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", atomic64_read(&mm->pinned_vm));
63 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
64 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
65 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
66 SEQ_PUT_DEC(" kB\nRssFile:\t", file);
67 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
68 SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
69 SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
70 seq_put_decimal_ull_width(m,
71 " kB\nVmExe:\t", text >> 10, 8);
72 seq_put_decimal_ull_width(m,
73 " kB\nVmLib:\t", lib >> 10, 8);
74 seq_put_decimal_ull_width(m,
75 " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
76 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
78 hugetlb_report_usage(m, mm);
82 unsigned long task_vsize(struct mm_struct *mm)
84 return PAGE_SIZE * mm->total_vm;
87 unsigned long task_statm(struct mm_struct *mm,
88 unsigned long *shared, unsigned long *text,
89 unsigned long *data, unsigned long *resident)
91 *shared = get_mm_counter(mm, MM_FILEPAGES) +
92 get_mm_counter(mm, MM_SHMEMPAGES);
93 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
95 *data = mm->data_vm + mm->stack_vm;
96 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
102 * Save get_task_policy() for show_numa_map().
104 static void hold_task_mempolicy(struct proc_maps_private *priv)
106 struct task_struct *task = priv->task;
109 priv->task_mempolicy = get_task_policy(task);
110 mpol_get(priv->task_mempolicy);
113 static void release_task_mempolicy(struct proc_maps_private *priv)
115 mpol_put(priv->task_mempolicy);
118 static void hold_task_mempolicy(struct proc_maps_private *priv)
121 static void release_task_mempolicy(struct proc_maps_private *priv)
126 static void 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 if (down_read_killable(&mm->mmap_sem)) {
171 return ERR_PTR(-EINTR);
174 hold_task_mempolicy(priv);
175 priv->tail_vma = get_gate_vma(mm);
178 vma = find_vma(mm, last_addr - 1);
179 if (vma && vma->vm_start <= last_addr)
180 vma = m_next_vma(priv, vma);
186 if (pos < mm->map_count) {
187 for (vma = mm->mmap; pos; pos--) {
188 m->version = vma->vm_start;
194 /* we do not bother to update m->version in this case */
195 if (pos == mm->map_count && priv->tail_vma)
196 return priv->tail_vma;
202 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
204 struct proc_maps_private *priv = m->private;
205 struct vm_area_struct *next;
208 next = m_next_vma(priv, v);
214 static void m_stop(struct seq_file *m, void *v)
216 struct proc_maps_private *priv = m->private;
218 if (!IS_ERR_OR_NULL(v))
221 put_task_struct(priv->task);
226 static int proc_maps_open(struct inode *inode, struct file *file,
227 const struct seq_operations *ops, int psize)
229 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
235 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
236 if (IS_ERR(priv->mm)) {
237 int err = PTR_ERR(priv->mm);
239 seq_release_private(inode, file);
246 static int proc_map_release(struct inode *inode, struct file *file)
248 struct seq_file *seq = file->private_data;
249 struct proc_maps_private *priv = seq->private;
254 return seq_release_private(inode, file);
257 static int do_maps_open(struct inode *inode, struct file *file,
258 const struct seq_operations *ops)
260 return proc_maps_open(inode, file, ops,
261 sizeof(struct proc_maps_private));
265 * Indicate if the VMA is a stack for the given task; for
266 * /proc/PID/maps that is the stack of the main task.
