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 - mm->reserved_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, ' ');
93 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
95 if (vma && vma != priv->tail_vma) {
96 struct mm_struct *mm = vma->vm_mm;
97 up_read(&mm->mmap_sem);
102 static void *m_start(struct seq_file *m, loff_t *pos)
104 struct proc_maps_private *priv = m->private;
105 unsigned long last_addr = m->version;
106 struct mm_struct *mm;
107 struct vm_area_struct *vma, *tail_vma = NULL;
110 /* Clear the per syscall fields in priv */
112 priv->tail_vma = NULL;
115 * We remember last_addr rather than next_addr to hit with
116 * mmap_cache most of the time. We have zero last_addr at
117 * the beginning and also after lseek. We will have -1 last_addr
118 * after the end of the vmas.
121 if (last_addr == -1UL)
124 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
126 return ERR_PTR(-ESRCH);
128 mm = mm_access(priv->task, PTRACE_MODE_READ);
129 if (!mm || IS_ERR(mm))
131 down_read(&mm->mmap_sem);
133 tail_vma = get_gate_vma(priv->task->mm);
134 priv->tail_vma = tail_vma;
136 /* Start with last addr hint */
137 vma = find_vma(mm, last_addr);
138 if (last_addr && vma) {
144 * Check the vma index is within the range and do
145 * sequential scan until m_index.
148 if ((unsigned long)l < mm->map_count) {
155 if (l != mm->map_count)
156 tail_vma = NULL; /* After gate vma */
162 /* End of vmas has been reached */
163 m->version = (tail_vma != NULL)? 0: -1UL;
164 up_read(&mm->mmap_sem);
169 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
171 struct proc_maps_private *priv = m->private;
172 struct vm_area_struct *vma = v;
173 struct vm_area_struct *tail_vma = priv->tail_vma;
176 if (vma && (vma != tail_vma) && vma->vm_next)
179 return (vma != tail_vma)? tail_vma: NULL;
182 static void m_stop(struct seq_file *m, void *v)
184 struct proc_maps_private *priv = m->private;
185 struct vm_area_struct *vma = v;
190 put_task_struct(priv->task);
193 static int do_maps_open(struct inode *inode, struct file *file,
194 const struct seq_operations *ops)
196 struct proc_maps_private *priv;
198 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
200 priv->pid = proc_pid(inode);
201 ret = seq_open(file, ops);
203 struct seq_file *m = file->private_data;
213 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
215 struct mm_struct *mm = vma->vm_mm;
216 struct file *file = vma->vm_file;
217 struct proc_maps_private *priv = m->private;
218 struct task_struct *task = priv->task;
219 vm_flags_t flags = vma->vm_flags;
220 unsigned long ino = 0;
221 unsigned long long pgoff = 0;
222 unsigned long start, end;
225 const char *name = NULL;
228 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
229 dev = inode->i_sb->s_dev;
231 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
234 /* We don't show the stack guard page in /proc/maps */
235 start = vma->vm_start;
236 if (stack_guard_page_start(vma, start))
239 if (stack_guard_page_end(vma, end))
242 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
245 flags & VM_READ ? 'r' : '-',
246 flags & VM_WRITE ? 'w' : '-',
247 flags & VM_EXEC ? 'x' : '-',
248 flags & VM_MAYSHARE ? 's' : 'p',
250 MAJOR(dev), MINOR(dev), ino, &len);
253 * Print the dentry name for named mappings, and a
254 * special [heap] marker for the heap:
257 pad_len_spaces(m, len);
258 seq_path(m, &file->f_path, "\n");
262 name = arch_vma_name(vma);
271 if (vma->vm_start <= mm->brk &&
272 vma->vm_end >= mm->start_brk) {
277 tid = vm_is_stack(task, vma, is_pid);
281 * Thread stack in /proc/PID/task/TID/maps or
282 * the main process stack.
