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 [ilog2(VM_LOCKED)] = "lo",
556 [ilog2(VM_IO)] = "io",
557 [ilog2(VM_SEQ_READ)] = "sr",
558 [ilog2(VM_RAND_READ)] = "rr",
559 [ilog2(VM_DONTCOPY)] = "dc",
560 [ilog2(VM_DONTEXPAND)] = "de",
561 [ilog2(VM_ACCOUNT)] = "ac",
562 [ilog2(VM_NORESERVE)] = "nr",
563 [ilog2(VM_HUGETLB)] = "ht",
564 [ilog2(VM_NONLINEAR)] = "nl",
565 [ilog2(VM_ARCH_1)] = "ar",
566 [ilog2(VM_DONTDUMP)] = "dd",
567 #ifdef CONFIG_MEM_SOFT_DIRTY
568 [ilog2(VM_SOFTDIRTY)] = "sd",
570 [ilog2(VM_MIXEDMAP)] = "mm",
571 [ilog2(VM_HUGEPAGE)] = "hg",
572 [ilog2(VM_NOHUGEPAGE)] = "nh",
573 [ilog2(VM_MERGEABLE)] = "mg",
577 seq_puts(m, "VmFlags: ");
578 for (i = 0; i < BITS_PER_LONG; i++) {
579 if (vma->vm_flags & (1UL << i)) {
580 seq_printf(m, "%c%c ",
581 mnemonics[i][0], mnemonics[i][1]);
587 static int show_smap(struct seq_file *m, void *v, int is_pid)
589 struct vm_area_struct *vma = v;
590 struct mem_size_stats mss;
591 struct mm_walk smaps_walk = {
592 .pmd_entry = smaps_pte_range,
597 memset(&mss, 0, sizeof mss);
599 /* mmap_sem is held in m_start */
600 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
601 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
603 show_map_vma(m, vma, is_pid);
609 "Shared_Clean: %8lu kB\n"
610 "Shared_Dirty: %8lu kB\n"
611 "Private_Clean: %8lu kB\n"
612 "Private_Dirty: %8lu kB\n"
613 "Referenced: %8lu kB\n"
614 "Anonymous: %8lu kB\n"
615 "AnonHugePages: %8lu kB\n"
617 "KernelPageSize: %8lu kB\n"
618 "MMUPageSize: %8lu kB\n"
620 (vma->vm_end - vma->vm_start) >> 10,
622 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
623 mss.shared_clean >> 10,
624 mss.shared_dirty >> 10,
625 mss.private_clean >> 10,
626 mss.private_dirty >> 10,
627 mss.referenced >> 10,
629 mss.anonymous_thp >> 10,
631 vma_kernel_pagesize(vma) >> 10,
632 vma_mmu_pagesize(vma) >> 10,
633 (vma->vm_flags & VM_LOCKED) ?
634 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
636 if (vma->vm_flags & VM_NONLINEAR)
637 seq_printf(m, "Nonlinear: %8lu kB\n",
638 mss.nonlinear >> 10);
640 show_smap_vma_flags(m, vma);
645 static int show_pid_smap(struct seq_file *m, void *v)
647 return show_smap(m, v, 1);
650 static int show_tid_smap(struct seq_file *m, void *v)
652 return show_smap(m, v, 0);
655 static const struct seq_operations proc_pid_smaps_op = {
659 .show = show_pid_smap
662 static const struct seq_operations proc_tid_smaps_op = {
666 .show = show_tid_smap
669 static int pid_smaps_open(struct inode *inode, struct file *file)
671 return do_maps_open(inode, file, &proc_pid_smaps_op);
674 static int tid_smaps_open(struct inode *inode, struct file *file)
676 return do_maps_open(inode, file, &proc_tid_smaps_op);
679 const struct file_operations proc_pid_smaps_operations = {
680 .open = pid_smaps_open,
683 .release = proc_map_release,
686 const struct file_operations proc_tid_smaps_operations = {
687 .open = tid_smaps_open,
690 .release = proc_map_release,
694 * We do not want to have constant page-shift bits sitting in
695 * pagemap entries and are about to reuse them some time soon.
