2 #include <linux/vmacache.h>
3 #include <linux/hugetlb.h>
4 #include <linux/huge_mm.h>
5 #include <linux/mount.h>
6 #include <linux/seq_file.h>
7 #include <linux/highmem.h>
8 #include <linux/ptrace.h>
9 #include <linux/slab.h>
10 #include <linux/pagemap.h>
11 #include <linux/mempolicy.h>
12 #include <linux/rmap.h>
13 #include <linux/swap.h>
14 #include <linux/swapops.h>
15 #include <linux/mmu_notifier.h>
18 #include <asm/uaccess.h>
19 #include <asm/tlbflush.h>
22 void task_mem(struct seq_file *m, struct mm_struct *mm)
24 unsigned long data, text, lib, swap;
25 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
28 * Note: to minimize their overhead, mm maintains hiwater_vm and
29 * hiwater_rss only when about to *lower* total_vm or rss. Any
30 * collector of these hiwater stats must therefore get total_vm
31 * and rss too, which will usually be the higher. Barriers? not
32 * worth the effort, such snapshots can always be inconsistent.
34 hiwater_vm = total_vm = mm->total_vm;
35 if (hiwater_vm < mm->hiwater_vm)
36 hiwater_vm = mm->hiwater_vm;
37 hiwater_rss = total_rss = get_mm_rss(mm);
38 if (hiwater_rss < mm->hiwater_rss)
39 hiwater_rss = mm->hiwater_rss;
41 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
42 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
43 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
44 swap = get_mm_counter(mm, MM_SWAPENTS);
58 hiwater_vm << (PAGE_SHIFT-10),
59 total_vm << (PAGE_SHIFT-10),
60 mm->locked_vm << (PAGE_SHIFT-10),
61 mm->pinned_vm << (PAGE_SHIFT-10),
62 hiwater_rss << (PAGE_SHIFT-10),
63 total_rss << (PAGE_SHIFT-10),
64 data << (PAGE_SHIFT-10),
65 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
66 (PTRS_PER_PTE * sizeof(pte_t) *
67 atomic_long_read(&mm->nr_ptes)) >> 10,
68 swap << (PAGE_SHIFT-10));
71 unsigned long task_vsize(struct mm_struct *mm)
73 return PAGE_SIZE * mm->total_vm;
76 unsigned long task_statm(struct mm_struct *mm,
77 unsigned long *shared, unsigned long *text,
78 unsigned long *data, unsigned long *resident)
80 *shared = get_mm_counter(mm, MM_FILEPAGES);
81 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
83 *data = mm->total_vm - mm->shared_vm;
84 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
90 * Save get_task_policy() for show_numa_map().
92 static void hold_task_mempolicy(struct proc_maps_private *priv)
94 struct task_struct *task = priv->task;
97 priv->task_mempolicy = get_task_policy(task);
98 mpol_get(priv->task_mempolicy);
101 static void release_task_mempolicy(struct proc_maps_private *priv)
103 mpol_put(priv->task_mempolicy);
106 static void hold_task_mempolicy(struct proc_maps_private *priv)
109 static void release_task_mempolicy(struct proc_maps_private *priv)
114 static void vma_stop(struct proc_maps_private *priv)
116 struct mm_struct *mm = priv->mm;
118 release_task_mempolicy(priv);
119 up_read(&mm->mmap_sem);
123 static struct vm_area_struct *
124 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
126 if (vma == priv->tail_vma)
128 return vma->vm_next ?: priv->tail_vma;
131 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
133 if (m->count < m->size) /* vma is copied successfully */
134 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
137 static void *m_start(struct seq_file *m, loff_t *ppos)
139 struct proc_maps_private *priv = m->private;
140 unsigned long last_addr = m->version;
141 struct mm_struct *mm;
142 struct vm_area_struct *vma;
143 unsigned int pos = *ppos;
145 /* See m_cache_vma(). Zero at the start or after lseek. */
146 if (last_addr == -1UL)
149 priv->task = get_proc_task(priv->inode);
151 return ERR_PTR(-ESRCH);
154 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
157 down_read(&mm->mmap_sem);
158 hold_task_mempolicy(priv);
159 priv->tail_vma = get_gate_vma(mm);
162 vma = find_vma(mm, last_addr);
163 if (vma && (vma = m_next_vma(priv, vma)))
168 if (pos < mm->map_count) {
169 for (vma = mm->mmap; pos; pos--) {
170 m->version = vma->vm_start;
176 /* we do not bother to update m->version in this case */
177 if (pos == mm->map_count && priv->tail_vma)
178 return priv->tail_vma;
184 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
186 struct proc_maps_private *priv = m->private;
