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 * These functions are for numa_maps but called in generic **maps seq_file
91 * ->start(), ->stop() ops.
93 * numa_maps scans all vmas under mmap_sem and checks their mempolicy.
94 * Each mempolicy object is controlled by reference counting. The problem here
95 * is how to avoid accessing dead mempolicy object.
97 * Because we're holding mmap_sem while reading seq_file, it's safe to access
98 * each vma's mempolicy, no vma objects will never drop refs to mempolicy.
100 * A task's mempolicy (task->mempolicy) has different behavior. task->mempolicy
101 * is set and replaced under mmap_sem but unrefed and cleared under task_lock().
102 * So, without task_lock(), we cannot trust get_vma_policy() because we cannot
103 * gurantee the task never exits under us. But taking task_lock() around
104 * get_vma_plicy() causes lock order problem.
106 * To access task->mempolicy without lock, we hold a reference count of an
107 * object pointed by task->mempolicy and remember it. This will guarantee
108 * that task->mempolicy points to an alive object or NULL in numa_maps accesses.
110 static void hold_task_mempolicy(struct proc_maps_private *priv)
112 struct task_struct *task = priv->task;
115 priv->task_mempolicy = task->mempolicy;
116 mpol_get(priv->task_mempolicy);
119 static void release_task_mempolicy(struct proc_maps_private *priv)
121 mpol_put(priv->task_mempolicy);
124 static void hold_task_mempolicy(struct proc_maps_private *priv)
127 static void release_task_mempolicy(struct proc_maps_private *priv)
132 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
134 if (vma && vma != priv->tail_vma) {
135 struct mm_struct *mm = vma->vm_mm;
136 release_task_mempolicy(priv);
137 up_read(&mm->mmap_sem);
142 static void *m_start(struct seq_file *m, loff_t *pos)
144 struct proc_maps_private *priv = m->private;
145 unsigned long last_addr = m->version;
146 struct mm_struct *mm;
147 struct vm_area_struct *vma, *tail_vma = NULL;
150 /* Clear the per syscall fields in priv */
152 priv->tail_vma = NULL;
155 * We remember last_addr rather than next_addr to hit with
156 * vmacache most of the time. We have zero last_addr at
157 * the beginning and also after lseek. We will have -1 last_addr
158 * after the end of the vmas.
161 if (last_addr == -1UL)
164 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
166 return ERR_PTR(-ESRCH);
169 if (!mm || !atomic_inc_not_zero(&mm->mm_users))
171 down_read(&mm->mmap_sem);
173 tail_vma = get_gate_vma(mm);
174 priv->tail_vma = tail_vma;
175 hold_task_mempolicy(priv);
176 /* Start with last addr hint */
177 vma = find_vma(mm, last_addr);
178 if (last_addr && vma) {
184 * Check the vma index is within the range and do
185 * sequential scan until m_index.
188 if ((unsigned long)l < mm->map_count) {
195 if (l != mm->map_count)
196 tail_vma = NULL; /* After gate vma */
202 release_task_mempolicy(priv);
203 /* End of vmas has been reached */
204 m->version = (tail_vma != NULL)? 0: -1UL;
205 up_read(&mm->mmap_sem);
210 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
212 struct proc_maps_private *priv = m->private;
213 struct vm_area_struct *vma = v;
214 struct vm_area_struct *tail_vma = priv->tail_vma;
217 if (vma && (vma != tail_vma) && vma->vm_next)
220 return (vma != tail_vma)? tail_vma: NULL;
223 static void m_stop(struct seq_file *m, void *v)
225 struct proc_maps_private *priv = m->private;
226 struct vm_area_struct *vma = v;
231 put_task_struct(priv->task);
234 static int proc_maps_open(struct inode *inode, struct file *file,
235 const struct seq_operations *ops, int psize)
237 struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
242 priv->pid = proc_pid(inode);
243 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
244 if (IS_ERR(priv->mm)) {
245 int err = PTR_ERR(priv->mm);
