2 #include <linux/hugetlb.h>
3 #include <linux/huge_mm.h>
4 #include <linux/mount.h>
5 #include <linux/seq_file.h>
6 #include <linux/highmem.h>
7 #include <linux/ptrace.h>
8 #include <linux/slab.h>
9 #include <linux/pagemap.h>
10 #include <linux/mempolicy.h>
11 #include <linux/rmap.h>
12 #include <linux/swap.h>
13 #include <linux/swapops.h>
16 #include <asm/uaccess.h>
17 #include <asm/tlbflush.h>
20 void task_mem(struct seq_file *m, struct mm_struct *mm)
22 unsigned long data, text, lib, swap;
23 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
26 * Note: to minimize their overhead, mm maintains hiwater_vm and
27 * hiwater_rss only when about to *lower* total_vm or rss. Any
28 * collector of these hiwater stats must therefore get total_vm
29 * and rss too, which will usually be the higher. Barriers? not
30 * worth the effort, such snapshots can always be inconsistent.
32 hiwater_vm = total_vm = mm->total_vm;
33 if (hiwater_vm < mm->hiwater_vm)
34 hiwater_vm = mm->hiwater_vm;
35 hiwater_rss = total_rss = get_mm_rss(mm);
36 if (hiwater_rss < mm->hiwater_rss)
37 hiwater_rss = mm->hiwater_rss;
39 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
40 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
41 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
42 swap = get_mm_counter(mm, MM_SWAPENTS);
56 hiwater_vm << (PAGE_SHIFT-10),
57 total_vm << (PAGE_SHIFT-10),
58 mm->locked_vm << (PAGE_SHIFT-10),
59 mm->pinned_vm << (PAGE_SHIFT-10),
60 hiwater_rss << (PAGE_SHIFT-10),
61 total_rss << (PAGE_SHIFT-10),
62 data << (PAGE_SHIFT-10),
63 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
64 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
65 swap << (PAGE_SHIFT-10));
68 unsigned long task_vsize(struct mm_struct *mm)
70 return PAGE_SIZE * mm->total_vm;
73 unsigned long task_statm(struct mm_struct *mm,
74 unsigned long *shared, unsigned long *text,
75 unsigned long *data, unsigned long *resident)
77 *shared = get_mm_counter(mm, MM_FILEPAGES);
78 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
80 *data = mm->total_vm - mm->shared_vm;
81 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
85 static void pad_len_spaces(struct seq_file *m, int len)
87 len = 25 + sizeof(void*) * 6 - len;
90 seq_printf(m, "%*c", len, ' ');
95 * These functions are for numa_maps but called in generic **maps seq_file
96 * ->start(), ->stop() ops.
98 * numa_maps scans all vmas under mmap_sem and checks their mempolicy.
99 * Each mempolicy object is controlled by reference counting. The problem here
100 * is how to avoid accessing dead mempolicy object.
102 * Because we're holding mmap_sem while reading seq_file, it's safe to access
103 * each vma's mempolicy, no vma objects will never drop refs to mempolicy.
105 * A task's mempolicy (task->mempolicy) has different behavior. task->mempolicy
106 * is set and replaced under mmap_sem but unrefed and cleared under task_lock().
107 * So, without task_lock(), we cannot trust get_vma_policy() because we cannot
108 * gurantee the task never exits under us. But taking task_lock() around
109 * get_vma_plicy() causes lock order problem.
111 * To access task->mempolicy without lock, we hold a reference count of an
112 * object pointed by task->mempolicy and remember it. This will guarantee
113 * that task->mempolicy points to an alive object or NULL in numa_maps accesses.
