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, ptes, pmds;
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);
45 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes);
46 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm);
61 hiwater_vm << (PAGE_SHIFT-10),
62 total_vm << (PAGE_SHIFT-10),
63 mm->locked_vm << (PAGE_SHIFT-10),
64 mm->pinned_vm << (PAGE_SHIFT-10),
65 hiwater_rss << (PAGE_SHIFT-10),
66 total_rss << (PAGE_SHIFT-10),
67 data << (PAGE_SHIFT-10),
68 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
71 swap << (PAGE_SHIFT-10));
74 unsigned long task_vsize(struct mm_struct *mm)
76 return PAGE_SIZE * mm->total_vm;
79 unsigned long task_statm(struct mm_struct *mm,
80 unsigned long *shared, unsigned long *text,
81 unsigned long *data, unsigned long *resident)
83 *shared = get_mm_counter(mm, MM_FILEPAGES);
84 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
86 *data = mm->total_vm - mm->shared_vm;
87 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
93 * Save get_task_policy() for show_numa_map().
95 static void hold_task_mempolicy(struct proc_maps_private *priv)
97 struct task_struct *task = priv->task;
100 priv->task_mempolicy = get_task_policy(task);
101 mpol_get(priv->task_mempolicy);
104 static void release_task_mempolicy(struct proc_maps_private *priv)
106 mpol_put(priv->task_mempolicy);
109 static void hold_task_mempolicy(struct proc_maps_private *priv)
112 static void release_task_mempolicy(struct proc_maps_private *priv)
117 static void vma_stop(struct proc_maps_private *priv)
119 struct mm_struct *mm = priv->mm;
121 release_task_mempolicy(priv);
122 up_read(&mm->mmap_sem);
126 static struct vm_area_struct *
127 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
129 if (vma == priv->tail_vma)
131 return vma->vm_next ?: priv->tail_vma;
134 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
136 if (m->count < m->size) /* vma is copied successfully */
137 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL;
140 static void *m_start(struct seq_file *m, loff_t *ppos)
142 struct proc_maps_private *priv = m->private;
143 unsigned long last_addr = m->version;
144 struct mm_struct *mm;
145 struct vm_area_struct *vma;
146 unsigned int pos = *ppos;
148 /* See m_cache_vma(). Zero at the start or after lseek. */
149 if (last_addr == -1UL)
152 priv->task = get_proc_task(priv->inode);
154 return ERR_PTR(-ESRCH);
157 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
160 down_read(&mm->mmap_sem);
161 hold_task_mempolicy(priv);
162 priv->tail_vma = get_gate_vma(mm);
165 vma = find_vma(mm, last_addr);
166 if (vma && (vma = m_next_vma(priv, vma)))
171 if (pos < mm->map_count) {
172 for (vma = mm->mmap; pos; pos--) {
173 m->version = vma->vm_start;
179 /* we do not bother to update m->version in this case */
180 if (pos == mm->map_count && priv->tail_vma)
181 return priv->tail_vma;
187 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
189 struct proc_maps_private *priv = m->private;
190 struct vm_area_struct *next;
193 next = m_next_vma(priv, v);
199 static void m_stop(struct seq_file *m, void *v)
201 struct proc_maps_private *priv = m->private;
203 if (!IS_ERR_OR_NULL(v))
206 put_task_struct(priv->task);
211 static int proc_maps_open(struct inode *inode, struct file *file,
212 const struct seq_operations *ops, int psize)
214 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
220 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
221 if (IS_ERR(priv->mm)) {
222 int err = PTR_ERR(priv->mm);
224 seq_release_private(inode, file);
231 static int proc_map_release(struct inode *inode, struct file *file)
233 struct seq_file *seq = file->private_data;
234 struct proc_maps_private *priv = seq->private;