284 if (!is_pid || (vma->vm_start <= mm->start_stack &&
285 vma->vm_end >= mm->start_stack)) {
288 /* Thread stack in /proc/PID/maps */
289 pad_len_spaces(m, len);
290 seq_printf(m, "[stack:%d]", tid);
297 pad_len_spaces(m, len);
303 static int show_map(struct seq_file *m, void *v, int is_pid)
305 struct vm_area_struct *vma = v;
306 struct proc_maps_private *priv = m->private;
307 struct task_struct *task = priv->task;
309 show_map_vma(m, vma, is_pid);
311 if (m->count < m->size) /* vma is copied successfully */
312 m->version = (vma != get_gate_vma(task->mm))
317 static int show_pid_map(struct seq_file *m, void *v)
319 return show_map(m, v, 1);
322 static int show_tid_map(struct seq_file *m, void *v)
324 return show_map(m, v, 0);
327 static const struct seq_operations proc_pid_maps_op = {
334 static const struct seq_operations proc_tid_maps_op = {
341 static int pid_maps_open(struct inode *inode, struct file *file)
343 return do_maps_open(inode, file, &proc_pid_maps_op);
346 static int tid_maps_open(struct inode *inode, struct file *file)
348 return do_maps_open(inode, file, &proc_tid_maps_op);
351 const struct file_operations proc_pid_maps_operations = {
352 .open = pid_maps_open,
355 .release = seq_release_private,
358 const struct file_operations proc_tid_maps_operations = {
359 .open = tid_maps_open,
362 .release = seq_release_private,
366 * Proportional Set Size(PSS): my share of RSS.
368 * PSS of a process is the count of pages it has in memory, where each
369 * page is divided by the number of processes sharing it. So if a
370 * process has 1000 pages all to itself, and 1000 shared with one other
371 * process, its PSS will be 1500.
373 * To keep (accumulated) division errors low, we adopt a 64bit
374 * fixed-point pss counter to minimize division errors. So (pss >>
375 * PSS_SHIFT) would be the real byte count.
377 * A shift of 12 before division means (assuming 4K page size):
378 * - 1M 3-user-pages add up to 8KB errors;
379 * - supports mapcount up to 2^24, or 16M;
380 * - supports PSS up to 2^52 bytes, or 4PB.
384 #ifdef CONFIG_PROC_PAGE_MONITOR
385 struct mem_size_stats {
386 struct vm_area_struct *vma;
387 unsigned long resident;
388 unsigned long shared_clean;
389 unsigned long shared_dirty;
390 unsigned long private_clean;
391 unsigned long private_dirty;
392 unsigned long referenced;
393 unsigned long anonymous;
394 unsigned long anonymous_thp;
400 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
401 unsigned long ptent_size, struct mm_walk *walk)
403 struct mem_size_stats *mss = walk->private;
404 struct vm_area_struct *vma = mss->vma;
405 struct page *page = NULL;
408 if (pte_present(ptent)) {
409 page = vm_normal_page(vma, addr, ptent);
410 } else if (is_swap_pte(ptent)) {
411 swp_entry_t swpent = pte_to_swp_entry(ptent);
413 if (!non_swap_entry(swpent))
414 mss->swap += ptent_size;
415 else if (is_migration_entry(swpent))
416 page = migration_entry_to_page(swpent);
423 mss->anonymous += ptent_size;
425 mss->resident += ptent_size;
426 /* Accumulate the size in pages that have been accessed. */
427 if (pte_young(ptent) || PageReferenced(page))
428 mss->referenced += ptent_size;
429 mapcount = page_mapcount(page);
431 if (pte_dirty(ptent) || PageDirty(page))
432 mss->shared_dirty += ptent_size;
434 mss->shared_clean += ptent_size;
435 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
437 if (pte_dirty(ptent) || PageDirty(page))
438 mss->private_dirty += ptent_size;
440 mss->private_clean += ptent_size;
441 mss->pss += (ptent_size << PSS_SHIFT);
445 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
446 struct mm_walk *walk)
448 struct mem_size_stats *mss = walk->private;
449 struct vm_area_struct *vma = mss->vma;
453 if (pmd_trans_huge_lock(pmd, vma) == 1) {
454 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
455 spin_unlock(&walk->mm->page_table_lock);
456 mss->anonymous_thp += HPAGE_PMD_SIZE;
460 if (pmd_trans_unstable(pmd))
463 * The mmap_sem held all the way back in m_start() is what
464 * keeps khugepaged out of here and from collapsing things
467 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
468 for (; addr != end; pte++, addr += PAGE_SIZE)
469 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
470 pte_unmap_unlock(pte - 1, ptl);
475 static int show_smap(struct seq_file *m, void *v, int is_pid)
477 struct proc_maps_private *priv = m->private;
478 struct task_struct *task = priv->task;
479 struct vm_area_struct *vma = v;
480 struct mem_size_stats mss;
481 struct mm_walk smaps_walk = {
482 .pmd_entry = smaps_pte_range,
487 memset(&mss, 0, sizeof mss);
489 /* mmap_sem is held in m_start */
490 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
491 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
493 show_map_vma(m, vma, is_pid);
499 "Shared_Clean: %8lu kB\n"
500 "Shared_Dirty: %8lu kB\n"
501 "Private_Clean: %8lu kB\n"
502 "Private_Dirty: %8lu kB\n"
503 "Referenced: %8lu kB\n"
504 "Anonymous: %8lu kB\n"
505 "AnonHugePages: %8lu kB\n"
507 "KernelPageSize: %8lu kB\n"
508 "MMUPageSize: %8lu kB\n"
510 (vma->vm_end - vma->vm_start) >> 10,
512 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
513 mss.shared_clean >> 10,
514 mss.shared_dirty >> 10,
515 mss.private_clean >> 10,
516 mss.private_dirty >> 10,
517 mss.referenced >> 10,
519 mss.anonymous_thp >> 10,
521 vma_kernel_pagesize(vma) >> 10,
522 vma_mmu_pagesize(vma) >> 10,
523 (vma->vm_flags & VM_LOCKED) ?
