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_for_maps(priv->task);
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;
212 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
214 struct mm_struct *mm = vma->vm_mm;
215 struct file *file = vma->vm_file;
216 vm_flags_t flags = vma->vm_flags;
217 unsigned long ino = 0;
218 unsigned long long pgoff = 0;
219 unsigned long start, end;
224 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
225 dev = inode->i_sb->s_dev;
227 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
230 /* We don't show the stack guard page in /proc/maps */
231 start = vma->vm_start;
232 if (stack_guard_page_start(vma, start))
235 if (stack_guard_page_end(vma, end))
238 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
241 flags & VM_READ ? 'r' : '-',
242 flags & VM_WRITE ? 'w' : '-',
243 flags & VM_EXEC ? 'x' : '-',
244 flags & VM_MAYSHARE ? 's' : 'p',
246 MAJOR(dev), MINOR(dev), ino, &len);
249 * Print the dentry name for named mappings, and a
250 * special [heap] marker for the heap:
253 pad_len_spaces(m, len);
254 seq_path(m, &file->f_path, "\n");
256 const char *name = arch_vma_name(vma);
259 if (vma->vm_start <= mm->brk &&
260 vma->vm_end >= mm->start_brk) {
262 } else if (vma->vm_start <= mm->start_stack &&
263 vma->vm_end >= mm->start_stack) {
271 pad_len_spaces(m, len);
278 static int show_map(struct seq_file *m, void *v)
280 struct vm_area_struct *vma = v;
281 struct proc_maps_private *priv = m->private;
282 struct task_struct *task = priv->task;
284 show_map_vma(m, vma);
286 if (m->count < m->size) /* vma is copied successfully */
287 m->version = (vma != get_gate_vma(task->mm))
292 static const struct seq_operations proc_pid_maps_op = {
299 static int maps_open(struct inode *inode, struct file *file)
301 return do_maps_open(inode, file, &proc_pid_maps_op);
304 const struct file_operations proc_maps_operations = {
308 .release = seq_release_private,
312 * Proportional Set Size(PSS): my share of RSS.
314 * PSS of a process is the count of pages it has in memory, where each
315 * page is divided by the number of processes sharing it. So if a
316 * process has 1000 pages all to itself, and 1000 shared with one other
317 * process, its PSS will be 1500.
319 * To keep (accumulated) division errors low, we adopt a 64bit
320 * fixed-point pss counter to minimize division errors. So (pss >>
321 * PSS_SHIFT) would be the real byte count.
323 * A shift of 12 before division means (assuming 4K page size):
324 * - 1M 3-user-pages add up to 8KB errors;
325 * - supports mapcount up to 2^24, or 16M;
326 * - supports PSS up to 2^52 bytes, or 4PB.
330 #ifdef CONFIG_PROC_PAGE_MONITOR
331 struct mem_size_stats {
332 struct vm_area_struct *vma;
333 unsigned long resident;
334 unsigned long shared_clean;
335 unsigned long shared_dirty;
336 unsigned long private_clean;
337 unsigned long private_dirty;
338 unsigned long referenced;
339 unsigned long anonymous;
340 unsigned long anonymous_thp;
346 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
347 unsigned long ptent_size, struct mm_walk *walk)
349 struct mem_size_stats *mss = walk->private;
350 struct vm_area_struct *vma = mss->vma;
354 if (is_swap_pte(ptent)) {
355 mss->swap += ptent_size;
359 if (!pte_present(ptent))
362 page = vm_normal_page(vma, addr, ptent);
367 mss->anonymous += ptent_size;
369 mss->resident += ptent_size;
370 /* Accumulate the size in pages that have been accessed. */
371 if (pte_young(ptent) || PageReferenced(page))
372 mss->referenced += ptent_size;
373 mapcount = page_mapcount(page);
375 if (pte_dirty(ptent) || PageDirty(page))
376 mss->shared_dirty += ptent_size;
378 mss->shared_clean += ptent_size;
379 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
381 if (pte_dirty(ptent) || PageDirty(page))
382 mss->private_dirty += ptent_size;
384 mss->private_clean += ptent_size;
385 mss->pss += (ptent_size << PSS_SHIFT);