239 return seq_release_private(inode, file);
242 static int do_maps_open(struct inode *inode, struct file *file,
243 const struct seq_operations *ops)
245 return proc_maps_open(inode, file, ops,
246 sizeof(struct proc_maps_private));
249 static pid_t pid_of_stack(struct proc_maps_private *priv,
250 struct vm_area_struct *vma, bool is_pid)
252 struct inode *inode = priv->inode;
253 struct task_struct *task;
257 task = pid_task(proc_pid(inode), PIDTYPE_PID);
259 task = task_of_stack(task, vma, is_pid);
261 ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info);
269 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
271 struct mm_struct *mm = vma->vm_mm;
272 struct file *file = vma->vm_file;
273 struct proc_maps_private *priv = m->private;
274 vm_flags_t flags = vma->vm_flags;
275 unsigned long ino = 0;
276 unsigned long long pgoff = 0;
277 unsigned long start, end;
279 const char *name = NULL;
282 struct inode *inode = file_inode(vma->vm_file);
283 dev = inode->i_sb->s_dev;
285 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
288 /* We don't show the stack guard page in /proc/maps */
289 start = vma->vm_start;
290 if (stack_guard_page_start(vma, start))
293 if (stack_guard_page_end(vma, end))
296 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
297 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
300 flags & VM_READ ? 'r' : '-',
301 flags & VM_WRITE ? 'w' : '-',
302 flags & VM_EXEC ? 'x' : '-',
303 flags & VM_MAYSHARE ? 's' : 'p',
305 MAJOR(dev), MINOR(dev), ino);
308 * Print the dentry name for named mappings, and a
309 * special [heap] marker for the heap:
313 seq_path(m, &file->f_path, "\n");
317 if (vma->vm_ops && vma->vm_ops->name) {
318 name = vma->vm_ops->name(vma);
323 name = arch_vma_name(vma);
332 if (vma->vm_start <= mm->brk &&
333 vma->vm_end >= mm->start_brk) {
338 tid = pid_of_stack(priv, vma, is_pid);
341 * Thread stack in /proc/PID/task/TID/maps or
342 * the main process stack.
344 if (!is_pid || (vma->vm_start <= mm->start_stack &&
345 vma->vm_end >= mm->start_stack)) {
348 /* Thread stack in /proc/PID/maps */
350 seq_printf(m, "[stack:%d]", tid);
363 static int show_map(struct seq_file *m, void *v, int is_pid)
365 show_map_vma(m, v, is_pid);
370 static int show_pid_map(struct seq_file *m, void *v)
372 return show_map(m, v, 1);
375 static int show_tid_map(struct seq_file *m, void *v)
377 return show_map(m, v, 0);
380 static const struct seq_operations proc_pid_maps_op = {
387 static const struct seq_operations proc_tid_maps_op = {
394 static int pid_maps_open(struct inode *inode, struct file *file)
396 return do_maps_open(inode, file, &proc_pid_maps_op);
399 static int tid_maps_open(struct inode *inode, struct file *file)
401 return do_maps_open(inode, file, &proc_tid_maps_op);
404 const struct file_operations proc_pid_maps_operations = {
405 .open = pid_maps_open,
408 .release = proc_map_release,
411 const struct file_operations proc_tid_maps_operations = {
412 .open = tid_maps_open,
415 .release = proc_map_release,
419 * Proportional Set Size(PSS): my share of RSS.
421 * PSS of a process is the count of pages it has in memory, where each
422 * page is divided by the number of processes sharing it. So if a
423 * process has 1000 pages all to itself, and 1000 shared with one other
424 * process, its PSS will be 1500.
426 * To keep (accumulated) division errors low, we adopt a 64bit
427 * fixed-point pss counter to minimize division errors. So (pss >>
428 * PSS_SHIFT) would be the real byte count.
430 * A shift of 12 before division means (assuming 4K page size):
431 * - 1M 3-user-pages add up to 8KB errors;
432 * - supports mapcount up to 2^24, or 16M;
433 * - supports PSS up to 2^52 bytes, or 4PB.
