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);
168 mm = mm_access(priv->task, PTRACE_MODE_READ);
169 if (!mm || IS_ERR(mm))
171 down_read(&mm->mmap_sem);
173 tail_vma = get_gate_vma(priv->task->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 do_maps_open(struct inode *inode, struct file *file,
235 const struct seq_operations *ops)
237 struct proc_maps_private *priv;
239 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
241 priv->pid = proc_pid(inode);
242 ret = seq_open(file, ops);
244 struct seq_file *m = file->private_data;
254 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
256 struct mm_struct *mm = vma->vm_mm;
257 struct file *file = vma->vm_file;
258 struct proc_maps_private *priv = m->private;
259 struct task_struct *task = priv->task;
260 vm_flags_t flags = vma->vm_flags;
261 unsigned long ino = 0;
262 unsigned long long pgoff = 0;
263 unsigned long start, end;
265 const char *name = NULL;
268 struct inode *inode = file_inode(vma->vm_file);
269 dev = inode->i_sb->s_dev;
271 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
274 /* We don't show the stack guard page in /proc/maps */
275 start = vma->vm_start;
276 if (stack_guard_page_start(vma, start))
279 if (stack_guard_page_end(vma, end))
282 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
283 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
286 flags & VM_READ ? 'r' : '-',
287 flags & VM_WRITE ? 'w' : '-',
288 flags & VM_EXEC ? 'x' : '-',
289 flags & VM_MAYSHARE ? 's' : 'p',
291 MAJOR(dev), MINOR(dev), ino);
294 * Print the dentry name for named mappings, and a
295 * special [heap] marker for the heap:
299 seq_path(m, &file->f_path, "\n");
303 name = arch_vma_name(vma);
312 if (vma->vm_start <= mm->brk &&
313 vma->vm_end >= mm->start_brk) {
318 tid = vm_is_stack(task, vma, is_pid);
322 * Thread stack in /proc/PID/task/TID/maps or
323 * the main process stack.
325 if (!is_pid || (vma->vm_start <= mm->start_stack &&
326 vma->vm_end >= mm->start_stack)) {
329 /* Thread stack in /proc/PID/maps */
331 seq_printf(m, "[stack:%d]", tid);
344 static int show_map(struct seq_file *m, void *v, int is_pid)
346 struct vm_area_struct *vma = v;
347 struct proc_maps_private *priv = m->private;
348 struct task_struct *task = priv->task;
350 show_map_vma(m, vma, is_pid);
352 if (m->count < m->size) /* vma is copied successfully */
353 m->version = (vma != get_gate_vma(task->mm))
358 static int show_pid_map(struct seq_file *m, void *v)
360 return show_map(m, v, 1);
363 static int show_tid_map(struct seq_file *m, void *v)
365 return show_map(m, v, 0);
368 static const struct seq_operations proc_pid_maps_op = {
375 static const struct seq_operations proc_tid_maps_op = {
382 static int pid_maps_open(struct inode *inode, struct file *file)
384 return do_maps_open(inode, file, &proc_pid_maps_op);
387 static int tid_maps_open(struct inode *inode, struct file *file)
389 return do_maps_open(inode, file, &proc_tid_maps_op);
392 const struct file_operations proc_pid_maps_operations = {
393 .open = pid_maps_open,
396 .release = seq_release_private,
399 const struct file_operations proc_tid_maps_operations = {
400 .open = tid_maps_open,
403 .release = seq_release_private,
407 * Proportional Set Size(PSS): my share of RSS.
409 * PSS of a process is the count of pages it has in memory, where each
410 * page is divided by the number of processes sharing it. So if a
411 * process has 1000 pages all to itself, and 1000 shared with one other
412 * process, its PSS will be 1500.
414 * To keep (accumulated) division errors low, we adopt a 64bit
415 * fixed-point pss counter to minimize division errors. So (pss >>
416 * PSS_SHIFT) would be the real byte count.
418 * A shift of 12 before division means (assuming 4K page size):
419 * - 1M 3-user-pages add up to 8KB errors;
420 * - supports mapcount up to 2^24, or 16M;
421 * - supports PSS up to 2^52 bytes, or 4PB.