187 struct vm_area_struct *next;
190 next = m_next_vma(priv, v);
196 static void m_stop(struct seq_file *m, void *v)
198 struct proc_maps_private *priv = m->private;
200 if (!IS_ERR_OR_NULL(v))
203 put_task_struct(priv->task);
208 static int proc_maps_open(struct inode *inode, struct file *file,
209 const struct seq_operations *ops, int psize)
211 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
217 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
218 if (IS_ERR(priv->mm)) {
219 int err = PTR_ERR(priv->mm);
221 seq_release_private(inode, file);
228 static int proc_map_release(struct inode *inode, struct file *file)
230 struct seq_file *seq = file->private_data;
231 struct proc_maps_private *priv = seq->private;
236 return seq_release_private(inode, file);
239 static int do_maps_open(struct inode *inode, struct file *file,
240 const struct seq_operations *ops)
242 return proc_maps_open(inode, file, ops,
243 sizeof(struct proc_maps_private));
246 static pid_t pid_of_stack(struct proc_maps_private *priv,
247 struct vm_area_struct *vma, bool is_pid)
249 struct inode *inode = priv->inode;
250 struct task_struct *task;
254 task = pid_task(proc_pid(inode), PIDTYPE_PID);
256 task = task_of_stack(task, vma, is_pid);
258 ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info);
266 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
268 struct mm_struct *mm = vma->vm_mm;
269 struct file *file = vma->vm_file;
270 struct proc_maps_private *priv = m->private;
271 vm_flags_t flags = vma->vm_flags;
272 unsigned long ino = 0;
273 unsigned long long pgoff = 0;
274 unsigned long start, end;
276 const char *name = NULL;
279 struct inode *inode = file_inode(vma->vm_file);
280 dev = inode->i_sb->s_dev;
282 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
285 /* We don't show the stack guard page in /proc/maps */
286 start = vma->vm_start;
287 if (stack_guard_page_start(vma, start))
290 if (stack_guard_page_end(vma, end))
293 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
294 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
297 flags & VM_READ ? 'r' : '-',
298 flags & VM_WRITE ? 'w' : '-',
299 flags & VM_EXEC ? 'x' : '-',
300 flags & VM_MAYSHARE ? 's' : 'p',
302 MAJOR(dev), MINOR(dev), ino);
305 * Print the dentry name for named mappings, and a
306 * special [heap] marker for the heap:
310 seq_path(m, &file->f_path, "\n");
314 if (vma->vm_ops && vma->vm_ops->name) {
315 name = vma->vm_ops->name(vma);
320 name = arch_vma_name(vma);
329 if (vma->vm_start <= mm->brk &&
330 vma->vm_end >= mm->start_brk) {
335 tid = pid_of_stack(priv, vma, is_pid);
338 * Thread stack in /proc/PID/task/TID/maps or
339 * the main process stack.
341 if (!is_pid || (vma->vm_start <= mm->start_stack &&
342 vma->vm_end >= mm->start_stack)) {
345 /* Thread stack in /proc/PID/maps */
347 seq_printf(m, "[stack:%d]", tid);
360 static int show_map(struct seq_file *m, void *v, int is_pid)
362 show_map_vma(m, v, is_pid);
367 static int show_pid_map(struct seq_file *m, void *v)
369 return show_map(m, v, 1);
372 static int show_tid_map(struct seq_file *m, void *v)
374 return show_map(m, v, 0);
377 static const struct seq_operations proc_pid_maps_op = {
384 static const struct seq_operations proc_tid_maps_op = {
391 static int pid_maps_open(struct inode *inode, struct file *file)
393 return do_maps_open(inode, file, &proc_pid_maps_op);
396 static int tid_maps_open(struct inode *inode, struct file *file)
398 return do_maps_open(inode, file, &proc_tid_maps_op);
401 const struct file_operations proc_pid_maps_operations = {
402 .open = pid_maps_open,
405 .release = proc_map_release,
408 const struct file_operations proc_tid_maps_operations = {
409 .open = tid_maps_open,
412 .release = proc_map_release,
416 * Proportional Set Size(PSS): my share of RSS.
418 * PSS of a process is the count of pages it has in memory, where each
419 * page is divided by the number of processes sharing it. So if a
420 * process has 1000 pages all to itself, and 1000 shared with one other
421 * process, its PSS will be 1500.
423 * To keep (accumulated) division errors low, we adopt a 64bit
424 * fixed-point pss counter to minimize division errors. So (pss >>
425 * PSS_SHIFT) would be the real byte count.
