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 vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
139 if (vma && vma != priv->tail_vma) {
140 struct mm_struct *mm = vma->vm_mm;
141 release_task_mempolicy(priv);
142 up_read(&mm->mmap_sem);
147 static void *m_start(struct seq_file *m, loff_t *pos)
149 struct proc_maps_private *priv = m->private;
150 unsigned long last_addr = m->version;
151 struct mm_struct *mm;
152 struct vm_area_struct *vma, *tail_vma = NULL;
155 /* Clear the per syscall fields in priv */
157 priv->tail_vma = NULL;
160 * We remember last_addr rather than next_addr to hit with
161 * mmap_cache most of the time. We have zero last_addr at
162 * the beginning and also after lseek. We will have -1 last_addr
163 * after the end of the vmas.
166 if (last_addr == -1UL)
169 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
171 return ERR_PTR(-ESRCH);
173 mm = mm_access(priv->task, PTRACE_MODE_READ);
174 if (!mm || IS_ERR(mm))
176 down_read(&mm->mmap_sem);
178 tail_vma = get_gate_vma(priv->task->mm);
179 priv->tail_vma = tail_vma;
180 hold_task_mempolicy(priv);
181 /* Start with last addr hint */
182 vma = find_vma(mm, last_addr);
183 if (last_addr && vma) {
189 * Check the vma index is within the range and do
190 * sequential scan until m_index.
193 if ((unsigned long)l < mm->map_count) {
200 if (l != mm->map_count)
201 tail_vma = NULL; /* After gate vma */
207 release_task_mempolicy(priv);
208 /* End of vmas has been reached */
209 m->version = (tail_vma != NULL)? 0: -1UL;
210 up_read(&mm->mmap_sem);
215 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
217 struct proc_maps_private *priv = m->private;
218 struct vm_area_struct *vma = v;
219 struct vm_area_struct *tail_vma = priv->tail_vma;
222 if (vma && (vma != tail_vma) && vma->vm_next)
225 return (vma != tail_vma)? tail_vma: NULL;
228 static void m_stop(struct seq_file *m, void *v)
230 struct proc_maps_private *priv = m->private;
231 struct vm_area_struct *vma = v;
236 put_task_struct(priv->task);
239 static int do_maps_open(struct inode *inode, struct file *file,
240 const struct seq_operations *ops)
242 struct proc_maps_private *priv;
244 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
246 priv->pid = proc_pid(inode);
247 ret = seq_open(file, ops);
249 struct seq_file *m = file->private_data;
259 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
261 struct mm_struct *mm = vma->vm_mm;
262 struct file *file = vma->vm_file;
263 struct proc_maps_private *priv = m->private;
264 struct task_struct *task = priv->task;
265 vm_flags_t flags = vma->vm_flags;
266 unsigned long ino = 0;
267 unsigned long long pgoff = 0;
268 unsigned long start, end;
271 const char *name = NULL;
274 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
275 dev = inode->i_sb->s_dev;
277 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
280 /* We don't show the stack guard page in /proc/maps */
281 start = vma->vm_start;
282 if (stack_guard_page_start(vma, start))
285 if (stack_guard_page_end(vma, end))
288 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
291 flags & VM_READ ? 'r' : '-',
292 flags & VM_WRITE ? 'w' : '-',
293 flags & VM_EXEC ? 'x' : '-',
294 flags & VM_MAYSHARE ? 's' : 'p',
296 MAJOR(dev), MINOR(dev), ino, &len);
299 * Print the dentry name for named mappings, and a
300 * special [heap] marker for the heap:
303 pad_len_spaces(m, len);
304 seq_path(m, &file->f_path, "\n");
308 name = arch_vma_name(vma);
317 if (vma->vm_start <= mm->brk &&
318 vma->vm_end >= mm->start_brk) {
323 tid = vm_is_stack(task, vma, is_pid);
327 * Thread stack in /proc/PID/task/TID/maps or
328 * the main process stack.