115 static void hold_task_mempolicy(struct proc_maps_private *priv)
117 struct task_struct *task = priv->task;
120 priv->task_mempolicy = task->mempolicy;
121 mpol_get(priv->task_mempolicy);
124 static void release_task_mempolicy(struct proc_maps_private *priv)
126 mpol_put(priv->task_mempolicy);
129 static void hold_task_mempolicy(struct proc_maps_private *priv)
132 static void release_task_mempolicy(struct proc_maps_private *priv)
137 static void seq_print_vma_name(struct seq_file *m, struct vm_area_struct *vma)
139 const char __user *name = vma_get_anon_name(vma);
140 struct mm_struct *mm = vma->vm_mm;
142 unsigned long page_start_vaddr;
143 unsigned long page_offset;
144 unsigned long num_pages;
145 unsigned long max_len = NAME_MAX;
148 page_start_vaddr = (unsigned long)name & PAGE_MASK;
149 page_offset = (unsigned long)name - page_start_vaddr;
150 num_pages = DIV_ROUND_UP(page_offset + max_len, PAGE_SIZE);
152 seq_puts(m, "[anon:");
154 for (i = 0; i < num_pages; i++) {
161 pages_pinned = get_user_pages(current, mm, page_start_vaddr,
162 1, 0, 0, &page, NULL);
163 if (pages_pinned < 1) {
164 seq_puts(m, "<fault>]");
168 kaddr = (const char *)kmap(page);
169 len = min(max_len, PAGE_SIZE - page_offset);
170 write_len = strnlen(kaddr + page_offset, len);
171 seq_write(m, kaddr + page_offset, write_len);
175 /* if strnlen hit a null terminator then we're done */
176 if (write_len != len)
181 page_start_vaddr += PAGE_SIZE;
187 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
189 if (vma && vma != priv->tail_vma) {
190 struct mm_struct *mm = vma->vm_mm;
191 release_task_mempolicy(priv);
192 up_read(&mm->mmap_sem);
197 static void *m_start(struct seq_file *m, loff_t *pos)
199 struct proc_maps_private *priv = m->private;
200 unsigned long last_addr = m->version;
201 struct mm_struct *mm;
202 struct vm_area_struct *vma, *tail_vma = NULL;
205 /* Clear the per syscall fields in priv */
207 priv->tail_vma = NULL;
210 * We remember last_addr rather than next_addr to hit with
211 * mmap_cache most of the time. We have zero last_addr at
212 * the beginning and also after lseek. We will have -1 last_addr
213 * after the end of the vmas.
216 if (last_addr == -1UL)
219 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
221 return ERR_PTR(-ESRCH);
223 mm = mm_access(priv->task, PTRACE_MODE_READ);
224 if (!mm || IS_ERR(mm))
226 down_read(&mm->mmap_sem);
228 tail_vma = get_gate_vma(priv->task->mm);
229 priv->tail_vma = tail_vma;
230 hold_task_mempolicy(priv);
231 /* Start with last addr hint */
232 vma = find_vma(mm, last_addr);
233 if (last_addr && vma) {
239 * Check the vma index is within the range and do
240 * sequential scan until m_index.
243 if ((unsigned long)l < mm->map_count) {
250 if (l != mm->map_count)
251 tail_vma = NULL; /* After gate vma */
257 release_task_mempolicy(priv);
258 /* End of vmas has been reached */
259 m->version = (tail_vma != NULL)? 0: -1UL;
260 up_read(&mm->mmap_sem);
265 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
267 struct proc_maps_private *priv = m->private;
268 struct vm_area_struct *vma = v;
269 struct vm_area_struct *tail_vma = priv->tail_vma;
272 if (vma && (vma != tail_vma) && vma->vm_next)
275 return (vma != tail_vma)? tail_vma: NULL;
278 static void m_stop(struct seq_file *m, void *v)
280 struct proc_maps_private *priv = m->private;
281 struct vm_area_struct *vma = v;
286 put_task_struct(priv->task);
289 static int do_maps_open(struct inode *inode, struct file *file,
290 const struct seq_operations *ops)
292 struct proc_maps_private *priv;
294 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
296 priv->pid = proc_pid(inode);
297 ret = seq_open(file, ops);
299 struct seq_file *m = file->private_data;
309 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
311 struct mm_struct *mm = vma->vm_mm;
312 struct file *file = vma->vm_file;
313 struct proc_maps_private *priv = m->private;
314 struct task_struct *task = priv->task;
315 vm_flags_t flags = vma->vm_flags;
316 unsigned long ino = 0;
317 unsigned long long pgoff = 0;
318 unsigned long start, end;
321 const char *name = NULL;
324 struct inode *inode = file_inode(vma->vm_file);
325 dev = inode->i_sb->s_dev;
327 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
330 /* We don't show the stack guard page in /proc/maps */
331 start = vma->vm_start;
332 if (stack_guard_page_start(vma, start))
335 if (stack_guard_page_end(vma, end))
338 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
341 flags & VM_READ ? 'r' : '-',
342 flags & VM_WRITE ? 'w' : '-',
343 flags & VM_EXEC ? 'x' : '-',
344 flags & VM_MAYSHARE ? 's' : 'p',
346 MAJOR(dev), MINOR(dev), ino, &len);
349 * Print the dentry name for named mappings, and a
350 * special [heap] marker for the heap:
353 pad_len_spaces(m, len);
354 seq_path(m, &file->f_path, "\n");
358 name = arch_vma_name(vma);
367 if (vma->vm_start <= mm->brk &&
368 vma->vm_end >= mm->start_brk) {
373 tid = vm_is_stack(task, vma, is_pid);
377 * Thread stack in /proc/PID/task/TID/maps or
378 * the main process stack.
