk->vma;
577 /* FOLL_DUMP will return -EFAULT on huge zero page */
578 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
579 if (IS_ERR_OR_NULL(page))
582 mss->anonymous_thp += HPAGE_PMD_SIZE;
583 else if (PageSwapBacked(page))
584 mss->shmem_thp += HPAGE_PMD_SIZE;
585 else if (is_zone_device_page(page))
588 VM_BUG_ON_PAGE(1, page);
589 smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
592 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
593 struct mm_walk *walk)
598 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
599 struct mm_walk *walk)
601 struct vm_area_struct *vma = walk->vma;
605 ptl = pmd_trans_huge_lock(pmd, vma);
607 if (pmd_present(*pmd))
608 smaps_pmd_entry(pmd, addr, walk);
613 if (pmd_trans_unstable(pmd))
616 * The mmap_sem held all the way back in m_start() is what
617 * keeps khugepaged out of here and from collapsing things
620 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
621 for (; addr != end; pte++, addr += PAGE_SIZE)
622 smaps_pte_entry(pte, addr, walk);
623 pte_unmap_unlock(pte - 1, ptl);
629 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
632 * Don't forget to update Documentation/ on changes.
634 static const char mnemonics[BITS_PER_LONG][2] = {
636 * In case if we meet a flag we don't know about.
638 [0 ... (BITS_PER_LONG-1)] = "??",
640 [ilog2(VM_READ)] = "rd",
641 [ilog2(VM_WRITE)] = "wr",
642 [ilog2(VM_EXEC)] = "ex",
643 [ilog2(VM_SHARED)] = "sh",
644 [ilog2(VM_MAYREAD)] = "mr",
645 [ilog2(VM_MAYWRITE)] = "mw",
646 [ilog2(VM_MAYEXEC)] = "me",
647 [ilog2(VM_MAYSHARE)] = "ms",
648 [ilog2(VM_GROWSDOWN)] = "gd",
649 [ilog2(VM_PFNMAP)] = "pf",
650 [ilog2(VM_DENYWRITE)] = "dw",
651 #ifdef CONFIG_X86_INTEL_MPX
652 [ilog2(VM_MPX)] = "mp",
654 [ilog2(VM_LOCKED)] = "lo",
655 [ilog2(VM_IO)] = "io",
656 [ilog2(VM_SEQ_READ)] = "sr",
657 [ilog2(VM_RAND_READ)] = "rr",
658 [ilog2(VM_DONTCOPY)] = "dc",
659 [ilog2(VM_DONTEXPAND)] = "de",
660 [ilog2(VM_ACCOUNT)] = "ac",
661 [ilog2(VM_NORESERVE)] = "nr",
662 [ilog2(VM_HUGETLB)] = "ht",
663 [ilog2(VM_SYNC)] = "sf",
664 [ilog2(VM_ARCH_1)] = "ar",
665 [ilog2(VM_WIPEONFORK)] = "wf",
666 [ilog2(VM_DONTDUMP)] = "dd",
667 #ifdef CONFIG_MEM_SOFT_DIRTY
668 [ilog2(VM_SOFTDIRTY)] = "sd",
670 [ilog2(VM_MIXEDMAP)] = "mm",
671 [ilog2(VM_HUGEPAGE)] = "hg",
672 [ilog2(VM_NOHUGEPAGE)] = "nh",
673 [ilog2(VM_MERGEABLE)] = "mg",
674 [ilog2(VM_UFFD_MISSING)]= "um",
675 [ilog2(VM_UFFD_WP)] = "uw",
676 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
677 /* These come out via ProtectionKey: */
678 [ilog2(VM_PKEY_BIT0)] = "",
679 [ilog2(VM_PKEY_BIT1)] = "",
680 [ilog2(VM_PKEY_BIT2)] = "",
681 [ilog2(VM_PKEY_BIT3)] = "",
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_migration_entry(swpent))
714 page = migration_entry_to_page(swpent);
715 else if (is_device_private_entry(swpent))
716 page = device_private_entry_to_page(swpent);
719 int mapcount = page_mapcount(page);
722 mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
724 mss->private_hugetlb += huge_page_size(hstate_vma(vma));
728 #endif /* HUGETLB_PAGE */
730 void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma)
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 arch_show_smap(m, vma);
839 show_smap_vma_flags(m, vma);
846 static int show_pid_smap(struct seq_file *m, void *v)
848 return show_smap(m, v, 1);
851 static int show_tid_smap(struct seq_file *m, void *v)
853 return show_smap(m, v, 0);
856 static const struct seq_operations proc_pid_smaps_op = {
860 .show = show_pid_smap
863 static const struct seq_operations proc_tid_smaps_op = {