247 seq_release_private(inode, file);
254 static int proc_map_release(struct inode *inode, struct file *file)
256 struct seq_file *seq = file->private_data;
257 struct proc_maps_private *priv = seq->private;
262 return seq_release_private(inode, file);
265 static int do_maps_open(struct inode *inode, struct file *file,
266 const struct seq_operations *ops)
268 return proc_maps_open(inode, file, ops,
269 sizeof(struct proc_maps_private));
273 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
275 struct mm_struct *mm = vma->vm_mm;
276 struct file *file = vma->vm_file;
277 struct proc_maps_private *priv = m->private;
278 struct task_struct *task = priv->task;
279 vm_flags_t flags = vma->vm_flags;
280 unsigned long ino = 0;
281 unsigned long long pgoff = 0;
282 unsigned long start, end;
284 const char *name = NULL;
287 struct inode *inode = file_inode(vma->vm_file);
288 dev = inode->i_sb->s_dev;
290 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
293 /* We don't show the stack guard page in /proc/maps */
294 start = vma->vm_start;
295 if (stack_guard_page_start(vma, start))
298 if (stack_guard_page_end(vma, end))
301 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
302 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
305 flags & VM_READ ? 'r' : '-',
306 flags & VM_WRITE ? 'w' : '-',
307 flags & VM_EXEC ? 'x' : '-',
308 flags & VM_MAYSHARE ? 's' : 'p',
310 MAJOR(dev), MINOR(dev), ino);
313 * Print the dentry name for named mappings, and a
314 * special [heap] marker for the heap:
318 seq_path(m, &file->f_path, "\n");
322 if (vma->vm_ops && vma->vm_ops->name) {
323 name = vma->vm_ops->name(vma);
328 name = arch_vma_name(vma);
337 if (vma->vm_start <= mm->brk &&
338 vma->vm_end >= mm->start_brk) {
343 tid = vm_is_stack(task, vma, is_pid);
347 * Thread stack in /proc/PID/task/TID/maps or
348 * the main process stack.
350 if (!is_pid || (vma->vm_start <= mm->start_stack &&
351 vma->vm_end >= mm->start_stack)) {
354 /* Thread stack in /proc/PID/maps */
356 seq_printf(m, "[stack:%d]", tid);
369 static int show_map(struct seq_file *m, void *v, int is_pid)
371 struct vm_area_struct *vma = v;
372 struct proc_maps_private *priv = m->private;
374 show_map_vma(m, vma, is_pid);
376 if (m->count < m->size) /* vma is copied successfully */
377 m->version = (vma != priv->tail_vma)
382 static int show_pid_map(struct seq_file *m, void *v)
384 return show_map(m, v, 1);
387 static int show_tid_map(struct seq_file *m, void *v)
389 return show_map(m, v, 0);
392 static const struct seq_operations proc_pid_maps_op = {
399 static const struct seq_operations proc_tid_maps_op = {
406 static int pid_maps_open(struct inode *inode, struct file *file)
408 return do_maps_open(inode, file, &proc_pid_maps_op);
411 static int tid_maps_open(struct inode *inode, struct file *file)
413 return do_maps_open(inode, file, &proc_tid_maps_op);
416 const struct file_operations proc_pid_maps_operations = {
417 .open = pid_maps_open,
420 .release = proc_map_release,
423 const struct file_operations proc_tid_maps_operations = {
424 .open = tid_maps_open,
427 .release = proc_map_release,
431 * Proportional Set Size(PSS): my share of RSS.
433 * PSS of a process is the count of pages it has in memory, where each
434 * page is divided by the number of processes sharing it. So if a
435 * process has 1000 pages all to itself, and 1000 shared with one other
436 * process, its PSS will be 1500.
438 * To keep (accumulated) division errors low, we adopt a 64bit
439 * fixed-point pss counter to minimize division errors. So (pss >>
440 * PSS_SHIFT) would be the real byte count.
442 * A shift of 12 before division means (assuming 4K page size):
443 * - 1M 3-user-pages add up to 8KB errors;
444 * - supports mapcount up to 2^24, or 16M;
445 * - supports PSS up to 2^52 bytes, or 4PB.