437 #ifdef CONFIG_PROC_PAGE_MONITOR
438 struct mem_size_stats {
439 unsigned long resident;
440 unsigned long shared_clean;
441 unsigned long shared_dirty;
442 unsigned long private_clean;
443 unsigned long private_dirty;
444 unsigned long referenced;
445 unsigned long anonymous;
446 unsigned long anonymous_thp;
451 static void smaps_account(struct mem_size_stats *mss, struct page *page,
452 unsigned long size, bool young, bool dirty)
457 mss->anonymous += size;
459 mss->resident += size;
460 /* Accumulate the size in pages that have been accessed. */
461 if (young || PageReferenced(page))
462 mss->referenced += size;
463 mapcount = page_mapcount(page);
467 if (dirty || PageDirty(page))
468 mss->shared_dirty += size;
470 mss->shared_clean += size;
471 pss_delta = (u64)size << PSS_SHIFT;
472 do_div(pss_delta, mapcount);
473 mss->pss += pss_delta;
475 if (dirty || PageDirty(page))
476 mss->private_dirty += size;
478 mss->private_clean += size;
479 mss->pss += (u64)size << PSS_SHIFT;
483 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
484 struct mm_walk *walk)
486 struct mem_size_stats *mss = walk->private;
487 struct vm_area_struct *vma = walk->vma;
488 struct page *page = NULL;
490 if (pte_present(*pte)) {
491 page = vm_normal_page(vma, addr, *pte);
492 } else if (is_swap_pte(*pte)) {
493 swp_entry_t swpent = pte_to_swp_entry(*pte);
495 if (!non_swap_entry(swpent))
496 mss->swap += PAGE_SIZE;
497 else if (is_migration_entry(swpent))
498 page = migration_entry_to_page(swpent);
503 smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte));
506 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
507 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
508 struct mm_walk *walk)
510 struct mem_size_stats *mss = walk->private;
511 struct vm_area_struct *vma = walk->vma;
514 /* FOLL_DUMP will return -EFAULT on huge zero page */
515 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
516 if (IS_ERR_OR_NULL(page))
518 mss->anonymous_thp += HPAGE_PMD_SIZE;
519 smaps_account(mss, page, HPAGE_PMD_SIZE,
520 pmd_young(*pmd), pmd_dirty(*pmd));
523 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
524 struct mm_walk *walk)
529 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
530 struct mm_walk *walk)
532 struct vm_area_struct *vma = walk->vma;
536 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
537 smaps_pmd_entry(pmd, addr, walk);
542 if (pmd_trans_unstable(pmd))
545 * The mmap_sem held all the way back in m_start() is what
546 * keeps khugepaged out of here and from collapsing things
549 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
550 for (; addr != end; pte++, addr += PAGE_SIZE)
551 smaps_pte_entry(pte, addr, walk);
552 pte_unmap_unlock(pte - 1, ptl);
557 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
560 * Don't forget to update Documentation/ on changes.
562 static const char mnemonics[BITS_PER_LONG][2] = {
564 * In case if we meet a flag we don't know about.
566 [0 ... (BITS_PER_LONG-1)] = "??",
568 [ilog2(VM_READ)] = "rd",
569 [ilog2(VM_WRITE)] = "wr",
570 [ilog2(VM_EXEC)] = "ex",
571 [ilog2(VM_SHARED)] = "sh",
572 [ilog2(VM_MAYREAD)] = "mr",
573 [ilog2(VM_MAYWRITE)] = "mw",
574 [ilog2(VM_MAYEXEC)] = "me",
575 [ilog2(VM_MAYSHARE)] = "ms",
576 [ilog2(VM_GROWSDOWN)] = "gd",
577 [ilog2(VM_PFNMAP)] = "pf",
578 [ilog2(VM_DENYWRITE)] = "dw",
579 #ifdef CONFIG_X86_INTEL_MPX
580 [ilog2(VM_MPX)] = "mp",
582 [ilog2(VM_LOCKED)] = "lo",
583 [ilog2(VM_IO)] = "io",
584 [ilog2(VM_SEQ_READ)] = "sr",
585 [ilog2(VM_RAND_READ)] = "rr",
586 [ilog2(VM_DONTCOPY)] = "dc",
587 [ilog2(VM_DONTEXPAND)] = "de",
588 [ilog2(VM_ACCOUNT)] = "ac",
589 [ilog2(VM_NORESERVE)] = "nr",
590 [ilog2(VM_HUGETLB)] = "ht",
591 [ilog2(VM_ARCH_1)] = "ar",
592 [ilog2(VM_DONTDUMP)] = "dd",
593 #ifdef CONFIG_MEM_SOFT_DIRTY
594 [ilog2(VM_SOFTDIRTY)] = "sd",
596 [ilog2(VM_MIXEDMAP)] = "mm",
597 [ilog2(VM_HUGEPAGE)] = "hg",
598 [ilog2(VM_NOHUGEPAGE)] = "nh",
599 [ilog2(VM_MERGEABLE)] = "mg",
603 seq_puts(m, "VmFlags: ");
604 for (i = 0; i < BITS_PER_LONG; i++) {
605 if (vma->vm_flags & (1UL << i)) {
606 seq_printf(m, "%c%c ",
607 mnemonics[i][0], mnemonics[i][1]);
613 static int show_smap(struct seq_file *m, void *v, int is_pid)
615 struct vm_area_struct *vma = v;
616 struct mem_size_stats mss;
617 struct mm_walk smaps_walk = {
618 .pmd_entry = smaps_pte_range,
623 memset(&mss, 0, sizeof mss);
624 /* mmap_sem is held in m_start */
625 walk_page_vma(vma, &smaps_walk);
627 show_map_vma(m, vma, is_pid);
633 "Shared_Clean: %8lu kB\n"
634 "Shared_Dirty: %8lu kB\n"
635 "Private_Clean: %8lu kB\n"
636 "Private_Dirty: %8lu kB\n"
637 "Referenced: %8lu kB\n"
638 "Anonymous: %8lu kB\n"
639 "AnonHugePages: %8lu kB\n"
641 "KernelPageSize: %8lu kB\n"
642 "MMUPageSize: %8lu kB\n"
644 (vma->vm_end - vma->vm_start) >> 10,
646 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
647 mss.shared_clean >> 10,
648 mss.shared_dirty >> 10,
649 mss.private_clean >> 10,
650 mss.private_dirty >> 10,
651 mss.referenced >> 10,
653 mss.anonymous_thp >> 10,
655 vma_kernel_pagesize(vma) >> 10,
656 vma_mmu_pagesize(vma) >> 10,
657 (vma->vm_flags & VM_LOCKED) ?