268 static int is_stack(struct vm_area_struct *vma)
271 * We make no effort to guess what a given thread considers to be
272 * its "stack". It's not even well-defined for programs written
275 return vma->vm_start <= vma->vm_mm->start_stack &&
276 vma->vm_end >= vma->vm_mm->start_stack;
279 static void show_vma_header_prefix(struct seq_file *m,
280 unsigned long start, unsigned long end,
281 vm_flags_t flags, unsigned long long pgoff,
282 dev_t dev, unsigned long ino)
284 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
285 seq_put_hex_ll(m, NULL, start, 8);
286 seq_put_hex_ll(m, "-", end, 8);
288 seq_putc(m, flags & VM_READ ? 'r' : '-');
289 seq_putc(m, flags & VM_WRITE ? 'w' : '-');
290 seq_putc(m, flags & VM_EXEC ? 'x' : '-');
291 seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
292 seq_put_hex_ll(m, " ", pgoff, 8);
293 seq_put_hex_ll(m, " ", MAJOR(dev), 2);
294 seq_put_hex_ll(m, ":", MINOR(dev), 2);
295 seq_put_decimal_ull(m, " ", ino);
300 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
302 struct mm_struct *mm = vma->vm_mm;
303 struct file *file = vma->vm_file;
304 vm_flags_t flags = vma->vm_flags;
305 unsigned long ino = 0;
306 unsigned long long pgoff = 0;
307 unsigned long start, end;
309 const char *name = NULL;
312 struct inode *inode = file_inode(vma->vm_file);
313 dev = inode->i_sb->s_dev;
315 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
318 start = vma->vm_start;
320 show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
323 * Print the dentry name for named mappings, and a
324 * special [heap] marker for the heap:
328 seq_file_path(m, file, "\n");
332 if (vma->vm_ops && vma->vm_ops->name) {
333 name = vma->vm_ops->name(vma);
338 name = arch_vma_name(vma);
345 if (vma->vm_start <= mm->brk &&
346 vma->vm_end >= mm->start_brk) {
363 static int show_map(struct seq_file *m, void *v)
370 static const struct seq_operations proc_pid_maps_op = {
377 static int pid_maps_open(struct inode *inode, struct file *file)
379 return do_maps_open(inode, file, &proc_pid_maps_op);
382 const struct file_operations proc_pid_maps_operations = {
383 .open = pid_maps_open,
386 .release = proc_map_release,
390 * Proportional Set Size(PSS): my share of RSS.
392 * PSS of a process is the count of pages it has in memory, where each
393 * page is divided by the number of processes sharing it. So if a
394 * process has 1000 pages all to itself, and 1000 shared with one other
395 * process, its PSS will be 1500.
397 * To keep (accumulated) division errors low, we adopt a 64bit
398 * fixed-point pss counter to minimize division errors. So (pss >>
399 * PSS_SHIFT) would be the real byte count.
401 * A shift of 12 before division means (assuming 4K page size):
402 * - 1M 3-user-pages add up to 8KB errors;
403 * - supports mapcount up to 2^24, or 16M;
404 * - supports PSS up to 2^52 bytes, or 4PB.
408 #ifdef CONFIG_PROC_PAGE_MONITOR
409 struct mem_size_stats {
410 unsigned long resident;
411 unsigned long shared_clean;
412 unsigned long shared_dirty;
413 unsigned long private_clean;
414 unsigned long private_dirty;
415 unsigned long referenced;
416 unsigned long anonymous;
417 unsigned long lazyfree;
418 unsigned long anonymous_thp;
419 unsigned long shmem_thp;
421 unsigned long shared_hugetlb;
422 unsigned long private_hugetlb;
429 bool check_shmem_swap;
432 static void smaps_page_accumulate(struct mem_size_stats *mss,
433 struct page *page, unsigned long size, unsigned long pss,
434 bool dirty, bool locked, bool private)
439 mss->pss_anon += pss;
440 else if (PageSwapBacked(page))
441 mss->pss_shmem += pss;
443 mss->pss_file += pss;
446 mss->pss_locked += pss;
448 if (dirty || PageDirty(page)) {
450 mss->private_dirty += size;
452 mss->shared_dirty += size;
455 mss->private_clean += size;
457 mss->shared_clean += size;
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;
468 * First accumulate quantities that depend only on |size| and the type
469 * of the compound page.
471 if (PageAnon(page)) {
472 mss->anonymous += size;
473 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
474 mss->lazyfree += size;
477 mss->resident += size;
478 /* Accumulate the size in pages that have been accessed. */
479 if (young || page_is_young(page) || PageReferenced(page))
480 mss->referenced += size;
483 * Then accumulate quantities that may depend on sharing, or that may
484 * differ page-by-page.
486 * page_count(page) == 1 guarantees the page is mapped exactly once.