449 #ifdef CONFIG_PROC_PAGE_MONITOR
450 struct mem_size_stats {
451 struct vm_area_struct *vma;
452 unsigned long resident;
453 unsigned long shared_clean;
454 unsigned long shared_dirty;
455 unsigned long private_clean;
456 unsigned long private_dirty;
457 unsigned long referenced;
458 unsigned long anonymous;
459 unsigned long anonymous_thp;
461 unsigned long nonlinear;
466 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
467 unsigned long ptent_size, struct mm_walk *walk)
469 struct mem_size_stats *mss = walk->private;
470 struct vm_area_struct *vma = mss->vma;
471 pgoff_t pgoff = linear_page_index(vma, addr);
472 struct page *page = NULL;
475 if (pte_present(ptent)) {
476 page = vm_normal_page(vma, addr, ptent);
477 } else if (is_swap_pte(ptent)) {
478 swp_entry_t swpent = pte_to_swp_entry(ptent);
480 if (!non_swap_entry(swpent))
481 mss->swap += ptent_size;
482 else if (is_migration_entry(swpent))
483 page = migration_entry_to_page(swpent);
484 } else if (pte_file(ptent)) {
485 if (pte_to_pgoff(ptent) != pgoff)
486 mss->nonlinear += ptent_size;
493 mss->anonymous += ptent_size;
495 if (page->index != pgoff)
496 mss->nonlinear += ptent_size;
498 mss->resident += ptent_size;
499 /* Accumulate the size in pages that have been accessed. */
500 if (pte_young(ptent) || PageReferenced(page))
501 mss->referenced += ptent_size;
502 mapcount = page_mapcount(page);
504 if (pte_dirty(ptent) || PageDirty(page))
505 mss->shared_dirty += ptent_size;
507 mss->shared_clean += ptent_size;
508 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
510 if (pte_dirty(ptent) || PageDirty(page))
511 mss->private_dirty += ptent_size;
513 mss->private_clean += ptent_size;
514 mss->pss += (ptent_size << PSS_SHIFT);
518 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
519 struct mm_walk *walk)
521 struct mem_size_stats *mss = walk->private;
522 struct vm_area_struct *vma = mss->vma;
526 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
527 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
529 mss->anonymous_thp += HPAGE_PMD_SIZE;
533 if (pmd_trans_unstable(pmd))
536 * The mmap_sem held all the way back in m_start() is what
537 * keeps khugepaged out of here and from collapsing things
540 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
541 for (; addr != end; pte++, addr += PAGE_SIZE)
542 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
543 pte_unmap_unlock(pte - 1, ptl);
548 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
551 * Don't forget to update Documentation/ on changes.
553 static const char mnemonics[BITS_PER_LONG][2] = {
555 * In case if we meet a flag we don't know about.
557 [0 ... (BITS_PER_LONG-1)] = "??",
559 [ilog2(VM_READ)] = "rd",
560 [ilog2(VM_WRITE)] = "wr",
561 [ilog2(VM_EXEC)] = "ex",
562 [ilog2(VM_SHARED)] = "sh",
563 [ilog2(VM_MAYREAD)] = "mr",
564 [ilog2(VM_MAYWRITE)] = "mw",
565 [ilog2(VM_MAYEXEC)] = "me",
566 [ilog2(VM_MAYSHARE)] = "ms",
567 [ilog2(VM_GROWSDOWN)] = "gd",
568 [ilog2(VM_PFNMAP)] = "pf",
569 [ilog2(VM_DENYWRITE)] = "dw",
570 [ilog2(VM_LOCKED)] = "lo",
571 [ilog2(VM_IO)] = "io",
572 [ilog2(VM_SEQ_READ)] = "sr",
573 [ilog2(VM_RAND_READ)] = "rr",
574 [ilog2(VM_DONTCOPY)] = "dc",
575 [ilog2(VM_DONTEXPAND)] = "de",
576 [ilog2(VM_ACCOUNT)] = "ac",
577 [ilog2(VM_NORESERVE)] = "nr",
578 [ilog2(VM_HUGETLB)] = "ht",
579 [ilog2(VM_NONLINEAR)] = "nl",
580 [ilog2(VM_ARCH_1)] = "ar",
581 [ilog2(VM_DONTDUMP)] = "dd",
582 #ifdef CONFIG_MEM_SOFT_DIRTY
583 [ilog2(VM_SOFTDIRTY)] = "sd",
585 [ilog2(VM_MIXEDMAP)] = "mm",
586 [ilog2(VM_HUGEPAGE)] = "hg",
587 [ilog2(VM_NOHUGEPAGE)] = "nh",
588 [ilog2(VM_MERGEABLE)] = "mg",
592 seq_puts(m, "VmFlags: ");
593 for (i = 0; i < BITS_PER_LONG; i++) {
594 if (vma->vm_flags & (1UL << i)) {
595 seq_printf(m, "%c%c ",
596 mnemonics[i][0], mnemonics[i][1]);
602 static int show_smap(struct seq_file *m, void *v, int is_pid)
604 struct proc_maps_private *priv = m->private;
605 struct vm_area_struct *vma = v;
606 struct mem_size_stats mss;
607 struct mm_walk smaps_walk = {
608 .pmd_entry = smaps_pte_range,
613 memset(&mss, 0, sizeof mss);
615 /* mmap_sem is held in m_start */
616 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
617 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
619 show_map_vma(m, vma, is_pid);
625 "Shared_Clean: %8lu kB\n"
626 "Shared_Dirty: %8lu kB\n"
627 "Private_Clean: %8lu kB\n"
628 "Private_Dirty: %8lu kB\n"
629 "Referenced: %8lu kB\n"
630 "Anonymous: %8lu kB\n"
631 "AnonHugePages: %8lu kB\n"
633 "KernelPageSize: %8lu kB\n"
634 "MMUPageSize: %8lu kB\n"
636 (vma->vm_end - vma->vm_start) >> 10,
638 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
639 mss.shared_clean >> 10,
640 mss.shared_dirty >> 10,
641 mss.private_clean >> 10,
642 mss.private_dirty >> 10,
643 mss.referenced >> 10,
645 mss.anonymous_thp >> 10,
647 vma_kernel_pagesize(vma) >> 10,
648 vma_mmu_pagesize(vma) >> 10,
649 (vma->vm_flags & VM_LOCKED) ?