867 .show = show_tid_smap
870 static int pid_smaps_open(struct inode *inode, struct file *file)
872 return do_maps_open(inode, file, &proc_pid_smaps_op);
875 static int pid_smaps_rollup_open(struct inode *inode, struct file *file)
877 struct seq_file *seq;
878 struct proc_maps_private *priv;
879 int ret = do_maps_open(inode, file, &proc_pid_smaps_op);
883 seq = file->private_data;
885 priv->rollup = kzalloc(sizeof(*priv->rollup), GFP_KERNEL);
887 proc_map_release(inode, file);
890 priv->rollup->first = true;
894 static int tid_smaps_open(struct inode *inode, struct file *file)
896 return do_maps_open(inode, file, &proc_tid_smaps_op);
899 const struct file_operations proc_pid_smaps_operations = {
900 .open = pid_smaps_open,
903 .release = proc_map_release,
906 const struct file_operations proc_pid_smaps_rollup_operations = {
907 .open = pid_smaps_rollup_open,
910 .release = proc_map_release,
913 const struct file_operations proc_tid_smaps_operations = {
914 .open = tid_smaps_open,
917 .release = proc_map_release,
920 enum clear_refs_types {
924 CLEAR_REFS_SOFT_DIRTY,
925 CLEAR_REFS_MM_HIWATER_RSS,
929 struct clear_refs_private {
930 enum clear_refs_types type;
933 #ifdef CONFIG_MEM_SOFT_DIRTY
934 static inline void clear_soft_dirty(struct vm_area_struct *vma,
935 unsigned long addr, pte_t *pte)
938 * The soft-dirty tracker uses #PF-s to catch writes
939 * to pages, so write-protect the pte as well. See the
940 * Documentation/vm/soft-dirty.txt for full description
941 * of how soft-dirty works.
945 if (pte_present(ptent)) {
946 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
947 ptent = pte_wrprotect(ptent);
948 ptent = pte_clear_soft_dirty(ptent);
949 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
950 } else if (is_swap_pte(ptent)) {
951 ptent = pte_swp_clear_soft_dirty(ptent);
952 set_pte_at(vma->vm_mm, addr, pte, ptent);
956 static inline void clear_soft_dirty(struct vm_area_struct *vma,
957 unsigned long addr, pte_t *pte)
962 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
963 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
964 unsigned long addr, pmd_t *pmdp)
966 pmd_t old, pmd = *pmdp;
968 if (pmd_present(pmd)) {
969 /* See comment in change_huge_pmd() */
970 old = pmdp_invalidate(vma, addr, pmdp);
972 pmd = pmd_mkdirty(pmd);
974 pmd = pmd_mkyoung(pmd);
976 pmd = pmd_wrprotect(pmd);
977 pmd = pmd_clear_soft_dirty(pmd);
979 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
980 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
981 pmd = pmd_swp_clear_soft_dirty(pmd);
982 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
986 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
987 unsigned long addr, pmd_t *pmdp)
992 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
993 unsigned long end, struct mm_walk *walk)
995 struct clear_refs_private *cp = walk->private;
996 struct vm_area_struct *vma = walk->vma;
1001 ptl = pmd_trans_huge_lock(pmd, vma);
1003 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1004 clear_soft_dirty_pmd(vma, addr, pmd);
1008 if (!pmd_present(*pmd))
1011 page = pmd_page(*pmd);
1013 /* Clear accessed and referenced bits. */
1014 pmdp_test_and_clear_young(vma, addr, pmd);
1015 test_and_clear_page_young(page);
1016 ClearPageReferenced(page);
1022 if (pmd_trans_unstable(pmd))
1025 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1026 for (; addr != end; pte++, addr += PAGE_SIZE) {
1029 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1030 clear_soft_dirty(vma, addr, pte);
