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);
55 hiwater_vm << (PAGE_SHIFT-10),
56 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
57 mm->locked_vm << (PAGE_SHIFT-10),
58 hiwater_rss << (PAGE_SHIFT-10),
59 total_rss << (PAGE_SHIFT-10),
60 data << (PAGE_SHIFT-10),
61 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
62 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
63 swap << (PAGE_SHIFT-10));
66 unsigned long task_vsize(struct mm_struct *mm)
68 return PAGE_SIZE * mm->total_vm;
71 unsigned long task_statm(struct mm_struct *mm,
72 unsigned long *shared, unsigned long *text,
73 unsigned long *data, unsigned long *resident)
75 *shared = get_mm_counter(mm, MM_FILEPAGES);
76 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
78 *data = mm->total_vm - mm->shared_vm;
79 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
83 static void pad_len_spaces(struct seq_file *m, int len)
85 len = 25 + sizeof(void*) * 6 - len;
88 seq_printf(m, "%*c", len, ' ');
91 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
93 if (vma && vma != priv->tail_vma) {
94 struct mm_struct *mm = vma->vm_mm;
95 up_read(&mm->mmap_sem);
100 static void *m_start(struct seq_file *m, loff_t *pos)
102 struct proc_maps_private *priv = m->private;
103 unsigned long last_addr = m->version;
104 struct mm_struct *mm;
105 struct vm_area_struct *vma, *tail_vma = NULL;
108 /* Clear the per syscall fields in priv */
110 priv->tail_vma = NULL;
113 * We remember last_addr rather than next_addr to hit with
114 * mmap_cache most of the time. We have zero last_addr at
115 * the beginning and also after lseek. We will have -1 last_addr
116 * after the end of the vmas.
119 if (last_addr == -1UL)
122 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
124 return ERR_PTR(-ESRCH);
126 mm = mm_for_maps(priv->task);
127 if (!mm || IS_ERR(mm))
129 down_read(&mm->mmap_sem);
131 tail_vma = get_gate_vma(priv->task->mm);
132 priv->tail_vma = tail_vma;
134 /* Start with last addr hint */
135 vma = find_vma(mm, last_addr);
136 if (last_addr && vma) {
142 * Check the vma index is within the range and do
143 * sequential scan until m_index.
146 if ((unsigned long)l < mm->map_count) {
153 if (l != mm->map_count)
154 tail_vma = NULL; /* After gate vma */
160 /* End of vmas has been reached */
161 m->version = (tail_vma != NULL)? 0: -1UL;
162 up_read(&mm->mmap_sem);
167 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
169 struct proc_maps_private *priv = m->private;
170 struct vm_area_struct *vma = v;
171 struct vm_area_struct *tail_vma = priv->tail_vma;
174 if (vma && (vma != tail_vma) && vma->vm_next)
177 return (vma != tail_vma)? tail_vma: NULL;
180 static void m_stop(struct seq_file *m, void *v)
182 struct proc_maps_private *priv = m->private;
183 struct vm_area_struct *vma = v;
188 put_task_struct(priv->task);
191 static int do_maps_open(struct inode *inode, struct file *file,
192 const struct seq_operations *ops)
194 struct proc_maps_private *priv;
196 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
198 priv->pid = proc_pid(inode);
199 ret = seq_open(file, ops);
201 struct seq_file *m = file->private_data;
210 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
212 struct mm_struct *mm = vma->vm_mm;
213 struct file *file = vma->vm_file;
214 vm_flags_t flags = vma->vm_flags;
215 unsigned long ino = 0;
216 unsigned long long pgoff = 0;
217 unsigned long start, end;
222 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
223 dev = inode->i_sb->s_dev;
225 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
228 /* We don't show the stack guard page in /proc/maps */
229 start = vma->vm_start;