697 * Here's the "migration strategy":
698 * 1. when the system boots these bits remain what they are,
699 * but a warning about future change is printed in log;
700 * 2. once anyone clears soft-dirty bits via clear_refs file,
701 * these flag is set to denote, that user is aware of the
702 * new API and those page-shift bits change their meaning.
703 * The respective warning is printed in dmesg;
704 * 3. In a couple of releases we will remove all the mentions
705 * of page-shift in pagemap entries.
708 static bool soft_dirty_cleared __read_mostly;
710 enum clear_refs_types {
714 CLEAR_REFS_SOFT_DIRTY,
718 struct clear_refs_private {
719 struct vm_area_struct *vma;
720 enum clear_refs_types type;
723 static inline void clear_soft_dirty(struct vm_area_struct *vma,
724 unsigned long addr, pte_t *pte)
726 #ifdef CONFIG_MEM_SOFT_DIRTY
728 * The soft-dirty tracker uses #PF-s to catch writes
729 * to pages, so write-protect the pte as well. See the
730 * Documentation/vm/soft-dirty.txt for full description
731 * of how soft-dirty works.
735 if (pte_present(ptent)) {
736 ptent = pte_wrprotect(ptent);
737 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY);
738 } else if (is_swap_pte(ptent)) {
739 ptent = pte_swp_clear_soft_dirty(ptent);
740 } else if (pte_file(ptent)) {
741 ptent = pte_file_clear_soft_dirty(ptent);
744 set_pte_at(vma->vm_mm, addr, pte, ptent);
748 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
749 unsigned long end, struct mm_walk *walk)
751 struct clear_refs_private *cp = walk->private;
752 struct vm_area_struct *vma = cp->vma;
757 split_huge_page_pmd(vma, addr, pmd);
758 if (pmd_trans_unstable(pmd))
761 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
762 for (; addr != end; pte++, addr += PAGE_SIZE) {
765 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
766 clear_soft_dirty(vma, addr, pte);
770 if (!pte_present(ptent))
773 page = vm_normal_page(vma, addr, ptent);
777 /* Clear accessed and referenced bits. */
778 ptep_test_and_clear_young(vma, addr, pte);
779 ClearPageReferenced(page);
781 pte_unmap_unlock(pte - 1, ptl);
786 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
787 size_t count, loff_t *ppos)
789 struct task_struct *task;
790 char buffer[PROC_NUMBUF];
791 struct mm_struct *mm;
792 struct vm_area_struct *vma;
793 enum clear_refs_types type;
797 memset(buffer, 0, sizeof(buffer));
798 if (count > sizeof(buffer) - 1)
799 count = sizeof(buffer) - 1;
800 if (copy_from_user(buffer, buf, count))
802 rv = kstrtoint(strstrip(buffer), 10, &itype);
805 type = (enum clear_refs_types)itype;
806 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
809 if (type == CLEAR_REFS_SOFT_DIRTY) {
810 soft_dirty_cleared = true;
811 pr_warn_once("The pagemap bits 55-60 has changed their meaning!"
812 " See the linux/Documentation/vm/pagemap.txt for "
816 task = get_proc_task(file_inode(file));
819 mm = get_task_mm(task);
821 struct clear_refs_private cp = {
824 struct mm_walk clear_refs_walk = {
825 .pmd_entry = clear_refs_pte_range,
829 down_read(&mm->mmap_sem);
830 if (type == CLEAR_REFS_SOFT_DIRTY) {
831 for (vma = mm->mmap; vma; vma = vma->vm_next) {
832 if (!(vma->vm_flags & VM_SOFTDIRTY))
834 up_read(&mm->mmap_sem);
835 down_write(&mm->mmap_sem);
836 for (vma = mm->mmap; vma; vma = vma->vm_next) {
837 vma->vm_flags &= ~VM_SOFTDIRTY;
838 vma_set_page_prot(vma);
840 downgrade_write(&mm->mmap_sem);
843 mmu_notifier_invalidate_range_start(mm, 0, -1);
845 for (vma = mm->mmap; vma; vma = vma->vm_next) {
847 if (is_vm_hugetlb_page(vma))
850 * Writing 1 to /proc/pid/clear_refs affects all pages.