1034 if (!pte_present(ptent))
1037 page = vm_normal_page(vma, addr, ptent);
1041 /* Clear accessed and referenced bits. */
1042 ptep_test_and_clear_young(vma, addr, pte);
1043 test_and_clear_page_young(page);
1044 ClearPageReferenced(page);
1046 pte_unmap_unlock(pte - 1, ptl);
1051 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1052 struct mm_walk *walk)
1054 struct clear_refs_private *cp = walk->private;
1055 struct vm_area_struct *vma = walk->vma;
1057 if (vma->vm_flags & VM_PFNMAP)
1061 * Writing 1 to /proc/pid/clear_refs affects all pages.
1062 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1063 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1064 * Writing 4 to /proc/pid/clear_refs affects all pages.
1066 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1068 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1073 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1074 size_t count, loff_t *ppos)
1076 struct task_struct *task;
1077 char buffer[PROC_NUMBUF];
1078 struct mm_struct *mm;
1079 struct vm_area_struct *vma;
1080 enum clear_refs_types type;
1081 struct mmu_gather tlb;
1085 memset(buffer, 0, sizeof(buffer));
1086 if (count > sizeof(buffer) - 1)
1087 count = sizeof(buffer) - 1;
1088 if (copy_from_user(buffer, buf, count))
1090 rv = kstrtoint(strstrip(buffer), 10, &itype);
1093 type = (enum clear_refs_types)itype;
1094 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1097 task = get_proc_task(file_inode(file));
1100 mm = get_task_mm(task);
1102 struct clear_refs_private cp = {
1105 struct mm_walk clear_refs_walk = {
1106 .pmd_entry = clear_refs_pte_range,
1107 .test_walk = clear_refs_test_walk,
1112 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1113 if (down_write_killable(&mm->mmap_sem)) {
1119 * Writing 5 to /proc/pid/clear_refs resets the peak
1120 * resident set size to this mm's current rss value.
1122 reset_mm_hiwater_rss(mm);
1123 up_write(&mm->mmap_sem);
1127 down_read(&mm->mmap_sem);
1128 tlb_gather_mmu(&tlb, mm, 0, -1);
1129 if (type == CLEAR_REFS_SOFT_DIRTY) {
1130 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1131 if (!(vma->vm_flags & VM_SOFTDIRTY))
1133 up_read(&mm->mmap_sem);
1134 if (down_write_killable(&mm->mmap_sem)) {
1138 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1139 vma->vm_flags &= ~VM_SOFTDIRTY;
1140 vma_set_page_prot(vma);
1142 downgrade_write(&mm->mmap_sem);
1145 mmu_notifier_invalidate_range_start(mm, 0, -1);
1147 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1148 if (type == CLEAR_REFS_SOFT_DIRTY)
1149 mmu_notifier_invalidate_range_end(mm, 0, -1);
1150 tlb_finish_mmu(&tlb, 0, -1);
1151 up_read(&mm->mmap_sem);
1155 put_task_struct(task);
1160 const struct file_operations proc_clear_refs_operations = {
1161 .write = clear_refs_write,
1162 .llseek = noop_llseek,
1169 struct pagemapread {
1170 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1171 pagemap_entry_t *buffer;
1175 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1176 #define PAGEMAP_WALK_MASK (PMD_MASK)
1178 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1179 #define PM_PFRAME_BITS 55
1180 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1181 #define PM_SOFT_DIRTY BIT_ULL(55)
1182 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1183 #define PM_FILE BIT_ULL(61)
1184 #define PM_SWAP BIT_ULL(62)
1185 #define PM_PRESENT BIT_ULL(63)
1187 #define PM_END_OF_BUFFER 1
1189 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1191 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1194 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1195 struct pagemapread *pm)
