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", transparent_hugepage_enabled(vma));
837 if (arch_pkeys_enabled())
838 seq_printf(m, "ProtectionKey: %8u\n", vma_pkey(vma));
839 show_smap_vma_flags(m, vma);
846 static int show_smaps_rollup(struct seq_file *m, void *v)
848 struct proc_maps_private *priv = m->private;
849 struct mem_size_stats mss;
850 struct mm_struct *mm;
851 struct vm_area_struct *vma;
852 unsigned long last_vma_end = 0;
855 priv->task = get_proc_task(priv->inode);
860 if (!mm || !mmget_not_zero(mm)) {
865 memset(&mss, 0, sizeof(mss));
867 ret = down_read_killable(&mm->mmap_sem);
871 hold_task_mempolicy(priv);
873 for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
874 smap_gather_stats(vma, &mss);
875 last_vma_end = vma->vm_end;
878 show_vma_header_prefix(m, priv->mm->mmap->vm_start,
879 last_vma_end, 0, 0, 0, 0);
881 seq_puts(m, "[rollup]\n");
883 __show_smap(m, &mss, true);
885 release_task_mempolicy(priv);
886 up_read(&mm->mmap_sem);
891 put_task_struct(priv->task);
898 static const struct seq_operations proc_pid_smaps_op = {
905 static int pid_smaps_open(struct inode *inode, struct file *file)
907 return do_maps_open(inode, file, &proc_pid_smaps_op);
910 static int smaps_rollup_open(struct inode *inode, struct file *file)
913 struct proc_maps_private *priv;
915 priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
919 ret = single_open(file, show_smaps_rollup, priv);
924 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
925 if (IS_ERR(priv->mm)) {
926 ret = PTR_ERR(priv->mm);
928 single_release(inode, file);
939 static int smaps_rollup_release(struct inode *inode, struct file *file)
941 struct seq_file *seq = file->private_data;
942 struct proc_maps_private *priv = seq->private;
948 return single_release(inode, file);
951 const struct file_operations proc_pid_smaps_operations = {
952 .open = pid_smaps_open,
955 .release = proc_map_release,
958 const struct file_operations proc_pid_smaps_rollup_operations = {
959 .open = smaps_rollup_open,
962 .release = smaps_rollup_release,
965 enum clear_refs_types {
969 CLEAR_REFS_SOFT_DIRTY,
970 CLEAR_REFS_MM_HIWATER_RSS,
974 struct clear_refs_private {
975 enum clear_refs_types type;
978 #ifdef CONFIG_MEM_SOFT_DIRTY
979 static inline void clear_soft_dirty(struct vm_area_struct *vma,
980 unsigned long addr, pte_t *pte)
983 * The soft-dirty tracker uses #PF-s to catch writes
984 * to pages, so write-protect the pte as well. See the
985 * Documentation/admin-guide/mm/soft-dirty.rst for full description
986 * of how soft-dirty works.