330 if (!is_pid || (vma->vm_start <= mm->start_stack &&
331 vma->vm_end >= mm->start_stack)) {
334 /* Thread stack in /proc/PID/maps */
335 pad_len_spaces(m, len);
336 seq_printf(m, "[stack:%d]", tid);
343 pad_len_spaces(m, len);
349 static int show_map(struct seq_file *m, void *v, int is_pid)
351 struct vm_area_struct *vma = v;
352 struct proc_maps_private *priv = m->private;
353 struct task_struct *task = priv->task;
355 show_map_vma(m, vma, is_pid);
357 if (m->count < m->size) /* vma is copied successfully */
358 m->version = (vma != get_gate_vma(task->mm))
363 static int show_pid_map(struct seq_file *m, void *v)
365 return show_map(m, v, 1);
368 static int show_tid_map(struct seq_file *m, void *v)
370 return show_map(m, v, 0);
373 static const struct seq_operations proc_pid_maps_op = {
380 static const struct seq_operations proc_tid_maps_op = {
387 static int pid_maps_open(struct inode *inode, struct file *file)
389 return do_maps_open(inode, file, &proc_pid_maps_op);
392 static int tid_maps_open(struct inode *inode, struct file *file)
394 return do_maps_open(inode, file, &proc_tid_maps_op);
397 const struct file_operations proc_pid_maps_operations = {
398 .open = pid_maps_open,
401 .release = seq_release_private,
404 const struct file_operations proc_tid_maps_operations = {
405 .open = tid_maps_open,
408 .release = seq_release_private,
412 * Proportional Set Size(PSS): my share of RSS.
414 * PSS of a process is the count of pages it has in memory, where each
415 * page is divided by the number of processes sharing it. So if a
416 * process has 1000 pages all to itself, and 1000 shared with one other
417 * process, its PSS will be 1500.
419 * To keep (accumulated) division errors low, we adopt a 64bit
420 * fixed-point pss counter to minimize division errors. So (pss >>
421 * PSS_SHIFT) would be the real byte count.
423 * A shift of 12 before division means (assuming 4K page size):
424 * - 1M 3-user-pages add up to 8KB errors;
425 * - supports mapcount up to 2^24, or 16M;
426 * - supports PSS up to 2^52 bytes, or 4PB.
430 #ifdef CONFIG_PROC_PAGE_MONITOR
431 struct mem_size_stats {
432 struct vm_area_struct *vma;
433 unsigned long resident;
434 unsigned long shared_clean;
435 unsigned long shared_dirty;
436 unsigned long private_clean;
437 unsigned long private_dirty;
438 unsigned long referenced;
439 unsigned long anonymous;
440 unsigned long anonymous_thp;
442 unsigned long nonlinear;
447 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
448 unsigned long ptent_size, struct mm_walk *walk)
450 struct mem_size_stats *mss = walk->private;
451 struct vm_area_struct *vma = mss->vma;
452 pgoff_t pgoff = linear_page_index(vma, addr);
453 struct page *page = NULL;
456 if (pte_present(ptent)) {
457 page = vm_normal_page(vma, addr, ptent);
458 } else if (is_swap_pte(ptent)) {
459 swp_entry_t swpent = pte_to_swp_entry(ptent);
461 if (!non_swap_entry(swpent))
462 mss->swap += ptent_size;
463 else if (is_migration_entry(swpent))
464 page = migration_entry_to_page(swpent);
465 } else if (pte_file(ptent)) {
466 if (pte_to_pgoff(ptent) != pgoff)
467 mss->nonlinear += ptent_size;
474 mss->anonymous += ptent_size;
476 if (page->index != pgoff)
477 mss->nonlinear += ptent_size;
479 mss->resident += ptent_size;
480 /* Accumulate the size in pages that have been accessed. */
481 if (pte_young(ptent) || PageReferenced(page))
482 mss->referenced += ptent_size;
483 mapcount = page_mapcount(page);
485 if (pte_dirty(ptent) || PageDirty(page))
486 mss->shared_dirty += ptent_size;
488 mss->shared_clean += ptent_size;
489 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
491 if (pte_dirty(ptent) || PageDirty(page))
492 mss->private_dirty += ptent_size;
494 mss->private_clean += ptent_size;
495 mss->pss += (ptent_size << PSS_SHIFT);
499 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
500 struct mm_walk *walk)
502 struct mem_size_stats *mss = walk->private;
503 struct vm_area_struct *vma = mss->vma;
507 if (pmd_trans_huge_lock(pmd, vma) == 1) {
508 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
509 spin_unlock(&walk->mm->page_table_lock);
510 mss->anonymous_thp += HPAGE_PMD_SIZE;
514 if (pmd_trans_unstable(pmd))
517 * The mmap_sem held all the way back in m_start() is what
518 * keeps khugepaged out of here and from collapsing things
521 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
522 for (; addr != end; pte++, addr += PAGE_SIZE)
523 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
524 pte_unmap_unlock(pte - 1, ptl);
529 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
532 * Don't forget to update Documentation/ on changes.