852 * Writing 2 to /proc/pid/clear_refs only affects
855 * Writing 3 to /proc/pid/clear_refs only affects file
858 * Writing 4 to /proc/pid/clear_refs affects all pages.
860 if (type == CLEAR_REFS_ANON && vma->vm_file)
862 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
864 walk_page_range(vma->vm_start, vma->vm_end,
867 if (type == CLEAR_REFS_SOFT_DIRTY)
868 mmu_notifier_invalidate_range_end(mm, 0, -1);
870 up_read(&mm->mmap_sem);
873 put_task_struct(task);
878 const struct file_operations proc_clear_refs_operations = {
879 .write = clear_refs_write,
880 .llseek = noop_llseek,
888 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
889 pagemap_entry_t *buffer;
893 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
894 #define PAGEMAP_WALK_MASK (PMD_MASK)
896 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
897 #define PM_STATUS_BITS 3
898 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
899 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
900 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
901 #define PM_PSHIFT_BITS 6
902 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
903 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
904 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
905 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
906 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
907 /* in "new" pagemap pshift bits are occupied with more status bits */
908 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT))
910 #define __PM_SOFT_DIRTY (1LL)
911 #define PM_PRESENT PM_STATUS(4LL)
912 #define PM_SWAP PM_STATUS(2LL)
913 #define PM_FILE PM_STATUS(1LL)
914 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0)
915 #define PM_END_OF_BUFFER 1
917 static inline pagemap_entry_t make_pme(u64 val)
919 return (pagemap_entry_t) { .pme = val };
922 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
923 struct pagemapread *pm)
925 pm->buffer[pm->pos++] = *pme;
926 if (pm->pos >= pm->len)
927 return PM_END_OF_BUFFER;
931 static int pagemap_pte_hole(unsigned long start, unsigned long end,
932 struct mm_walk *walk)
934 struct pagemapread *pm = walk->private;
935 unsigned long addr = start;
939 struct vm_area_struct *vma = find_vma(walk->mm, addr);
940 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
941 /* End of address space hole, which we mark as non-present. */
942 unsigned long hole_end;
945 hole_end = min(end, vma->vm_start);
949 for (; addr < hole_end; addr += PAGE_SIZE) {
950 err = add_to_pagemap(addr, &pme, pm);
958 /* Addresses in the VMA. */
959 if (vma->vm_flags & VM_SOFTDIRTY)
960 pme.pme |= PM_STATUS2(pm->v2, __PM_SOFT_DIRTY);
961 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
962 err = add_to_pagemap(addr, &pme, pm);
971 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
972 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
975 struct page *page = NULL;
978 if (pte_present(pte)) {
979 frame = pte_pfn(pte);
981 page = vm_normal_page(vma, addr, pte);
982 if (pte_soft_dirty(pte))
983 flags2 |= __PM_SOFT_DIRTY;
984 } else if (is_swap_pte(pte)) {
986 if (pte_swp_soft_dirty(pte))
987 flags2 |= __PM_SOFT_DIRTY;
988 entry = pte_to_swp_entry(pte);
989 frame = swp_type(entry) |
990 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
992 if (is_migration_entry(entry))
993 page = migration_entry_to_page(entry);
995 if (vma->vm_flags & VM_SOFTDIRTY)
996 flags2 |= __PM_SOFT_DIRTY;
997 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
1001 if (page && !PageAnon(page))
1003 if ((vma->vm_flags & VM_SOFTDIRTY))
1004 flags2 |= __PM_SOFT_DIRTY;
1006 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags);
1009 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1010 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1011 pmd_t pmd, int offset, int pmd_flags2)
1014 * Currently pmd for thp is always present because thp can not be
1015 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
1016 * This if-check is just to prepare for future implementation.