990 if (pte_present(ptent)) {
993 old_pte = ptep_modify_prot_start(vma, addr, pte);
994 ptent = pte_wrprotect(old_pte);
995 ptent = pte_clear_soft_dirty(ptent);
996 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
997 } else if (is_swap_pte(ptent)) {
998 ptent = pte_swp_clear_soft_dirty(ptent);
999 set_pte_at(vma->vm_mm, addr, pte, ptent);
1003 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1004 unsigned long addr, pte_t *pte)
1009 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1010 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1011 unsigned long addr, pmd_t *pmdp)
1013 pmd_t old, pmd = *pmdp;
1015 if (pmd_present(pmd)) {
1016 /* See comment in change_huge_pmd() */
1017 old = pmdp_invalidate(vma, addr, pmdp);
1019 pmd = pmd_mkdirty(pmd);
1021 pmd = pmd_mkyoung(pmd);
1023 pmd = pmd_wrprotect(pmd);
1024 pmd = pmd_clear_soft_dirty(pmd);
1026 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1027 } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1028 pmd = pmd_swp_clear_soft_dirty(pmd);
1029 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1033 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1034 unsigned long addr, pmd_t *pmdp)
1039 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1040 unsigned long end, struct mm_walk *walk)
1042 struct clear_refs_private *cp = walk->private;
1043 struct vm_area_struct *vma = walk->vma;
1048 ptl = pmd_trans_huge_lock(pmd, vma);
1050 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1051 clear_soft_dirty_pmd(vma, addr, pmd);
1055 if (!pmd_present(*pmd))
1058 page = pmd_page(*pmd);
1060 /* Clear accessed and referenced bits. */
1061 pmdp_test_and_clear_young(vma, addr, pmd);
1062 test_and_clear_page_young(page);
1063 ClearPageReferenced(page);
1069 if (pmd_trans_unstable(pmd))
1072 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1073 for (; addr != end; pte++, addr += PAGE_SIZE) {
1076 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1077 clear_soft_dirty(vma, addr, pte);
1081 if (!pte_present(ptent))
1084 page = vm_normal_page(vma, addr, ptent);
1088 /* Clear accessed and referenced bits. */
1089 ptep_test_and_clear_young(vma, addr, pte);
1090 test_and_clear_page_young(page);
1091 ClearPageReferenced(page);
1093 pte_unmap_unlock(pte - 1, ptl);
1098 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1099 struct mm_walk *walk)
1101 struct clear_refs_private *cp = walk->private;
1102 struct vm_area_struct *vma = walk->vma;
1104 if (vma->vm_flags & VM_PFNMAP)
1108 * Writing 1 to /proc/pid/clear_refs affects all pages.
1109 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1110 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1111 * Writing 4 to /proc/pid/clear_refs affects all pages.
1113 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1115 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1120 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1121 size_t count, loff_t *ppos)
1123 struct task_struct *task;
1124 char buffer[PROC_NUMBUF];
1125 struct mm_struct *mm;
1126 struct vm_area_struct *vma;
1127 enum clear_refs_types type;
1128 struct mmu_gather tlb;
1132 memset(buffer, 0, sizeof(buffer));
1133 if (count > sizeof(buffer) - 1)
1134 count = sizeof(buffer) - 1;
1135 if (copy_from_user(buffer, buf, count))
1137 rv = kstrtoint(strstrip(buffer), 10, &itype);
1140 type = (enum clear_refs_types)itype;
1141 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1144 task = get_proc_task(file_inode(file));
1147 mm = get_task_mm(task);
1149 struct mmu_notifier_range range;
1150 struct clear_refs_private cp = {
1153 struct mm_walk clear_refs_walk = {
1154 .pmd_entry = clear_refs_pte_range,
1155 .test_walk = clear_refs_test_walk,
1160 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1161 if (down_write_killable(&mm->mmap_sem)) {
1167 * Writing 5 to /proc/pid/clear_refs resets the peak
1168 * resident set size to this mm's current rss value.
1170 reset_mm_hiwater_rss(mm);
1171 up_write(&mm->mmap_sem);
1175 if (down_read_killable(&mm->mmap_sem)) {
1179 tlb_gather_mmu(&tlb, mm, 0, -1);
1180 if (type == CLEAR_REFS_SOFT_DIRTY) {
1181 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1182 if (!(vma->vm_flags & VM_SOFTDIRTY))
1184 up_read(&mm->mmap_sem);
1185 if (down_write_killable(&mm->mmap_sem)) {
1190 * Avoid to modify vma->vm_flags
1191 * without locked ops while the
1192 * coredump reads the vm_flags.
1194 if (!mmget_still_valid(mm)) {
1196 * Silently return "count"
1197 * like if get_task_mm()
1198 * failed. FIXME: should this
1199 * function have returned
1200 * -ESRCH if get_task_mm()
1202 * get_proc_task() fails?