425 #ifdef CONFIG_PROC_PAGE_MONITOR
426 struct mem_size_stats {
427 struct vm_area_struct *vma;
428 unsigned long resident;
429 unsigned long shared_clean;
430 unsigned long shared_dirty;
431 unsigned long private_clean;
432 unsigned long private_dirty;
433 unsigned long referenced;
434 unsigned long anonymous;
435 unsigned long anonymous_thp;
437 unsigned long nonlinear;
442 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
443 unsigned long ptent_size, struct mm_walk *walk)
445 struct mem_size_stats *mss = walk->private;
446 struct vm_area_struct *vma = mss->vma;
447 pgoff_t pgoff = linear_page_index(vma, addr);
448 struct page *page = NULL;
451 if (pte_present(ptent)) {
452 page = vm_normal_page(vma, addr, ptent);
453 } else if (is_swap_pte(ptent)) {
454 swp_entry_t swpent = pte_to_swp_entry(ptent);
456 if (!non_swap_entry(swpent))
457 mss->swap += ptent_size;
458 else if (is_migration_entry(swpent))
459 page = migration_entry_to_page(swpent);
460 } else if (pte_file(ptent)) {
461 if (pte_to_pgoff(ptent) != pgoff)
462 mss->nonlinear += ptent_size;
469 mss->anonymous += ptent_size;
471 if (page->index != pgoff)
472 mss->nonlinear += ptent_size;
474 mss->resident += ptent_size;
475 /* Accumulate the size in pages that have been accessed. */
476 if (pte_young(ptent) || PageReferenced(page))
477 mss->referenced += ptent_size;
478 mapcount = page_mapcount(page);
480 if (pte_dirty(ptent) || PageDirty(page))
481 mss->shared_dirty += ptent_size;
483 mss->shared_clean += ptent_size;
484 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
486 if (pte_dirty(ptent) || PageDirty(page))
487 mss->private_dirty += ptent_size;
489 mss->private_clean += ptent_size;
490 mss->pss += (ptent_size << PSS_SHIFT);
494 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
495 struct mm_walk *walk)
497 struct mem_size_stats *mss = walk->private;
498 struct vm_area_struct *vma = mss->vma;
502 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
503 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
505 mss->anonymous_thp += HPAGE_PMD_SIZE;
509 if (pmd_trans_unstable(pmd))
512 * The mmap_sem held all the way back in m_start() is what
513 * keeps khugepaged out of here and from collapsing things
516 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
517 for (; addr != end; pte++, addr += PAGE_SIZE)
518 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
519 pte_unmap_unlock(pte - 1, ptl);
524 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
527 * Don't forget to update Documentation/ on changes.
529 static const char mnemonics[BITS_PER_LONG][2] = {
531 * In case if we meet a flag we don't know about.
533 [0 ... (BITS_PER_LONG-1)] = "??",
535 [ilog2(VM_READ)] = "rd",
536 [ilog2(VM_WRITE)] = "wr",
537 [ilog2(VM_EXEC)] = "ex",
538 [ilog2(VM_SHARED)] = "sh",
539 [ilog2(VM_MAYREAD)] = "mr",
540 [ilog2(VM_MAYWRITE)] = "mw",
541 [ilog2(VM_MAYEXEC)] = "me",
542 [ilog2(VM_MAYSHARE)] = "ms",
543 [ilog2(VM_GROWSDOWN)] = "gd",
544 [ilog2(VM_PFNMAP)] = "pf",
545 [ilog2(VM_DENYWRITE)] = "dw",
546 [ilog2(VM_LOCKED)] = "lo",
547 [ilog2(VM_IO)] = "io",
548 [ilog2(VM_SEQ_READ)] = "sr",
549 [ilog2(VM_RAND_READ)] = "rr",
550 [ilog2(VM_DONTCOPY)] = "dc",
551 [ilog2(VM_DONTEXPAND)] = "de",
552 [ilog2(VM_ACCOUNT)] = "ac",
553 [ilog2(VM_NORESERVE)] = "nr",
554 [ilog2(VM_HUGETLB)] = "ht",
555 [ilog2(VM_NONLINEAR)] = "nl",
556 [ilog2(VM_ARCH_1)] = "ar",
557 [ilog2(VM_DONTDUMP)] = "dd",
558 #ifdef CONFIG_MEM_SOFT_DIRTY
559 [ilog2(VM_SOFTDIRTY)] = "sd",
561 [ilog2(VM_MIXEDMAP)] = "mm",
562 [ilog2(VM_HUGEPAGE)] = "hg",
563 [ilog2(VM_NOHUGEPAGE)] = "nh",
564 [ilog2(VM_MERGEABLE)] = "mg",
568 seq_puts(m, "VmFlags: ");