389 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
390 struct mm_walk *walk)
392 struct mem_size_stats *mss = walk->private;
393 struct vm_area_struct *vma = mss->vma;
397 spin_lock(&walk->mm->page_table_lock);
398 if (pmd_trans_huge(*pmd)) {
399 if (pmd_trans_splitting(*pmd)) {
400 spin_unlock(&walk->mm->page_table_lock);
401 wait_split_huge_page(vma->anon_vma, pmd);
403 smaps_pte_entry(*(pte_t *)pmd, addr,
404 HPAGE_PMD_SIZE, walk);
405 spin_unlock(&walk->mm->page_table_lock);
406 mss->anonymous_thp += HPAGE_PMD_SIZE;
410 spin_unlock(&walk->mm->page_table_lock);
413 * The mmap_sem held all the way back in m_start() is what
414 * keeps khugepaged out of here and from collapsing things
417 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
418 for (; addr != end; pte++, addr += PAGE_SIZE)
419 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
420 pte_unmap_unlock(pte - 1, ptl);
425 static int show_smap(struct seq_file *m, void *v)
427 struct proc_maps_private *priv = m->private;
428 struct task_struct *task = priv->task;
429 struct vm_area_struct *vma = v;
430 struct mem_size_stats mss;
431 struct mm_walk smaps_walk = {
432 .pmd_entry = smaps_pte_range,
437 memset(&mss, 0, sizeof mss);
439 /* mmap_sem is held in m_start */
440 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
441 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
443 show_map_vma(m, vma);
449 "Shared_Clean: %8lu kB\n"
450 "Shared_Dirty: %8lu kB\n"
451 "Private_Clean: %8lu kB\n"
452 "Private_Dirty: %8lu kB\n"
453 "Referenced: %8lu kB\n"
454 "Anonymous: %8lu kB\n"
455 "AnonHugePages: %8lu kB\n"
457 "KernelPageSize: %8lu kB\n"
458 "MMUPageSize: %8lu kB\n"
460 (vma->vm_end - vma->vm_start) >> 10,
462 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
463 mss.shared_clean >> 10,
464 mss.shared_dirty >> 10,
465 mss.private_clean >> 10,
466 mss.private_dirty >> 10,
467 mss.referenced >> 10,
469 mss.anonymous_thp >> 10,
471 vma_kernel_pagesize(vma) >> 10,
472 vma_mmu_pagesize(vma) >> 10,
473 (vma->vm_flags & VM_LOCKED) ?
474 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
476 if (m->count < m->size) /* vma is copied successfully */
477 m->version = (vma != get_gate_vma(task->mm))
482 static const struct seq_operations proc_pid_smaps_op = {
489 static int smaps_open(struct inode *inode, struct file *file)
491 return do_maps_open(inode, file, &proc_pid_smaps_op);
494 const struct file_operations proc_smaps_operations = {
498 .release = seq_release_private,
501 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
502 unsigned long end, struct mm_walk *walk)
504 struct vm_area_struct *vma = walk->private;
509 split_huge_page_pmd(walk->mm, pmd);
511 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
512 for (; addr != end; pte++, addr += PAGE_SIZE) {
514 if (!pte_present(ptent))
517 page = vm_normal_page(vma, addr, ptent);
521 /* Clear accessed and referenced bits. */
522 ptep_test_and_clear_young(vma, addr, pte);
523 ClearPageReferenced(page);
525 pte_unmap_unlock(pte - 1, ptl);
530 #define CLEAR_REFS_ALL 1
531 #define CLEAR_REFS_ANON 2
532 #define CLEAR_REFS_MAPPED 3
534 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
535 size_t count, loff_t *ppos)
537 struct task_struct *task;
538 char buffer[PROC_NUMBUF];
539 struct mm_struct *mm;
540 struct vm_area_struct *vma;
544 memset(buffer, 0, sizeof(buffer));
545 if (count > sizeof(buffer) - 1)
546 count = sizeof(buffer) - 1;
547 if (copy_from_user(buffer, buf, count))
549 rv = kstrtoint(strstrip(buffer), 10, &type);
552 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
554 task = get_proc_task(file->f_path.dentry->d_inode);
557 mm = get_task_mm(task);
559 struct mm_walk clear_refs_walk = {
560 .pmd_entry = clear_refs_pte_range,
563 down_read(&mm->mmap_sem);
564 for (vma = mm->mmap; vma; vma = vma->vm_next) {
565 clear_refs_walk.private = vma;
566 if (is_vm_hugetlb_page(vma))
569 * Writing 1 to /proc/pid/clear_refs affects all pages.