534 static const char mnemonics[BITS_PER_LONG][2] = {
536 * In case if we meet a flag we don't know about.
538 [0 ... (BITS_PER_LONG-1)] = "??",
540 [ilog2(VM_READ)] = "rd",
541 [ilog2(VM_WRITE)] = "wr",
542 [ilog2(VM_EXEC)] = "ex",
543 [ilog2(VM_SHARED)] = "sh",
544 [ilog2(VM_MAYREAD)] = "mr",
545 [ilog2(VM_MAYWRITE)] = "mw",
546 [ilog2(VM_MAYEXEC)] = "me",
547 [ilog2(VM_MAYSHARE)] = "ms",
548 [ilog2(VM_GROWSDOWN)] = "gd",
549 [ilog2(VM_PFNMAP)] = "pf",
550 [ilog2(VM_DENYWRITE)] = "dw",
551 [ilog2(VM_LOCKED)] = "lo",
552 [ilog2(VM_IO)] = "io",
553 [ilog2(VM_SEQ_READ)] = "sr",
554 [ilog2(VM_RAND_READ)] = "rr",
555 [ilog2(VM_DONTCOPY)] = "dc",
556 [ilog2(VM_DONTEXPAND)] = "de",
557 [ilog2(VM_ACCOUNT)] = "ac",
558 [ilog2(VM_NORESERVE)] = "nr",
559 [ilog2(VM_HUGETLB)] = "ht",
560 [ilog2(VM_NONLINEAR)] = "nl",
561 [ilog2(VM_ARCH_1)] = "ar",
562 [ilog2(VM_DONTDUMP)] = "dd",
563 [ilog2(VM_MIXEDMAP)] = "mm",
564 [ilog2(VM_HUGEPAGE)] = "hg",
565 [ilog2(VM_NOHUGEPAGE)] = "nh",
566 [ilog2(VM_MERGEABLE)] = "mg",
570 seq_puts(m, "VmFlags: ");
571 for (i = 0; i < BITS_PER_LONG; i++) {
572 if (vma->vm_flags & (1UL << i)) {
573 seq_printf(m, "%c%c ",
574 mnemonics[i][0], mnemonics[i][1]);
580 static int show_smap(struct seq_file *m, void *v, int is_pid)
582 struct proc_maps_private *priv = m->private;
583 struct task_struct *task = priv->task;
584 struct vm_area_struct *vma = v;
585 struct mem_size_stats mss;
586 struct mm_walk smaps_walk = {
587 .pmd_entry = smaps_pte_range,
592 memset(&mss, 0, sizeof mss);
594 /* mmap_sem is held in m_start */
595 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
596 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
598 show_map_vma(m, vma, is_pid);
604 "Shared_Clean: %8lu kB\n"
605 "Shared_Dirty: %8lu kB\n"
606 "Private_Clean: %8lu kB\n"
607 "Private_Dirty: %8lu kB\n"
608 "Referenced: %8lu kB\n"
609 "Anonymous: %8lu kB\n"
610 "AnonHugePages: %8lu kB\n"
612 "KernelPageSize: %8lu kB\n"
613 "MMUPageSize: %8lu kB\n"
615 (vma->vm_end - vma->vm_start) >> 10,
617 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
618 mss.shared_clean >> 10,
619 mss.shared_dirty >> 10,
620 mss.private_clean >> 10,
621 mss.private_dirty >> 10,
622 mss.referenced >> 10,
624 mss.anonymous_thp >> 10,
626 vma_kernel_pagesize(vma) >> 10,
627 vma_mmu_pagesize(vma) >> 10,
628 (vma->vm_flags & VM_LOCKED) ?