1197 pm->buffer[pm->pos++] = *pme;
1198 if (pm->pos >= pm->len)
1199 return PM_END_OF_BUFFER;
1203 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1204 struct mm_walk *walk)
1206 struct pagemapread *pm = walk->private;
1207 unsigned long addr = start;
1210 while (addr < end) {
1211 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1212 pagemap_entry_t pme = make_pme(0, 0);
1213 /* End of address space hole, which we mark as non-present. */
1214 unsigned long hole_end;
1217 hole_end = min(end, vma->vm_start);
1221 for (; addr < hole_end; addr += PAGE_SIZE) {
1222 err = add_to_pagemap(addr, &pme, pm);
1230 /* Addresses in the VMA. */
1231 if (vma->vm_flags & VM_SOFTDIRTY)
1232 pme = make_pme(0, PM_SOFT_DIRTY);
1233 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1234 err = add_to_pagemap(addr, &pme, pm);
1243 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1244 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1246 u64 frame = 0, flags = 0;
1247 struct page *page = NULL;
1249 if (pte_present(pte)) {
1251 frame = pte_pfn(pte);
1252 flags |= PM_PRESENT;
1253 page = _vm_normal_page(vma, addr, pte, true);
1254 if (pte_soft_dirty(pte))
1255 flags |= PM_SOFT_DIRTY;
1256 } else if (is_swap_pte(pte)) {
1258 if (pte_swp_soft_dirty(pte))
1259 flags |= PM_SOFT_DIRTY;
1260 entry = pte_to_swp_entry(pte);
1261 frame = swp_type(entry) |
1262 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1264 if (is_migration_entry(entry))
1265 page = migration_entry_to_page(entry);
1267 if (is_device_private_entry(entry))
1268 page = device_private_entry_to_page(entry);
1271 if (page && !PageAnon(page))
1273 if (page && page_mapcount(page) == 1)
1274 flags |= PM_MMAP_EXCLUSIVE;
1275 if (vma->vm_flags & VM_SOFTDIRTY)
1276 flags |= PM_SOFT_DIRTY;
1278 return make_pme(frame, flags);
1281 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1282 struct mm_walk *walk)
1284 struct vm_area_struct *vma = walk->vma;
1285 struct pagemapread *pm = walk->private;
1287 pte_t *pte, *orig_pte;
1290 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1291 ptl = pmd_trans_huge_lock(pmdp, vma);
1293 u64 flags = 0, frame = 0;
1295 struct page *page = NULL;
1297 if (vma->vm_flags & VM_SOFTDIRTY)
1298 flags |= PM_SOFT_DIRTY;
1300 if (pmd_present(pmd)) {
1301 page = pmd_page(pmd);
1303 flags |= PM_PRESENT;
1304 if (pmd_soft_dirty(pmd))
1305 flags |= PM_SOFT_DIRTY;
1307 frame = pmd_pfn(pmd) +
1308 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1310 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1311 else if (is_swap_pmd(pmd)) {
1312 swp_entry_t entry = pmd_to_swp_entry(pmd);
1313 unsigned long offset = swp_offset(entry);
1315 offset += (addr & ~PMD_MASK) >> PAGE_SHIFT;
1316 frame = swp_type(entry) |
1317 (offset << MAX_SWAPFILES_SHIFT);
1319 if (pmd_swp_soft_dirty(pmd))
1320 flags |= PM_SOFT_DIRTY;
1321 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1322 page = migration_entry_to_page(entry);
1326 if (page && page_mapcount(page) == 1)
1327 flags |= PM_MMAP_EXCLUSIVE;
1329 for (; addr != end; addr += PAGE_SIZE) {
1330 pagemap_entry_t pme = make_pme(frame, flags);
1332 err = add_to_pagemap(addr, &pme, pm);
1335 if (pm->show_pfn && (flags & PM_PRESENT))
1337 else if (flags & PM_SWAP)
1338 frame += (1 << MAX_SWAPFILES_SHIFT);
1344 if (pmd_trans_unstable(pmdp))
1346 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1349 * We can assume that @vma always points to a valid one and @end never
1350 * goes beyond vma->vm_end.