658 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
660 show_smap_vma_flags(m, vma);
665 static int show_pid_smap(struct seq_file *m, void *v)
667 return show_smap(m, v, 1);
670 static int show_tid_smap(struct seq_file *m, void *v)
672 return show_smap(m, v, 0);
675 static const struct seq_operations proc_pid_smaps_op = {
679 .show = show_pid_smap
682 static const struct seq_operations proc_tid_smaps_op = {
686 .show = show_tid_smap
689 static int pid_smaps_open(struct inode *inode, struct file *file)
691 return do_maps_open(inode, file, &proc_pid_smaps_op);
694 static int tid_smaps_open(struct inode *inode, struct file *file)
696 return do_maps_open(inode, file, &proc_tid_smaps_op);
699 const struct file_operations proc_pid_smaps_operations = {
700 .open = pid_smaps_open,
703 .release = proc_map_release,
706 const struct file_operations proc_tid_smaps_operations = {
707 .open = tid_smaps_open,
710 .release = proc_map_release,
714 * We do not want to have constant page-shift bits sitting in
715 * pagemap entries and are about to reuse them some time soon.
717 * Here's the "migration strategy":
718 * 1. when the system boots these bits remain what they are,
719 * but a warning about future change is printed in log;
720 * 2. once anyone clears soft-dirty bits via clear_refs file,
721 * these flag is set to denote, that user is aware of the
722 * new API and those page-shift bits change their meaning.
723 * The respective warning is printed in dmesg;
724 * 3. In a couple of releases we will remove all the mentions
725 * of page-shift in pagemap entries.
728 static bool soft_dirty_cleared __read_mostly;
730 enum clear_refs_types {
734 CLEAR_REFS_SOFT_DIRTY,
735 CLEAR_REFS_MM_HIWATER_RSS,
739 struct clear_refs_private {
740 enum clear_refs_types type;
743 #ifdef CONFIG_MEM_SOFT_DIRTY
744 static inline void clear_soft_dirty(struct vm_area_struct *vma,
745 unsigned long addr, pte_t *pte)
748 * The soft-dirty tracker uses #PF-s to catch writes
749 * to pages, so write-protect the pte as well. See the
750 * Documentation/vm/soft-dirty.txt for full description
751 * of how soft-dirty works.
755 if (pte_present(ptent)) {
756 ptent = pte_wrprotect(ptent);
757 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY);
758 } else if (is_swap_pte(ptent)) {
759 ptent = pte_swp_clear_soft_dirty(ptent);
762 set_pte_at(vma->vm_mm, addr, pte, ptent);
765 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
766 unsigned long addr, pmd_t *pmdp)
770 pmd = pmd_wrprotect(pmd);
771 pmd = pmd_clear_flags(pmd, _PAGE_SOFT_DIRTY);
773 if (vma->vm_flags & VM_SOFTDIRTY)
774 vma->vm_flags &= ~VM_SOFTDIRTY;
776 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
781 static inline void clear_soft_dirty(struct vm_area_struct *vma,
782 unsigned long addr, pte_t *pte)
786 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
787 unsigned long addr, pmd_t *pmdp)
792 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
793 unsigned long end, struct mm_walk *walk)
795 struct clear_refs_private *cp = walk->private;
796 struct vm_area_struct *vma = walk->vma;
801 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
802 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
803 clear_soft_dirty_pmd(vma, addr, pmd);
807 page = pmd_page(*pmd);
809 /* Clear accessed and referenced bits. */
810 pmdp_test_and_clear_young(vma, addr, pmd);
811 ClearPageReferenced(page);
817 if (pmd_trans_unstable(pmd))
820 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
821 for (; addr != end; pte++, addr += PAGE_SIZE) {
824 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
825 clear_soft_dirty(vma, addr, pte);
829 if (!pte_present(ptent))
832 page = vm_normal_page(vma, addr, ptent);
836 /* Clear accessed and referenced bits. */
837 ptep_test_and_clear_young(vma, addr, pte);
838 ClearPageReferenced(page);
840 pte_unmap_unlock(pte - 1, ptl);
845 static int clear_refs_test_walk(unsigned long start, unsigned long end,
846 struct mm_walk *walk)
848 struct clear_refs_private *cp = walk->private;
849 struct vm_area_struct *vma = walk->vma;
851 if (vma->vm_flags & VM_PFNMAP)
855 * Writing 1 to /proc/pid/clear_refs affects all pages.