427 * A shift of 12 before division means (assuming 4K page size):
428 * - 1M 3-user-pages add up to 8KB errors;
429 * - supports mapcount up to 2^24, or 16M;
430 * - supports PSS up to 2^52 bytes, or 4PB.
434 #ifdef CONFIG_PROC_PAGE_MONITOR
435 struct mem_size_stats {
436 struct vm_area_struct *vma;
437 unsigned long resident;
438 unsigned long shared_clean;
439 unsigned long shared_dirty;
440 unsigned long private_clean;
441 unsigned long private_dirty;
442 unsigned long referenced;
443 unsigned long anonymous;
444 unsigned long anonymous_thp;
446 unsigned long nonlinear;
451 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
452 unsigned long ptent_size, struct mm_walk *walk)
454 struct mem_size_stats *mss = walk->private;
455 struct vm_area_struct *vma = mss->vma;
456 pgoff_t pgoff = linear_page_index(vma, addr);
457 struct page *page = NULL;
460 if (pte_present(ptent)) {
461 page = vm_normal_page(vma, addr, ptent);
462 } else if (is_swap_pte(ptent)) {
463 swp_entry_t swpent = pte_to_swp_entry(ptent);
465 if (!non_swap_entry(swpent))
466 mss->swap += ptent_size;
467 else if (is_migration_entry(swpent))
468 page = migration_entry_to_page(swpent);
469 } else if (pte_file(ptent)) {
470 if (pte_to_pgoff(ptent) != pgoff)
471 mss->nonlinear += ptent_size;
478 mss->anonymous += ptent_size;
480 if (page->index != pgoff)
481 mss->nonlinear += ptent_size;
483 mss->resident += ptent_size;
484 /* Accumulate the size in pages that have been accessed. */
485 if (pte_young(ptent) || PageReferenced(page))
486 mss->referenced += ptent_size;
487 mapcount = page_mapcount(page);
489 if (pte_dirty(ptent) || PageDirty(page))
490 mss->shared_dirty += ptent_size;
492 mss->shared_clean += ptent_size;
493 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
495 if (pte_dirty(ptent) || PageDirty(page))
496 mss->private_dirty += ptent_size;
498 mss->private_clean += ptent_size;
499 mss->pss += (ptent_size << PSS_SHIFT);
503 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
504 struct mm_walk *walk)
506 struct mem_size_stats *mss = walk->private;
507 struct vm_area_struct *vma = mss->vma;
511 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
512 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
514 mss->anonymous_thp += HPAGE_PMD_SIZE;
518 if (pmd_trans_unstable(pmd))
521 * The mmap_sem held all the way back in m_start() is what
522 * keeps khugepaged out of here and from collapsing things
525 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
526 for (; addr != end; pte++, addr += PAGE_SIZE)
527 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
528 pte_unmap_unlock(pte - 1, ptl);
533 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
536 * Don't forget to update Documentation/ on changes.
538 static const char mnemonics[BITS_PER_LONG][2] = {
540 * In case if we meet a flag we don't know about.
542 [0 ... (BITS_PER_LONG-1)] = "??",
544 [ilog2(VM_READ)] = "rd",
545 [ilog2(VM_WRITE)] = "wr",
546 [ilog2(VM_EXEC)] = "ex",
547 [ilog2(VM_SHARED)] = "sh",
548 [ilog2(VM_MAYREAD)] = "mr",
549 [ilog2(VM_MAYWRITE)] = "mw",
550 [ilog2(VM_MAYEXEC)] = "me",
551 [ilog2(VM_MAYSHARE)] = "ms",
552 [ilog2(VM_GROWSDOWN)] = "gd",
553 [ilog2(VM_PFNMAP)] = "pf",
554 [ilog2(VM_DENYWRITE)] = "dw",
555 #ifdef CONFIG_X86_INTEL_MPX
556 [ilog2(VM_MPX)] = "mp",
558 [ilog2(VM_LOCKED)] = "lo",
559 [ilog2(VM_IO)] = "io",
560 [ilog2(VM_SEQ_READ)] = "sr",
561 [ilog2(VM_RAND_READ)] = "rr",
562 [ilog2(VM_DONTCOPY)] = "dc",
563 [ilog2(VM_DONTEXPAND)] = "de",
564 [ilog2(VM_ACCOUNT)] = "ac",
565 [ilog2(VM_NORESERVE)] = "nr",
566 [ilog2(VM_HUGETLB)] = "ht",
567 [ilog2(VM_NONLINEAR)] = "nl",
568 [ilog2(VM_ARCH_1)] = "ar",
569 [ilog2(VM_DONTDUMP)] = "dd",