230 if (stack_guard_page_start(vma, start))
233 if (stack_guard_page_end(vma, end))
236 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
239 flags & VM_READ ? 'r' : '-',
240 flags & VM_WRITE ? 'w' : '-',
241 flags & VM_EXEC ? 'x' : '-',
242 flags & VM_MAYSHARE ? 's' : 'p',
244 MAJOR(dev), MINOR(dev), ino, &len);
247 * Print the dentry name for named mappings, and a
248 * special [heap] marker for the heap:
251 pad_len_spaces(m, len);
252 seq_path(m, &file->f_path, "\n");
254 const char *name = arch_vma_name(vma);
257 if (vma->vm_start <= mm->brk &&
258 vma->vm_end >= mm->start_brk) {
260 } else if (vma->vm_start <= mm->start_stack &&
261 vma->vm_end >= mm->start_stack) {
269 pad_len_spaces(m, len);
276 static int show_map(struct seq_file *m, void *v)
278 struct vm_area_struct *vma = v;
279 struct proc_maps_private *priv = m->private;
280 struct task_struct *task = priv->task;
282 show_map_vma(m, vma);
284 if (m->count < m->size) /* vma is copied successfully */
285 m->version = (vma != get_gate_vma(task->mm))
290 static const struct seq_operations proc_pid_maps_op = {
297 static int maps_open(struct inode *inode, struct file *file)
299 return do_maps_open(inode, file, &proc_pid_maps_op);
302 const struct file_operations proc_maps_operations = {
306 .release = seq_release_private,
310 * Proportional Set Size(PSS): my share of RSS.
312 * PSS of a process is the count of pages it has in memory, where each
313 * page is divided by the number of processes sharing it. So if a
314 * process has 1000 pages all to itself, and 1000 shared with one other
315 * process, its PSS will be 1500.
317 * To keep (accumulated) division errors low, we adopt a 64bit
318 * fixed-point pss counter to minimize division errors. So (pss >>
319 * PSS_SHIFT) would be the real byte count.
321 * A shift of 12 before division means (assuming 4K page size):
322 * - 1M 3-user-pages add up to 8KB errors;
323 * - supports mapcount up to 2^24, or 16M;
324 * - supports PSS up to 2^52 bytes, or 4PB.
328 #ifdef CONFIG_PROC_PAGE_MONITOR
329 struct mem_size_stats {
330 struct vm_area_struct *vma;
331 unsigned long resident;
332 unsigned long shared_clean;
333 unsigned long shared_dirty;
334 unsigned long private_clean;
335 unsigned long private_dirty;
336 unsigned long referenced;
337 unsigned long anonymous;
338 unsigned long anonymous_thp;
344 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
345 unsigned long ptent_size, struct mm_walk *walk)
347 struct mem_size_stats *mss = walk->private;
348 struct vm_area_struct *vma = mss->vma;
352 if (is_swap_pte(ptent)) {
353 mss->swap += ptent_size;
357 if (!pte_present(ptent))
360 page = vm_normal_page(vma, addr, ptent);
365 mss->anonymous += ptent_size;
367 mss->resident += ptent_size;
368 /* Accumulate the size in pages that have been accessed. */
369 if (pte_young(ptent) || PageReferenced(page))
370 mss->referenced += ptent_size;
371 mapcount = page_mapcount(page);
373 if (pte_dirty(ptent) || PageDirty(page))
374 mss->shared_dirty += ptent_size;
376 mss->shared_clean += ptent_size;
377 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
379 if (pte_dirty(ptent) || PageDirty(page))
380 mss->private_dirty += ptent_size;
382 mss->private_clean += ptent_size;
383 mss->pss += (ptent_size << PSS_SHIFT);
387 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
388 struct mm_walk *walk)
390 struct mem_size_stats *mss = walk->private;
391 struct vm_area_struct *vma = mss->vma;
395 spin_lock(&walk->mm->page_table_lock);
396 if (pmd_trans_huge(*pmd)) {
397 if (pmd_trans_splitting(*pmd)) {
398 spin_unlock(&walk->mm->page_table_lock);
399 wait_split_huge_page(vma->anon_vma, pmd);