380 if (!is_pid || (vma->vm_start <= mm->start_stack &&
381 vma->vm_end >= mm->start_stack)) {
384 /* Thread stack in /proc/PID/maps */
385 pad_len_spaces(m, len);
386 seq_printf(m, "[stack:%d]", tid);
391 if (vma_get_anon_name(vma)) {
392 pad_len_spaces(m, len);
393 seq_print_vma_name(m, vma);
399 pad_len_spaces(m, len);
405 static int show_map(struct seq_file *m, void *v, int is_pid)
407 struct vm_area_struct *vma = v;
408 struct proc_maps_private *priv = m->private;
409 struct task_struct *task = priv->task;
411 show_map_vma(m, vma, is_pid);
413 if (m->count < m->size) /* vma is copied successfully */
414 m->version = (vma != get_gate_vma(task->mm))
419 static int show_pid_map(struct seq_file *m, void *v)
421 return show_map(m, v, 1);
424 static int show_tid_map(struct seq_file *m, void *v)
426 return show_map(m, v, 0);
429 static const struct seq_operations proc_pid_maps_op = {
436 static const struct seq_operations proc_tid_maps_op = {
443 static int pid_maps_open(struct inode *inode, struct file *file)
445 return do_maps_open(inode, file, &proc_pid_maps_op);
448 static int tid_maps_open(struct inode *inode, struct file *file)
450 return do_maps_open(inode, file, &proc_tid_maps_op);
453 const struct file_operations proc_pid_maps_operations = {
454 .open = pid_maps_open,
457 .release = seq_release_private,
460 const struct file_operations proc_tid_maps_operations = {
461 .open = tid_maps_open,
464 .release = seq_release_private,
468 * Proportional Set Size(PSS): my share of RSS.
470 * PSS of a process is the count of pages it has in memory, where each
471 * page is divided by the number of processes sharing it. So if a
472 * process has 1000 pages all to itself, and 1000 shared with one other
473 * process, its PSS will be 1500.
475 * To keep (accumulated) division errors low, we adopt a 64bit
476 * fixed-point pss counter to minimize division errors. So (pss >>
477 * PSS_SHIFT) would be the real byte count.
479 * A shift of 12 before division means (assuming 4K page size):
480 * - 1M 3-user-pages add up to 8KB errors;
481 * - supports mapcount up to 2^24, or 16M;
482 * - supports PSS up to 2^52 bytes, or 4PB.
486 #ifdef CONFIG_PROC_PAGE_MONITOR
487 struct mem_size_stats {
488 struct vm_area_struct *vma;
489 unsigned long resident;
490 unsigned long shared_clean;
491 unsigned long shared_dirty;
492 unsigned long private_clean;
493 unsigned long private_dirty;
494 unsigned long referenced;
495 unsigned long anonymous;
496 unsigned long anonymous_thp;
498 unsigned long nonlinear;
503 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
504 unsigned long ptent_size, struct mm_walk *walk)
506 struct mem_size_stats *mss = walk->private;
507 struct vm_area_struct *vma = mss->vma;
508 pgoff_t pgoff = linear_page_index(vma, addr);
509 struct page *page = NULL;
512 if (pte_present(ptent)) {
513 page = vm_normal_page(vma, addr, ptent);
514 } else if (is_swap_pte(ptent)) {
515 swp_entry_t swpent = pte_to_swp_entry(ptent);
517 if (!non_swap_entry(swpent))
518 mss->swap += ptent_size;
519 else if (is_migration_entry(swpent))
520 page = migration_entry_to_page(swpent);
521 } else if (pte_file(ptent)) {
522 if (pte_to_pgoff(ptent) != pgoff)
523 mss->nonlinear += ptent_size;
530 mss->anonymous += ptent_size;
532 if (page->index != pgoff)
533 mss->nonlinear += ptent_size;
535 mss->resident += ptent_size;
536 /* Accumulate the size in pages that have been accessed. */
537 if (pte_young(ptent) || PageReferenced(page))
538 mss->referenced += ptent_size;
539 mapcount = page_mapcount(page);
541 if (pte_dirty(ptent) || PageDirty(page))
542 mss->shared_dirty += ptent_size;
544 mss->shared_clean += ptent_size;
545 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
547 if (pte_dirty(ptent) || PageDirty(page))
548 mss->private_dirty += ptent_size;
550 mss->private_clean += ptent_size;
551 mss->pss += (ptent_size << PSS_SHIFT);
555 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
556 struct mm_walk *walk)
558 struct mem_size_stats *mss = walk->private;
559 struct vm_area_struct *vma = mss->vma;
563 if (pmd_trans_huge_lock(pmd, vma) == 1) {
564 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
565 spin_unlock(&walk->mm->page_table_lock);
566 mss->anonymous_thp += HPAGE_PMD_SIZE;
570 if (pmd_trans_unstable(pmd))
573 * The mmap_sem held all the way back in m_start() is what
574 * keeps khugepaged out of here and from collapsing things
577 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
578 for (; addr != end; pte++, addr += PAGE_SIZE)
579 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
580 pte_unmap_unlock(pte - 1, ptl);
585 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
588 * Don't forget to update Documentation/ on changes.