570 #ifdef CONFIG_MEM_SOFT_DIRTY
571 [ilog2(VM_SOFTDIRTY)] = "sd",
573 [ilog2(VM_MIXEDMAP)] = "mm",
574 [ilog2(VM_HUGEPAGE)] = "hg",
575 [ilog2(VM_NOHUGEPAGE)] = "nh",
576 [ilog2(VM_MERGEABLE)] = "mg",
580 seq_puts(m, "VmFlags: ");
581 for (i = 0; i < BITS_PER_LONG; i++) {
582 if (vma->vm_flags & (1UL << i)) {
583 seq_printf(m, "%c%c ",
584 mnemonics[i][0], mnemonics[i][1]);
590 static int show_smap(struct seq_file *m, void *v, int is_pid)
592 struct vm_area_struct *vma = v;
593 struct mem_size_stats mss;
594 struct mm_walk smaps_walk = {
595 .pmd_entry = smaps_pte_range,
600 memset(&mss, 0, sizeof mss);
602 /* mmap_sem is held in m_start */
603 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
604 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
606 show_map_vma(m, vma, is_pid);
612 "Shared_Clean: %8lu kB\n"
613 "Shared_Dirty: %8lu kB\n"
614 "Private_Clean: %8lu kB\n"
615 "Private_Dirty: %8lu kB\n"
616 "Referenced: %8lu kB\n"
617 "Anonymous: %8lu kB\n"
618 "AnonHugePages: %8lu kB\n"
620 "KernelPageSize: %8lu kB\n"
621 "MMUPageSize: %8lu kB\n"
623 (vma->vm_end - vma->vm_start) >> 10,
625 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
626 mss.shared_clean >> 10,
627 mss.shared_dirty >> 10,
628 mss.private_clean >> 10,
629 mss.private_dirty >> 10,
630 mss.referenced >> 10,
632 mss.anonymous_thp >> 10,
634 vma_kernel_pagesize(vma) >> 10,
635 vma_mmu_pagesize(vma) >> 10,
636 (vma->vm_flags & VM_LOCKED) ?
637 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
639 if (vma->vm_flags & VM_NONLINEAR)
640 seq_printf(m, "Nonlinear: %8lu kB\n",
641 mss.nonlinear >> 10);
643 show_smap_vma_flags(m, vma);
648 static int show_pid_smap(struct seq_file *m, void *v)
650 return show_smap(m, v, 1);
653 static int show_tid_smap(struct seq_file *m, void *v)
655 return show_smap(m, v, 0);
658 static const struct seq_operations proc_pid_smaps_op = {
662 .show = show_pid_smap
665 static const struct seq_operations proc_tid_smaps_op = {
669 .show = show_tid_smap
672 static int pid_smaps_open(struct inode *inode, struct file *file)
674 return do_maps_open(inode, file, &proc_pid_smaps_op);
677 static int tid_smaps_open(struct inode *inode, struct file *file)
679 return do_maps_open(inode, file, &proc_tid_smaps_op);
682 const struct file_operations proc_pid_smaps_operations = {
683 .open = pid_smaps_open,
686 .release = proc_map_release,
689 const struct file_operations proc_tid_smaps_operations = {
690 .open = tid_smaps_open,
693 .release = proc_map_release,
697 * We do not want to have constant page-shift bits sitting in
698 * pagemap entries and are about to reuse them some time soon.
700 * Here's the "migration strategy":
701 * 1. when the system boots these bits remain what they are,
702 * but a warning about future change is printed in log;
703 * 2. once anyone clears soft-dirty bits via clear_refs file,
704 * these flag is set to denote, that user is aware of the
705 * new API and those page-shift bits change their meaning.
706 * The respective warning is printed in dmesg;
707 * 3. In a couple of releases we will remove all the mentions
708 * of page-shift in pagemap entries.
711 static bool soft_dirty_cleared __read_mostly;
713 enum clear_refs_types {
717 CLEAR_REFS_SOFT_DIRTY,
721 struct clear_refs_private {
722 struct vm_area_struct *vma;
723 enum clear_refs_types type;
726 static inline void clear_soft_dirty(struct vm_area_struct *vma,
727 unsigned long addr, pte_t *pte)
729 #ifdef CONFIG_MEM_SOFT_DIRTY
731 * The soft-dirty tracker uses #PF-s to catch writes
732 * to pages, so write-protect the pte as well. See the
733 * Documentation/vm/soft-dirty.txt for full description
734 * of how soft-dirty works.