571 * Writing 2 to /proc/pid/clear_refs only affects
574 * Writing 3 to /proc/pid/clear_refs only affects file
577 if (type == CLEAR_REFS_ANON && vma->vm_file)
579 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
581 walk_page_range(vma->vm_start, vma->vm_end,
585 up_read(&mm->mmap_sem);
588 put_task_struct(task);
593 const struct file_operations proc_clear_refs_operations = {
594 .write = clear_refs_write,
595 .llseek = noop_llseek,
603 #define PM_ENTRY_BYTES sizeof(u64)
604 #define PM_STATUS_BITS 3
605 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
606 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
607 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
608 #define PM_PSHIFT_BITS 6
609 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
610 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
611 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
612 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
613 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
615 #define PM_PRESENT PM_STATUS(4LL)
616 #define PM_SWAP PM_STATUS(2LL)
617 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
618 #define PM_END_OF_BUFFER 1
620 static int add_to_pagemap(unsigned long addr, u64 pfn,
621 struct pagemapread *pm)
623 pm->buffer[pm->pos++] = pfn;
624 if (pm->pos >= pm->len)
625 return PM_END_OF_BUFFER;
629 static int pagemap_pte_hole(unsigned long start, unsigned long end,
630 struct mm_walk *walk)
632 struct pagemapread *pm = walk->private;
635 for (addr = start; addr < end; addr += PAGE_SIZE) {
636 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
643 static u64 swap_pte_to_pagemap_entry(pte_t pte)
645 swp_entry_t e = pte_to_swp_entry(pte);
646 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
649 static u64 pte_to_pagemap_entry(pte_t pte)
652 if (is_swap_pte(pte))
653 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
654 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
655 else if (pte_present(pte))
656 pme = PM_PFRAME(pte_pfn(pte))
657 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
661 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
662 struct mm_walk *walk)
664 struct vm_area_struct *vma;
665 struct pagemapread *pm = walk->private;
669 split_huge_page_pmd(walk->mm, pmd);
671 /* find the first VMA at or above 'addr' */
672 vma = find_vma(walk->mm, addr);
673 for (; addr != end; addr += PAGE_SIZE) {
674 u64 pfn = PM_NOT_PRESENT;
676 /* check to see if we've left 'vma' behind
677 * and need a new, higher one */
678 if (vma && (addr >= vma->vm_end))
679 vma = find_vma(walk->mm, addr);
681 /* check that 'vma' actually covers this address,
682 * and that it isn't a huge page vma */
683 if (vma && (vma->vm_start <= addr) &&
684 !is_vm_hugetlb_page(vma)) {
685 pte = pte_offset_map(pmd, addr);
686 pfn = pte_to_pagemap_entry(*pte);
687 /* unmap before userspace copy */
690 err = add_to_pagemap(addr, pfn, pm);
700 #ifdef CONFIG_HUGETLB_PAGE
701 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
704 if (pte_present(pte))
705 pme = PM_PFRAME(pte_pfn(pte) + offset)
706 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
710 /* This function walks within one hugetlb entry in the single call */
711 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
712 unsigned long addr, unsigned long end,
713 struct mm_walk *walk)
715 struct pagemapread *pm = walk->private;
719 for (; addr != end; addr += PAGE_SIZE) {
720 int offset = (addr & ~hmask) >> PAGE_SHIFT;
721 pfn = huge_pte_to_pagemap_entry(*pte, offset);
722 err = add_to_pagemap(addr, pfn, pm);
731 #endif /* HUGETLB_PAGE */
734 * /proc/pid/pagemap - an array mapping virtual pages to pfns
736 * For each page in the address space, this file contains one 64-bit entry
737 * consisting of the following:
739 * Bits 0-55 page frame number (PFN) if present
740 * Bits 0-4 swap type if swapped
741 * Bits 5-55 swap offset if swapped
742 * Bits 55-60 page shift (page size = 1<<page shift)
743 * Bit 61 reserved for future use
744 * Bit 62 page swapped
745 * Bit 63 page present
747 * If the page is not present but in swap, then the PFN contains an
748 * encoding of the swap file number and the page's offset into the
749 * swap. Unmapped pages return a null PFN. This allows determining
750 * precisely which pages are mapped (or in swap) and comparing mapped
751 * pages between processes.