629 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
631 if (vma->vm_flags & VM_NONLINEAR)
632 seq_printf(m, "Nonlinear: %8lu kB\n",
633 mss.nonlinear >> 10);
635 show_smap_vma_flags(m, vma);
637 if (m->count < m->size) /* vma is copied successfully */
638 m->version = (vma != get_gate_vma(task->mm))
643 static int show_pid_smap(struct seq_file *m, void *v)
645 return show_smap(m, v, 1);
648 static int show_tid_smap(struct seq_file *m, void *v)
650 return show_smap(m, v, 0);
653 static const struct seq_operations proc_pid_smaps_op = {
657 .show = show_pid_smap
660 static const struct seq_operations proc_tid_smaps_op = {
664 .show = show_tid_smap
667 static int pid_smaps_open(struct inode *inode, struct file *file)
669 return do_maps_open(inode, file, &proc_pid_smaps_op);
672 static int tid_smaps_open(struct inode *inode, struct file *file)
674 return do_maps_open(inode, file, &proc_tid_smaps_op);
677 const struct file_operations proc_pid_smaps_operations = {
678 .open = pid_smaps_open,
681 .release = seq_release_private,
684 const struct file_operations proc_tid_smaps_operations = {
685 .open = tid_smaps_open,
688 .release = seq_release_private,
691 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
692 unsigned long end, struct mm_walk *walk)
694 struct vm_area_struct *vma = walk->private;
699 split_huge_page_pmd(vma, addr, pmd);
700 if (pmd_trans_unstable(pmd))
703 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
704 for (; addr != end; pte++, addr += PAGE_SIZE) {
706 if (!pte_present(ptent))
709 page = vm_normal_page(vma, addr, ptent);
713 /* Clear accessed and referenced bits. */
714 ptep_test_and_clear_young(vma, addr, pte);
715 ClearPageReferenced(page);
717 pte_unmap_unlock(pte - 1, ptl);
722 #define CLEAR_REFS_ALL 1
723 #define CLEAR_REFS_ANON 2
724 #define CLEAR_REFS_MAPPED 3
726 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
727 size_t count, loff_t *ppos)
729 struct task_struct *task;
730 char buffer[PROC_NUMBUF];
731 struct mm_struct *mm;
732 struct vm_area_struct *vma;
736 memset(buffer, 0, sizeof(buffer));
737 if (count > sizeof(buffer) - 1)
738 count = sizeof(buffer) - 1;
739 if (copy_from_user(buffer, buf, count))
741 rv = kstrtoint(strstrip(buffer), 10, &type);
744 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
746 task = get_proc_task(file->f_path.dentry->d_inode);
749 mm = get_task_mm(task);
751 struct mm_walk clear_refs_walk = {
752 .pmd_entry = clear_refs_pte_range,
755 down_read(&mm->mmap_sem);
756 for (vma = mm->mmap; vma; vma = vma->vm_next) {
757 clear_refs_walk.private = vma;
758 if (is_vm_hugetlb_page(vma))
761 * Writing 1 to /proc/pid/clear_refs affects all pages.