401 smaps_pte_entry(*(pte_t *)pmd, addr,
402 HPAGE_PMD_SIZE, walk);
403 spin_unlock(&walk->mm->page_table_lock);
404 mss->anonymous_thp += HPAGE_PMD_SIZE;
408 spin_unlock(&walk->mm->page_table_lock);
411 if (pmd_trans_unstable(pmd))
414 * The mmap_sem held all the way back in m_start() is what
415 * keeps khugepaged out of here and from collapsing things
418 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
419 for (; addr != end; pte++, addr += PAGE_SIZE)
420 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
421 pte_unmap_unlock(pte - 1, ptl);
426 static int show_smap(struct seq_file *m, void *v)
428 struct proc_maps_private *priv = m->private;
429 struct task_struct *task = priv->task;
430 struct vm_area_struct *vma = v;
431 struct mem_size_stats mss;
432 struct mm_walk smaps_walk = {
433 .pmd_entry = smaps_pte_range,
438 memset(&mss, 0, sizeof mss);
440 /* mmap_sem is held in m_start */
441 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
442 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
444 show_map_vma(m, vma);
450 "Shared_Clean: %8lu kB\n"
451 "Shared_Dirty: %8lu kB\n"
452 "Private_Clean: %8lu kB\n"
453 "Private_Dirty: %8lu kB\n"
454 "Referenced: %8lu kB\n"
455 "Anonymous: %8lu kB\n"
456 "AnonHugePages: %8lu kB\n"
458 "KernelPageSize: %8lu kB\n"
459 "MMUPageSize: %8lu kB\n"
461 (vma->vm_end - vma->vm_start) >> 10,
463 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
464 mss.shared_clean >> 10,
465 mss.shared_dirty >> 10,
466 mss.private_clean >> 10,
467 mss.private_dirty >> 10,
468 mss.referenced >> 10,
470 mss.anonymous_thp >> 10,
472 vma_kernel_pagesize(vma) >> 10,
473 vma_mmu_pagesize(vma) >> 10,
474 (vma->vm_flags & VM_LOCKED) ?
475 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
477 if (m->count < m->size) /* vma is copied successfully */
478 m->version = (vma != get_gate_vma(task->mm))
483 static const struct seq_operations proc_pid_smaps_op = {
490 static int smaps_open(struct inode *inode, struct file *file)
492 return do_maps_open(inode, file, &proc_pid_smaps_op);
495 const struct file_operations proc_smaps_operations = {
499 .release = seq_release_private,
502 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
503 unsigned long end, struct mm_walk *walk)
505 struct vm_area_struct *vma = walk->private;
510 split_huge_page_pmd(walk->mm, pmd);
511 if (pmd_trans_unstable(pmd))
514 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
515 for (; addr != end; pte++, addr += PAGE_SIZE) {
517 if (!pte_present(ptent))
520 page = vm_normal_page(vma, addr, ptent);
524 if (PageReserved(page))
527 /* Clear accessed and referenced bits. */
528 ptep_test_and_clear_young(vma, addr, pte);
529 ClearPageReferenced(page);
531 pte_unmap_unlock(pte - 1, ptl);
536 #define CLEAR_REFS_ALL 1
537 #define CLEAR_REFS_ANON 2
538 #define CLEAR_REFS_MAPPED 3
540 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
541 size_t count, loff_t *ppos)
543 struct task_struct *task;
544 char buffer[PROC_NUMBUF];
545 struct mm_struct *mm;
546 struct vm_area_struct *vma;
550 memset(buffer, 0, sizeof(buffer));
551 if (count > sizeof(buffer) - 1)
552 count = sizeof(buffer) - 1;
553 if (copy_from_user(buffer, buf, count))
555 rv = kstrtoint(strstrip(buffer), 10, &type);
558 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
560 task = get_proc_task(file->f_path.dentry->d_inode);
563 mm = get_task_mm(task);
565 struct mm_walk clear_refs_walk = {
566 .pmd_entry = clear_refs_pte_range,
569 down_read(&mm->mmap_sem);
570 for (vma = mm->mmap; vma; vma = vma->vm_next) {
571 clear_refs_walk.private = vma;
572 if (is_vm_hugetlb_page(vma))
575 * Writing 1 to /proc/pid/clear_refs affects all pages.