650 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
652 if (vma->vm_flags & VM_NONLINEAR)
653 seq_printf(m, "Nonlinear: %8lu kB\n",
654 mss.nonlinear >> 10);
656 show_smap_vma_flags(m, vma);
658 if (m->count < m->size) /* vma is copied successfully */
659 m->version = (vma != priv->tail_vma)
664 static int show_pid_smap(struct seq_file *m, void *v)
666 return show_smap(m, v, 1);
669 static int show_tid_smap(struct seq_file *m, void *v)
671 return show_smap(m, v, 0);
674 static const struct seq_operations proc_pid_smaps_op = {
678 .show = show_pid_smap
681 static const struct seq_operations proc_tid_smaps_op = {
685 .show = show_tid_smap
688 static int pid_smaps_open(struct inode *inode, struct file *file)
690 return do_maps_open(inode, file, &proc_pid_smaps_op);
693 static int tid_smaps_open(struct inode *inode, struct file *file)
695 return do_maps_open(inode, file, &proc_tid_smaps_op);
698 const struct file_operations proc_pid_smaps_operations = {
699 .open = pid_smaps_open,
702 .release = proc_map_release,
705 const struct file_operations proc_tid_smaps_operations = {
706 .open = tid_smaps_open,
709 .release = proc_map_release,
713 * We do not want to have constant page-shift bits sitting in
714 * pagemap entries and are about to reuse them some time soon.
716 * Here's the "migration strategy":
717 * 1. when the system boots these bits remain what they are,
718 * but a warning about future change is printed in log;
719 * 2. once anyone clears soft-dirty bits via clear_refs file,
720 * these flag is set to denote, that user is aware of the
721 * new API and those page-shift bits change their meaning.
722 * The respective warning is printed in dmesg;
723 * 3. In a couple of releases we will remove all the mentions
724 * of page-shift in pagemap entries.
727 static bool soft_dirty_cleared __read_mostly;
729 enum clear_refs_types {
733 CLEAR_REFS_SOFT_DIRTY,
737 struct clear_refs_private {
738 struct vm_area_struct *vma;
739 enum clear_refs_types type;
742 static inline void clear_soft_dirty(struct vm_area_struct *vma,
743 unsigned long addr, pte_t *pte)
745 #ifdef CONFIG_MEM_SOFT_DIRTY
747 * The soft-dirty tracker uses #PF-s to catch writes
748 * to pages, so write-protect the pte as well. See the
749 * Documentation/vm/soft-dirty.txt for full description
750 * of how soft-dirty works.