590 static const char mnemonics[BITS_PER_LONG][2] = {
592 * In case if we meet a flag we don't know about.
594 [0 ... (BITS_PER_LONG-1)] = "??",
596 [ilog2(VM_READ)] = "rd",
597 [ilog2(VM_WRITE)] = "wr",
598 [ilog2(VM_EXEC)] = "ex",
599 [ilog2(VM_SHARED)] = "sh",
600 [ilog2(VM_MAYREAD)] = "mr",
601 [ilog2(VM_MAYWRITE)] = "mw",
602 [ilog2(VM_MAYEXEC)] = "me",
603 [ilog2(VM_MAYSHARE)] = "ms",
604 [ilog2(VM_GROWSDOWN)] = "gd",
605 [ilog2(VM_PFNMAP)] = "pf",
606 [ilog2(VM_DENYWRITE)] = "dw",
607 [ilog2(VM_LOCKED)] = "lo",
608 [ilog2(VM_IO)] = "io",
609 [ilog2(VM_SEQ_READ)] = "sr",
610 [ilog2(VM_RAND_READ)] = "rr",
611 [ilog2(VM_DONTCOPY)] = "dc",
612 [ilog2(VM_DONTEXPAND)] = "de",
613 [ilog2(VM_ACCOUNT)] = "ac",
614 [ilog2(VM_NORESERVE)] = "nr",
615 [ilog2(VM_HUGETLB)] = "ht",
616 [ilog2(VM_NONLINEAR)] = "nl",
617 [ilog2(VM_ARCH_1)] = "ar",
618 [ilog2(VM_DONTDUMP)] = "dd",
619 [ilog2(VM_MIXEDMAP)] = "mm",
620 [ilog2(VM_HUGEPAGE)] = "hg",
621 [ilog2(VM_NOHUGEPAGE)] = "nh",
622 [ilog2(VM_MERGEABLE)] = "mg",
626 seq_puts(m, "VmFlags: ");
627 for (i = 0; i < BITS_PER_LONG; i++) {
628 if (vma->vm_flags & (1UL << i)) {
629 seq_printf(m, "%c%c ",
630 mnemonics[i][0], mnemonics[i][1]);
636 static int show_smap(struct seq_file *m, void *v, int is_pid)
638 struct proc_maps_private *priv = m->private;
639 struct task_struct *task = priv->task;
640 struct vm_area_struct *vma = v;
641 struct mem_size_stats mss;
642 struct mm_walk smaps_walk = {
643 .pmd_entry = smaps_pte_range,
648 memset(&mss, 0, sizeof mss);
650 /* mmap_sem is held in m_start */
651 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
652 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
654 show_map_vma(m, vma, is_pid);
660 "Shared_Clean: %8lu kB\n"
661 "Shared_Dirty: %8lu kB\n"
662 "Private_Clean: %8lu kB\n"
663 "Private_Dirty: %8lu kB\n"
664 "Referenced: %8lu kB\n"
665 "Anonymous: %8lu kB\n"
666 "AnonHugePages: %8lu kB\n"
668 "KernelPageSize: %8lu kB\n"
669 "MMUPageSize: %8lu kB\n"
671 (vma->vm_end - vma->vm_start) >> 10,
673 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
674 mss.shared_clean >> 10,
675 mss.shared_dirty >> 10,
676 mss.private_clean >> 10,
677 mss.private_dirty >> 10,
678 mss.referenced >> 10,
680 mss.anonymous_thp >> 10,
682 vma_kernel_pagesize(vma) >> 10,
683 vma_mmu_pagesize(vma) >> 10,
684 (vma->vm_flags & VM_LOCKED) ?