763 * Writing 2 to /proc/pid/clear_refs only affects
766 * Writing 3 to /proc/pid/clear_refs only affects file
769 if (type == CLEAR_REFS_ANON && vma->vm_file)
771 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
773 walk_page_range(vma->vm_start, vma->vm_end,
777 up_read(&mm->mmap_sem);
780 put_task_struct(task);
785 const struct file_operations proc_clear_refs_operations = {
786 .write = clear_refs_write,
787 .llseek = noop_llseek,
796 pagemap_entry_t *buffer;
799 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
800 #define PAGEMAP_WALK_MASK (PMD_MASK)
802 #define PM_ENTRY_BYTES sizeof(u64)
803 #define PM_STATUS_BITS 3
804 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
805 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
806 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
807 #define PM_PSHIFT_BITS 6
808 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
809 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
810 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
811 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
812 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
814 #define PM_PRESENT PM_STATUS(4LL)
815 #define PM_SWAP PM_STATUS(2LL)
816 #define PM_FILE PM_STATUS(1LL)
817 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
818 #define PM_END_OF_BUFFER 1
820 static inline pagemap_entry_t make_pme(u64 val)
822 return (pagemap_entry_t) { .pme = val };
825 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
826 struct pagemapread *pm)
828 pm->buffer[pm->pos++] = *pme;
829 if (pm->pos >= pm->len)
830 return PM_END_OF_BUFFER;
834 static int pagemap_pte_hole(unsigned long start, unsigned long end,
835 struct mm_walk *walk)
837 struct pagemapread *pm = walk->private;
840 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
842 for (addr = start; addr < end; addr += PAGE_SIZE) {
843 err = add_to_pagemap(addr, &pme, pm);
850 static void pte_to_pagemap_entry(pagemap_entry_t *pme,
851 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
854 struct page *page = NULL;
856 if (pte_present(pte)) {
857 frame = pte_pfn(pte);
859 page = vm_normal_page(vma, addr, pte);
860 } else if (is_swap_pte(pte)) {
861 swp_entry_t entry = pte_to_swp_entry(pte);
863 frame = swp_type(entry) |
864 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
866 if (is_migration_entry(entry))
867 page = migration_entry_to_page(entry);
869 *pme = make_pme(PM_NOT_PRESENT);
873 if (page && !PageAnon(page))
876 *pme = make_pme(PM_PFRAME(frame) | PM_PSHIFT(PAGE_SHIFT) | flags);
879 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
880 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
881 pmd_t pmd, int offset)
884 * Currently pmd for thp is always present because thp can not be
885 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
886 * This if-check is just to prepare for future implementation.
888 if (pmd_present(pmd))
889 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
890 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
892 *pme = make_pme(PM_NOT_PRESENT);
895 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
896 pmd_t pmd, int offset)
901 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
902 struct mm_walk *walk)
904 struct vm_area_struct *vma;
905 struct pagemapread *pm = walk->private;
908 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
910 /* find the first VMA at or above 'addr' */
911 vma = find_vma(walk->mm, addr);
912 if (vma && pmd_trans_huge_lock(pmd, vma) == 1) {
913 for (; addr != end; addr += PAGE_SIZE) {
914 unsigned long offset;
916 offset = (addr & ~PAGEMAP_WALK_MASK) >>
918 thp_pmd_to_pagemap_entry(&pme, *pmd, offset);
919 err = add_to_pagemap(addr, &pme, pm);
923 spin_unlock(&walk->mm->page_table_lock);
927 if (pmd_trans_unstable(pmd))
929 for (; addr != end; addr += PAGE_SIZE) {
931 /* check to see if we've left 'vma' behind
932 * and need a new, higher one */
933 if (vma && (addr >= vma->vm_end)) {
934 vma = find_vma(walk->mm, addr);
935 pme = make_pme(PM_NOT_PRESENT);
938 /* check that 'vma' actually covers this address,
939 * and that it isn't a huge page vma */
940 if (vma && (vma->vm_start <= addr) &&
941 !is_vm_hugetlb_page(vma)) {
942 pte = pte_offset_map(pmd, addr);
943 pte_to_pagemap_entry(&pme, vma, addr, *pte);
944 /* unmap before userspace copy */
947 err = add_to_pagemap(addr, &pme, pm);
957 #ifdef CONFIG_HUGETLB_PAGE
958 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme,
959 pte_t pte, int offset)
961 if (pte_present(pte))
962 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset)
963 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
965 *pme = make_pme(PM_NOT_PRESENT);
968 /* This function walks within one hugetlb entry in the single call */
969 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
970 unsigned long addr, unsigned long end,
971 struct mm_walk *walk)
973 struct pagemapread *pm = walk->private;
977 for (; addr != end; addr += PAGE_SIZE) {
978 int offset = (addr & ~hmask) >> PAGE_SHIFT;
979 huge_pte_to_pagemap_entry(&pme, *pte, offset);
980 err = add_to_pagemap(addr, &pme, pm);
989 #endif /* HUGETLB_PAGE */
992 * /proc/pid/pagemap - an array mapping virtual pages to pfns
994 * For each page in the address space, this file contains one 64-bit entry
995 * consisting of the following:
997 * Bits 0-54 page frame number (PFN) if present
998 * Bits 0-4 swap type if swapped
999 * Bits 5-54 swap offset if swapped
1000 * Bits 55-60 page shift (page size = 1<<page shift)
1001 * Bit 61 page is file-page or shared-anon
1002 * Bit 62 page swapped
1003 * Bit 63 page present
1005 * If the page is not present but in swap, then the PFN contains an
1006 * encoding of the swap file number and the page's offset into the
1007 * swap. Unmapped pages return a null PFN. This allows determining
1008 * precisely which pages are mapped (or in swap) and comparing mapped
1009 * pages between processes.