487 * If any subpage of the compound page mapped with PTE it would elevate
490 if (page_count(page) == 1) {
491 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
495 for (i = 0; i < nr; i++, page++) {
496 int mapcount = page_mapcount(page);
497 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
500 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
506 static int smaps_pte_hole(unsigned long addr, unsigned long end,
507 struct mm_walk *walk)
509 struct mem_size_stats *mss = walk->private;
511 mss->swap += shmem_partial_swap_usage(
512 walk->vma->vm_file->f_mapping, addr, end);
518 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
519 struct mm_walk *walk)
521 struct mem_size_stats *mss = walk->private;
522 struct vm_area_struct *vma = walk->vma;
523 bool locked = !!(vma->vm_flags & VM_LOCKED);
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 (xa_is_value(page))
556 mss->swap += PAGE_SIZE;
566 smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
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;
575 bool locked = !!(vma->vm_flags & VM_LOCKED);
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), locked);
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 static void smap_gather_stats(struct vm_area_struct *vma,
735 struct mem_size_stats *mss)
737 struct mm_walk smaps_walk = {
738 .pmd_entry = smaps_pte_range,
739 #ifdef CONFIG_HUGETLB_PAGE
740 .hugetlb_entry = smaps_hugetlb_range,
745 smaps_walk.private = mss;
748 /* In case of smaps_rollup, reset the value from previous vma */
749 mss->check_shmem_swap = false;
750 if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
752 * For shared or readonly shmem mappings we know that all
753 * swapped out pages belong to the shmem object, and we can
754 * obtain the swap value much more efficiently. For private
755 * writable mappings, we might have COW pages that are
756 * not affected by the parent swapped out pages of the shmem
757 * object, so we have to distinguish them during the page walk.
758 * Unless we know that the shmem object (or the part mapped by
759 * our VMA) has no swapped out pages at all.
761 unsigned long shmem_swapped = shmem_swap_usage(vma);
763 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
764 !(vma->vm_flags & VM_WRITE)) {
765 mss->swap += shmem_swapped;
767 mss->check_shmem_swap = true;
768 smaps_walk.pte_hole = smaps_pte_hole;
772 /* mmap_sem is held in m_start */
773 walk_page_vma(vma, &smaps_walk);
776 #define SEQ_PUT_DEC(str, val) \
777 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
779 /* Show the contents common for smaps and smaps_rollup */
780 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
783 SEQ_PUT_DEC("Rss: ", mss->resident);
784 SEQ_PUT_DEC(" kB\nPss: ", mss->pss >> PSS_SHIFT);
787 * These are meaningful only for smaps_rollup, otherwise two of
788 * them are zero, and the other one is the same as Pss.
790 SEQ_PUT_DEC(" kB\nPss_Anon: ",
791 mss->pss_anon >> PSS_SHIFT);
792 SEQ_PUT_DEC(" kB\nPss_File: ",
793 mss->pss_file >> PSS_SHIFT);
794 SEQ_PUT_DEC(" kB\nPss_Shmem: ",
795 mss->pss_shmem >> PSS_SHIFT);
797 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss->shared_clean);
798 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss->shared_dirty);
799 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss->private_clean);
800 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss->private_dirty);
801 SEQ_PUT_DEC(" kB\nReferenced: ", mss->referenced);
802 SEQ_PUT_DEC(" kB\nAnonymous: ", mss->anonymous);
803 SEQ_PUT_DEC(" kB\nLazyFree: ", mss->lazyfree);
804 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss->anonymous_thp);
805 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
806 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
807 seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
808 mss->private_hugetlb >> 10, 7);
809 SEQ_PUT_DEC(" kB\nSwap: ", mss->swap);
810 SEQ_PUT_DEC(" kB\nSwapPss: ",
811 mss->swap_pss >> PSS_SHIFT);
812 SEQ_PUT_DEC(" kB\nLocked: ",
813 mss->pss_locked >> PSS_SHIFT);
814 seq_puts(m, " kB\n");
817 static int show_smap(struct seq_file *m, void *v)
819 struct vm_area_struct *vma = v;
820 struct mem_size_stats mss;
822 memset(&mss, 0, sizeof(mss));
824 smap_gather_stats(vma, &mss);
826 show_map_vma(m, vma);
828 SEQ_PUT_DEC("Size: ", vma->vm_end - vma->vm_start);
829 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
830 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma));
831 seq_puts(m, " kB\n");
833 __show_smap(m, &mss, false);
835 seq_printf(m, "THPeligible: %d\n",
836 transparent_hugepage_enabled(vma));
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_smaps_rollup(struct seq_file *m, void *v)
849 struct proc_maps_private *priv = m->private;
850 struct mem_size_stats mss;