685 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
687 if (vma->vm_flags & VM_NONLINEAR)
688 seq_printf(m, "Nonlinear: %8lu kB\n",
689 mss.nonlinear >> 10);
691 show_smap_vma_flags(m, vma);
693 if (vma_get_anon_name(vma)) {
694 seq_puts(m, "Name: ");
695 seq_print_vma_name(m, vma);
699 if (m->count < m->size) /* vma is copied successfully */
700 m->version = (vma != get_gate_vma(task->mm))
705 static int show_pid_smap(struct seq_file *m, void *v)
707 return show_smap(m, v, 1);
710 static int show_tid_smap(struct seq_file *m, void *v)
712 return show_smap(m, v, 0);
715 static const struct seq_operations proc_pid_smaps_op = {
719 .show = show_pid_smap
722 static const struct seq_operations proc_tid_smaps_op = {
726 .show = show_tid_smap
729 static int pid_smaps_open(struct inode *inode, struct file *file)
731 return do_maps_open(inode, file, &proc_pid_smaps_op);
734 static int tid_smaps_open(struct inode *inode, struct file *file)
736 return do_maps_open(inode, file, &proc_tid_smaps_op);
739 const struct file_operations proc_pid_smaps_operations = {
740 .open = pid_smaps_open,
743 .release = seq_release_private,
746 const struct file_operations proc_tid_smaps_operations = {
747 .open = tid_smaps_open,
750 .release = seq_release_private,
753 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
754 unsigned long end, struct mm_walk *walk)
756 struct vm_area_struct *vma = walk->private;
761 split_huge_page_pmd(vma, addr, pmd);
762 if (pmd_trans_unstable(pmd))
765 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
766 for (; addr != end; pte++, addr += PAGE_SIZE) {
768 if (!pte_present(ptent))
771 page = vm_normal_page(vma, addr, ptent);
775 /* Clear accessed and referenced bits. */
776 ptep_test_and_clear_young(vma, addr, pte);
777 ClearPageReferenced(page);
779 pte_unmap_unlock(pte - 1, ptl);
784 #define CLEAR_REFS_ALL 1
785 #define CLEAR_REFS_ANON 2
786 #define CLEAR_REFS_MAPPED 3
788 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
789 size_t count, loff_t *ppos)
791 struct task_struct *task;
792 char buffer[PROC_NUMBUF];
793 struct mm_struct *mm;
794 struct vm_area_struct *vma;
798 memset(buffer, 0, sizeof(buffer));
799 if (count > sizeof(buffer) - 1)
800 count = sizeof(buffer) - 1;
801 if (copy_from_user(buffer, buf, count))
803 rv = kstrtoint(strstrip(buffer), 10, &type);
806 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
808 task = get_proc_task(file_inode(file));
811 mm = get_task_mm(task);
813 struct mm_walk clear_refs_walk = {
814 .pmd_entry = clear_refs_pte_range,
817 down_read(&mm->mmap_sem);
818 for (vma = mm->mmap; vma; vma = vma->vm_next) {
819 clear_refs_walk.private = vma;
820 if (is_vm_hugetlb_page(vma))
823 * Writing 1 to /proc/pid/clear_refs affects all pages.