1011 * Efficient users of this interface will use /proc/pid/maps to
1012 * determine which areas of memory are actually mapped and llseek to
1013 * skip over unmapped regions.
1015 static ssize_t pagemap_read(struct file *file, char __user *buf,
1016 size_t count, loff_t *ppos)
1018 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1019 struct mm_struct *mm;
1020 struct pagemapread pm;
1022 struct mm_walk pagemap_walk = {};
1024 unsigned long svpfn;
1025 unsigned long start_vaddr;
1026 unsigned long end_vaddr;
1033 /* file position must be aligned */
1034 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1041 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1042 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
1047 mm = mm_access(task, PTRACE_MODE_READ);
1049 if (!mm || IS_ERR(mm))
1052 pagemap_walk.pmd_entry = pagemap_pte_range;
1053 pagemap_walk.pte_hole = pagemap_pte_hole;
1054 #ifdef CONFIG_HUGETLB_PAGE
1055 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1057 pagemap_walk.mm = mm;
1058 pagemap_walk.private = ±
1061 svpfn = src / PM_ENTRY_BYTES;
1062 start_vaddr = svpfn << PAGE_SHIFT;
1063 end_vaddr = TASK_SIZE_OF(task);
1065 /* watch out for wraparound */
1066 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1067 start_vaddr = end_vaddr;
1070 * The odds are that this will stop walking way
1071 * before end_vaddr, because the length of the
1072 * user buffer is tracked in "pm", and the walk
1073 * will stop when we hit the end of the buffer.
1076 while (count && (start_vaddr < end_vaddr)) {
1081 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1083 if (end < start_vaddr || end > end_vaddr)
1085 down_read(&mm->mmap_sem);
1086 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1087 up_read(&mm->mmap_sem);
1090 len = min(count, PM_ENTRY_BYTES * pm.pos);
1091 if (copy_to_user(buf, pm.buffer, len)) {
1100 if (!ret || ret == PM_END_OF_BUFFER)
1108 put_task_struct(task);
1113 const struct file_operations proc_pagemap_operations = {
1114 .llseek = mem_lseek, /* borrow this */
1115 .read = pagemap_read,
1117 #endif /* CONFIG_PROC_PAGE_MONITOR */
1122 struct vm_area_struct *vma;
1123 unsigned long pages;
1125 unsigned long active;
1126 unsigned long writeback;
1127 unsigned long mapcount_max;
1128 unsigned long dirty;
1129 unsigned long swapcache;
1130 unsigned long node[MAX_NUMNODES];
1133 struct numa_maps_private {
1134 struct proc_maps_private proc_maps;
1135 struct numa_maps md;
1138 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1139 unsigned long nr_pages)
1141 int count = page_mapcount(page);
1143 md->pages += nr_pages;
1144 if (pte_dirty || PageDirty(page))
1145 md->dirty += nr_pages;
1147 if (PageSwapCache(page))
1148 md->swapcache += nr_pages;
1150 if (PageActive(page) || PageUnevictable(page))
1151 md->active += nr_pages;
1153 if (PageWriteback(page))
1154 md->writeback += nr_pages;
1157 md->anon += nr_pages;
1159 if (count > md->mapcount_max)
1160 md->mapcount_max = count;
1162 md->node[page_to_nid(page)] += nr_pages;
1165 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1171 if (!pte_present(pte))
1174 page = vm_normal_page(vma, addr, pte);
1178 if (PageReserved(page))
1181 nid = page_to_nid(page);
1182 if (!node_isset(nid, node_states[N_MEMORY]))
1188 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1189 unsigned long end, struct mm_walk *walk)
1191 struct numa_maps *md;
1198 if (pmd_trans_huge_lock(pmd, md->vma) == 1) {
1199 pte_t huge_pte = *(pte_t *)pmd;
1202 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1204 gather_stats(page, md, pte_dirty(huge_pte),
1205 HPAGE_PMD_SIZE/PAGE_SIZE);
1206 spin_unlock(&walk->mm->page_table_lock);
1210 if (pmd_trans_unstable(pmd))
1212 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1214 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1217 gather_stats(page, md, pte_dirty(*pte), 1);
1219 } while (pte++, addr += PAGE_SIZE, addr != end);
1220 pte_unmap_unlock(orig_pte, ptl);
1223 #ifdef CONFIG_HUGETLB_PAGE
1224 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1225 unsigned long addr, unsigned long end, struct mm_walk *walk)
1227 struct numa_maps *md;
1233 page = pte_page(*pte);
1238 gather_stats(page, md, pte_dirty(*pte), 1);
1243 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1244 unsigned long addr, unsigned long end, struct mm_walk *walk)
1251 * Display pages allocated per node and memory policy via /proc.