569 for (i = 0; i < BITS_PER_LONG; i++) {
570 if (vma->vm_flags & (1UL << i)) {
571 seq_printf(m, "%c%c ",
572 mnemonics[i][0], mnemonics[i][1]);
578 static int show_smap(struct seq_file *m, void *v, int is_pid)
580 struct proc_maps_private *priv = m->private;
581 struct task_struct *task = priv->task;
582 struct vm_area_struct *vma = v;
583 struct mem_size_stats mss;
584 struct mm_walk smaps_walk = {
585 .pmd_entry = smaps_pte_range,
590 memset(&mss, 0, sizeof mss);
592 /* mmap_sem is held in m_start */
593 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
594 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
596 show_map_vma(m, vma, is_pid);
602 "Shared_Clean: %8lu kB\n"
603 "Shared_Dirty: %8lu kB\n"
604 "Private_Clean: %8lu kB\n"
605 "Private_Dirty: %8lu kB\n"
606 "Referenced: %8lu kB\n"
607 "Anonymous: %8lu kB\n"
608 "AnonHugePages: %8lu kB\n"
610 "KernelPageSize: %8lu kB\n"
611 "MMUPageSize: %8lu kB\n"
613 (vma->vm_end - vma->vm_start) >> 10,
615 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
616 mss.shared_clean >> 10,
617 mss.shared_dirty >> 10,
618 mss.private_clean >> 10,
619 mss.private_dirty >> 10,
620 mss.referenced >> 10,
622 mss.anonymous_thp >> 10,
624 vma_kernel_pagesize(vma) >> 10,
625 vma_mmu_pagesize(vma) >> 10,
626 (vma->vm_flags & VM_LOCKED) ?
627 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
629 if (vma->vm_flags & VM_NONLINEAR)
630 seq_printf(m, "Nonlinear: %8lu kB\n",
631 mss.nonlinear >> 10);
633 show_smap_vma_flags(m, vma);
635 if (m->count < m->size) /* vma is copied successfully */
636 m->version = (vma != get_gate_vma(task->mm))
641 static int show_pid_smap(struct seq_file *m, void *v)
643 return show_smap(m, v, 1);
646 static int show_tid_smap(struct seq_file *m, void *v)
648 return show_smap(m, v, 0);
651 static const struct seq_operations proc_pid_smaps_op = {
655 .show = show_pid_smap
658 static const struct seq_operations proc_tid_smaps_op = {
662 .show = show_tid_smap
665 static int pid_smaps_open(struct inode *inode, struct file *file)
667 return do_maps_open(inode, file, &proc_pid_smaps_op);
670 static int tid_smaps_open(struct inode *inode, struct file *file)
672 return do_maps_open(inode, file, &proc_tid_smaps_op);
675 const struct file_operations proc_pid_smaps_operations = {
676 .open = pid_smaps_open,
679 .release = seq_release_private,
682 const struct file_operations proc_tid_smaps_operations = {
683 .open = tid_smaps_open,
686 .release = seq_release_private,
690 * We do not want to have constant page-shift bits sitting in
691 * pagemap entries and are about to reuse them some time soon.
693 * Here's the "migration strategy":
694 * 1. when the system boots these bits remain what they are,
695 * but a warning about future change is printed in log;
696 * 2. once anyone clears soft-dirty bits via clear_refs file,
697 * these flag is set to denote, that user is aware of the
698 * new API and those page-shift bits change their meaning.
699 * The respective warning is printed in dmesg;
700 * 3. In a couple of releases we will remove all the mentions
701 * of page-shift in pagemap entries.
704 static bool soft_dirty_cleared __read_mostly;
706 enum clear_refs_types {
710 CLEAR_REFS_SOFT_DIRTY,
714 struct clear_refs_private {
715 struct vm_area_struct *vma;
716 enum clear_refs_types type;
719 static inline void clear_soft_dirty(struct vm_area_struct *vma,
720 unsigned long addr, pte_t *pte)
722 #ifdef CONFIG_MEM_SOFT_DIRTY
724 * The soft-dirty tracker uses #PF-s to catch writes
725 * to pages, so write-protect the pte as well. See the
726 * Documentation/vm/soft-dirty.txt for full description
727 * of how soft-dirty works.