825 * Writing 2 to /proc/pid/clear_refs only affects
828 * Writing 3 to /proc/pid/clear_refs only affects file
831 if (type == CLEAR_REFS_ANON && vma->vm_file)
833 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
835 walk_page_range(vma->vm_start, vma->vm_end,
839 up_read(&mm->mmap_sem);
842 put_task_struct(task);
847 const struct file_operations proc_clear_refs_operations = {
848 .write = clear_refs_write,
849 .llseek = noop_llseek,
857 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
858 pagemap_entry_t *buffer;
861 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
862 #define PAGEMAP_WALK_MASK (PMD_MASK)
864 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
865 #define PM_STATUS_BITS 3
866 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
867 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
868 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
869 #define PM_PSHIFT_BITS 6
870 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
871 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
872 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
873 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
874 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
876 #define PM_PRESENT PM_STATUS(4LL)
877 #define PM_SWAP PM_STATUS(2LL)
878 #define PM_FILE PM_STATUS(1LL)
879 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
880 #define PM_END_OF_BUFFER 1
882 static inline pagemap_entry_t make_pme(u64 val)
884 return (pagemap_entry_t) { .pme = val };
887 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
888 struct pagemapread *pm)
890 pm->buffer[pm->pos++] = *pme;
891 if (pm->pos >= pm->len)
892 return PM_END_OF_BUFFER;
896 static int pagemap_pte_hole(unsigned long start, unsigned long end,
897 struct mm_walk *walk)
899 struct pagemapread *pm = walk->private;
902 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
904 for (addr = start; addr < end; addr += PAGE_SIZE) {
905 err = add_to_pagemap(addr, &pme, pm);
912 static void pte_to_pagemap_entry(pagemap_entry_t *pme,
913 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
916 struct page *page = NULL;
918 if (pte_present(pte)) {
919 frame = pte_pfn(pte);
921 page = vm_normal_page(vma, addr, pte);
922 } else if (is_swap_pte(pte)) {
923 swp_entry_t entry = pte_to_swp_entry(pte);
925 frame = swp_type(entry) |
926 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
928 if (is_migration_entry(entry))
929 page = migration_entry_to_page(entry);
931 *pme = make_pme(PM_NOT_PRESENT);
935 if (page && !PageAnon(page))
938 *pme = make_pme(PM_PFRAME(frame) | PM_PSHIFT(PAGE_SHIFT) | flags);
941 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
942 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
943 pmd_t pmd, int offset)
946 * Currently pmd for thp is always present because thp can not be
947 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
948 * This if-check is just to prepare for future implementation.
950 if (pmd_present(pmd))
951 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
952 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
954 *pme = make_pme(PM_NOT_PRESENT);
957 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
958 pmd_t pmd, int offset)
963 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
964 struct mm_walk *walk)
966 struct vm_area_struct *vma;
967 struct pagemapread *pm = walk->private;
970 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
972 /* find the first VMA at or above 'addr' */
973 vma = find_vma(walk->mm, addr);
974 if (vma && pmd_trans_huge_lock(pmd, vma) == 1) {
975 for (; addr != end; addr += PAGE_SIZE) {
976 unsigned long offset;
978 offset = (addr & ~PAGEMAP_WALK_MASK) >>
980 thp_pmd_to_pagemap_entry(&pme, *pmd, offset);
981 err = add_to_pagemap(addr, &pme, pm);
985 spin_unlock(&walk->mm->page_table_lock);
989 if (pmd_trans_unstable(pmd))
991 for (; addr != end; addr += PAGE_SIZE) {
993 /* check to see if we've left 'vma' behind
994 * and need a new, higher one */
995 if (vma && (addr >= vma->vm_end)) {
996 vma = find_vma(walk->mm, addr);
997 pme = make_pme(PM_NOT_PRESENT);
1000 /* check that 'vma' actually covers this address,
1001 * and that it isn't a huge page vma */
1002 if (vma && (vma->vm_start <= addr) &&
1003 !is_vm_hugetlb_page(vma)) {
1004 pte = pte_offset_map(pmd, addr);
1005 pte_to_pagemap_entry(&pme, vma, addr, *pte);
1006 /* unmap before userspace copy */
1009 err = add_to_pagemap(addr, &pme, pm);
1019 #ifdef CONFIG_HUGETLB_PAGE
1020 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme,
1021 pte_t pte, int offset)
1023 if (pte_present(pte))
1024 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset)
1025 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
1027 *pme = make_pme(PM_NOT_PRESENT);
1030 /* This function walks within one hugetlb entry in the single call */
1031 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1032 unsigned long addr, unsigned long end,
1033 struct mm_walk *walk)
1035 struct pagemapread *pm = walk->private;
1037 pagemap_entry_t pme;
1039 for (; addr != end; addr += PAGE_SIZE) {
1040 int offset = (addr & ~hmask) >> PAGE_SHIFT;
1041 huge_pte_to_pagemap_entry(&pme, *pte, offset);
1042 err = add_to_pagemap(addr, &pme, pm);
1051 #endif /* HUGETLB_PAGE */
1054 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1056 * For each page in the address space, this file contains one 64-bit entry
1057 * consisting of the following:
1059 * Bits 0-54 page frame number (PFN) if present
1060 * Bits 0-4 swap type if swapped
1061 * Bits 5-54 swap offset if swapped
1062 * Bits 55-60 page shift (page size = 1<<page shift)
1063 * Bit 61 page is file-page or shared-anon
1064 * Bit 62 page swapped
1065 * Bit 63 page present
1067 * If the page is not present but in swap, then the PFN contains an
1068 * encoding of the swap file number and the page's offset into the
1069 * swap. Unmapped pages return a null PFN. This allows determining
1070 * precisely which pages are mapped (or in swap) and comparing mapped
1071 * pages between processes.