1204 up_write(&mm->mmap_sem);
1207 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1208 vma->vm_flags &= ~VM_SOFTDIRTY;
1209 vma_set_page_prot(vma);
1211 downgrade_write(&mm->mmap_sem);
1215 mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1216 0, NULL, mm, 0, -1UL);
1217 mmu_notifier_invalidate_range_start(&range);
1219 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1220 if (type == CLEAR_REFS_SOFT_DIRTY)
1221 mmu_notifier_invalidate_range_end(&range);
1222 tlb_finish_mmu(&tlb, 0, -1);
1223 up_read(&mm->mmap_sem);
1227 put_task_struct(task);
1232 const struct file_operations proc_clear_refs_operations = {
1233 .write = clear_refs_write,
1234 .llseek = noop_llseek,
1241 struct pagemapread {
1242 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
1243 pagemap_entry_t *buffer;
1247 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1248 #define PAGEMAP_WALK_MASK (PMD_MASK)
1250 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1251 #define PM_PFRAME_BITS 55
1252 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1253 #define PM_SOFT_DIRTY BIT_ULL(55)
1254 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1255 #define PM_FILE BIT_ULL(61)
1256 #define PM_SWAP BIT_ULL(62)
1257 #define PM_PRESENT BIT_ULL(63)
1259 #define PM_END_OF_BUFFER 1
1261 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1263 return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1266 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1267 struct pagemapread *pm)
1269 pm->buffer[pm->pos++] = *pme;
1270 if (pm->pos >= pm->len)
1271 return PM_END_OF_BUFFER;
1275 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1276 struct mm_walk *walk)
1278 struct pagemapread *pm = walk->private;
1279 unsigned long addr = start;
1282 while (addr < end) {
1283 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1284 pagemap_entry_t pme = make_pme(0, 0);
1285 /* End of address space hole, which we mark as non-present. */
1286 unsigned long hole_end;
1289 hole_end = min(end, vma->vm_start);
1293 for (; addr < hole_end; addr += PAGE_SIZE) {
1294 err = add_to_pagemap(addr, &pme, pm);
1302 /* Addresses in the VMA. */
1303 if (vma->vm_flags & VM_SOFTDIRTY)
1304 pme = make_pme(0, PM_SOFT_DIRTY);
1305 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1306 err = add_to_pagemap(addr, &pme, pm);
1315 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1316 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1318 u64 frame = 0, flags = 0;
1319 struct page *page = NULL;
1321 if (pte_present(pte)) {
1323 frame = pte_pfn(pte);
1324 flags |= PM_PRESENT;
1325 page = vm_normal_page(vma, addr, pte);
1326 if (pte_soft_dirty(pte))
1327 flags |= PM_SOFT_DIRTY;
1328 } else if (is_swap_pte(pte)) {
1330 if (pte_swp_soft_dirty(pte))
1331 flags |= PM_SOFT_DIRTY;
1332 entry = pte_to_swp_entry(pte);
1334 frame = swp_type(entry) |
1335 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1337 if (is_migration_entry(entry))
1338 page = migration_entry_to_page(entry);
1340 if (is_device_private_entry(entry))
1341 page = device_private_entry_to_page(entry);
1344 if (page && !PageAnon(page))
1346 if (page && page_mapcount(page) == 1)
1347 flags |= PM_MMAP_EXCLUSIVE;
1348 if (vma->vm_flags & VM_SOFTDIRTY)
1349 flags |= PM_SOFT_DIRTY;
1351 return make_pme(frame, flags);
1354 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1355 struct mm_walk *walk)
1357 struct vm_area_struct *vma = walk->vma;
1358 struct pagemapread *pm = walk->private;
1360 pte_t *pte, *orig_pte;