1018 if (pmd_present(pmd))
1019 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
1020 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT);
1022 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2));
1025 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1026 pmd_t pmd, int offset, int pmd_flags2)
1031 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
1032 struct mm_walk *walk)
1034 struct vm_area_struct *vma;
1035 struct pagemapread *pm = walk->private;
1040 /* find the first VMA at or above 'addr' */
1041 vma = find_vma(walk->mm, addr);
1042 if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1045 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd))
1046 pmd_flags2 = __PM_SOFT_DIRTY;
1050 for (; addr != end; addr += PAGE_SIZE) {
1051 unsigned long offset;
1052 pagemap_entry_t pme;
1054 offset = (addr & ~PAGEMAP_WALK_MASK) >>
1056 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2);
1057 err = add_to_pagemap(addr, &pme, pm);
1065 if (pmd_trans_unstable(pmd))
1069 /* End of address space hole, which we mark as non-present. */
1070 unsigned long hole_end;
1073 hole_end = min(end, vma->vm_start);
1077 for (; addr < hole_end; addr += PAGE_SIZE) {
1078 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
1080 err = add_to_pagemap(addr, &pme, pm);
1085 if (!vma || vma->vm_start >= end)
1088 * We can't possibly be in a hugetlb VMA. In general,
1089 * for a mm_walk with a pmd_entry and a hugetlb_entry,
1090 * the pmd_entry can only be called on addresses in a
1091 * hugetlb if the walk starts in a non-hugetlb VMA and
1092 * spans a hugepage VMA. Since pagemap_read walks are
1093 * PMD-sized and PMD-aligned, this will never be true.
1095 BUG_ON(is_vm_hugetlb_page(vma));
1097 /* Addresses in the VMA. */
1098 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1099 pagemap_entry_t pme;
1100 pte = pte_offset_map(pmd, addr);
1101 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte);
1103 err = add_to_pagemap(addr, &pme, pm);
1111 vma = find_vma(walk->mm, addr);
1119 #ifdef CONFIG_HUGETLB_PAGE
1120 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1121 pte_t pte, int offset, int flags2)
1123 if (pte_present(pte))
1124 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) |
1125 PM_STATUS2(pm->v2, flags2) |
1128 *pme = make_pme(PM_NOT_PRESENT(pm->v2) |
1129 PM_STATUS2(pm->v2, flags2));
1132 /* This function walks within one hugetlb entry in the single call */
1133 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1134 unsigned long addr, unsigned long end,
1135 struct mm_walk *walk)
1137 struct pagemapread *pm = walk->private;
1138 struct vm_area_struct *vma;
1141 pagemap_entry_t pme;
1143 vma = find_vma(walk->mm, addr);
1146 if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1147 flags2 = __PM_SOFT_DIRTY;
1151 for (; addr != end; addr += PAGE_SIZE) {
1152 int offset = (addr & ~hmask) >> PAGE_SHIFT;
1153 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2);
1154 err = add_to_pagemap(addr, &pme, pm);
1163 #endif /* HUGETLB_PAGE */
1166 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1168 * For each page in the address space, this file contains one 64-bit entry
1169 * consisting of the following:
1171 * Bits 0-54 page frame number (PFN) if present
1172 * Bits 0-4 swap type if swapped
1173 * Bits 5-54 swap offset if swapped
1174 * Bits 55-60 page shift (page size = 1<<page shift)
1175 * Bit 61 page is file-page or shared-anon
1176 * Bit 62 page swapped
1177 * Bit 63 page present
1179 * If the page is not present but in swap, then the PFN contains an
1180 * encoding of the swap file number and the page's offset into the
1181 * swap. Unmapped pages return a null PFN. This allows determining
1182 * precisely which pages are mapped (or in swap) and comparing mapped
1183 * pages between processes.
1185 * Efficient users of this interface will use /proc/pid/maps to
1186 * determine which areas of memory are actually mapped and llseek to
1187 * skip over unmapped regions.