856 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
857 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
858 * Writing 4 to /proc/pid/clear_refs affects all pages.
860 if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
862 if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
867 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
868 size_t count, loff_t *ppos)
870 struct task_struct *task;
871 char buffer[PROC_NUMBUF];
872 struct mm_struct *mm;
873 struct vm_area_struct *vma;
874 enum clear_refs_types type;
878 memset(buffer, 0, sizeof(buffer));
879 if (count > sizeof(buffer) - 1)
880 count = sizeof(buffer) - 1;
881 if (copy_from_user(buffer, buf, count))
883 rv = kstrtoint(strstrip(buffer), 10, &itype);
886 type = (enum clear_refs_types)itype;
887 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
890 if (type == CLEAR_REFS_SOFT_DIRTY) {
891 soft_dirty_cleared = true;
892 pr_warn_once("The pagemap bits 55-60 has changed their meaning!"
893 " See the linux/Documentation/vm/pagemap.txt for "
897 task = get_proc_task(file_inode(file));
900 mm = get_task_mm(task);
902 struct clear_refs_private cp = {
905 struct mm_walk clear_refs_walk = {
906 .pmd_entry = clear_refs_pte_range,
907 .test_walk = clear_refs_test_walk,
912 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
914 * Writing 5 to /proc/pid/clear_refs resets the peak
915 * resident set size to this mm's current rss value.
917 down_write(&mm->mmap_sem);
918 reset_mm_hiwater_rss(mm);
919 up_write(&mm->mmap_sem);
923 down_read(&mm->mmap_sem);
924 if (type == CLEAR_REFS_SOFT_DIRTY) {
925 for (vma = mm->mmap; vma; vma = vma->vm_next) {
926 if (!(vma->vm_flags & VM_SOFTDIRTY))
928 up_read(&mm->mmap_sem);
929 down_write(&mm->mmap_sem);
930 for (vma = mm->mmap; vma; vma = vma->vm_next) {
931 vma->vm_flags &= ~VM_SOFTDIRTY;
932 vma_set_page_prot(vma);
934 downgrade_write(&mm->mmap_sem);
937 mmu_notifier_invalidate_range_start(mm, 0, -1);
939 walk_page_range(0, ~0UL, &clear_refs_walk);
940 if (type == CLEAR_REFS_SOFT_DIRTY)
941 mmu_notifier_invalidate_range_end(mm, 0, -1);
943 up_read(&mm->mmap_sem);
947 put_task_struct(task);
952 const struct file_operations proc_clear_refs_operations = {
953 .write = clear_refs_write,
954 .llseek = noop_llseek,
962 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
963 pagemap_entry_t *buffer;
967 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
968 #define PAGEMAP_WALK_MASK (PMD_MASK)
970 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
971 #define PM_STATUS_BITS 3
972 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
973 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
974 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
975 #define PM_PSHIFT_BITS 6
976 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
977 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
978 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
979 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
980 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
981 /* in "new" pagemap pshift bits are occupied with more status bits */
982 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT))
984 #define __PM_SOFT_DIRTY (1LL)
985 #define PM_PRESENT PM_STATUS(4LL)
986 #define PM_SWAP PM_STATUS(2LL)
987 #define PM_FILE PM_STATUS(1LL)
988 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0)
989 #define PM_END_OF_BUFFER 1
991 static inline pagemap_entry_t make_pme(u64 val)
993 return (pagemap_entry_t) { .pme = val };
996 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
997 struct pagemapread *pm)
999 pm->buffer[pm->pos++] = *pme;
1000 if (pm->pos >= pm->len)
1001 return PM_END_OF_BUFFER;
1005 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1006 struct mm_walk *walk)
1008 struct pagemapread *pm = walk->private;
1009 unsigned long addr = start;
1012 while (addr < end) {
1013 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1014 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
1015 /* End of address space hole, which we mark as non-present. */
1016 unsigned long hole_end;
1019 hole_end = min(end, vma->vm_start);
1023 for (; addr < hole_end; addr += PAGE_SIZE) {
1024 err = add_to_pagemap(addr, &pme, pm);
1032 /* Addresses in the VMA. */
1033 if (vma->vm_flags & VM_SOFTDIRTY)
1034 pme.pme |= PM_STATUS2(pm->v2, __PM_SOFT_DIRTY);
1035 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1036 err = add_to_pagemap(addr, &pme, pm);