1073 * Efficient users of this interface will use /proc/pid/maps to
1074 * determine which areas of memory are actually mapped and llseek to
1075 * skip over unmapped regions.
1077 static ssize_t pagemap_read(struct file *file, char __user *buf,
1078 size_t count, loff_t *ppos)
1080 struct task_struct *task = get_proc_task(file_inode(file));
1081 struct mm_struct *mm;
1082 struct pagemapread pm;
1084 struct mm_walk pagemap_walk = {};
1086 unsigned long svpfn;
1087 unsigned long start_vaddr;
1088 unsigned long end_vaddr;
1095 /* file position must be aligned */
1096 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1103 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1104 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1109 mm = mm_access(task, PTRACE_MODE_READ);
1111 if (!mm || IS_ERR(mm))
1114 pagemap_walk.pmd_entry = pagemap_pte_range;
1115 pagemap_walk.pte_hole = pagemap_pte_hole;
1116 #ifdef CONFIG_HUGETLB_PAGE
1117 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1119 pagemap_walk.mm = mm;
1120 pagemap_walk.private = ±
1123 svpfn = src / PM_ENTRY_BYTES;
1124 start_vaddr = svpfn << PAGE_SHIFT;
1125 end_vaddr = TASK_SIZE_OF(task);
1127 /* watch out for wraparound */
1128 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1129 start_vaddr = end_vaddr;
1132 * The odds are that this will stop walking way
1133 * before end_vaddr, because the length of the
1134 * user buffer is tracked in "pm", and the walk
1135 * will stop when we hit the end of the buffer.
1138 while (count && (start_vaddr < end_vaddr)) {
1143 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1145 if (end < start_vaddr || end > end_vaddr)
1147 down_read(&mm->mmap_sem);
1148 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1149 up_read(&mm->mmap_sem);
1152 len = min(count, PM_ENTRY_BYTES * pm.pos);
1153 if (copy_to_user(buf, pm.buffer, len)) {
1162 if (!ret || ret == PM_END_OF_BUFFER)
1170 put_task_struct(task);
1175 const struct file_operations proc_pagemap_operations = {
1176 .llseek = mem_lseek, /* borrow this */
1177 .read = pagemap_read,
1179 #endif /* CONFIG_PROC_PAGE_MONITOR */
1184 struct vm_area_struct *vma;
1185 unsigned long pages;
1187 unsigned long active;
1188 unsigned long writeback;
1189 unsigned long mapcount_max;
1190 unsigned long dirty;
1191 unsigned long swapcache;
1192 unsigned long node[MAX_NUMNODES];
1195 struct numa_maps_private {
1196 struct proc_maps_private proc_maps;
1197 struct numa_maps md;
1200 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1201 unsigned long nr_pages)
1203 int count = page_mapcount(page);
1205 md->pages += nr_pages;
1206 if (pte_dirty || PageDirty(page))
1207 md->dirty += nr_pages;
1209 if (PageSwapCache(page))
1210 md->swapcache += nr_pages;
1212 if (PageActive(page) || PageUnevictable(page))
1213 md->active += nr_pages;
1215 if (PageWriteback(page))
1216 md->writeback += nr_pages;
1219 md->anon += nr_pages;
1221 if (count > md->mapcount_max)
1222 md->mapcount_max = count;
1224 md->node[page_to_nid(page)] += nr_pages;
1227 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1233 if (!pte_present(pte))
1236 page = vm_normal_page(vma, addr, pte);
1240 if (PageReserved(page))
1243 nid = page_to_nid(page);
1244 if (!node_isset(nid, node_states[N_MEMORY]))
1250 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1251 unsigned long end, struct mm_walk *walk)
1253 struct numa_maps *md;
1260 if (pmd_trans_huge_lock(pmd, md->vma) == 1) {
1261 pte_t huge_pte = *(pte_t *)pmd;
1264 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1266 gather_stats(page, md, pte_dirty(huge_pte),
1267 HPAGE_PMD_SIZE/PAGE_SIZE);
1268 spin_unlock(&walk->mm->page_table_lock);
1272 if (pmd_trans_unstable(pmd))
1274 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1276 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1279 gather_stats(page, md, pte_dirty(*pte), 1);
1281 } while (pte++, addr += PAGE_SIZE, addr != end);
1282 pte_unmap_unlock(orig_pte, ptl);
1285 #ifdef CONFIG_HUGETLB_PAGE
1286 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1287 unsigned long addr, unsigned long end, struct mm_walk *walk)
1289 struct numa_maps *md;
1295 page = pte_page(*pte);
1300 gather_stats(page, md, pte_dirty(*pte), 1);
1305 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1306 unsigned long addr, unsigned long end, struct mm_walk *walk)
1313 * Display pages allocated per node and memory policy via /proc.