731 if (pte_present(ptent)) {
732 ptent = pte_wrprotect(ptent);
733 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY);
734 } else if (is_swap_pte(ptent)) {
735 ptent = pte_swp_clear_soft_dirty(ptent);
736 } else if (pte_file(ptent)) {
737 ptent = pte_file_clear_soft_dirty(ptent);
740 if (vma->vm_flags & VM_SOFTDIRTY)
741 vma->vm_flags &= ~VM_SOFTDIRTY;
743 set_pte_at(vma->vm_mm, addr, pte, ptent);
747 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
748 unsigned long end, struct mm_walk *walk)
750 struct clear_refs_private *cp = walk->private;
751 struct vm_area_struct *vma = cp->vma;
756 split_huge_page_pmd(vma, addr, pmd);
757 if (pmd_trans_unstable(pmd))
760 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
761 for (; addr != end; pte++, addr += PAGE_SIZE) {
764 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
765 clear_soft_dirty(vma, addr, pte);
769 if (!pte_present(ptent))
772 page = vm_normal_page(vma, addr, ptent);
776 /* Clear accessed and referenced bits. */
777 ptep_test_and_clear_young(vma, addr, pte);
778 ClearPageReferenced(page);
780 pte_unmap_unlock(pte - 1, ptl);
785 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
786 size_t count, loff_t *ppos)
788 struct task_struct *task;
789 char buffer[PROC_NUMBUF];
790 struct mm_struct *mm;
791 struct vm_area_struct *vma;
792 enum clear_refs_types type;
796 memset(buffer, 0, sizeof(buffer));
797 if (count > sizeof(buffer) - 1)
798 count = sizeof(buffer) - 1;
799 if (copy_from_user(buffer, buf, count))
801 rv = kstrtoint(strstrip(buffer), 10, &itype);
804 type = (enum clear_refs_types)itype;
805 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
808 if (type == CLEAR_REFS_SOFT_DIRTY) {
809 soft_dirty_cleared = true;
810 pr_warn_once("The pagemap bits 55-60 has changed their meaning! "
811 "See the linux/Documentation/vm/pagemap.txt for details.\n");
814 task = get_proc_task(file_inode(file));
817 mm = get_task_mm(task);
819 struct clear_refs_private cp = {
822 struct mm_walk clear_refs_walk = {
823 .pmd_entry = clear_refs_pte_range,
827 down_read(&mm->mmap_sem);
828 if (type == CLEAR_REFS_SOFT_DIRTY)
829 mmu_notifier_invalidate_range_start(mm, 0, -1);
830 for (vma = mm->mmap; vma; vma = vma->vm_next) {
832 if (is_vm_hugetlb_page(vma))
835 * Writing 1 to /proc/pid/clear_refs affects all pages.
837 * Writing 2 to /proc/pid/clear_refs only affects
840 * Writing 3 to /proc/pid/clear_refs only affects file
843 if (type == CLEAR_REFS_ANON && vma->vm_file)
845 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
847 walk_page_range(vma->vm_start, vma->vm_end,
850 if (type == CLEAR_REFS_SOFT_DIRTY)
851 mmu_notifier_invalidate_range_end(mm, 0, -1);
853 up_read(&mm->mmap_sem);
856 put_task_struct(task);
861 const struct file_operations proc_clear_refs_operations = {
862 .write = clear_refs_write,
863 .llseek = noop_llseek,
871 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
872 pagemap_entry_t *buffer;
876 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
877 #define PAGEMAP_WALK_MASK (PMD_MASK)
879 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
880 #define PM_STATUS_BITS 3
881 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
882 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
883 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
884 #define PM_PSHIFT_BITS 6
885 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
886 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
887 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
888 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
889 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
890 /* in "new" pagemap pshift bits are occupied with more status bits */
891 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT))
893 #define __PM_SOFT_DIRTY (1LL)
894 #define PM_PRESENT PM_STATUS(4LL)
895 #define PM_SWAP PM_STATUS(2LL)
896 #define PM_FILE PM_STATUS(1LL)
897 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0)
898 #define PM_END_OF_BUFFER 1
900 static inline pagemap_entry_t make_pme(u64 val)
902 return (pagemap_entry_t) { .pme = val };
905 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
906 struct pagemapread *pm)
908 pm->buffer[pm->pos++] = *pme;
909 if (pm->pos >= pm->len)
910 return PM_END_OF_BUFFER;
914 static int pagemap_pte_hole(unsigned long start, unsigned long end,
915 struct mm_walk *walk)
917 struct pagemapread *pm = walk->private;
920 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
922 for (addr = start; addr < end; addr += PAGE_SIZE) {
923 err = add_to_pagemap(addr, &pme, pm);
930 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
931 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