1045 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1046 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1049 struct page *page = NULL;
1052 if (pte_present(pte)) {
1053 frame = pte_pfn(pte);
1055 page = vm_normal_page(vma, addr, pte);
1056 if (pte_soft_dirty(pte))
1057 flags2 |= __PM_SOFT_DIRTY;
1058 } else if (is_swap_pte(pte)) {
1060 if (pte_swp_soft_dirty(pte))
1061 flags2 |= __PM_SOFT_DIRTY;
1062 entry = pte_to_swp_entry(pte);
1063 frame = swp_type(entry) |
1064 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1066 if (is_migration_entry(entry))
1067 page = migration_entry_to_page(entry);
1069 if (vma->vm_flags & VM_SOFTDIRTY)
1070 flags2 |= __PM_SOFT_DIRTY;
1071 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
1075 if (page && !PageAnon(page))
1077 if ((vma->vm_flags & VM_SOFTDIRTY))
1078 flags2 |= __PM_SOFT_DIRTY;
1080 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags);
1083 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1084 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1085 pmd_t pmd, int offset, int pmd_flags2)
1088 * Currently pmd for thp is always present because thp can not be
1089 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
1090 * This if-check is just to prepare for future implementation.
1092 if (pmd_present(pmd))
1093 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
1094 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT);
1096 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2));
1099 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1100 pmd_t pmd, int offset, int pmd_flags2)
1105 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
1106 struct mm_walk *walk)
1108 struct vm_area_struct *vma = walk->vma;
1109 struct pagemapread *pm = walk->private;
1111 pte_t *pte, *orig_pte;
1114 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1117 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd))
1118 pmd_flags2 = __PM_SOFT_DIRTY;
1122 for (; addr != end; addr += PAGE_SIZE) {
1123 unsigned long offset;
1124 pagemap_entry_t pme;
1126 offset = (addr & ~PAGEMAP_WALK_MASK) >>
1128 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2);
1129 err = add_to_pagemap(addr, &pme, pm);
1137 if (pmd_trans_unstable(pmd))
1141 * We can assume that @vma always points to a valid one and @end never
1142 * goes beyond vma->vm_end.
1144 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1145 for (; addr < end; pte++, addr += PAGE_SIZE) {
1146 pagemap_entry_t pme;
1148 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte);
1149 err = add_to_pagemap(addr, &pme, pm);
1153 pte_unmap_unlock(orig_pte, ptl);
1160 #ifdef CONFIG_HUGETLB_PAGE
1161 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1162 pte_t pte, int offset, int flags2)
1164 if (pte_present(pte))
1165 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) |
1166 PM_STATUS2(pm->v2, flags2) |
1169 *pme = make_pme(PM_NOT_PRESENT(pm->v2) |
1170 PM_STATUS2(pm->v2, flags2));
1173 /* This function walks within one hugetlb entry in the single call */
1174 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1175 unsigned long addr, unsigned long end,
1176 struct mm_walk *walk)
1178 struct pagemapread *pm = walk->private;
1179 struct vm_area_struct *vma = walk->vma;
1182 pagemap_entry_t pme;
1184 if (vma->vm_flags & VM_SOFTDIRTY)
1185 flags2 = __PM_SOFT_DIRTY;
1189 for (; addr != end; addr += PAGE_SIZE) {
1190 int offset = (addr & ~hmask) >> PAGE_SHIFT;
1191 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2);
1192 err = add_to_pagemap(addr, &pme, pm);
1201 #endif /* HUGETLB_PAGE */
1204 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1206 * For each page in the address space, this file contains one 64-bit entry
1207 * consisting of the following:
1209 * Bits 0-54 page frame number (PFN) if present
1210 * Bits 0-4 swap type if swapped
1211 * Bits 5-54 swap offset if swapped
1212 * Bits 55-60 page shift (page size = 1<<page shift)
1213 * Bit 61 page is file-page or shared-anon
1214 * Bit 62 page swapped
1215 * Bit 63 page present
1217 * If the page is not present but in swap, then the PFN contains an
1218 * encoding of the swap file number and the page's offset into the
1219 * swap. Unmapped pages return a null PFN. This allows determining
1220 * precisely which pages are mapped (or in swap) and comparing mapped
1221 * pages between processes.