738 if (pte_present(ptent)) {
739 ptent = pte_wrprotect(ptent);
740 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY);
741 } else if (is_swap_pte(ptent)) {
742 ptent = pte_swp_clear_soft_dirty(ptent);
743 } else if (pte_file(ptent)) {
744 ptent = pte_file_clear_soft_dirty(ptent);
747 set_pte_at(vma->vm_mm, addr, pte, ptent);
751 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
752 unsigned long end, struct mm_walk *walk)
754 struct clear_refs_private *cp = walk->private;
755 struct vm_area_struct *vma = cp->vma;
760 split_huge_page_pmd(vma, addr, pmd);
761 if (pmd_trans_unstable(pmd))
764 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
765 for (; addr != end; pte++, addr += PAGE_SIZE) {
768 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
769 clear_soft_dirty(vma, addr, pte);
773 if (!pte_present(ptent))
776 page = vm_normal_page(vma, addr, ptent);
780 /* Clear accessed and referenced bits. */
781 ptep_test_and_clear_young(vma, addr, pte);
782 ClearPageReferenced(page);
784 pte_unmap_unlock(pte - 1, ptl);
789 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
790 size_t count, loff_t *ppos)
792 struct task_struct *task;
793 char buffer[PROC_NUMBUF];
794 struct mm_struct *mm;
795 struct vm_area_struct *vma;
796 enum clear_refs_types type;
800 memset(buffer, 0, sizeof(buffer));
801 if (count > sizeof(buffer) - 1)
802 count = sizeof(buffer) - 1;
803 if (copy_from_user(buffer, buf, count))
805 rv = kstrtoint(strstrip(buffer), 10, &itype);
808 type = (enum clear_refs_types)itype;
809 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
812 if (type == CLEAR_REFS_SOFT_DIRTY) {
813 soft_dirty_cleared = true;
814 pr_warn_once("The pagemap bits 55-60 has changed their meaning!"
815 " See the linux/Documentation/vm/pagemap.txt for "
819 task = get_proc_task(file_inode(file));
822 mm = get_task_mm(task);
824 struct clear_refs_private cp = {
827 struct mm_walk clear_refs_walk = {
828 .pmd_entry = clear_refs_pte_range,
832 down_read(&mm->mmap_sem);
833 if (type == CLEAR_REFS_SOFT_DIRTY) {
834 for (vma = mm->mmap; vma; vma = vma->vm_next) {
835 if (!(vma->vm_flags & VM_SOFTDIRTY))
837 up_read(&mm->mmap_sem);
838 down_write(&mm->mmap_sem);
839 for (vma = mm->mmap; vma; vma = vma->vm_next) {
840 vma->vm_flags &= ~VM_SOFTDIRTY;
841 vma_set_page_prot(vma);
843 downgrade_write(&mm->mmap_sem);
846 mmu_notifier_invalidate_range_start(mm, 0, -1);
848 for (vma = mm->mmap; vma; vma = vma->vm_next) {
850 if (is_vm_hugetlb_page(vma))
853 * Writing 1 to /proc/pid/clear_refs affects all pages.
855 * Writing 2 to /proc/pid/clear_refs only affects
858 * Writing 3 to /proc/pid/clear_refs only affects file
861 * Writing 4 to /proc/pid/clear_refs affects all pages.
863 if (type == CLEAR_REFS_ANON && vma->vm_file)
865 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
867 walk_page_range(vma->vm_start, vma->vm_end,
870 if (type == CLEAR_REFS_SOFT_DIRTY)
871 mmu_notifier_invalidate_range_end(mm, 0, -1);
873 up_read(&mm->mmap_sem);
876 put_task_struct(task);
881 const struct file_operations proc_clear_refs_operations = {
882 .write = clear_refs_write,
883 .llseek = noop_llseek,
891 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
892 pagemap_entry_t *buffer;
896 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
897 #define PAGEMAP_WALK_MASK (PMD_MASK)
899 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
900 #define PM_STATUS_BITS 3
901 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
902 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
903 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
904 #define PM_PSHIFT_BITS 6
905 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
906 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
907 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
908 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
909 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
910 /* in "new" pagemap pshift bits are occupied with more status bits */
911 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT))
913 #define __PM_SOFT_DIRTY (1LL)
914 #define PM_PRESENT PM_STATUS(4LL)