754 if (pte_present(ptent)) {
755 ptent = pte_wrprotect(ptent);
756 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY);
757 } else if (is_swap_pte(ptent)) {
758 ptent = pte_swp_clear_soft_dirty(ptent);
759 } else if (pte_file(ptent)) {
760 ptent = pte_file_clear_soft_dirty(ptent);
763 set_pte_at(vma->vm_mm, addr, pte, ptent);
767 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
768 unsigned long end, struct mm_walk *walk)
770 struct clear_refs_private *cp = walk->private;
771 struct vm_area_struct *vma = cp->vma;
776 split_huge_page_pmd(vma, addr, pmd);
777 if (pmd_trans_unstable(pmd))
780 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
781 for (; addr != end; pte++, addr += PAGE_SIZE) {
784 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
785 clear_soft_dirty(vma, addr, pte);
789 if (!pte_present(ptent))
792 page = vm_normal_page(vma, addr, ptent);
796 /* Clear accessed and referenced bits. */
797 ptep_test_and_clear_young(vma, addr, pte);
798 ClearPageReferenced(page);
800 pte_unmap_unlock(pte - 1, ptl);
805 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
806 size_t count, loff_t *ppos)
808 struct task_struct *task;
809 char buffer[PROC_NUMBUF];
810 struct mm_struct *mm;
811 struct vm_area_struct *vma;
812 enum clear_refs_types type;
816 memset(buffer, 0, sizeof(buffer));
817 if (count > sizeof(buffer) - 1)
818 count = sizeof(buffer) - 1;
819 if (copy_from_user(buffer, buf, count))
821 rv = kstrtoint(strstrip(buffer), 10, &itype);
824 type = (enum clear_refs_types)itype;
825 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
828 if (type == CLEAR_REFS_SOFT_DIRTY) {
829 soft_dirty_cleared = true;
830 pr_warn_once("The pagemap bits 55-60 has changed their meaning!"
831 " See the linux/Documentation/vm/pagemap.txt for "
835 task = get_proc_task(file_inode(file));
838 mm = get_task_mm(task);
840 struct clear_refs_private cp = {
843 struct mm_walk clear_refs_walk = {
844 .pmd_entry = clear_refs_pte_range,
848 down_read(&mm->mmap_sem);
849 if (type == CLEAR_REFS_SOFT_DIRTY)
850 mmu_notifier_invalidate_range_start(mm, 0, -1);
851 for (vma = mm->mmap; vma; vma = vma->vm_next) {
853 if (is_vm_hugetlb_page(vma))
856 * Writing 1 to /proc/pid/clear_refs affects all pages.
858 * Writing 2 to /proc/pid/clear_refs only affects
861 * Writing 3 to /proc/pid/clear_refs only affects file
864 * Writing 4 to /proc/pid/clear_refs affects all pages.
866 if (type == CLEAR_REFS_ANON && vma->vm_file)
868 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
870 if (type == CLEAR_REFS_SOFT_DIRTY) {
871 if (vma->vm_flags & VM_SOFTDIRTY)
872 vma->vm_flags &= ~VM_SOFTDIRTY;
874 walk_page_range(vma->vm_start, vma->vm_end,
877 if (type == CLEAR_REFS_SOFT_DIRTY)
878 mmu_notifier_invalidate_range_end(mm, 0, -1);
880 up_read(&mm->mmap_sem);
883 put_task_struct(task);
888 const struct file_operations proc_clear_refs_operations = {
889 .write = clear_refs_write,
890 .llseek = noop_llseek,
898 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
899 pagemap_entry_t *buffer;
903 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
904 #define PAGEMAP_WALK_MASK (PMD_MASK)
906 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
907 #define PM_STATUS_BITS 3
908 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
909 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
910 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
911 #define PM_PSHIFT_BITS 6
912 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
913 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
914 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
915 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
916 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
917 /* in "new" pagemap pshift bits are occupied with more status bits */
918 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT))
920 #define __PM_SOFT_DIRTY (1LL)
921 #define PM_PRESENT PM_STATUS(4LL)
922 #define PM_SWAP PM_STATUS(2LL)
923 #define PM_FILE PM_STATUS(1LL)
924 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0)
925 #define PM_END_OF_BUFFER 1
927 static inline pagemap_entry_t make_pme(u64 val)
929 return (pagemap_entry_t) { .pme = val };
932 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
933 struct pagemapread *pm)
935 pm->buffer[pm->pos++] = *pme;
936 if (pm->pos >= pm->len)
937 return PM_END_OF_BUFFER;
941 static int pagemap_pte_hole(unsigned long start, unsigned long end,
942 struct mm_walk *walk)
944 struct pagemapread *pm = walk->private;
945 unsigned long addr = start;
949 struct vm_area_struct *vma = find_vma(walk->mm, addr);
950 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
951 /* End of address space hole, which we mark as non-present. */
952 unsigned long hole_end;
955 hole_end = min(end, vma->vm_start);
959 for (; addr < hole_end; addr += PAGE_SIZE) {
960 err = add_to_pagemap(addr, &pme, pm);
968 /* Addresses in the VMA. */
969 if (vma->vm_flags & VM_SOFTDIRTY)