577 * Writing 2 to /proc/pid/clear_refs only affects
580 * Writing 3 to /proc/pid/clear_refs only affects file
583 if (type == CLEAR_REFS_ANON && vma->vm_file)
585 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
587 walk_page_range(vma->vm_start, vma->vm_end,
591 up_read(&mm->mmap_sem);
594 put_task_struct(task);
599 const struct file_operations proc_clear_refs_operations = {
600 .write = clear_refs_write,
601 .llseek = noop_llseek,
609 #define PM_ENTRY_BYTES sizeof(u64)
610 #define PM_STATUS_BITS 3
611 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
612 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
613 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
614 #define PM_PSHIFT_BITS 6
615 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
616 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
617 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
618 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
619 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
621 #define PM_PRESENT PM_STATUS(4LL)
622 #define PM_SWAP PM_STATUS(2LL)
623 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
624 #define PM_END_OF_BUFFER 1
626 static int add_to_pagemap(unsigned long addr, u64 pfn,
627 struct pagemapread *pm)
629 pm->buffer[pm->pos++] = pfn;
630 if (pm->pos >= pm->len)
631 return PM_END_OF_BUFFER;
635 static int pagemap_pte_hole(unsigned long start, unsigned long end,
636 struct mm_walk *walk)
638 struct pagemapread *pm = walk->private;
641 for (addr = start; addr < end; addr += PAGE_SIZE) {
642 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
649 static u64 swap_pte_to_pagemap_entry(pte_t pte)
651 swp_entry_t e = pte_to_swp_entry(pte);
652 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
655 static u64 pte_to_pagemap_entry(pte_t pte)
658 if (is_swap_pte(pte))
659 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
660 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
661 else if (pte_present(pte))
662 pme = PM_PFRAME(pte_pfn(pte))
663 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
667 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
668 struct mm_walk *walk)
670 struct vm_area_struct *vma;
671 struct pagemapread *pm = walk->private;
675 split_huge_page_pmd(walk->mm, pmd);
676 if (pmd_trans_unstable(pmd))
679 /* find the first VMA at or above 'addr' */
680 vma = find_vma(walk->mm, addr);
681 for (; addr != end; addr += PAGE_SIZE) {
682 u64 pfn = PM_NOT_PRESENT;
684 /* check to see if we've left 'vma' behind
685 * and need a new, higher one */
686 if (vma && (addr >= vma->vm_end))
687 vma = find_vma(walk->mm, addr);
689 /* check that 'vma' actually covers this address,
690 * and that it isn't a huge page vma */
691 if (vma && (vma->vm_start <= addr) &&
692 !is_vm_hugetlb_page(vma)) {
693 pte = pte_offset_map(pmd, addr);
694 pfn = pte_to_pagemap_entry(*pte);
695 /* unmap before userspace copy */
698 err = add_to_pagemap(addr, pfn, pm);
708 #ifdef CONFIG_HUGETLB_PAGE
709 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
712 if (pte_present(pte))
713 pme = PM_PFRAME(pte_pfn(pte) + offset)
714 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
718 /* This function walks within one hugetlb entry in the single call */
719 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
720 unsigned long addr, unsigned long end,
721 struct mm_walk *walk)
723 struct pagemapread *pm = walk->private;
727 for (; addr != end; addr += PAGE_SIZE) {
728 int offset = (addr & ~hmask) >> PAGE_SHIFT;
729 pfn = huge_pte_to_pagemap_entry(*pte, offset);
730 err = add_to_pagemap(addr, pfn, pm);
739 #endif /* HUGETLB_PAGE */
742 * /proc/pid/pagemap - an array mapping virtual pages to pfns
744 * For each page in the address space, this file contains one 64-bit entry
745 * consisting of the following:
747 * Bits 0-55 page frame number (PFN) if present
748 * Bits 0-4 swap type if swapped
749 * Bits 5-55 swap offset if swapped
750 * Bits 55-60 page shift (page size = 1<<page shift)
751 * Bit 61 reserved for future use
752 * Bit 62 page swapped
753 * Bit 63 page present
755 * If the page is not present but in swap, then the PFN contains an
756 * encoding of the swap file number and the page's offset into the
757 * swap. Unmapped pages return a null PFN. This allows determining
758 * precisely which pages are mapped (or in swap) and comparing mapped
759 * pages between processes.