915 #define PM_SWAP PM_STATUS(2LL)
916 #define PM_FILE PM_STATUS(1LL)
917 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0)
918 #define PM_END_OF_BUFFER 1
920 static inline pagemap_entry_t make_pme(u64 val)
922 return (pagemap_entry_t) { .pme = val };
925 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
926 struct pagemapread *pm)
928 pm->buffer[pm->pos++] = *pme;
929 if (pm->pos >= pm->len)
930 return PM_END_OF_BUFFER;
934 static int pagemap_pte_hole(unsigned long start, unsigned long end,
935 struct mm_walk *walk)
937 struct pagemapread *pm = walk->private;
938 unsigned long addr = start;
942 struct vm_area_struct *vma = find_vma(walk->mm, addr);
943 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
944 /* End of address space hole, which we mark as non-present. */
945 unsigned long hole_end;
948 hole_end = min(end, vma->vm_start);
952 for (; addr < hole_end; addr += PAGE_SIZE) {
953 err = add_to_pagemap(addr, &pme, pm);
961 /* Addresses in the VMA. */
962 if (vma->vm_flags & VM_SOFTDIRTY)
963 pme.pme |= PM_STATUS2(pm->v2, __PM_SOFT_DIRTY);
964 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
965 err = add_to_pagemap(addr, &pme, pm);
974 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
975 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
978 struct page *page = NULL;
981 if (pte_present(pte)) {
982 frame = pte_pfn(pte);
984 page = vm_normal_page(vma, addr, pte);
985 if (pte_soft_dirty(pte))
986 flags2 |= __PM_SOFT_DIRTY;
987 } else if (is_swap_pte(pte)) {
989 if (pte_swp_soft_dirty(pte))
990 flags2 |= __PM_SOFT_DIRTY;
991 entry = pte_to_swp_entry(pte);
992 frame = swp_type(entry) |
993 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
995 if (is_migration_entry(entry))
996 page = migration_entry_to_page(entry);
998 if (vma->vm_flags & VM_SOFTDIRTY)
999 flags2 |= __PM_SOFT_DIRTY;
1000 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
1004 if (page && !PageAnon(page))
1006 if ((vma->vm_flags & VM_SOFTDIRTY))
1007 flags2 |= __PM_SOFT_DIRTY;
1009 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags);
1012 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1013 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1014 pmd_t pmd, int offset, int pmd_flags2)
1017 * Currently pmd for thp is always present because thp can not be
1018 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
1019 * This if-check is just to prepare for future implementation.
1021 if (pmd_present(pmd))
1022 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
1023 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT);
1025 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2));
1028 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1029 pmd_t pmd, int offset, int pmd_flags2)
1034 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
1035 struct mm_walk *walk)
1037 struct vm_area_struct *vma;
1038 struct pagemapread *pm = walk->private;
1043 /* find the first VMA at or above 'addr' */
1044 vma = find_vma(walk->mm, addr);
1045 if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1048 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd))
1049 pmd_flags2 = __PM_SOFT_DIRTY;
1053 for (; addr != end; addr += PAGE_SIZE) {
1054 unsigned long offset;
1055 pagemap_entry_t pme;
1057 offset = (addr & ~PAGEMAP_WALK_MASK) >>
1059 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2);
1060 err = add_to_pagemap(addr, &pme, pm);
1068 if (pmd_trans_unstable(pmd))
1072 /* End of address space hole, which we mark as non-present. */
1073 unsigned long hole_end;
1076 hole_end = min(end, vma->vm_start);
1080 for (; addr < hole_end; addr += PAGE_SIZE) {
1081 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
1083 err = add_to_pagemap(addr, &pme, pm);
1088 if (!vma || vma->vm_start >= end)
1091 * We can't possibly be in a hugetlb VMA. In general,
1092 * for a mm_walk with a pmd_entry and a hugetlb_entry,
1093 * the pmd_entry can only be called on addresses in a
1094 * hugetlb if the walk starts in a non-hugetlb VMA and
1095 * spans a hugepage VMA. Since pagemap_read walks are
1096 * PMD-sized and PMD-aligned, this will never be true.