970 pme.pme |= PM_STATUS2(pm->v2, __PM_SOFT_DIRTY);
971 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
972 err = add_to_pagemap(addr, &pme, pm);
981 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
982 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
985 struct page *page = NULL;
988 if (pte_present(pte)) {
989 frame = pte_pfn(pte);
991 page = vm_normal_page(vma, addr, pte);
992 if (pte_soft_dirty(pte))
993 flags2 |= __PM_SOFT_DIRTY;
994 } else if (is_swap_pte(pte)) {
996 if (pte_swp_soft_dirty(pte))
997 flags2 |= __PM_SOFT_DIRTY;
998 entry = pte_to_swp_entry(pte);
999 frame = swp_type(entry) |
1000 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1002 if (is_migration_entry(entry))
1003 page = migration_entry_to_page(entry);
1005 if (vma->vm_flags & VM_SOFTDIRTY)
1006 flags2 |= __PM_SOFT_DIRTY;
1007 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
1011 if (page && !PageAnon(page))
1013 if ((vma->vm_flags & VM_SOFTDIRTY))
1014 flags2 |= __PM_SOFT_DIRTY;
1016 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags);
1019 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1020 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1021 pmd_t pmd, int offset, int pmd_flags2)
1024 * Currently pmd for thp is always present because thp can not be
1025 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
1026 * This if-check is just to prepare for future implementation.
1028 if (pmd_present(pmd))
1029 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
1030 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT);
1032 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2));
1035 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1036 pmd_t pmd, int offset, int pmd_flags2)
1041 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
1042 struct mm_walk *walk)
1044 struct vm_area_struct *vma;
1045 struct pagemapread *pm = walk->private;
1049 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
1051 /* find the first VMA at or above 'addr' */
1052 vma = find_vma(walk->mm, addr);
1053 if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1056 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd))
1057 pmd_flags2 = __PM_SOFT_DIRTY;
1061 for (; addr != end; addr += PAGE_SIZE) {
1062 unsigned long offset;
1064 offset = (addr & ~PAGEMAP_WALK_MASK) >>
1066 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2);
1067 err = add_to_pagemap(addr, &pme, pm);
1075 if (pmd_trans_unstable(pmd))
1077 for (; addr != end; addr += PAGE_SIZE) {
1080 /* check to see if we've left 'vma' behind
1081 * and need a new, higher one */
1082 if (vma && (addr >= vma->vm_end)) {
1083 vma = find_vma(walk->mm, addr);
1084 if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1085 flags2 = __PM_SOFT_DIRTY;
1088 pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
1091 /* check that 'vma' actually covers this address,
1092 * and that it isn't a huge page vma */
1093 if (vma && (vma->vm_start <= addr) &&
1094 !is_vm_hugetlb_page(vma)) {
1095 pte = pte_offset_map(pmd, addr);
1096 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte);
1097 /* unmap before userspace copy */
1100 err = add_to_pagemap(addr, &pme, pm);
1110 #ifdef CONFIG_HUGETLB_PAGE
1111 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1112 pte_t pte, int offset, int flags2)
1114 if (pte_present(pte))
1115 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) |
1116 PM_STATUS2(pm->v2, flags2) |
1119 *pme = make_pme(PM_NOT_PRESENT(pm->v2) |
1120 PM_STATUS2(pm->v2, flags2));
1123 /* This function walks within one hugetlb entry in the single call */
1124 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1125 unsigned long addr, unsigned long end,
1126 struct mm_walk *walk)
1128 struct pagemapread *pm = walk->private;
1129 struct vm_area_struct *vma;
1132 pagemap_entry_t pme;
1134 vma = find_vma(walk->mm, addr);
1137 if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1138 flags2 = __PM_SOFT_DIRTY;
1142 for (; addr != end; addr += PAGE_SIZE) {
1143 int offset = (addr & ~hmask) >> PAGE_SHIFT;
1144 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2);
1145 err = add_to_pagemap(addr, &pme, pm);
1154 #endif /* HUGETLB_PAGE */
1157 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1159 * For each page in the address space, this file contains one 64-bit entry
1160 * consisting of the following:
1162 * Bits 0-54 page frame number (PFN) if present
1163 * Bits 0-4 swap type if swapped
1164 * Bits 5-54 swap offset if swapped
1165 * Bits 55-60 page shift (page size = 1<<page shift)
1166 * Bit 61 page is file-page or shared-anon
1167 * Bit 62 page swapped
1168 * Bit 63 page present
1170 * If the page is not present but in swap, then the PFN contains an
1171 * encoding of the swap file number and the page's offset into the
1172 * swap. Unmapped pages return a null PFN. This allows determining
1173 * precisely which pages are mapped (or in swap) and comparing mapped
1174 * pages between processes.