851 struct mm_struct *mm;
852 struct vm_area_struct *vma;
853 unsigned long last_vma_end = 0;
856 priv->task = get_proc_task(priv->inode);
861 if (!mm || !mmget_not_zero(mm)) {
866 memset(&mss, 0, sizeof(mss));
868 ret = down_read_killable(&mm->mmap_sem);
872 hold_task_mempolicy(priv);
874 for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
875 smap_gather_stats(vma, &mss);
876 last_vma_end = vma->vm_end;
879 show_vma_header_prefix(m, priv->mm->mmap->vm_start,
880 last_vma_end, 0, 0, 0, 0);
882 seq_puts(m, "[rollup]\n");
884 __show_smap(m, &mss, true);
886 release_task_mempolicy(priv);
887 up_read(&mm->mmap_sem);
892 put_task_struct(priv->task);
899 static const struct seq_operations proc_pid_smaps_op = {
906 static int pid_smaps_open(struct inode *inode, struct file *file)
908 return do_maps_open(inode, file, &proc_pid_smaps_op);
911 static int smaps_rollup_open(struct inode *inode, struct file *file)
914 struct proc_maps_private *priv;
916 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
920 ret = single_open(file, show_smaps_rollup, priv);
925 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
926 if (IS_ERR(priv->mm)) {
927 ret = PTR_ERR(priv->mm);
929 single_release(inode, file);
940 static int smaps_rollup_release(struct inode *inode, struct file *file)
942 struct seq_file *seq = file->private_data;
943 struct proc_maps_private *priv = seq->private;
949 return single_release(inode, file);
952 const struct file_operations proc_pid_smaps_operations = {
953 .open = pid_smaps_open,
956 .release = proc_map_release,
959 const struct file_operations proc_pid_smaps_rollup_operations = {
960 .open = smaps_rollup_open,
963 .release = smaps_rollup_release,
966 enum clear_refs_types {
970 CLEAR_REFS_SOFT_DIRTY,
971 CLEAR_REFS_MM_HIWATER_RSS,
975 struct clear_refs_private {
976 enum clear_refs_types type;
979 #ifdef CONFIG_MEM_SOFT_DIRTY
980 static inline void clear_soft_dirty(struct vm_area_struct *vma,
981 unsigned long addr, pte_t *pte)
984 * The soft-dirty tracker uses #PF-s to catch writes
985 * to pages, so write-protect the pte as well. See the
986 * Documentation/admin-guide/mm/soft-dirty.rst for full description
987 * of how soft-dirty works.
991 if (pte_present(ptent)) {
994 old_pte = ptep_modify_prot_start(vma, addr, pte);
995 ptent = pte_wrprotect(old_pte);
996 ptent = pte_clear_soft_dirty(ptent);
997 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
998 } else if (is_swap_pte(ptent)) {
999 ptent = pte_swp_clear_soft_dirty(ptent);
1000 set_pte_at(vma->vm_mm, addr, pte, ptent);
1004 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1005 unsigned long addr, pte_t *pte)
1010 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1011 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1012 unsigned long addr, pmd_t *pmdp)
1014 pmd_t old, pmd = *pmdp;
1016 if (pmd_present(pmd)) {
1017 /* See comment in change_huge_pmd() */
1018 old = pmdp_invalidate(vma, addr, pmdp);
1020 pmd = pmd_mkdirty(pmd);
1022 pmd = pmd_mkyoung(pmd);
1024 pmd = pmd_wrprotect(pmd);
1025 pmd = pmd_clear_soft_dirty(pmd);
1027 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1028 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1029 pmd = pmd_swp_clear_soft_dirty(pmd);
1030 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1034 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1035 unsigned long addr, pmd_t *pmdp)
1040 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1041 unsigned long end, struct mm_walk *walk)
1043 struct clear_refs_private *cp = walk->private;
1044 struct vm_area_struct *vma = walk->vma;
1049 ptl = pmd_trans_huge_lock(pmd, vma);
1051 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1052 clear_soft_dirty_pmd(vma, addr, pmd);
1056 if (!pmd_present(*pmd))
1059 page = pmd_page(*pmd);
1061 /* Clear accessed and referenced bits. */
1062 pmdp_test_and_clear_young(vma, addr, pmd);
1063 test_and_clear_page_young(page);
1064 ClearPageReferenced(page);
1070 if (pmd_trans_unstable(pmd))
1073 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1074 for (; addr != end; pte++, addr += PAGE_SIZE) {
1077 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1078 clear_soft_dirty(vma, addr, pte);
1082 if (!pte_present(ptent))
1085 page = vm_normal_page(vma, addr, ptent);
1089 /* Clear accessed and referenced bits. */
1090 ptep_test_and_clear_young(vma, addr, pte);
1091 test_and_clear_page_young(page);
1092 ClearPageReferenced(page);
1094 pte_unmap_unlock(pte - 1, ptl);
1099 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1100 struct mm_walk *walk)
1102 struct clear_refs_private *cp = walk->private;
1103 struct vm_area_struct *vma = walk->vma;
1105 if (vma->vm_flags & VM_PFNMAP)
1109 * Writing 1 to /proc/pid/clear_refs affects all pages.