934 struct page *page = NULL;
937 if (pte_present(pte)) {
938 frame = pte_pfn(pte);
940 page = vm_normal_page(vma, addr, pte);
941 if (pte_soft_dirty(pte))
942 flags2 |= __PM_SOFT_DIRTY;
943 } else if (is_swap_pte(pte)) {
945 if (pte_swp_soft_dirty(pte))
946 flags2 |= __PM_SOFT_DIRTY;
947 entry = pte_to_swp_entry(pte);
948 frame = swp_type(entry) |
949 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
951 if (is_migration_entry(entry))
952 page = migration_entry_to_page(entry);
954 if (vma->vm_flags & VM_SOFTDIRTY)
955 flags2 |= __PM_SOFT_DIRTY;
956 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
960 if (page && !PageAnon(page))
962 if ((vma->vm_flags & VM_SOFTDIRTY))
963 flags2 |= __PM_SOFT_DIRTY;
965 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags);
968 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
969 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
970 pmd_t pmd, int offset, int pmd_flags2)
973 * Currently pmd for thp is always present because thp can not be
974 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
975 * This if-check is just to prepare for future implementation.
977 if (pmd_present(pmd))
978 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
979 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT);
981 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2));
984 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
985 pmd_t pmd, int offset, int pmd_flags2)
990 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
991 struct mm_walk *walk)
993 struct vm_area_struct *vma;
994 struct pagemapread *pm = walk->private;
998 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
1000 /* find the first VMA at or above 'addr' */
1001 vma = find_vma(walk->mm, addr);
1002 if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1005 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd))
1006 pmd_flags2 = __PM_SOFT_DIRTY;
1010 for (; addr != end; addr += PAGE_SIZE) {
1011 unsigned long offset;
1013 offset = (addr & ~PAGEMAP_WALK_MASK) >>
1015 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2);
1016 err = add_to_pagemap(addr, &pme, pm);
1024 if (pmd_trans_unstable(pmd))
1026 for (; addr != end; addr += PAGE_SIZE) {
1029 /* check to see if we've left 'vma' behind
1030 * and need a new, higher one */
1031 if (vma && (addr >= vma->vm_end)) {
1032 vma = find_vma(walk->mm, addr);
1033 if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1034 flags2 = __PM_SOFT_DIRTY;
1037 pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
1040 /* check that 'vma' actually covers this address,
1041 * and that it isn't a huge page vma */
1042 if (vma && (vma->vm_start <= addr) &&
1043 !is_vm_hugetlb_page(vma)) {
1044 pte = pte_offset_map(pmd, addr);
1045 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte);
1046 /* unmap before userspace copy */
1049 err = add_to_pagemap(addr, &pme, pm);
1059 #ifdef CONFIG_HUGETLB_PAGE
1060 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1061 pte_t pte, int offset, int flags2)
1063 if (pte_present(pte))
1064 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) |
1065 PM_STATUS2(pm->v2, flags2) |
1068 *pme = make_pme(PM_NOT_PRESENT(pm->v2) |
1069 PM_STATUS2(pm->v2, flags2));
1072 /* This function walks within one hugetlb entry in the single call */
1073 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1074 unsigned long addr, unsigned long end,
1075 struct mm_walk *walk)
1077 struct pagemapread *pm = walk->private;
1078 struct vm_area_struct *vma;
1081 pagemap_entry_t pme;
1083 vma = find_vma(walk->mm, addr);
1086 if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1087 flags2 = __PM_SOFT_DIRTY;
1091 for (; addr != end; addr += PAGE_SIZE) {
1092 int offset = (addr & ~hmask) >> PAGE_SHIFT;
1093 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2);
1094 err = add_to_pagemap(addr, &pme, pm);
1103 #endif /* HUGETLB_PAGE */
1106 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1108 * For each page in the address space, this file contains one 64-bit entry
1109 * consisting of the following:
1111 * Bits 0-54 page frame number (PFN) if present
1112 * Bits 0-4 swap type if swapped
1113 * Bits 5-54 swap offset if swapped
1114 * Bits 55-60 page shift (page size = 1<<page shift)
1115 * Bit 61 page is file-page or shared-anon
1116 * Bit 62 page swapped
1117 * Bit 63 page present
1119 * If the page is not present but in swap, then the PFN contains an
1120 * encoding of the swap file number and the page's offset into the
1121 * swap. Unmapped pages return a null PFN. This allows determining
1122 * precisely which pages are mapped (or in swap) and comparing mapped
1123 * pages between processes.