1189 static ssize_t pagemap_read(struct file *file, char __user *buf,
1190 size_t count, loff_t *ppos)
1192 struct task_struct *task = get_proc_task(file_inode(file));
1193 struct mm_struct *mm;
1194 struct pagemapread pm;
1196 struct mm_walk pagemap_walk = {};
1198 unsigned long svpfn;
1199 unsigned long start_vaddr;
1200 unsigned long end_vaddr;
1207 /* file position must be aligned */
1208 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1215 pm.v2 = soft_dirty_cleared;
1216 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1217 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1222 mm = mm_access(task, PTRACE_MODE_READ);
1224 if (!mm || IS_ERR(mm))
1227 pagemap_walk.pmd_entry = pagemap_pte_range;
1228 pagemap_walk.pte_hole = pagemap_pte_hole;
1229 #ifdef CONFIG_HUGETLB_PAGE
1230 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1232 pagemap_walk.mm = mm;
1233 pagemap_walk.private = ±
1236 svpfn = src / PM_ENTRY_BYTES;
1237 start_vaddr = svpfn << PAGE_SHIFT;
1238 end_vaddr = TASK_SIZE_OF(task);
1240 /* watch out for wraparound */
1241 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1242 start_vaddr = end_vaddr;
1245 * The odds are that this will stop walking way
1246 * before end_vaddr, because the length of the
1247 * user buffer is tracked in "pm", and the walk
1248 * will stop when we hit the end of the buffer.
1251 while (count && (start_vaddr < end_vaddr)) {
1256 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1258 if (end < start_vaddr || end > end_vaddr)
1260 down_read(&mm->mmap_sem);
1261 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1262 up_read(&mm->mmap_sem);
1265 len = min(count, PM_ENTRY_BYTES * pm.pos);
1266 if (copy_to_user(buf, pm.buffer, len)) {
1275 if (!ret || ret == PM_END_OF_BUFFER)
1283 put_task_struct(task);
1288 static int pagemap_open(struct inode *inode, struct file *file)
1290 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about "
1291 "to stop being page-shift some time soon. See the "
1292 "linux/Documentation/vm/pagemap.txt for details.\n");
1296 const struct file_operations proc_pagemap_operations = {
1297 .llseek = mem_lseek, /* borrow this */
1298 .read = pagemap_read,
1299 .open = pagemap_open,
1301 #endif /* CONFIG_PROC_PAGE_MONITOR */
1306 struct vm_area_struct *vma;
1307 unsigned long pages;
1309 unsigned long active;
1310 unsigned long writeback;
1311 unsigned long mapcount_max;
1312 unsigned long dirty;
1313 unsigned long swapcache;
1314 unsigned long node[MAX_NUMNODES];
1317 struct numa_maps_private {
1318 struct proc_maps_private proc_maps;
1319 struct numa_maps md;
1322 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1323 unsigned long nr_pages)
1325 int count = page_mapcount(page);
1327 md->pages += nr_pages;
1328 if (pte_dirty || PageDirty(page))
1329 md->dirty += nr_pages;
1331 if (PageSwapCache(page))
1332 md->swapcache += nr_pages;
1334 if (PageActive(page) || PageUnevictable(page))
1335 md->active += nr_pages;
1337 if (PageWriteback(page))
1338 md->writeback += nr_pages;
1341 md->anon += nr_pages;
1343 if (count > md->mapcount_max)
1344 md->mapcount_max = count;
1346 md->node[page_to_nid(page)] += nr_pages;
1349 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1355 if (!pte_present(pte))
1358 page = vm_normal_page(vma, addr, pte);
1362 if (PageReserved(page))
1365 nid = page_to_nid(page);
1366 if (!node_isset(nid, node_states[N_MEMORY]))
1372 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1373 unsigned long end, struct mm_walk *walk)
1375 struct numa_maps *md;
1382 if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) {
1383 pte_t huge_pte = *(pte_t *)pmd;
1386 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1388 gather_stats(page, md, pte_dirty(huge_pte),
1389 HPAGE_PMD_SIZE/PAGE_SIZE);
1394 if (pmd_trans_unstable(pmd))
1396 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1398 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1401 gather_stats(page, md, pte_dirty(*pte), 1);
1403 } while (pte++, addr += PAGE_SIZE, addr != end);
1404 pte_unmap_unlock(orig_pte, ptl);
1407 #ifdef CONFIG_HUGETLB_PAGE
1408 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1409 unsigned long addr, unsigned long end, struct mm_walk *walk)
1411 struct numa_maps *md;
1414 if (!pte_present(*pte))
1417 page = pte_page(*pte);
1422 gather_stats(page, md, pte_dirty(*pte), 1);
1427 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1428 unsigned long addr, unsigned long end, struct mm_walk *walk)
1435 * Display pages allocated per node and memory policy via /proc.