1253 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1255 struct numa_maps_private *numa_priv = m->private;
1256 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1257 struct vm_area_struct *vma = v;
1258 struct numa_maps *md = &numa_priv->md;
1259 struct file *file = vma->vm_file;
1260 struct task_struct *task = proc_priv->task;
1261 struct mm_struct *mm = vma->vm_mm;
1262 struct mm_walk walk = {};
1263 struct mempolicy *pol;
1270 /* Ensure we start with an empty set of numa_maps statistics. */
1271 memset(md, 0, sizeof(*md));
1275 walk.hugetlb_entry = gather_hugetbl_stats;
1276 walk.pmd_entry = gather_pte_stats;
1280 pol = get_vma_policy(task, vma, vma->vm_start);
1281 mpol_to_str(buffer, sizeof(buffer), pol);
1284 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1287 seq_printf(m, " file=");
1288 seq_path(m, &file->f_path, "\n\t= ");
1289 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1290 seq_printf(m, " heap");
1292 pid_t tid = vm_is_stack(task, vma, is_pid);
1295 * Thread stack in /proc/PID/task/TID/maps or
1296 * the main process stack.
1298 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1299 vma->vm_end >= mm->start_stack))
1300 seq_printf(m, " stack");
1302 seq_printf(m, " stack:%d", tid);
1306 if (is_vm_hugetlb_page(vma))
1307 seq_printf(m, " huge");
1309 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1315 seq_printf(m, " anon=%lu", md->anon);
1318 seq_printf(m, " dirty=%lu", md->dirty);
1320 if (md->pages != md->anon && md->pages != md->dirty)
1321 seq_printf(m, " mapped=%lu", md->pages);
1323 if (md->mapcount_max > 1)
1324 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1327 seq_printf(m, " swapcache=%lu", md->swapcache);
1329 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1330 seq_printf(m, " active=%lu", md->active);
1333 seq_printf(m, " writeback=%lu", md->writeback);
1335 for_each_node_state(n, N_MEMORY)
1337 seq_printf(m, " N%d=%lu", n, md->node[n]);
1341 if (m->count < m->size)
1342 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1346 static int show_pid_numa_map(struct seq_file *m, void *v)
1348 return show_numa_map(m, v, 1);
1351 static int show_tid_numa_map(struct seq_file *m, void *v)
1353 return show_numa_map(m, v, 0);
1356 static const struct seq_operations proc_pid_numa_maps_op = {
1360 .show = show_pid_numa_map,
1363 static const struct seq_operations proc_tid_numa_maps_op = {
1367 .show = show_tid_numa_map,
1370 static int numa_maps_open(struct inode *inode, struct file *file,
1371 const struct seq_operations *ops)
1373 struct numa_maps_private *priv;
1375 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1377 priv->proc_maps.pid = proc_pid(inode);
1378 ret = seq_open(file, ops);
1380 struct seq_file *m = file->private_data;
1389 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1391 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1394 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1396 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1399 const struct file_operations proc_pid_numa_maps_operations = {
1400 .open = pid_numa_maps_open,
1402 .llseek = seq_lseek,
1403 .release = seq_release_private,
1406 const struct file_operations proc_tid_numa_maps_operations = {
1407 .open = tid_numa_maps_open,
1409 .llseek = seq_lseek,
1410 .release = seq_release_private,
1412 #endif /* CONFIG_NUMA */