1125 * Efficient users of this interface will use /proc/pid/maps to
1126 * determine which areas of memory are actually mapped and llseek to
1127 * skip over unmapped regions.
1129 static ssize_t pagemap_read(struct file *file, char __user *buf,
1130 size_t count, loff_t *ppos)
1132 struct task_struct *task = get_proc_task(file_inode(file));
1133 struct mm_struct *mm;
1134 struct pagemapread pm;
1136 struct mm_walk pagemap_walk = {};
1138 unsigned long svpfn;
1139 unsigned long start_vaddr;
1140 unsigned long end_vaddr;
1147 /* file position must be aligned */
1148 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1155 pm.v2 = soft_dirty_cleared;
1156 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1157 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1162 mm = mm_access(task, PTRACE_MODE_READ);
1164 if (!mm || IS_ERR(mm))
1167 pagemap_walk.pmd_entry = pagemap_pte_range;
1168 pagemap_walk.pte_hole = pagemap_pte_hole;
1169 #ifdef CONFIG_HUGETLB_PAGE
1170 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1172 pagemap_walk.mm = mm;
1173 pagemap_walk.private = ±
1176 svpfn = src / PM_ENTRY_BYTES;
1177 start_vaddr = svpfn << PAGE_SHIFT;
1178 end_vaddr = TASK_SIZE_OF(task);
1180 /* watch out for wraparound */
1181 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1182 start_vaddr = end_vaddr;
1185 * The odds are that this will stop walking way
1186 * before end_vaddr, because the length of the
1187 * user buffer is tracked in "pm", and the walk
1188 * will stop when we hit the end of the buffer.
1191 while (count && (start_vaddr < end_vaddr)) {
1196 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1198 if (end < start_vaddr || end > end_vaddr)
1200 down_read(&mm->mmap_sem);
1201 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1202 up_read(&mm->mmap_sem);
1205 len = min(count, PM_ENTRY_BYTES * pm.pos);
1206 if (copy_to_user(buf, pm.buffer, len)) {
1215 if (!ret || ret == PM_END_OF_BUFFER)
1223 put_task_struct(task);
1228 static int pagemap_open(struct inode *inode, struct file *file)
1230 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about "
1231 "to stop being page-shift some time soon. See the "
1232 "linux/Documentation/vm/pagemap.txt for details.\n");
1236 const struct file_operations proc_pagemap_operations = {
1237 .llseek = mem_lseek, /* borrow this */
1238 .read = pagemap_read,
1239 .open = pagemap_open,
1241 #endif /* CONFIG_PROC_PAGE_MONITOR */
1246 struct vm_area_struct *vma;
1247 unsigned long pages;
1249 unsigned long active;
1250 unsigned long writeback;
1251 unsigned long mapcount_max;
1252 unsigned long dirty;
1253 unsigned long swapcache;
1254 unsigned long node[MAX_NUMNODES];
1257 struct numa_maps_private {
1258 struct proc_maps_private proc_maps;
1259 struct numa_maps md;
1262 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1263 unsigned long nr_pages)
1265 int count = page_mapcount(page);
1267 md->pages += nr_pages;
1268 if (pte_dirty || PageDirty(page))
1269 md->dirty += nr_pages;
1271 if (PageSwapCache(page))
1272 md->swapcache += nr_pages;
1274 if (PageActive(page) || PageUnevictable(page))
1275 md->active += nr_pages;
1277 if (PageWriteback(page))
1278 md->writeback += nr_pages;
1281 md->anon += nr_pages;
1283 if (count > md->mapcount_max)
1284 md->mapcount_max = count;
1286 md->node[page_to_nid(page)] += nr_pages;
1289 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1295 if (!pte_present(pte))
1298 page = vm_normal_page(vma, addr, pte);
1302 if (PageReserved(page))
1305 nid = page_to_nid(page);
1306 if (!node_isset(nid, node_states[N_MEMORY]))
1312 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1313 unsigned long end, struct mm_walk *walk)
1315 struct numa_maps *md;
1322 if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) {
1323 pte_t huge_pte = *(pte_t *)pmd;
1326 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1328 gather_stats(page, md, pte_dirty(huge_pte),
1329 HPAGE_PMD_SIZE/PAGE_SIZE);
1334 if (pmd_trans_unstable(pmd))
1336 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1338 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1341 gather_stats(page, md, pte_dirty(*pte), 1);
1343 } while (pte++, addr += PAGE_SIZE, addr != end);
1344 pte_unmap_unlock(orig_pte, ptl);
1347 #ifdef CONFIG_HUGETLB_PAGE
1348 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1349 unsigned long addr, unsigned long end, struct mm_walk *walk)
1351 struct numa_maps *md;
1357 page = pte_page(*pte);
1362 gather_stats(page, md, pte_dirty(*pte), 1);
1367 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1368 unsigned long addr, unsigned long end, struct mm_walk *walk)
1375 * Display pages allocated per node and memory policy via /proc.