1176 * Efficient users of this interface will use /proc/pid/maps to
1177 * determine which areas of memory are actually mapped and llseek to
1178 * skip over unmapped regions.
1180 static ssize_t pagemap_read(struct file *file, char __user *buf,
1181 size_t count, loff_t *ppos)
1183 struct task_struct *task = get_proc_task(file_inode(file));
1184 struct mm_struct *mm;
1185 struct pagemapread pm;
1187 struct mm_walk pagemap_walk = {};
1189 unsigned long svpfn;
1190 unsigned long start_vaddr;
1191 unsigned long end_vaddr;
1198 /* file position must be aligned */
1199 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1206 pm.v2 = soft_dirty_cleared;
1207 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1208 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1213 mm = mm_access(task, PTRACE_MODE_READ);
1215 if (!mm || IS_ERR(mm))
1218 pagemap_walk.pmd_entry = pagemap_pte_range;
1219 pagemap_walk.pte_hole = pagemap_pte_hole;
1220 #ifdef CONFIG_HUGETLB_PAGE
1221 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1223 pagemap_walk.mm = mm;
1224 pagemap_walk.private = ±
1227 svpfn = src / PM_ENTRY_BYTES;
1228 start_vaddr = svpfn << PAGE_SHIFT;
1229 end_vaddr = TASK_SIZE_OF(task);
1231 /* watch out for wraparound */
1232 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1233 start_vaddr = end_vaddr;
1236 * The odds are that this will stop walking way
1237 * before end_vaddr, because the length of the
1238 * user buffer is tracked in "pm", and the walk
1239 * will stop when we hit the end of the buffer.
1242 while (count && (start_vaddr < end_vaddr)) {
1247 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1249 if (end < start_vaddr || end > end_vaddr)
1251 down_read(&mm->mmap_sem);
1252 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1253 up_read(&mm->mmap_sem);
1256 len = min(count, PM_ENTRY_BYTES * pm.pos);
1257 if (copy_to_user(buf, pm.buffer, len)) {
1266 if (!ret || ret == PM_END_OF_BUFFER)
1274 put_task_struct(task);
1279 static int pagemap_open(struct inode *inode, struct file *file)
1281 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about "
1282 "to stop being page-shift some time soon. See the "
1283 "linux/Documentation/vm/pagemap.txt for details.\n");
1287 const struct file_operations proc_pagemap_operations = {
1288 .llseek = mem_lseek, /* borrow this */
1289 .read = pagemap_read,
1290 .open = pagemap_open,
1292 #endif /* CONFIG_PROC_PAGE_MONITOR */
1297 struct vm_area_struct *vma;
1298 unsigned long pages;
1300 unsigned long active;
1301 unsigned long writeback;
1302 unsigned long mapcount_max;
1303 unsigned long dirty;
1304 unsigned long swapcache;
1305 unsigned long node[MAX_NUMNODES];
1308 struct numa_maps_private {
1309 struct proc_maps_private proc_maps;
1310 struct numa_maps md;
1313 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1314 unsigned long nr_pages)
1316 int count = page_mapcount(page);
1318 md->pages += nr_pages;
1319 if (pte_dirty || PageDirty(page))
1320 md->dirty += nr_pages;
1322 if (PageSwapCache(page))
1323 md->swapcache += nr_pages;
1325 if (PageActive(page) || PageUnevictable(page))
1326 md->active += nr_pages;
1328 if (PageWriteback(page))
1329 md->writeback += nr_pages;
1332 md->anon += nr_pages;
1334 if (count > md->mapcount_max)
1335 md->mapcount_max = count;
1337 md->node[page_to_nid(page)] += nr_pages;
1340 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1346 if (!pte_present(pte))
1349 page = vm_normal_page(vma, addr, pte);
1353 if (PageReserved(page))
1356 nid = page_to_nid(page);
1357 if (!node_isset(nid, node_states[N_MEMORY]))
1363 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1364 unsigned long end, struct mm_walk *walk)
1366 struct numa_maps *md;
1373 if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) {
1374 pte_t huge_pte = *(pte_t *)pmd;
1377 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1379 gather_stats(page, md, pte_dirty(huge_pte),
1380 HPAGE_PMD_SIZE/PAGE_SIZE);
1385 if (pmd_trans_unstable(pmd))
1387 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1389 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1392 gather_stats(page, md, pte_dirty(*pte), 1);
1394 } while (pte++, addr += PAGE_SIZE, addr != end);
1395 pte_unmap_unlock(orig_pte, ptl);
1398 #ifdef CONFIG_HUGETLB_PAGE
1399 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1400 unsigned long addr, unsigned long end, struct mm_walk *walk)
1402 struct numa_maps *md;
1405 if (!pte_present(*pte))
1408 page = pte_page(*pte);
1413 gather_stats(page, md, pte_dirty(*pte), 1);
1418 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1419 unsigned long addr, unsigned long end, struct mm_walk *walk)
1426 * Display pages allocated per node and memory policy via /proc.