1352 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1353 for (; addr < end; pte++, addr += PAGE_SIZE) {
1354 pagemap_entry_t pme;
1356 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1357 err = add_to_pagemap(addr, &pme, pm);
1361 pte_unmap_unlock(orig_pte, ptl);
1368 #ifdef CONFIG_HUGETLB_PAGE
1369 /* This function walks within one hugetlb entry in the single call */
1370 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1371 unsigned long addr, unsigned long end,
1372 struct mm_walk *walk)
1374 struct pagemapread *pm = walk->private;
1375 struct vm_area_struct *vma = walk->vma;
1376 u64 flags = 0, frame = 0;
1380 if (vma->vm_flags & VM_SOFTDIRTY)
1381 flags |= PM_SOFT_DIRTY;
1383 pte = huge_ptep_get(ptep);
1384 if (pte_present(pte)) {
1385 struct page *page = pte_page(pte);
1387 if (!PageAnon(page))
1390 if (page_mapcount(page) == 1)
1391 flags |= PM_MMAP_EXCLUSIVE;
1393 flags |= PM_PRESENT;
1395 frame = pte_pfn(pte) +
1396 ((addr & ~hmask) >> PAGE_SHIFT);
1399 for (; addr != end; addr += PAGE_SIZE) {
1400 pagemap_entry_t pme = make_pme(frame, flags);
1402 err = add_to_pagemap(addr, &pme, pm);
1405 if (pm->show_pfn && (flags & PM_PRESENT))
1413 #endif /* HUGETLB_PAGE */
1416 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1418 * For each page in the address space, this file contains one 64-bit entry
1419 * consisting of the following:
1421 * Bits 0-54 page frame number (PFN) if present
1422 * Bits 0-4 swap type if swapped
1423 * Bits 5-54 swap offset if swapped
1424 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1425 * Bit 56 page exclusively mapped
1427 * Bit 61 page is file-page or shared-anon
1428 * Bit 62 page swapped
1429 * Bit 63 page present
1431 * If the page is not present but in swap, then the PFN contains an
1432 * encoding of the swap file number and the page's offset into the
1433 * swap. Unmapped pages return a null PFN. This allows determining
1434 * precisely which pages are mapped (or in swap) and comparing mapped
1435 * pages between processes.
1437 * Efficient users of this interface will use /proc/pid/maps to
1438 * determine which areas of memory are actually mapped and llseek to
1439 * skip over unmapped regions.