1223 * Efficient users of this interface will use /proc/pid/maps to
1224 * determine which areas of memory are actually mapped and llseek to
1225 * skip over unmapped regions.
1227 static ssize_t pagemap_read(struct file *file, char __user *buf,
1228 size_t count, loff_t *ppos)
1230 struct task_struct *task = get_proc_task(file_inode(file));
1231 struct mm_struct *mm;
1232 struct pagemapread pm;
1234 struct mm_walk pagemap_walk = {};
1236 unsigned long svpfn;
1237 unsigned long start_vaddr;
1238 unsigned long end_vaddr;
1245 /* file position must be aligned */
1246 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1253 pm.v2 = soft_dirty_cleared;
1254 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1255 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1260 mm = mm_access(task, PTRACE_MODE_READ);
1262 if (!mm || IS_ERR(mm))
1265 pagemap_walk.pmd_entry = pagemap_pte_range;
1266 pagemap_walk.pte_hole = pagemap_pte_hole;
1267 #ifdef CONFIG_HUGETLB_PAGE
1268 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1270 pagemap_walk.mm = mm;
1271 pagemap_walk.private = ±
1274 svpfn = src / PM_ENTRY_BYTES;
1275 start_vaddr = svpfn << PAGE_SHIFT;
1276 end_vaddr = TASK_SIZE_OF(task);
1278 /* watch out for wraparound */
1279 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1280 start_vaddr = end_vaddr;
1283 * The odds are that this will stop walking way
1284 * before end_vaddr, because the length of the
1285 * user buffer is tracked in "pm", and the walk
1286 * will stop when we hit the end of the buffer.
1289 while (count && (start_vaddr < end_vaddr)) {
1294 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1296 if (end < start_vaddr || end > end_vaddr)
1298 down_read(&mm->mmap_sem);
1299 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1300 up_read(&mm->mmap_sem);
1303 len = min(count, PM_ENTRY_BYTES * pm.pos);
1304 if (copy_to_user(buf, pm.buffer, len)) {
1313 if (!ret || ret == PM_END_OF_BUFFER)
1321 put_task_struct(task);
1326 static int pagemap_open(struct inode *inode, struct file *file)
1328 /* do not disclose physical addresses: attack vector */
1329 if (!capable(CAP_SYS_ADMIN))
1331 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about "
1332 "to stop being page-shift some time soon. See the "
1333 "linux/Documentation/vm/pagemap.txt for details.\n");
1337 const struct file_operations proc_pagemap_operations = {
1338 .llseek = mem_lseek, /* borrow this */
1339 .read = pagemap_read,
1340 .open = pagemap_open,
1342 #endif /* CONFIG_PROC_PAGE_MONITOR */
1347 unsigned long pages;
1349 unsigned long active;
1350 unsigned long writeback;
1351 unsigned long mapcount_max;
1352 unsigned long dirty;
1353 unsigned long swapcache;
1354 unsigned long node[MAX_NUMNODES];
1357 struct numa_maps_private {
1358 struct proc_maps_private proc_maps;
1359 struct numa_maps md;
1362 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1363 unsigned long nr_pages)
1365 int count = page_mapcount(page);
1367 md->pages += nr_pages;
1368 if (pte_dirty || PageDirty(page))
1369 md->dirty += nr_pages;
1371 if (PageSwapCache(page))
1372 md->swapcache += nr_pages;
1374 if (PageActive(page) || PageUnevictable(page))
1375 md->active += nr_pages;
1377 if (PageWriteback(page))
1378 md->writeback += nr_pages;
1381 md->anon += nr_pages;
1383 if (count > md->mapcount_max)
1384 md->mapcount_max = count;
1386 md->node[page_to_nid(page)] += nr_pages;
1389 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1395 if (!pte_present(pte))
1398 page = vm_normal_page(vma, addr, pte);
1402 if (PageReserved(page))
1405 nid = page_to_nid(page);
1406 if (!node_isset(nid, node_states[N_MEMORY]))
1412 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1413 unsigned long end, struct mm_walk *walk)
1415 struct numa_maps *md = walk->private;
1416 struct vm_area_struct *vma = walk->vma;
1421 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1422 pte_t huge_pte = *(pte_t *)pmd;
1425 page = can_gather_numa_stats(huge_pte, vma, addr);
1427 gather_stats(page, md, pte_dirty(huge_pte),
1428 HPAGE_PMD_SIZE/PAGE_SIZE);
1433 if (pmd_trans_unstable(pmd))
1435 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1437 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1440 gather_stats(page, md, pte_dirty(*pte), 1);
1442 } while (pte++, addr += PAGE_SIZE, addr != end);
1443 pte_unmap_unlock(orig_pte, ptl);
1446 #ifdef CONFIG_HUGETLB_PAGE
1447 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1448 unsigned long addr, unsigned long end, struct mm_walk *walk)
1450 struct numa_maps *md;
1453 if (!pte_present(*pte))
1456 page = pte_page(*pte);
1461 gather_stats(page, md, pte_dirty(*pte), 1);
1466 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1467 unsigned long addr, unsigned long end, struct mm_walk *walk)
1474 * Display pages allocated per node and memory policy via /proc.