761 * Efficient users of this interface will use /proc/pid/maps to
762 * determine which areas of memory are actually mapped and llseek to
763 * skip over unmapped regions.
765 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
766 #define PAGEMAP_WALK_MASK (PMD_MASK)
767 static ssize_t pagemap_read(struct file *file, char __user *buf,
768 size_t count, loff_t *ppos)
770 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
771 struct mm_struct *mm;
772 struct pagemapread pm;
774 struct mm_walk pagemap_walk = {};
777 unsigned long start_vaddr;
778 unsigned long end_vaddr;
785 /* file position must be aligned */
786 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
793 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
794 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
799 mm = mm_for_maps(task);
801 if (!mm || IS_ERR(mm))
804 pagemap_walk.pmd_entry = pagemap_pte_range;
805 pagemap_walk.pte_hole = pagemap_pte_hole;
806 #ifdef CONFIG_HUGETLB_PAGE
807 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
809 pagemap_walk.mm = mm;
810 pagemap_walk.private = ±
813 svpfn = src / PM_ENTRY_BYTES;
814 start_vaddr = svpfn << PAGE_SHIFT;
815 end_vaddr = TASK_SIZE_OF(task);
817 /* watch out for wraparound */
818 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
819 start_vaddr = end_vaddr;
822 * The odds are that this will stop walking way
823 * before end_vaddr, because the length of the
824 * user buffer is tracked in "pm", and the walk
825 * will stop when we hit the end of the buffer.
828 while (count && (start_vaddr < end_vaddr)) {
833 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
835 if (end < start_vaddr || end > end_vaddr)
837 down_read(&mm->mmap_sem);
838 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
839 up_read(&mm->mmap_sem);
842 len = min(count, PM_ENTRY_BYTES * pm.pos);
843 if (copy_to_user(buf, pm.buffer, len)) {
852 if (!ret || ret == PM_END_OF_BUFFER)
860 put_task_struct(task);
865 const struct file_operations proc_pagemap_operations = {
866 .llseek = mem_lseek, /* borrow this */
867 .read = pagemap_read,
869 #endif /* CONFIG_PROC_PAGE_MONITOR */
874 struct vm_area_struct *vma;
877 unsigned long active;
878 unsigned long writeback;
879 unsigned long mapcount_max;
881 unsigned long swapcache;
882 unsigned long node[MAX_NUMNODES];
885 struct numa_maps_private {
886 struct proc_maps_private proc_maps;
890 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
891 unsigned long nr_pages)
893 int count = page_mapcount(page);
895 md->pages += nr_pages;
896 if (pte_dirty || PageDirty(page))
897 md->dirty += nr_pages;
899 if (PageSwapCache(page))
900 md->swapcache += nr_pages;
902 if (PageActive(page) || PageUnevictable(page))
903 md->active += nr_pages;
905 if (PageWriteback(page))
906 md->writeback += nr_pages;
909 md->anon += nr_pages;
911 if (count > md->mapcount_max)
912 md->mapcount_max = count;
914 md->node[page_to_nid(page)] += nr_pages;
917 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
923 if (!pte_present(pte))
926 page = vm_normal_page(vma, addr, pte);
930 if (PageReserved(page))
933 nid = page_to_nid(page);
934 if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
940 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
941 unsigned long end, struct mm_walk *walk)
943 struct numa_maps *md;
949 spin_lock(&walk->mm->page_table_lock);
950 if (pmd_trans_huge(*pmd)) {
951 if (pmd_trans_splitting(*pmd)) {
952 spin_unlock(&walk->mm->page_table_lock);