1363 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1364 ptl = pmd_trans_huge_lock(pmdp, vma);
1366 u64 flags = 0, frame = 0;
1368 struct page *page = NULL;
1370 if (vma->vm_flags & VM_SOFTDIRTY)
1371 flags |= PM_SOFT_DIRTY;
1373 if (pmd_present(pmd)) {
1374 page = pmd_page(pmd);
1376 flags |= PM_PRESENT;
1377 if (pmd_soft_dirty(pmd))
1378 flags |= PM_SOFT_DIRTY;
1380 frame = pmd_pfn(pmd) +
1381 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1383 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1384 else if (is_swap_pmd(pmd)) {
1385 swp_entry_t entry = pmd_to_swp_entry(pmd);
1386 unsigned long offset;
1389 offset = swp_offset(entry) +
1390 ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1391 frame = swp_type(entry) |
1392 (offset << MAX_SWAPFILES_SHIFT);
1395 if (pmd_swp_soft_dirty(pmd))
1396 flags |= PM_SOFT_DIRTY;
1397 VM_BUG_ON(!is_pmd_migration_entry(pmd));
1398 page = migration_entry_to_page(entry);
1402 if (page && page_mapcount(page) == 1)
1403 flags |= PM_MMAP_EXCLUSIVE;
1405 for (; addr != end; addr += PAGE_SIZE) {
1406 pagemap_entry_t pme = make_pme(frame, flags);
1408 err = add_to_pagemap(addr, &pme, pm);
1412 if (flags & PM_PRESENT)
1414 else if (flags & PM_SWAP)
1415 frame += (1 << MAX_SWAPFILES_SHIFT);
1422 if (pmd_trans_unstable(pmdp))
1424 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1427 * We can assume that @vma always points to a valid one and @end never
1428 * goes beyond vma->vm_end.
1430 orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1431 for (; addr < end; pte++, addr += PAGE_SIZE) {
1432 pagemap_entry_t pme;
1434 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1435 err = add_to_pagemap(addr, &pme, pm);
1439 pte_unmap_unlock(orig_pte, ptl);
1446 #ifdef CONFIG_HUGETLB_PAGE
1447 /* This function walks within one hugetlb entry in the single call */
1448 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1449 unsigned long addr, unsigned long end,
1450 struct mm_walk *walk)
1452 struct pagemapread *pm = walk->private;
1453 struct vm_area_struct *vma = walk->vma;
1454 u64 flags = 0, frame = 0;
1458 if (vma->vm_flags & VM_SOFTDIRTY)
1459 flags |= PM_SOFT_DIRTY;
1461 pte = huge_ptep_get(ptep);
1462 if (pte_present(pte)) {
1463 struct page *page = pte_page(pte);
1465 if (!PageAnon(page))
1468 if (page_mapcount(page) == 1)
1469 flags |= PM_MMAP_EXCLUSIVE;
1471 flags |= PM_PRESENT;
1473 frame = pte_pfn(pte) +
1474 ((addr & ~hmask) >> PAGE_SHIFT);
1477 for (; addr != end; addr += PAGE_SIZE) {
1478 pagemap_entry_t pme = make_pme(frame, flags);
1480 err = add_to_pagemap(addr, &pme, pm);
1483 if (pm->show_pfn && (flags & PM_PRESENT))
1491 #endif /* HUGETLB_PAGE */
1494 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1496 * For each page in the address space, this file contains one 64-bit entry
1497 * consisting of the following:
1499 * Bits 0-54 page frame number (PFN) if present
1500 * Bits 0-4 swap type if swapped
1501 * Bits 5-54 swap offset if swapped
1502 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1503 * Bit 56 page exclusively mapped
1505 * Bit 61 page is file-page or shared-anon
1506 * Bit 62 page swapped
1507 * Bit 63 page present
1509 * If the page is not present but in swap, then the PFN contains an
1510 * encoding of the swap file number and the page's offset into the
1511 * swap. Unmapped pages return a null PFN. This allows determining
1512 * precisely which pages are mapped (or in swap) and comparing mapped
1513 * pages between processes.