1110 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1111 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1112 * Writing 4 to /proc/pid/clear_refs affects all pages.
1114 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1116 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1121 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1122 size_t count, loff_t *ppos)
1124 struct task_struct *task;
1125 char buffer[PROC_NUMBUF];
1126 struct mm_struct *mm;
1127 struct vm_area_struct *vma;
1128 enum clear_refs_types type;
1129 struct mmu_gather tlb;
1133 memset(buffer, 0, sizeof(buffer));
1134 if (count > sizeof(buffer) - 1)
1135 count = sizeof(buffer) - 1;
1136 if (copy_from_user(buffer, buf, count))
1138 rv = kstrtoint(strstrip(buffer), 10, &itype);
1141 type = (enum clear_refs_types)itype;
1142 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1145 task = get_proc_task(file_inode(file));
1148 mm = get_task_mm(task);
1150 struct mmu_notifier_range range;
1151 struct clear_refs_private cp = {
1154 struct mm_walk clear_refs_walk = {
1155 .pmd_entry = clear_refs_pte_range,
1156 .test_walk = clear_refs_test_walk,
1161 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1162 if (down_write_killable(&mm->mmap_sem)) {
1168 * Writing 5 to /proc/pid/clear_refs resets the peak
1169 * resident set size to this mm's current rss value.
1171 reset_mm_hiwater_rss(mm);
1172 up_write(&mm->mmap_sem);
1176 if (down_read_killable(&mm->mmap_sem)) {
1180 tlb_gather_mmu(&tlb, mm, 0, -1);
1181 if (type == CLEAR_REFS_SOFT_DIRTY) {
1182 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1183 if (!(vma->vm_flags & VM_SOFTDIRTY))
1185 up_read(&mm->mmap_sem);
1186 if (down_write_killable(&mm->mmap_sem)) {
1191 * Avoid to modify vma->vm_flags
1192 * without locked ops while the
1193 * coredump reads the vm_flags.
1195 if (!mmget_still_valid(mm)) {
1197 * Silently return "count"
1198 * like if get_task_mm()
1199 * failed. FIXME: should this
1200 * function have returned
1201 * -ESRCH if get_task_mm()
1203 * get_proc_task() fails?
1205 up_write(&mm->mmap_sem);
1208 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1209 vma->vm_flags &= ~VM_SOFTDIRTY;
1210 vma_set_page_prot(vma);
1212 downgrade_write(&mm->mmap_sem);
1216 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1217 0, NULL, mm, 0, -1UL);
1218 mmu_notifier_invalidate_range_start(&range);
1220 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1221 if (type == CLEAR_REFS_SOFT_DIRTY)
1222 mmu_notifier_invalidate_range_end(&range);
1223 tlb_finish_mmu(&tlb, 0, -1);
1224 up_read(&mm->mmap_sem);
1228 put_task_struct(task);
1233 const struct file_operations proc_clear_refs_operations = {
1234 .write = clear_refs_write,
1235 .llseek = noop_llseek,
1242 struct pagemapread {
1243 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1244 pagemap_entry_t *buffer;
1248 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1249 #define PAGEMAP_WALK_MASK (PMD_MASK)
1251 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1252 #define PM_PFRAME_BITS 55
1253 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1254 #define PM_SOFT_DIRTY BIT_ULL(55)
1255 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1256 #define PM_FILE BIT_ULL(61)
1257 #define PM_SWAP BIT_ULL(62)
1258 #define PM_PRESENT BIT_ULL(63)
1260 #define PM_END_OF_BUFFER 1
1262 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1264 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1267 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1268 struct pagemapread *pm)