1098 BUG_ON(is_vm_hugetlb_page(vma));
1100 /* Addresses in the VMA. */
1101 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1102 pagemap_entry_t pme;
1103 pte = pte_offset_map(pmd, addr);
1104 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte);
1106 err = add_to_pagemap(addr, &pme, pm);
1114 vma = find_vma(walk->mm, addr);
1122 #ifdef CONFIG_HUGETLB_PAGE
1123 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1124 pte_t pte, int offset, int flags2)
1126 if (pte_present(pte))
1127 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) |
1128 PM_STATUS2(pm->v2, flags2) |
1131 *pme = make_pme(PM_NOT_PRESENT(pm->v2) |
1132 PM_STATUS2(pm->v2, flags2));
1135 /* This function walks within one hugetlb entry in the single call */
1136 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1137 unsigned long addr, unsigned long end,
1138 struct mm_walk *walk)
1140 struct pagemapread *pm = walk->private;
1141 struct vm_area_struct *vma;
1144 pagemap_entry_t pme;
1146 vma = find_vma(walk->mm, addr);
1149 if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1150 flags2 = __PM_SOFT_DIRTY;
1154 for (; addr != end; addr += PAGE_SIZE) {
1155 int offset = (addr & ~hmask) >> PAGE_SHIFT;
1156 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2);
1157 err = add_to_pagemap(addr, &pme, pm);
1166 #endif /* HUGETLB_PAGE */
1169 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1171 * For each page in the address space, this file contains one 64-bit entry
1172 * consisting of the following:
1174 * Bits 0-54 page frame number (PFN) if present
1175 * Bits 0-4 swap type if swapped
1176 * Bits 5-54 swap offset if swapped
1177 * Bits 55-60 page shift (page size = 1<<page shift)
1178 * Bit 61 page is file-page or shared-anon
1179 * Bit 62 page swapped
1180 * Bit 63 page present
1182 * If the page is not present but in swap, then the PFN contains an
1183 * encoding of the swap file number and the page's offset into the
1184 * swap. Unmapped pages return a null PFN. This allows determining
1185 * precisely which pages are mapped (or in swap) and comparing mapped
1186 * pages between processes.
1188 * Efficient users of this interface will use /proc/pid/maps to
1189 * determine which areas of memory are actually mapped and llseek to
1190 * skip over unmapped regions.
1192 static ssize_t pagemap_read(struct file *file, char __user *buf,
1193 size_t count, loff_t *ppos)
1195 struct task_struct *task = get_proc_task(file_inode(file));
1196 struct mm_struct *mm;
1197 struct pagemapread pm;
1199 struct mm_walk pagemap_walk = {};
1201 unsigned long svpfn;
1202 unsigned long start_vaddr;
1203 unsigned long end_vaddr;
1210 /* file position must be aligned */
1211 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1218 pm.v2 = soft_dirty_cleared;
1219 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1220 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1225 mm = mm_access(task, PTRACE_MODE_READ);
1227 if (!mm || IS_ERR(mm))
1230 pagemap_walk.pmd_entry = pagemap_pte_range;
1231 pagemap_walk.pte_hole = pagemap_pte_hole;
1232 #ifdef CONFIG_HUGETLB_PAGE
1233 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1235 pagemap_walk.mm = mm;
1236 pagemap_walk.private = ±
1239 svpfn = src / PM_ENTRY_BYTES;
1240 start_vaddr = svpfn << PAGE_SHIFT;
1241 end_vaddr = TASK_SIZE_OF(task);
1243 /* watch out for wraparound */
1244 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1245 start_vaddr = end_vaddr;
1248 * The odds are that this will stop walking way
1249 * before end_vaddr, because the length of the
1250 * user buffer is tracked in "pm", and the walk
1251 * will stop when we hit the end of the buffer.
1254 while (count && (start_vaddr < end_vaddr)) {
1259 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1261 if (end < start_vaddr || end > end_vaddr)
1263 down_read(&mm->mmap_sem);
1264 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1265 up_read(&mm->mmap_sem);
1268 len = min(count, PM_ENTRY_BYTES * pm.pos);
1269 if (copy_to_user(buf, pm.buffer, len)) {
1278 if (!ret || ret == PM_END_OF_BUFFER)
1286 put_task_struct(task);
1291 static int pagemap_open(struct inode *inode, struct file *file)
1293 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about "
1294 "to stop being page-shift some time soon. See the "
1295 "linux/Documentation/vm/pagemap.txt for details.\n");
1299 const struct file_operations proc_pagemap_operations = {
1300 .llseek = mem_lseek, /* borrow this */
1301 .read = pagemap_read,
1302 .open = pagemap_open,
1304 #endif /* CONFIG_PROC_PAGE_MONITOR */
1309 struct vm_area_struct *vma;
1310 unsigned long pages;
1312 unsigned long active;
1313 unsigned long writeback;
1314 unsigned long mapcount_max;
1315 unsigned long dirty;
1316 unsigned long swapcache;
1317 unsigned long node[MAX_NUMNODES];
1320 struct numa_maps_private {
1321 struct proc_maps_private proc_maps;
1322 struct numa_maps md;
1325 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1326 unsigned long nr_pages)
1328 int count = page_mapcount(page);