524 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
526 if (m->count < m->size) /* vma is copied successfully */
527 m->version = (vma != get_gate_vma(task->mm))
532 static int show_pid_smap(struct seq_file *m, void *v)
534 return show_smap(m, v, 1);
537 static int show_tid_smap(struct seq_file *m, void *v)
539 return show_smap(m, v, 0);
542 static const struct seq_operations proc_pid_smaps_op = {
546 .show = show_pid_smap
549 static const struct seq_operations proc_tid_smaps_op = {
553 .show = show_tid_smap
556 static int pid_smaps_open(struct inode *inode, struct file *file)
558 return do_maps_open(inode, file, &proc_pid_smaps_op);
561 static int tid_smaps_open(struct inode *inode, struct file *file)
563 return do_maps_open(inode, file, &proc_tid_smaps_op);
566 const struct file_operations proc_pid_smaps_operations = {
567 .open = pid_smaps_open,
570 .release = seq_release_private,
573 const struct file_operations proc_tid_smaps_operations = {
574 .open = tid_smaps_open,
577 .release = seq_release_private,
580 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
581 unsigned long end, struct mm_walk *walk)
583 struct vm_area_struct *vma = walk->private;
588 split_huge_page_pmd(walk->mm, pmd);
589 if (pmd_trans_unstable(pmd))
592 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
593 for (; addr != end; pte++, addr += PAGE_SIZE) {
595 if (!pte_present(ptent))
598 page = vm_normal_page(vma, addr, ptent);
602 /* Clear accessed and referenced bits. */
603 ptep_test_and_clear_young(vma, addr, pte);
604 ClearPageReferenced(page);
606 pte_unmap_unlock(pte - 1, ptl);
611 #define CLEAR_REFS_ALL 1
612 #define CLEAR_REFS_ANON 2
613 #define CLEAR_REFS_MAPPED 3
615 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
616 size_t count, loff_t *ppos)
618 struct task_struct *task;
619 char buffer[PROC_NUMBUF];
620 struct mm_struct *mm;
621 struct vm_area_struct *vma;
625 memset(buffer, 0, sizeof(buffer));
626 if (count > sizeof(buffer) - 1)
627 count = sizeof(buffer) - 1;
628 if (copy_from_user(buffer, buf, count))
630 rv = kstrtoint(strstrip(buffer), 10, &type);
633 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
635 task = get_proc_task(file->f_path.dentry->d_inode);
638 mm = get_task_mm(task);
640 struct mm_walk clear_refs_walk = {
641 .pmd_entry = clear_refs_pte_range,
644 down_read(&mm->mmap_sem);
645 for (vma = mm->mmap; vma; vma = vma->vm_next) {
646 clear_refs_walk.private = vma;
647 if (is_vm_hugetlb_page(vma))
650 * Writing 1 to /proc/pid/clear_refs affects all pages.