1377 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1379 struct numa_maps_private *numa_priv = m->private;
1380 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1381 struct vm_area_struct *vma = v;
1382 struct numa_maps *md = &numa_priv->md;
1383 struct file *file = vma->vm_file;
1384 struct task_struct *task = proc_priv->task;
1385 struct mm_struct *mm = vma->vm_mm;
1386 struct mm_walk walk = {};
1387 struct mempolicy *pol;
1394 /* Ensure we start with an empty set of numa_maps statistics. */
1395 memset(md, 0, sizeof(*md));
1399 walk.hugetlb_entry = gather_hugetbl_stats;
1400 walk.pmd_entry = gather_pte_stats;
1404 pol = get_vma_policy(task, vma, vma->vm_start);
1405 mpol_to_str(buffer, sizeof(buffer), pol);
1408 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1411 seq_printf(m, " file=");
1412 seq_path(m, &file->f_path, "\n\t= ");
1413 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1414 seq_printf(m, " heap");
1416 pid_t tid = vm_is_stack(task, vma, is_pid);
1419 * Thread stack in /proc/PID/task/TID/maps or
1420 * the main process stack.
1422 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1423 vma->vm_end >= mm->start_stack))
1424 seq_printf(m, " stack");
1426 seq_printf(m, " stack:%d", tid);
1430 if (is_vm_hugetlb_page(vma))
1431 seq_printf(m, " huge");
1433 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1439 seq_printf(m, " anon=%lu", md->anon);
1442 seq_printf(m, " dirty=%lu", md->dirty);
1444 if (md->pages != md->anon && md->pages != md->dirty)
1445 seq_printf(m, " mapped=%lu", md->pages);
1447 if (md->mapcount_max > 1)
1448 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1451 seq_printf(m, " swapcache=%lu", md->swapcache);
1453 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1454 seq_printf(m, " active=%lu", md->active);
1457 seq_printf(m, " writeback=%lu", md->writeback);
1459 for_each_node_state(nid, N_MEMORY)
1461 seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1465 if (m->count < m->size)
1466 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1470 static int show_pid_numa_map(struct seq_file *m, void *v)
1472 return show_numa_map(m, v, 1);
1475 static int show_tid_numa_map(struct seq_file *m, void *v)
1477 return show_numa_map(m, v, 0);
1480 static const struct seq_operations proc_pid_numa_maps_op = {
1484 .show = show_pid_numa_map,
1487 static const struct seq_operations proc_tid_numa_maps_op = {
1491 .show = show_tid_numa_map,
1494 static int numa_maps_open(struct inode *inode, struct file *file,
1495 const struct seq_operations *ops)
1497 struct numa_maps_private *priv;
1499 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1501 priv->proc_maps.pid = proc_pid(inode);
1502 ret = seq_open(file, ops);
1504 struct seq_file *m = file->private_data;
1513 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1515 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1518 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1520 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1523 const struct file_operations proc_pid_numa_maps_operations = {
1524 .open = pid_numa_maps_open,
1526 .llseek = seq_lseek,
1527 .release = seq_release_private,
1530 const struct file_operations proc_tid_numa_maps_operations = {
1531 .open = tid_numa_maps_open,
1533 .llseek = seq_lseek,
1534 .release = seq_release_private,
1536 #endif /* CONFIG_NUMA */