953 wait_split_huge_page(md->vma->anon_vma, pmd);
955 pte_t huge_pte = *(pte_t *)pmd;
958 page = can_gather_numa_stats(huge_pte, md->vma, addr);
960 gather_stats(page, md, pte_dirty(huge_pte),
961 HPAGE_PMD_SIZE/PAGE_SIZE);
962 spin_unlock(&walk->mm->page_table_lock);
966 spin_unlock(&walk->mm->page_table_lock);
969 if (pmd_trans_unstable(pmd))
971 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
973 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
976 gather_stats(page, md, pte_dirty(*pte), 1);
978 } while (pte++, addr += PAGE_SIZE, addr != end);
979 pte_unmap_unlock(orig_pte, ptl);
982 #ifdef CONFIG_HUGETLB_PAGE
983 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
984 unsigned long addr, unsigned long end, struct mm_walk *walk)
986 struct numa_maps *md;
992 page = pte_page(*pte);
997 gather_stats(page, md, pte_dirty(*pte), 1);
1002 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1003 unsigned long addr, unsigned long end, struct mm_walk *walk)
1010 * Display pages allocated per node and memory policy via /proc.
1012 static int show_numa_map(struct seq_file *m, void *v)
1014 struct numa_maps_private *numa_priv = m->private;
1015 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1016 struct vm_area_struct *vma = v;
1017 struct numa_maps *md = &numa_priv->md;
1018 struct file *file = vma->vm_file;
1019 struct mm_struct *mm = vma->vm_mm;
1020 struct mm_walk walk = {};
1021 struct mempolicy *pol;
1028 /* Ensure we start with an empty set of numa_maps statistics. */
1029 memset(md, 0, sizeof(*md));
1033 walk.hugetlb_entry = gather_hugetbl_stats;
1034 walk.pmd_entry = gather_pte_stats;
1038 pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
1039 mpol_to_str(buffer, sizeof(buffer), pol, 0);
1042 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1045 seq_printf(m, " file=");
1046 seq_path(m, &file->f_path, "\n\t= ");
1047 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1048 seq_printf(m, " heap");
1049 } else if (vma->vm_start <= mm->start_stack &&
1050 vma->vm_end >= mm->start_stack) {
1051 seq_printf(m, " stack");
1054 if (is_vm_hugetlb_page(vma))
1055 seq_printf(m, " huge");
1057 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1063 seq_printf(m, " anon=%lu", md->anon);
1066 seq_printf(m, " dirty=%lu", md->dirty);
1068 if (md->pages != md->anon && md->pages != md->dirty)
1069 seq_printf(m, " mapped=%lu", md->pages);
1071 if (md->mapcount_max > 1)
1072 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1075 seq_printf(m, " swapcache=%lu", md->swapcache);
1077 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1078 seq_printf(m, " active=%lu", md->active);
1081 seq_printf(m, " writeback=%lu", md->writeback);
1083 for_each_node_state(n, N_HIGH_MEMORY)
1085 seq_printf(m, " N%d=%lu", n, md->node[n]);
1089 if (m->count < m->size)
1090 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1094 static const struct seq_operations proc_pid_numa_maps_op = {
1098 .show = show_numa_map,
1101 static int numa_maps_open(struct inode *inode, struct file *file)
1103 struct numa_maps_private *priv;
1105 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1107 priv->proc_maps.pid = proc_pid(inode);
1108 ret = seq_open(file, &proc_pid_numa_maps_op);
1110 struct seq_file *m = file->private_data;
1119 const struct file_operations proc_numa_maps_operations = {
1120 .open = numa_maps_open,
1122 .llseek = seq_lseek,
1123 .release = seq_release_private,
1125 #endif /* CONFIG_NUMA */