652 * Writing 2 to /proc/pid/clear_refs only affects
655 * Writing 3 to /proc/pid/clear_refs only affects file
658 if (type == CLEAR_REFS_ANON && vma->vm_file)
660 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
662 walk_page_range(vma->vm_start, vma->vm_end,
666 up_read(&mm->mmap_sem);
669 put_task_struct(task);
674 const struct file_operations proc_clear_refs_operations = {
675 .write = clear_refs_write,
676 .llseek = noop_llseek,
685 pagemap_entry_t *buffer;
688 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
689 #define PAGEMAP_WALK_MASK (PMD_MASK)
691 #define PM_ENTRY_BYTES sizeof(u64)
692 #define PM_STATUS_BITS 3
693 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
694 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
695 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
696 #define PM_PSHIFT_BITS 6
697 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
698 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
699 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
700 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
701 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
703 #define PM_PRESENT PM_STATUS(4LL)
704 #define PM_SWAP PM_STATUS(2LL)
705 #define PM_FILE PM_STATUS(1LL)
706 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
707 #define PM_END_OF_BUFFER 1
709 static inline pagemap_entry_t make_pme(u64 val)
711 return (pagemap_entry_t) { .pme = val };
714 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
715 struct pagemapread *pm)
717 pm->buffer[pm->pos++] = *pme;
718 if (pm->pos >= pm->len)
719 return PM_END_OF_BUFFER;
723 static int pagemap_pte_hole(unsigned long start, unsigned long end,
724 struct mm_walk *walk)
726 struct pagemapread *pm = walk->private;
729 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
731 for (addr = start; addr < end; addr += PAGE_SIZE) {
732 err = add_to_pagemap(addr, &pme, pm);
739 static void pte_to_pagemap_entry(pagemap_entry_t *pme,
740 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
743 struct page *page = NULL;
745 if (pte_present(pte)) {
746 frame = pte_pfn(pte);
748 page = vm_normal_page(vma, addr, pte);
749 } else if (is_swap_pte(pte)) {
750 swp_entry_t entry = pte_to_swp_entry(pte);
752 frame = swp_type(entry) |
753 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
755 if (is_migration_entry(entry))
756 page = migration_entry_to_page(entry);
758 *pme = make_pme(PM_NOT_PRESENT);
762 if (page && !PageAnon(page))
765 *pme = make_pme(PM_PFRAME(frame) | PM_PSHIFT(PAGE_SHIFT) | flags);
768 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
769 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
770 pmd_t pmd, int offset)
773 * Currently pmd for thp is always present because thp can not be
774 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
775 * This if-check is just to prepare for future implementation.
777 if (pmd_present(pmd))
778 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
779 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
781 *pme = make_pme(PM_NOT_PRESENT);
784 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
785 pmd_t pmd, int offset)
790 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
791 struct mm_walk *walk)
793 struct vm_area_struct *vma;
794 struct pagemapread *pm = walk->private;
797 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
799 /* find the first VMA at or above 'addr' */
800 vma = find_vma(walk->mm, addr);
801 if (vma && pmd_trans_huge_lock(pmd, vma) == 1) {
802 for (; addr != end; addr += PAGE_SIZE) {
803 unsigned long offset;
805 offset = (addr & ~PAGEMAP_WALK_MASK) >>
807 thp_pmd_to_pagemap_entry(&pme, *pmd, offset);
808 err = add_to_pagemap(addr, &pme, pm);
812 spin_unlock(&walk->mm->page_table_lock);
816 if (pmd_trans_unstable(pmd))
818 for (; addr != end; addr += PAGE_SIZE) {
820 /* check to see if we've left 'vma' behind
821 * and need a new, higher one */
822 if (vma && (addr >= vma->vm_end)) {
823 vma = find_vma(walk->mm, addr);
824 pme = make_pme(PM_NOT_PRESENT);
827 /* check that 'vma' actually covers this address,
828 * and that it isn't a huge page vma */
829 if (vma && (vma->vm_start <= addr) &&
830 !is_vm_hugetlb_page(vma)) {
831 pte = pte_offset_map(pmd, addr);
832 pte_to_pagemap_entry(&pme, vma, addr, *pte);
833 /* unmap before userspace copy */
836 err = add_to_pagemap(addr, &pme, pm);
846 #ifdef CONFIG_HUGETLB_PAGE
847 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme,
848 pte_t pte, int offset)
850 if (pte_present(pte))
851 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset)
852 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
854 *pme = make_pme(PM_NOT_PRESENT);
857 /* This function walks within one hugetlb entry in the single call */
858 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
859 unsigned long addr, unsigned long end,
860 struct mm_walk *walk)
862 struct pagemapread *pm = walk->private;
866 for (; addr != end; addr += PAGE_SIZE) {
867 int offset = (addr & ~hmask) >> PAGE_SHIFT;
868 huge_pte_to_pagemap_entry(&pme, *pte, offset);
869 err = add_to_pagemap(addr, &pme, pm);
878 #endif /* HUGETLB_PAGE */
881 * /proc/pid/pagemap - an array mapping virtual pages to pfns
883 * For each page in the address space, this file contains one 64-bit entry
884 * consisting of the following:
886 * Bits 0-54 page frame number (PFN) if present
887 * Bits 0-4 swap type if swapped
888 * Bits 5-54 swap offset if swapped
889 * Bits 55-60 page shift (page size = 1<<page shift)
890 * Bit 61 page is file-page or shared-anon
891 * Bit 62 page swapped
892 * Bit 63 page present
894 * If the page is not present but in swap, then the PFN contains an
895 * encoding of the swap file number and the page's offset into the
896 * swap. Unmapped pages return a null PFN. This allows determining
897 * precisely which pages are mapped (or in swap) and comparing mapped
898 * pages between processes.