1270 pm->buffer[pm->pos++] = *pme;
1271 if (pm->pos >= pm->len)
1272 return PM_END_OF_BUFFER;
1276 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1277 struct mm_walk *walk)
1279 struct pagemapread *pm = walk->private;
1280 unsigned long addr = start;
1283 while (addr < end) {
1284 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1285 pagemap_entry_t pme = make_pme(0, 0);
1286 /* End of address space hole, which we mark as non-present. */
1287 unsigned long hole_end;
1290 hole_end = min(end, vma->vm_start);
1294 for (; addr < hole_end; addr += PAGE_SIZE) {
1295 err = add_to_pagemap(addr, &pme, pm);
1303 /* Addresses in the VMA. */
1304 if (vma->vm_flags & VM_SOFTDIRTY)
1305 pme = make_pme(0, PM_SOFT_DIRTY);
1306 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1307 err = add_to_pagemap(addr, &pme, pm);
1316 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1317 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1319 u64 frame = 0, flags = 0;
1320 struct page *page = NULL;
1322 if (pte_present(pte)) {
1324 frame = pte_pfn(pte);
1325 flags |= PM_PRESENT;
1326 page = vm_normal_page(vma, addr, pte);
1327 if (pte_soft_dirty(pte))
1328 flags |= PM_SOFT_DIRTY;
1329 } else if (is_swap_pte(pte)) {
1331 if (pte_swp_soft_dirty(pte))
1332 flags |= PM_SOFT_DIRTY;
1333 entry = pte_to_swp_entry(pte);
1335 frame = swp_type(entry) |
1336 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1338 if (is_migration_entry(entry))
1339 page = migration_entry_to_page(entry);
1341 if (is_device_private_entry(entry))
1342 page = device_private_entry_to_page(entry);
1345 if (page && !PageAnon(page))
1347 if (page && page_mapcount(page) == 1)
1348 flags |= PM_MMAP_EXCLUSIVE;
1349 if (vma->vm_flags & VM_SOFTDIRTY)
1350 flags |= PM_SOFT_DIRTY;
1352 return make_pme(frame, flags);
1355 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1356 struct mm_walk *walk)
1358 struct vm_area_struct *vma = walk->vma;
1359 struct pagemapread *pm = walk->private;
1361 pte_t *pte, *orig_pte;
1364 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1365 ptl = pmd_trans_huge_lock(pmdp, vma);
1367 u64 flags = 0, frame = 0;
1369 struct page *page = NULL;
1371 if (vma->vm_flags & VM_SOFTDIRTY)
1372 flags |= PM_SOFT_DIRTY;
1374 if (pmd_present(pmd)) {
1375 page = pmd_page(pmd);
1377 flags |= PM_PRESENT;
1378 if (pmd_soft_dirty(pmd))
1379 flags |= PM_SOFT_DIRTY;
1381 frame = pmd_pfn(pmd) +
1382 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1384 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1385 else if (is_swap_pmd(pmd)) {
1386 swp_entry_t entry = pmd_to_swp_entry(pmd);
1387 unsigned long offset;
1390 offset = swp_offset(entry) +
1391 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1392 frame = swp_type(entry) |
1393 (offset << MAX_SWAPFILES_SHIFT);
1396 if (pmd_swp_soft_dirty(pmd))
1397 flags |= PM_SOFT_DIRTY;
1398 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1399 page = migration_entry_to_page(entry);
1403 if (page && page_mapcount(page) == 1)
1404 flags |= PM_MMAP_EXCLUSIVE;
1406 for (; addr != end; addr += PAGE_SIZE) {
1407 pagemap_entry_t pme = make_pme(frame, flags);
1409 err = add_to_pagemap(addr, &pme, pm);
1413 if (flags & PM_PRESENT)
1415 else if (flags & PM_SWAP)
1416 frame += (1 << MAX_SWAPFILES_SHIFT);
1423 if (pmd_trans_unstable(pmdp))
1425 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1428 * We can assume that @vma always points to a valid one and @end never
1429 * goes beyond vma->vm_end.