753 * Efficient users of this interface will use /proc/pid/maps to
754 * determine which areas of memory are actually mapped and llseek to
755 * skip over unmapped regions.
757 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
758 #define PAGEMAP_WALK_MASK (PMD_MASK)
759 static ssize_t pagemap_read(struct file *file, char __user *buf,
760 size_t count, loff_t *ppos)
762 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
763 struct mm_struct *mm;
764 struct pagemapread pm;
766 struct mm_walk pagemap_walk = {};
769 unsigned long start_vaddr;
770 unsigned long end_vaddr;
777 /* file position must be aligned */
778 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
785 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
786 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
791 mm = mm_for_maps(task);
793 if (!mm || IS_ERR(mm))
796 pagemap_walk.pmd_entry = pagemap_pte_range;
797 pagemap_walk.pte_hole = pagemap_pte_hole;
798 #ifdef CONFIG_HUGETLB_PAGE
799 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
801 pagemap_walk.mm = mm;
802 pagemap_walk.private = ±
805 svpfn = src / PM_ENTRY_BYTES;
806 start_vaddr = svpfn << PAGE_SHIFT;
807 end_vaddr = TASK_SIZE_OF(task);
809 /* watch out for wraparound */
810 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
811 start_vaddr = end_vaddr;
814 * The odds are that this will stop walking way
815 * before end_vaddr, because the length of the
816 * user buffer is tracked in "pm", and the walk
817 * will stop when we hit the end of the buffer.
820 while (count && (start_vaddr < end_vaddr)) {
825 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
827 if (end < start_vaddr || end > end_vaddr)
829 down_read(&mm->mmap_sem);
830 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
831 up_read(&mm->mmap_sem);
834 len = min(count, PM_ENTRY_BYTES * pm.pos);
835 if (copy_to_user(buf, pm.buffer, len)) {
844 if (!ret || ret == PM_END_OF_BUFFER)
852 put_task_struct(task);
857 const struct file_operations proc_pagemap_operations = {
858 .llseek = mem_lseek, /* borrow this */
859 .read = pagemap_read,
861 #endif /* CONFIG_PROC_PAGE_MONITOR */
866 struct vm_area_struct *vma;
869 unsigned long active;
870 unsigned long writeback;
871 unsigned long mapcount_max;
873 unsigned long swapcache;
874 unsigned long node[MAX_NUMNODES];
877 struct numa_maps_private {
878 struct proc_maps_private proc_maps;
882 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
883 unsigned long nr_pages)
885 int count = page_mapcount(page);
887 md->pages += nr_pages;
888 if (pte_dirty || PageDirty(page))
889 md->dirty += nr_pages;
891 if (PageSwapCache(page))
892 md->swapcache += nr_pages;
894 if (PageActive(page) || PageUnevictable(page))
895 md->active += nr_pages;
897 if (PageWriteback(page))
898 md->writeback += nr_pages;
901 md->anon += nr_pages;
903 if (count > md->mapcount_max)
904 md->mapcount_max = count;
906 md->node[page_to_nid(page)] += nr_pages;
909 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
915 if (!pte_present(pte))
918 page = vm_normal_page(vma, addr, pte);
922 if (PageReserved(page))
925 nid = page_to_nid(page);
926 if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
932 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
933 unsigned long end, struct mm_walk *walk)
935 struct numa_maps *md;
941 spin_lock(&walk->mm->page_table_lock);
942 if (pmd_trans_huge(*pmd)) {
943 if (pmd_trans_splitting(*pmd)) {
944 spin_unlock(&walk->mm->page_table_lock);