900 * Efficient users of this interface will use /proc/pid/maps to
901 * determine which areas of memory are actually mapped and llseek to
902 * skip over unmapped regions.
904 static ssize_t pagemap_read(struct file *file, char __user *buf,
905 size_t count, loff_t *ppos)
907 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
908 struct mm_struct *mm;
909 struct pagemapread pm;
911 struct mm_walk pagemap_walk = {};
914 unsigned long start_vaddr;
915 unsigned long end_vaddr;
922 /* file position must be aligned */
923 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
930 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
931 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
936 mm = mm_access(task, PTRACE_MODE_READ);
938 if (!mm || IS_ERR(mm))
941 pagemap_walk.pmd_entry = pagemap_pte_range;
942 pagemap_walk.pte_hole = pagemap_pte_hole;
943 #ifdef CONFIG_HUGETLB_PAGE
944 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
946 pagemap_walk.mm = mm;
947 pagemap_walk.private = ±
950 svpfn = src / PM_ENTRY_BYTES;
951 start_vaddr = svpfn << PAGE_SHIFT;
952 end_vaddr = TASK_SIZE_OF(task);
954 /* watch out for wraparound */
955 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
956 start_vaddr = end_vaddr;
959 * The odds are that this will stop walking way
960 * before end_vaddr, because the length of the
961 * user buffer is tracked in "pm", and the walk
962 * will stop when we hit the end of the buffer.
965 while (count && (start_vaddr < end_vaddr)) {
970 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
972 if (end < start_vaddr || end > end_vaddr)
974 down_read(&mm->mmap_sem);
975 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
976 up_read(&mm->mmap_sem);
979 len = min(count, PM_ENTRY_BYTES * pm.pos);
980 if (copy_to_user(buf, pm.buffer, len)) {
989 if (!ret || ret == PM_END_OF_BUFFER)
997 put_task_struct(task);
1002 const struct file_operations proc_pagemap_operations = {
1003 .llseek = mem_lseek, /* borrow this */
1004 .read = pagemap_read,
1006 #endif /* CONFIG_PROC_PAGE_MONITOR */
1011 struct vm_area_struct *vma;
1012 unsigned long pages;
1014 unsigned long active;
1015 unsigned long writeback;
1016 unsigned long mapcount_max;
1017 unsigned long dirty;
1018 unsigned long swapcache;
1019 unsigned long node[MAX_NUMNODES];
1022 struct numa_maps_private {
1023 struct proc_maps_private proc_maps;
1024 struct numa_maps md;
1027 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1028 unsigned long nr_pages)
1030 int count = page_mapcount(page);
1032 md->pages += nr_pages;
1033 if (pte_dirty || PageDirty(page))
1034 md->dirty += nr_pages;
1036 if (PageSwapCache(page))
1037 md->swapcache += nr_pages;
1039 if (PageActive(page) || PageUnevictable(page))
1040 md->active += nr_pages;
1042 if (PageWriteback(page))
1043 md->writeback += nr_pages;
1046 md->anon += nr_pages;
1048 if (count > md->mapcount_max)
1049 md->mapcount_max = count;
1051 md->node[page_to_nid(page)] += nr_pages;
1054 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1060 if (!pte_present(pte))
1063 page = vm_normal_page(vma, addr, pte);
1067 if (PageReserved(page))
1070 nid = page_to_nid(page);
1071 if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
1077 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1078 unsigned long end, struct mm_walk *walk)
1080 struct numa_maps *md;
1087 if (pmd_trans_huge_lock(pmd, md->vma) == 1) {
1088 pte_t huge_pte = *(pte_t *)pmd;
1091 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1093 gather_stats(page, md, pte_dirty(huge_pte),
1094 HPAGE_PMD_SIZE/PAGE_SIZE);
1095 spin_unlock(&walk->mm->page_table_lock);
1099 if (pmd_trans_unstable(pmd))
1101 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1103 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1106 gather_stats(page, md, pte_dirty(*pte), 1);
1108 } while (pte++, addr += PAGE_SIZE, addr != end);
1109 pte_unmap_unlock(orig_pte, ptl);
1112 #ifdef CONFIG_HUGETLB_PAGE
1113 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1114 unsigned long addr, unsigned long end, struct mm_walk *walk)
1116 struct numa_maps *md;
1122 page = pte_page(*pte);
1127 gather_stats(page, md, pte_dirty(*pte), 1);
1132 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1133 unsigned long addr, unsigned long end, struct mm_walk *walk)
1140 * Display pages allocated per node and memory policy via /proc.