1330 md->pages += nr_pages;
1331 if (pte_dirty || PageDirty(page))
1332 md->dirty += nr_pages;
1334 if (PageSwapCache(page))
1335 md->swapcache += nr_pages;
1337 if (PageActive(page) || PageUnevictable(page))
1338 md->active += nr_pages;
1340 if (PageWriteback(page))
1341 md->writeback += nr_pages;
1344 md->anon += nr_pages;
1346 if (count > md->mapcount_max)
1347 md->mapcount_max = count;
1349 md->node[page_to_nid(page)] += nr_pages;
1352 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1358 if (!pte_present(pte))
1361 page = vm_normal_page(vma, addr, pte);
1365 if (PageReserved(page))
1368 nid = page_to_nid(page);
1369 if (!node_isset(nid, node_states[N_MEMORY]))
1375 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1376 unsigned long end, struct mm_walk *walk)
1378 struct numa_maps *md;
1385 if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) {
1386 pte_t huge_pte = *(pte_t *)pmd;
1389 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1391 gather_stats(page, md, pte_dirty(huge_pte),
1392 HPAGE_PMD_SIZE/PAGE_SIZE);
1397 if (pmd_trans_unstable(pmd))
1399 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1401 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1404 gather_stats(page, md, pte_dirty(*pte), 1);
1406 } while (pte++, addr += PAGE_SIZE, addr != end);
1407 pte_unmap_unlock(orig_pte, ptl);
1410 #ifdef CONFIG_HUGETLB_PAGE
1411 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1412 unsigned long addr, unsigned long end, struct mm_walk *walk)
1414 struct numa_maps *md;
1417 if (!pte_present(*pte))
1420 page = pte_page(*pte);
1425 gather_stats(page, md, pte_dirty(*pte), 1);
1430 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1431 unsigned long addr, unsigned long end, struct mm_walk *walk)
1438 * Display pages allocated per node and memory policy via /proc.
1440 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1442 struct numa_maps_private *numa_priv = m->private;
1443 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1444 struct vm_area_struct *vma = v;
1445 struct numa_maps *md = &numa_priv->md;
1446 struct file *file = vma->vm_file;
1447 struct mm_struct *mm = vma->vm_mm;
1448 struct mm_walk walk = {};
1449 struct mempolicy *pol;
1456 /* Ensure we start with an empty set of numa_maps statistics. */
1457 memset(md, 0, sizeof(*md));
1461 walk.hugetlb_entry = gather_hugetbl_stats;
1462 walk.pmd_entry = gather_pte_stats;
1466 pol = __get_vma_policy(vma, vma->vm_start);
1468 mpol_to_str(buffer, sizeof(buffer), pol);
1471 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1474 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1477 seq_puts(m, " file=");
1478 seq_path(m, &file->f_path, "\n\t= ");
1479 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1480 seq_puts(m, " heap");
1482 pid_t tid = pid_of_stack(proc_priv, vma, is_pid);
1485 * Thread stack in /proc/PID/task/TID/maps or
1486 * the main process stack.
1488 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1489 vma->vm_end >= mm->start_stack))
1490 seq_puts(m, " stack");
1492 seq_printf(m, " stack:%d", tid);
1496 if (is_vm_hugetlb_page(vma))
1497 seq_puts(m, " huge");
1499 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1505 seq_printf(m, " anon=%lu", md->anon);
1508 seq_printf(m, " dirty=%lu", md->dirty);
1510 if (md->pages != md->anon && md->pages != md->dirty)
1511 seq_printf(m, " mapped=%lu", md->pages);
1513 if (md->mapcount_max > 1)
1514 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1517 seq_printf(m, " swapcache=%lu", md->swapcache);
1519 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1520 seq_printf(m, " active=%lu", md->active);
1523 seq_printf(m, " writeback=%lu", md->writeback);
1525 for_each_node_state(nid, N_MEMORY)
1527 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1530 m_cache_vma(m, vma);
1534 static int show_pid_numa_map(struct seq_file *m, void *v)
1536 return show_numa_map(m, v, 1);
1539 static int show_tid_numa_map(struct seq_file *m, void *v)
1541 return show_numa_map(m, v, 0);
1544 static const struct seq_operations proc_pid_numa_maps_op = {
1548 .show = show_pid_numa_map,
1551 static const struct seq_operations proc_tid_numa_maps_op = {
1555 .show = show_tid_numa_map,
1558 static int numa_maps_open(struct inode *inode, struct file *file,
1559 const struct seq_operations *ops)
1561 return proc_maps_open(inode, file, ops,
1562 sizeof(struct numa_maps_private));
1565 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1567 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1570 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1572 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1575 const struct file_operations proc_pid_numa_maps_operations = {
1576 .open = pid_numa_maps_open,
1578 .llseek = seq_lseek,
1579 .release = proc_map_release,
1582 const struct file_operations proc_tid_numa_maps_operations = {
1583 .open = tid_numa_maps_open,
1585 .llseek = seq_lseek,
1586 .release = proc_map_release,
1588 #endif /* CONFIG_NUMA */