1441 static ssize_t pagemap_read(struct file *file, char __user *buf,
1442 size_t count, loff_t *ppos)
1444 struct mm_struct *mm = file->private_data;
1445 struct pagemapread pm;
1446 struct mm_walk pagemap_walk = {};
1448 unsigned long svpfn;
1449 unsigned long start_vaddr;
1450 unsigned long end_vaddr;
1451 int ret = 0, copied = 0;
1453 if (!mm || !mmget_not_zero(mm))
1457 /* file position must be aligned */
1458 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1465 /* do not disclose physical addresses: attack vector */
1466 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1468 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1469 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_KERNEL);
1474 pagemap_walk.pmd_entry = pagemap_pmd_range;
1475 pagemap_walk.pte_hole = pagemap_pte_hole;
1476 #ifdef CONFIG_HUGETLB_PAGE
1477 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1479 pagemap_walk.mm = mm;
1480 pagemap_walk.private = ±
1483 svpfn = src / PM_ENTRY_BYTES;
1484 start_vaddr = svpfn << PAGE_SHIFT;
1485 end_vaddr = mm->task_size;
1487 /* watch out for wraparound */
1488 if (svpfn > mm->task_size >> PAGE_SHIFT)
1489 start_vaddr = end_vaddr;
1492 * The odds are that this will stop walking way
1493 * before end_vaddr, because the length of the
1494 * user buffer is tracked in "pm", and the walk
1495 * will stop when we hit the end of the buffer.
1498 while (count && (start_vaddr < end_vaddr)) {
1503 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1505 if (end < start_vaddr || end > end_vaddr)
1507 down_read(&mm->mmap_sem);
1508 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1509 up_read(&mm->mmap_sem);
1512 len = min(count, PM_ENTRY_BYTES * pm.pos);
1513 if (copy_to_user(buf, pm.buffer, len)) {
1522 if (!ret || ret == PM_END_OF_BUFFER)
1533 static int pagemap_open(struct inode *inode, struct file *file)
1535 struct mm_struct *mm;
1537 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1540 file->private_data = mm;
1544 static int pagemap_release(struct inode *inode, struct file *file)
1546 struct mm_struct *mm = file->private_data;
1553 const struct file_operations proc_pagemap_operations = {
1554 .llseek = mem_lseek, /* borrow this */
1555 .read = pagemap_read,
1556 .open = pagemap_open,
1557 .release = pagemap_release,
1559 #endif /* CONFIG_PROC_PAGE_MONITOR */
1564 unsigned long pages;
1566 unsigned long active;
1567 unsigned long writeback;
1568 unsigned long mapcount_max;
1569 unsigned long dirty;
1570 unsigned long swapcache;
1571 unsigned long node[MAX_NUMNODES];
1574 struct numa_maps_private {
1575 struct proc_maps_private proc_maps;
1576 struct numa_maps md;
1579 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1580 unsigned long nr_pages)
1582 int count = page_mapcount(page);
1584 md->pages += nr_pages;
1585 if (pte_dirty || PageDirty(page))
1586 md->dirty += nr_pages;
1588 if (PageSwapCache(page))
1589 md->swapcache += nr_pages;
1591 if (PageActive(page) || PageUnevictable(page))
1592 md->active += nr_pages;
1594 if (PageWriteback(page))
1595 md->writeback += nr_pages;
1598 md->anon += nr_pages;
1600 if (count > md->mapcount_max)
1601 md->mapcount_max = count;
1603 md->node[page_to_nid(page)] += nr_pages;
1606 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1612 if (!pte_present(pte))
1615 page = vm_normal_page(vma, addr, pte);
1619 if (PageReserved(page))
1622 nid = page_to_nid(page);
1623 if (!node_isset(nid, node_states[N_MEMORY]))
1629 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1630 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1631 struct vm_area_struct *vma,
1637 if (!pmd_present(pmd))
1640 page = vm_normal_page_pmd(vma, addr, pmd);
1644 if (PageReserved(page))
1647 nid = page_to_nid(page);
1648 if (!node_isset(nid, node_states[N_MEMORY]))
1655 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1656 unsigned long end, struct mm_walk *walk)
1658 struct numa_maps *md = walk->private;
1659 struct vm_area_struct *vma = walk->vma;
1664 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1665 ptl = pmd_trans_huge_lock(pmd, vma);
1669 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1671 gather_stats(page, md, pmd_dirty(*pmd),
1672 HPAGE_PMD_SIZE/PAGE_SIZE);
1677 if (pmd_trans_unstable(pmd))
1680 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1682 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1685 gather_stats(page, md, pte_dirty(*pte), 1);
1687 } while (pte++, addr += PAGE_SIZE, addr != end);
1688 pte_unmap_unlock(orig_pte, ptl);
1692 #ifdef CONFIG_HUGETLB_PAGE
1693 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1694 unsigned long addr, unsigned long end, struct mm_walk *walk)
1696 pte_t huge_pte = huge_ptep_get(pte);
1697 struct numa_maps *md;
1700 if (!pte_present(huge_pte))
1703 page = pte_page(huge_pte);
1708 gather_stats(page, md, pte_dirty(huge_pte), 1);
1713 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1714 unsigned long addr, unsigned long end, struct mm_walk *walk)
1721 * Display pages allocated per node and memory policy via /proc.