1428 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1430 struct numa_maps_private *numa_priv = m->private;
1431 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1432 struct vm_area_struct *vma = v;
1433 struct numa_maps *md = &numa_priv->md;
1434 struct file *file = vma->vm_file;
1435 struct task_struct *task = proc_priv->task;
1436 struct mm_struct *mm = vma->vm_mm;
1437 struct mm_walk walk = {};
1438 struct mempolicy *pol;
1445 /* Ensure we start with an empty set of numa_maps statistics. */
1446 memset(md, 0, sizeof(*md));
1450 walk.hugetlb_entry = gather_hugetbl_stats;
1451 walk.pmd_entry = gather_pte_stats;
1455 pol = get_vma_policy(task, vma, vma->vm_start);
1456 mpol_to_str(buffer, sizeof(buffer), pol);
1459 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1462 seq_puts(m, " file=");
1463 seq_path(m, &file->f_path, "\n\t= ");
1464 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1465 seq_puts(m, " heap");
1467 pid_t tid = vm_is_stack(task, vma, is_pid);
1470 * Thread stack in /proc/PID/task/TID/maps or
1471 * the main process stack.
1473 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1474 vma->vm_end >= mm->start_stack))
1475 seq_puts(m, " stack");
1477 seq_printf(m, " stack:%d", tid);
1481 if (is_vm_hugetlb_page(vma))
1482 seq_puts(m, " huge");
1484 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1490 seq_printf(m, " anon=%lu", md->anon);
1493 seq_printf(m, " dirty=%lu", md->dirty);
1495 if (md->pages != md->anon && md->pages != md->dirty)
1496 seq_printf(m, " mapped=%lu", md->pages);
1498 if (md->mapcount_max > 1)
1499 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1502 seq_printf(m, " swapcache=%lu", md->swapcache);
1504 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1505 seq_printf(m, " active=%lu", md->active);
1508 seq_printf(m, " writeback=%lu", md->writeback);
1510 for_each_node_state(nid, N_MEMORY)
1512 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1516 if (m->count < m->size)
1517 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1521 static int show_pid_numa_map(struct seq_file *m, void *v)
1523 return show_numa_map(m, v, 1);
1526 static int show_tid_numa_map(struct seq_file *m, void *v)
1528 return show_numa_map(m, v, 0);
1531 static const struct seq_operations proc_pid_numa_maps_op = {
1535 .show = show_pid_numa_map,
1538 static const struct seq_operations proc_tid_numa_maps_op = {
1542 .show = show_tid_numa_map,
1545 static int numa_maps_open(struct inode *inode, struct file *file,
1546 const struct seq_operations *ops)
1548 return proc_maps_open(inode, file, ops,
1549 sizeof(struct numa_maps_private));
1552 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1554 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1557 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1559 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1562 const struct file_operations proc_pid_numa_maps_operations = {
1563 .open = pid_numa_maps_open,
1565 .llseek = seq_lseek,
1566 .release = proc_map_release,
1569 const struct file_operations proc_tid_numa_maps_operations = {
1570 .open = tid_numa_maps_open,
1572 .llseek = seq_lseek,
1573 .release = proc_map_release,
1575 #endif /* CONFIG_NUMA */