1476 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1478 struct numa_maps_private *numa_priv = m->private;
1479 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1480 struct vm_area_struct *vma = v;
1481 struct numa_maps *md = &numa_priv->md;
1482 struct file *file = vma->vm_file;
1483 struct mm_struct *mm = vma->vm_mm;
1484 struct mm_walk walk = {
1485 .hugetlb_entry = gather_hugetlb_stats,
1486 .pmd_entry = gather_pte_stats,
1490 struct mempolicy *pol;
1497 /* Ensure we start with an empty set of numa_maps statistics. */
1498 memset(md, 0, sizeof(*md));
1500 pol = __get_vma_policy(vma, vma->vm_start);
1502 mpol_to_str(buffer, sizeof(buffer), pol);
1505 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1508 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1511 seq_puts(m, " file=");
1512 seq_path(m, &file->f_path, "\n\t= ");
1513 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1514 seq_puts(m, " heap");
1516 pid_t tid = pid_of_stack(proc_priv, vma, is_pid);
1519 * Thread stack in /proc/PID/task/TID/maps or
1520 * the main process stack.
1522 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1523 vma->vm_end >= mm->start_stack))
1524 seq_puts(m, " stack");
1526 seq_printf(m, " stack:%d", tid);
1530 if (is_vm_hugetlb_page(vma))
1531 seq_puts(m, " huge");
1533 /* mmap_sem is held by m_start */
1534 walk_page_vma(vma, &walk);
1540 seq_printf(m, " anon=%lu", md->anon);
1543 seq_printf(m, " dirty=%lu", md->dirty);
1545 if (md->pages != md->anon && md->pages != md->dirty)
1546 seq_printf(m, " mapped=%lu", md->pages);
1548 if (md->mapcount_max > 1)
1549 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1552 seq_printf(m, " swapcache=%lu", md->swapcache);
1554 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1555 seq_printf(m, " active=%lu", md->active);
1558 seq_printf(m, " writeback=%lu", md->writeback);
1560 for_each_node_state(nid, N_MEMORY)
1562 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1564 seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1567 m_cache_vma(m, vma);
1571 static int show_pid_numa_map(struct seq_file *m, void *v)
1573 return show_numa_map(m, v, 1);
1576 static int show_tid_numa_map(struct seq_file *m, void *v)
1578 return show_numa_map(m, v, 0);
1581 static const struct seq_operations proc_pid_numa_maps_op = {
1585 .show = show_pid_numa_map,
1588 static const struct seq_operations proc_tid_numa_maps_op = {
1592 .show = show_tid_numa_map,
1595 static int numa_maps_open(struct inode *inode, struct file *file,
1596 const struct seq_operations *ops)
1598 return proc_maps_open(inode, file, ops,
1599 sizeof(struct numa_maps_private));
1602 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1604 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1607 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1609 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1612 const struct file_operations proc_pid_numa_maps_operations = {
1613 .open = pid_numa_maps_open,
1615 .llseek = seq_lseek,
1616 .release = proc_map_release,
1619 const struct file_operations proc_tid_numa_maps_operations = {
1620 .open = tid_numa_maps_open,
1622 .llseek = seq_lseek,
1623 .release = proc_map_release,
1625 #endif /* CONFIG_NUMA */