1515 * Efficient users of this interface will use /proc/pid/maps to
1516 * determine which areas of memory are actually mapped and llseek to
1517 * skip over unmapped regions.
1519 static ssize_t pagemap_read(struct file *file, char __user *buf,
1520 size_t count, loff_t *ppos)
1522 struct mm_struct *mm = file->private_data;
1523 struct pagemapread pm;
1524 struct mm_walk pagemap_walk = {};
1526 unsigned long svpfn;
1527 unsigned long start_vaddr;
1528 unsigned long end_vaddr;
1529 int ret = 0, copied = 0;
1531 if (!mm || !mmget_not_zero(mm))
1535 /* file position must be aligned */
1536 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1543 /* do not disclose physical addresses: attack vector */
1544 pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1546 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1547 pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1552 pagemap_walk.pmd_entry = pagemap_pmd_range;
1553 pagemap_walk.pte_hole = pagemap_pte_hole;
1554 #ifdef CONFIG_HUGETLB_PAGE
1555 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1557 pagemap_walk.mm = mm;
1558 pagemap_walk.private = ±
1561 svpfn = src / PM_ENTRY_BYTES;
1562 start_vaddr = svpfn << PAGE_SHIFT;
1563 end_vaddr = mm->task_size;
1565 /* watch out for wraparound */
1566 if (svpfn > mm->task_size >> PAGE_SHIFT)
1567 start_vaddr = end_vaddr;
1570 * The odds are that this will stop walking way
1571 * before end_vaddr, because the length of the
1572 * user buffer is tracked in "pm", and the walk
1573 * will stop when we hit the end of the buffer.
1576 while (count && (start_vaddr < end_vaddr)) {
1581 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1583 if (end < start_vaddr || end > end_vaddr)
1585 ret = down_read_killable(&mm->mmap_sem);
1588 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1589 up_read(&mm->mmap_sem);
1592 len = min(count, PM_ENTRY_BYTES * pm.pos);
1593 if (copy_to_user(buf, pm.buffer, len)) {
1602 if (!ret || ret == PM_END_OF_BUFFER)
1613 static int pagemap_open(struct inode *inode, struct file *file)
1615 struct mm_struct *mm;
1617 mm = proc_mem_open(inode, PTRACE_MODE_READ);
1620 file->private_data = mm;
1624 static int pagemap_release(struct inode *inode, struct file *file)
1626 struct mm_struct *mm = file->private_data;
1633 const struct file_operations proc_pagemap_operations = {
1634 .llseek = mem_lseek, /* borrow this */
1635 .read = pagemap_read,
1636 .open = pagemap_open,
1637 .release = pagemap_release,
1639 #endif /* CONFIG_PROC_PAGE_MONITOR */
1644 unsigned long pages;
1646 unsigned long active;
1647 unsigned long writeback;
1648 unsigned long mapcount_max;
1649 unsigned long dirty;
1650 unsigned long swapcache;
1651 unsigned long node[MAX_NUMNODES];
1654 struct numa_maps_private {
1655 struct proc_maps_private proc_maps;
1656 struct numa_maps md;
1659 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1660 unsigned long nr_pages)
1662 int count = page_mapcount(page);
1664 md->pages += nr_pages;
1665 if (pte_dirty || PageDirty(page))
1666 md->dirty += nr_pages;
1668 if (PageSwapCache(page))
1669 md->swapcache += nr_pages;
1671 if (PageActive(page) || PageUnevictable(page))
1672 md->active += nr_pages;
1674 if (PageWriteback(page))
1675 md->writeback += nr_pages;
1678 md->anon += nr_pages;
1680 if (count > md->mapcount_max)
1681 md->mapcount_max = count;
1683 md->node[page_to_nid(page)] += nr_pages;
1686 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1692 if (!pte_present(pte))
1695 page = vm_normal_page(vma, addr, pte);
1699 if (PageReserved(page))
1702 nid = page_to_nid(page);
1703 if (!node_isset(nid, node_states[N_MEMORY]))
1709 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1710 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1711 struct vm_area_struct *vma,