1437 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1439 struct numa_maps_private *numa_priv = m->private;
1440 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1441 struct vm_area_struct *vma = v;
1442 struct numa_maps *md = &numa_priv->md;
1443 struct file *file = vma->vm_file;
1444 struct mm_struct *mm = vma->vm_mm;
1445 struct mm_walk walk = {};
1446 struct mempolicy *pol;
1453 /* Ensure we start with an empty set of numa_maps statistics. */
1454 memset(md, 0, sizeof(*md));
1458 walk.hugetlb_entry = gather_hugetbl_stats;
1459 walk.pmd_entry = gather_pte_stats;
1463 pol = __get_vma_policy(vma, vma->vm_start);
1465 mpol_to_str(buffer, sizeof(buffer), pol);
1468 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1471 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1474 seq_puts(m, " file=");
1475 seq_path(m, &file->f_path, "\n\t= ");
1476 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1477 seq_puts(m, " heap");
1479 pid_t tid = pid_of_stack(proc_priv, vma, is_pid);
1482 * Thread stack in /proc/PID/task/TID/maps or
1483 * the main process stack.
1485 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1486 vma->vm_end >= mm->start_stack))
1487 seq_puts(m, " stack");
1489 seq_printf(m, " stack:%d", tid);
1493 if (is_vm_hugetlb_page(vma))
1494 seq_puts(m, " huge");
1496 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1502 seq_printf(m, " anon=%lu", md->anon);
1505 seq_printf(m, " dirty=%lu", md->dirty);
1507 if (md->pages != md->anon && md->pages != md->dirty)
1508 seq_printf(m, " mapped=%lu", md->pages);
1510 if (md->mapcount_max > 1)
1511 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1514 seq_printf(m, " swapcache=%lu", md->swapcache);
1516 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1517 seq_printf(m, " active=%lu", md->active);
1520 seq_printf(m, " writeback=%lu", md->writeback);
1522 for_each_node_state(nid, N_MEMORY)
1524 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1527 m_cache_vma(m, vma);
1531 static int show_pid_numa_map(struct seq_file *m, void *v)
1533 return show_numa_map(m, v, 1);
1536 static int show_tid_numa_map(struct seq_file *m, void *v)
1538 return show_numa_map(m, v, 0);
1541 static const struct seq_operations proc_pid_numa_maps_op = {
1545 .show = show_pid_numa_map,
1548 static const struct seq_operations proc_tid_numa_maps_op = {
1552 .show = show_tid_numa_map,
1555 static int numa_maps_open(struct inode *inode, struct file *file,
1556 const struct seq_operations *ops)
1558 return proc_maps_open(inode, file, ops,
1559 sizeof(struct numa_maps_private));
1562 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1564 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1567 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1569 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1572 const struct file_operations proc_pid_numa_maps_operations = {
1573 .open = pid_numa_maps_open,
1575 .llseek = seq_lseek,
1576 .release = proc_map_release,
1579 const struct file_operations proc_tid_numa_maps_operations = {
1580 .open = tid_numa_maps_open,
1582 .llseek = seq_lseek,
1583 .release = proc_map_release,
1585 #endif /* CONFIG_NUMA */