1431 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1432 for (; addr < end; pte++, addr += PAGE_SIZE) {
1433 pagemap_entry_t pme;
1435 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1436 err = add_to_pagemap(addr, &pme, pm);
1440 pte_unmap_unlock(orig_pte, ptl);
1447 #ifdef CONFIG_HUGETLB_PAGE
1448 /* This function walks within one hugetlb entry in the single call */
1449 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1450 unsigned long addr, unsigned long end,
1451 struct mm_walk *walk)
1453 struct pagemapread *pm = walk->private;
1454 struct vm_area_struct *vma = walk->vma;
1455 u64 flags = 0, frame = 0;
1459 if (vma->vm_flags & VM_SOFTDIRTY)
1460 flags |= PM_SOFT_DIRTY;
1462 pte = huge_ptep_get(ptep);
1463 if (pte_present(pte)) {
1464 struct page *page = pte_page(pte);
1466 if (!PageAnon(page))
1469 if (page_mapcount(page) == 1)
1470 flags |= PM_MMAP_EXCLUSIVE;
1472 flags |= PM_PRESENT;
1474 frame = pte_pfn(pte) +
1475 ((addr & ~hmask) >> PAGE_SHIFT);
1478 for (; addr != end; addr += PAGE_SIZE) {
1479 pagemap_entry_t pme = make_pme(frame, flags);
1481 err = add_to_pagemap(addr, &pme, pm);
1484 if (pm->show_pfn && (flags & PM_PRESENT))
1492 #endif /* HUGETLB_PAGE */
1495 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1497 * For each page in the address space, this file contains one 64-bit entry
1498 * consisting of the following:
1500 * Bits 0-54 page frame number (PFN) if present
1501 * Bits 0-4 swap type if swapped
1502 * Bits 5-54 swap offset if swapped
1503 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1504 * Bit 56 page exclusively mapped
1506 * Bit 61 page is file-page or shared-anon
1507 * Bit 62 page swapped
1508 * Bit 63 page present
1510 * If the page is not present but in swap, then the PFN contains an
1511 * encoding of the swap file number and the page's offset into the
1512 * swap. Unmapped pages return a null PFN. This allows determining
1513 * precisely which pages are mapped (or in swap) and comparing mapped
1514 * pages between processes.
1516 * Efficient users of this interface will use /proc/pid/maps to
1517 * determine which areas of memory are actually mapped and llseek to
1518 * skip over unmapped regions.
1520 static ssize_t pagemap_read(struct file *file, char __user *buf,
1521 size_t count, loff_t *ppos)
1523 struct mm_struct *mm = file->private_data;
1524 struct pagemapread pm;
1525 struct mm_walk pagemap_walk = {};
1527 unsigned long svpfn;
1528 unsigned long start_vaddr;
1529 unsigned long end_vaddr;
1530 int ret = 0, copied = 0;
1532 if (!mm || !mmget_not_zero(mm))
1536 /* file position must be aligned */
1537 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1544 /* do not disclose physical addresses: attack vector */
1545 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1547 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1548 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1553 pagemap_walk.pmd_entry = pagemap_pmd_range;
1554 pagemap_walk.pte_hole = pagemap_pte_hole;
1555 #ifdef CONFIG_HUGETLB_PAGE
1556 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1558 pagemap_walk.mm = mm;
1559 pagemap_walk.private = ±
1562 svpfn = src / PM_ENTRY_BYTES;
1563 start_vaddr = svpfn << PAGE_SHIFT;
1564 end_vaddr = mm->task_size;
1566 /* watch out for wraparound */
1567 if (svpfn > mm->task_size >> PAGE_SHIFT)
1568 start_vaddr = end_vaddr;
1571 * The odds are that this will stop walking way
1572 * before end_vaddr, because the length of the
1573 * user buffer is tracked in "pm", and the walk
1574 * will stop when we hit the end of the buffer.
1577 while (count && (start_vaddr < end_vaddr)) {
1582 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1584 if (end < start_vaddr || end > end_vaddr)
1586 ret = down_read_killable(&mm->mmap_sem);
1589 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1590 up_read(&mm->mmap_sem);
1593 len = min(count, PM_ENTRY_BYTES * pm.pos);
1594 if (copy_to_user(buf, pm.buffer, len)) {
1603 if (!ret || ret == PM_END_OF_BUFFER)
1614 static int pagemap_open(struct inode *inode, struct file *file)
1616 struct mm_struct *mm;
1618 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1621 file->private_data = mm;
1625 static int pagemap_release(struct inode *inode, struct file *file)
1627 struct mm_struct *mm = file->private_data;
1634 const struct file_operations proc_pagemap_operations = {
1635 .llseek = mem_lseek, /* borrow this */
1636 .read = pagemap_read,
1637 .open = pagemap_open,
1638 .release = pagemap_release,
1640 #endif /* CONFIG_PROC_PAGE_MONITOR */
1645 unsigned long pages;
1647 unsigned long active;
1648 unsigned long writeback;
1649 unsigned long mapcount_max;
1650 unsigned long dirty;
1651 unsigned long swapcache;
1652 unsigned long node[MAX_NUMNODES];