1723 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1725 struct numa_maps_private *numa_priv = m->private;
1726 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1727 struct vm_area_struct *vma = v;
1728 struct numa_maps *md = &numa_priv->md;
1729 struct file *file = vma->vm_file;
1730 struct mm_struct *mm = vma->vm_mm;
1731 struct mm_walk walk = {
1732 .hugetlb_entry = gather_hugetlb_stats,
1733 .pmd_entry = gather_pte_stats,
1737 struct mempolicy *pol;
1744 /* Ensure we start with an empty set of numa_maps statistics. */
1745 memset(md, 0, sizeof(*md));
1747 pol = __get_vma_policy(vma, vma->vm_start);
1749 mpol_to_str(buffer, sizeof(buffer), pol);
1752 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1755 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1758 seq_puts(m, " file=");
1759 seq_file_path(m, file, "\n\t= ");
1760 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1761 seq_puts(m, " heap");
1762 } else if (is_stack(vma)) {
1763 seq_puts(m, " stack");
1766 if (is_vm_hugetlb_page(vma))
1767 seq_puts(m, " huge");
1769 /* mmap_sem is held by m_start */
1770 walk_page_vma(vma, &walk);
1776 seq_printf(m, " anon=%lu", md->anon);
1779 seq_printf(m, " dirty=%lu", md->dirty);
1781 if (md->pages != md->anon && md->pages != md->dirty)
1782 seq_printf(m, " mapped=%lu", md->pages);
1784 if (md->mapcount_max > 1)
1785 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1788 seq_printf(m, " swapcache=%lu", md->swapcache);
1790 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1791 seq_printf(m, " active=%lu", md->active);
1794 seq_printf(m, " writeback=%lu", md->writeback);
1796 for_each_node_state(nid, N_MEMORY)
1798 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1800 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1803 m_cache_vma(m, vma);
1807 static int show_pid_numa_map(struct seq_file *m, void *v)
1809 return show_numa_map(m, v, 1);
1812 static int show_tid_numa_map(struct seq_file *m, void *v)
1814 return show_numa_map(m, v, 0);
1817 static const struct seq_operations proc_pid_numa_maps_op = {
1821 .show = show_pid_numa_map,
1824 static const struct seq_operations proc_tid_numa_maps_op = {
1828 .show = show_tid_numa_map,
1831 static int numa_maps_open(struct inode *inode, struct file *file,
1832 const struct seq_operations *ops)
1834 return proc_maps_open(inode, file, ops,
1835 sizeof(struct numa_maps_private));
1838 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1840 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1843 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1845 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1848 const struct file_operations proc_pid_numa_maps_operations = {
1849 .open = pid_numa_maps_open,
1851 .llseek = seq_lseek,
1852 .release = proc_map_release,
1855 const struct file_operations proc_tid_numa_maps_operations = {
1856 .open = tid_numa_maps_open,
1858 .llseek = seq_lseek,
1859 .release = proc_map_release,
1861 #endif /* CONFIG_NUMA */