1142 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1144 struct numa_maps_private *numa_priv = m->private;
1145 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1146 struct vm_area_struct *vma = v;
1147 struct numa_maps *md = &numa_priv->md;
1148 struct file *file = vma->vm_file;
1149 struct mm_struct *mm = vma->vm_mm;
1150 struct mm_walk walk = {};
1151 struct mempolicy *pol;
1158 /* Ensure we start with an empty set of numa_maps statistics. */
1159 memset(md, 0, sizeof(*md));
1163 walk.hugetlb_entry = gather_hugetbl_stats;
1164 walk.pmd_entry = gather_pte_stats;
1168 pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
1169 mpol_to_str(buffer, sizeof(buffer), pol, 0);
1172 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1175 seq_printf(m, " file=");
1176 seq_path(m, &file->f_path, "\n\t= ");
1177 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1178 seq_printf(m, " heap");
1180 pid_t tid = vm_is_stack(proc_priv->task, vma, is_pid);
1183 * Thread stack in /proc/PID/task/TID/maps or
1184 * the main process stack.
1186 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1187 vma->vm_end >= mm->start_stack))
1188 seq_printf(m, " stack");
1190 seq_printf(m, " stack:%d", tid);
1194 if (is_vm_hugetlb_page(vma))
1195 seq_printf(m, " huge");
1197 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1203 seq_printf(m, " anon=%lu", md->anon);
1206 seq_printf(m, " dirty=%lu", md->dirty);
1208 if (md->pages != md->anon && md->pages != md->dirty)
1209 seq_printf(m, " mapped=%lu", md->pages);
1211 if (md->mapcount_max > 1)
1212 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1215 seq_printf(m, " swapcache=%lu", md->swapcache);
1217 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1218 seq_printf(m, " active=%lu", md->active);
1221 seq_printf(m, " writeback=%lu", md->writeback);
1223 for_each_node_state(n, N_HIGH_MEMORY)
1225 seq_printf(m, " N%d=%lu", n, md->node[n]);
1229 if (m->count < m->size)
1230 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1234 static int show_pid_numa_map(struct seq_file *m, void *v)
1236 return show_numa_map(m, v, 1);
1239 static int show_tid_numa_map(struct seq_file *m, void *v)
1241 return show_numa_map(m, v, 0);
1244 static const struct seq_operations proc_pid_numa_maps_op = {
1248 .show = show_pid_numa_map,
1251 static const struct seq_operations proc_tid_numa_maps_op = {
1255 .show = show_tid_numa_map,
1258 static int numa_maps_open(struct inode *inode, struct file *file,
1259 const struct seq_operations *ops)
1261 struct numa_maps_private *priv;
1263 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1265 priv->proc_maps.pid = proc_pid(inode);
1266 ret = seq_open(file, ops);
1268 struct seq_file *m = file->private_data;
1277 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1279 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1282 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1284 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1287 const struct file_operations proc_pid_numa_maps_operations = {
1288 .open = pid_numa_maps_open,
1290 .llseek = seq_lseek,
1291 .release = seq_release_private,
1294 const struct file_operations proc_tid_numa_maps_operations = {
1295 .open = tid_numa_maps_open,
1297 .llseek = seq_lseek,
1298 .release = seq_release_private,
1300 #endif /* CONFIG_NUMA */