945 wait_split_huge_page(md->vma->anon_vma, pmd);
947 pte_t huge_pte = *(pte_t *)pmd;
950 page = can_gather_numa_stats(huge_pte, md->vma, addr);
952 gather_stats(page, md, pte_dirty(huge_pte),
953 HPAGE_PMD_SIZE/PAGE_SIZE);
954 spin_unlock(&walk->mm->page_table_lock);
958 spin_unlock(&walk->mm->page_table_lock);
961 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
963 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
966 gather_stats(page, md, pte_dirty(*pte), 1);
968 } while (pte++, addr += PAGE_SIZE, addr != end);
969 pte_unmap_unlock(orig_pte, ptl);
972 #ifdef CONFIG_HUGETLB_PAGE
973 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
974 unsigned long addr, unsigned long end, struct mm_walk *walk)
976 struct numa_maps *md;
982 page = pte_page(*pte);
987 gather_stats(page, md, pte_dirty(*pte), 1);
992 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
993 unsigned long addr, unsigned long end, struct mm_walk *walk)
1000 * Display pages allocated per node and memory policy via /proc.
1002 static int show_numa_map(struct seq_file *m, void *v)
1004 struct numa_maps_private *numa_priv = m->private;
1005 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1006 struct vm_area_struct *vma = v;
1007 struct numa_maps *md = &numa_priv->md;
1008 struct file *file = vma->vm_file;
1009 struct mm_struct *mm = vma->vm_mm;
1010 struct mm_walk walk = {};
1011 struct mempolicy *pol;
1018 /* Ensure we start with an empty set of numa_maps statistics. */
1019 memset(md, 0, sizeof(*md));
1023 walk.hugetlb_entry = gather_hugetbl_stats;
1024 walk.pmd_entry = gather_pte_stats;
1028 pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
1029 mpol_to_str(buffer, sizeof(buffer), pol, 0);
1032 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1035 seq_printf(m, " file=");
1036 seq_path(m, &file->f_path, "\n\t= ");
1037 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1038 seq_printf(m, " heap");
1039 } else if (vma->vm_start <= mm->start_stack &&
1040 vma->vm_end >= mm->start_stack) {
1041 seq_printf(m, " stack");
1044 if (is_vm_hugetlb_page(vma))
1045 seq_printf(m, " huge");
1047 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1053 seq_printf(m, " anon=%lu", md->anon);
1056 seq_printf(m, " dirty=%lu", md->dirty);
1058 if (md->pages != md->anon && md->pages != md->dirty)
1059 seq_printf(m, " mapped=%lu", md->pages);
1061 if (md->mapcount_max > 1)
1062 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1065 seq_printf(m, " swapcache=%lu", md->swapcache);
1067 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1068 seq_printf(m, " active=%lu", md->active);
1071 seq_printf(m, " writeback=%lu", md->writeback);
1073 for_each_node_state(n, N_HIGH_MEMORY)
1075 seq_printf(m, " N%d=%lu", n, md->node[n]);
1079 if (m->count < m->size)
1080 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1084 static const struct seq_operations proc_pid_numa_maps_op = {
1088 .show = show_numa_map,
1091 static int numa_maps_open(struct inode *inode, struct file *file)
1093 struct numa_maps_private *priv;
1095 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1097 priv->proc_maps.pid = proc_pid(inode);
1098 ret = seq_open(file, &proc_pid_numa_maps_op);
1100 struct seq_file *m = file->private_data;
1109 const struct file_operations proc_numa_maps_operations = {
1110 .open = numa_maps_open,
1112 .llseek = seq_lseek,
1113 .release = seq_release_private,
1115 #endif /* CONFIG_NUMA */