1315 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1317 struct numa_maps_private *numa_priv = m->private;
1318 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1319 struct vm_area_struct *vma = v;
1320 struct numa_maps *md = &numa_priv->md;
1321 struct file *file = vma->vm_file;
1322 struct task_struct *task = proc_priv->task;
1323 struct mm_struct *mm = vma->vm_mm;
1324 struct mm_walk walk = {};
1325 struct mempolicy *pol;
1332 /* Ensure we start with an empty set of numa_maps statistics. */
1333 memset(md, 0, sizeof(*md));
1337 walk.hugetlb_entry = gather_hugetbl_stats;
1338 walk.pmd_entry = gather_pte_stats;
1342 pol = get_vma_policy(task, vma, vma->vm_start);
1343 mpol_to_str(buffer, sizeof(buffer), pol);
1346 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1349 seq_printf(m, " file=");
1350 seq_path(m, &file->f_path, "\n\t= ");
1351 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1352 seq_printf(m, " heap");
1354 pid_t tid = vm_is_stack(task, vma, is_pid);
1357 * Thread stack in /proc/PID/task/TID/maps or
1358 * the main process stack.
1360 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1361 vma->vm_end >= mm->start_stack))
1362 seq_printf(m, " stack");
1364 seq_printf(m, " stack:%d", tid);
1368 if (is_vm_hugetlb_page(vma))
1369 seq_printf(m, " huge");
1371 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1377 seq_printf(m, " anon=%lu", md->anon);
1380 seq_printf(m, " dirty=%lu", md->dirty);
1382 if (md->pages != md->anon && md->pages != md->dirty)
1383 seq_printf(m, " mapped=%lu", md->pages);
1385 if (md->mapcount_max > 1)
1386 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1389 seq_printf(m, " swapcache=%lu", md->swapcache);
1391 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1392 seq_printf(m, " active=%lu", md->active);
1395 seq_printf(m, " writeback=%lu", md->writeback);
1397 for_each_node_state(n, N_MEMORY)
1399 seq_printf(m, " N%d=%lu", n, md->node[n]);
1403 if (m->count < m->size)
1404 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1408 static int show_pid_numa_map(struct seq_file *m, void *v)
1410 return show_numa_map(m, v, 1);
1413 static int show_tid_numa_map(struct seq_file *m, void *v)
1415 return show_numa_map(m, v, 0);
1418 static const struct seq_operations proc_pid_numa_maps_op = {
1422 .show = show_pid_numa_map,
1425 static const struct seq_operations proc_tid_numa_maps_op = {
1429 .show = show_tid_numa_map,
1432 static int numa_maps_open(struct inode *inode, struct file *file,
1433 const struct seq_operations *ops)
1435 struct numa_maps_private *priv;
1437 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1439 priv->proc_maps.pid = proc_pid(inode);
1440 ret = seq_open(file, ops);
1442 struct seq_file *m = file->private_data;
1451 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1453 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1456 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1458 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1461 const struct file_operations proc_pid_numa_maps_operations = {
1462 .open = pid_numa_maps_open,
1464 .llseek = seq_lseek,
1465 .release = seq_release_private,
1468 const struct file_operations proc_tid_numa_maps_operations = {
1469 .open = tid_numa_maps_open,
1471 .llseek = seq_lseek,
1472 .release = seq_release_private,
1474 #endif /* CONFIG_NUMA */