1717 if (!pmd_present(pmd))
1720 page = vm_normal_page_pmd(vma, addr, pmd);
1724 if (PageReserved(page))
1727 nid = page_to_nid(page);
1728 if (!node_isset(nid, node_states[N_MEMORY]))
1735 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1736 unsigned long end, struct mm_walk *walk)
1738 struct numa_maps *md = walk->private;
1739 struct vm_area_struct *vma = walk->vma;
1744 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1745 ptl = pmd_trans_huge_lock(pmd, vma);
1749 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1751 gather_stats(page, md, pmd_dirty(*pmd),
1752 HPAGE_PMD_SIZE/PAGE_SIZE);
1757 if (pmd_trans_unstable(pmd))
1760 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1762 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1765 gather_stats(page, md, pte_dirty(*pte), 1);
1767 } while (pte++, addr += PAGE_SIZE, addr != end);
1768 pte_unmap_unlock(orig_pte, ptl);
1772 #ifdef CONFIG_HUGETLB_PAGE
1773 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1774 unsigned long addr, unsigned long end, struct mm_walk *walk)
1776 pte_t huge_pte = huge_ptep_get(pte);
1777 struct numa_maps *md;
1780 if (!pte_present(huge_pte))
1783 page = pte_page(huge_pte);
1788 gather_stats(page, md, pte_dirty(huge_pte), 1);
1793 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1794 unsigned long addr, unsigned long end, struct mm_walk *walk)
1801 * Display pages allocated per node and memory policy via /proc.
1803 static int show_numa_map(struct seq_file *m, void *v)
1805 struct numa_maps_private *numa_priv = m->private;
1806 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1807 struct vm_area_struct *vma = v;
1808 struct numa_maps *md = &numa_priv->md;
1809 struct file *file = vma->vm_file;
1810 struct mm_struct *mm = vma->vm_mm;
1811 struct mm_walk walk = {
1812 .hugetlb_entry = gather_hugetlb_stats,
1813 .pmd_entry = gather_pte_stats,
1817 struct mempolicy *pol;
1824 /* Ensure we start with an empty set of numa_maps statistics. */
1825 memset(md, 0, sizeof(*md));
1827 pol = __get_vma_policy(vma, vma->vm_start);
1829 mpol_to_str(buffer, sizeof(buffer), pol);
1832 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1835 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1838 seq_puts(m, " file=");
1839 seq_file_path(m, file, "\n\t= ");
1840 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1841 seq_puts(m, " heap");
1842 } else if (is_stack(vma)) {
1843 seq_puts(m, " stack");
1846 if (is_vm_hugetlb_page(vma))
1847 seq_puts(m, " huge");
1849 /* mmap_sem is held by m_start */
1850 walk_page_vma(vma, &walk);
1856 seq_printf(m, " anon=%lu", md->anon);
1859 seq_printf(m, " dirty=%lu", md->dirty);
1861 if (md->pages != md->anon && md->pages != md->dirty)
1862 seq_printf(m, " mapped=%lu", md->pages);
1864 if (md->mapcount_max > 1)
1865 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1868 seq_printf(m, " swapcache=%lu", md->swapcache);
1870 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1871 seq_printf(m, " active=%lu", md->active);
1874 seq_printf(m, " writeback=%lu", md->writeback);
1876 for_each_node_state(nid, N_MEMORY)
1878 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1880 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1883 m_cache_vma(m, vma);
1887 static const struct seq_operations proc_pid_numa_maps_op = {
1891 .show = show_numa_map,
1894 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1896 return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1897 sizeof(struct numa_maps_private));
1900 const struct file_operations proc_pid_numa_maps_operations = {
1901 .open = pid_numa_maps_open,
1903 .llseek = seq_lseek,
1904 .release = proc_map_release,
1907 #endif /* CONFIG_NUMA */