1655 struct numa_maps_private {
1656 struct proc_maps_private proc_maps;
1657 struct numa_maps md;
1660 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1661 unsigned long nr_pages)
1663 int count = page_mapcount(page);
1665 md->pages += nr_pages;
1666 if (pte_dirty || PageDirty(page))
1667 md->dirty += nr_pages;
1669 if (PageSwapCache(page))
1670 md->swapcache += nr_pages;
1672 if (PageActive(page) || PageUnevictable(page))
1673 md->active += nr_pages;
1675 if (PageWriteback(page))
1676 md->writeback += nr_pages;
1679 md->anon += nr_pages;
1681 if (count > md->mapcount_max)
1682 md->mapcount_max = count;
1684 md->node[page_to_nid(page)] += nr_pages;
1687 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1693 if (!pte_present(pte))
1696 page = vm_normal_page(vma, addr, pte);
1700 if (PageReserved(page))
1703 nid = page_to_nid(page);
1704 if (!node_isset(nid, node_states[N_MEMORY]))
1710 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1711 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1712 struct vm_area_struct *vma,
1718 if (!pmd_present(pmd))
1721 page = vm_normal_page_pmd(vma, addr, pmd);
1725 if (PageReserved(page))
1728 nid = page_to_nid(page);
1729 if (!node_isset(nid, node_states[N_MEMORY]))
1736 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1737 unsigned long end, struct mm_walk *walk)
1739 struct numa_maps *md = walk->private;
1740 struct vm_area_struct *vma = walk->vma;
1745 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1746 ptl = pmd_trans_huge_lock(pmd, vma);
1750 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1752 gather_stats(page, md, pmd_dirty(*pmd),
1753 HPAGE_PMD_SIZE/PAGE_SIZE);
1758 if (pmd_trans_unstable(pmd))
1761 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1763 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1766 gather_stats(page, md, pte_dirty(*pte), 1);
1768 } while (pte++, addr += PAGE_SIZE, addr != end);
1769 pte_unmap_unlock(orig_pte, ptl);
1773 #ifdef CONFIG_HUGETLB_PAGE
1774 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1775 unsigned long addr, unsigned long end, struct mm_walk *walk)
1777 pte_t huge_pte = huge_ptep_get(pte);
1778 struct numa_maps *md;
1781 if (!pte_present(huge_pte))
1784 page = pte_page(huge_pte);
1789 gather_stats(page, md, pte_dirty(huge_pte), 1);
1794 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1795 unsigned long addr, unsigned long end, struct mm_walk *walk)
1802 * Display pages allocated per node and memory policy via /proc.
1804 static int show_numa_map(struct seq_file *m, void *v)
1806 struct numa_maps_private *numa_priv = m->private;
1807 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1808 struct vm_area_struct *vma = v;
1809 struct numa_maps *md = &numa_priv->md;
1810 struct file *file = vma->vm_file;
1811 struct mm_struct *mm = vma->vm_mm;
1812 struct mm_walk walk = {
1813 .hugetlb_entry = gather_hugetlb_stats,
1814 .pmd_entry = gather_pte_stats,
1818 struct mempolicy *pol;
1825 /* Ensure we start with an empty set of numa_maps statistics. */
1826 memset(md, 0, sizeof(*md));
1828 pol = __get_vma_policy(vma, vma->vm_start);
1830 mpol_to_str(buffer, sizeof(buffer), pol);
1833 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1836 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1839 seq_puts(m, " file=");
1840 seq_file_path(m, file, "\n\t= ");
1841 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1842 seq_puts(m, " heap");
1843 } else if (is_stack(vma)) {
1844 seq_puts(m, " stack");
1847 if (is_vm_hugetlb_page(vma))
1848 seq_puts(m, " huge");
1850 /* mmap_sem is held by m_start */
1851 walk_page_vma(vma, &walk);
1857 seq_printf(m, " anon=%lu", md->anon);
1860 seq_printf(m, " dirty=%lu", md->dirty);
1862 if (md->pages != md->anon && md->pages != md->dirty)
1863 seq_printf(m, " mapped=%lu", md->pages);
1865 if (md->mapcount_max > 1)
1866 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1869 seq_printf(m, " swapcache=%lu", md->swapcache);
1871 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1872 seq_printf(m, " active=%lu", md->active);
1875 seq_printf(m, " writeback=%lu", md->writeback);
1877 for_each_node_state(nid, N_MEMORY)
1879 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1881 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1884 m_cache_vma(m, vma);
1888 static const struct seq_operations proc_pid_numa_maps_op = {
1892 .show = show_numa_map,
1895 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1897 return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1898 sizeof(struct numa_maps_private));
1901 const struct file_operations proc_pid_numa_maps_operations = {
1902 .open = pid_numa_maps_open,
1904 .llseek = seq_lseek,
1905 .release = proc_map_release,
1908 #endif /* CONFIG_NUMA */