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;
213 static void show_arm_map_vma(struct seq_file *m, struct vm_area_struct *vma)
215 static const char *cache_attrs4[4] = { "noC", "WB-WA", "WT-noWA", "WB-noWA" };
216 struct mm_struct *mm = vma->vm_mm;
217 struct file *file = vma->vm_file;
218 vm_flags_t flags = vma->vm_flags;
219 unsigned long start, end, end_b;
220 const char *name = NULL;
239 name = arch_vma_name(vma);
241 if (vma->vm_start <= mm->brk &&
242 vma->vm_end >= mm->start_brk) {
244 } else if (vma->vm_start <= mm->start_stack &&
245 vma->vm_end >= mm->start_stack) {
251 arm_pgd = (u32 *)mm->pgd;
253 asm ("mrc p15, 0, %0, c1, c0, 0" : "=r"(control));
254 asm ("mrc p15, 0, %0, c10, c2, 0" : "=r"(prrr)); // primary region RR
255 asm ("mrc p15, 0, %0, c10, c2, 1" : "=r"(nmrr)); // normal memory RR
257 start = vma->vm_start;
260 while (start < end) {
263 desc1 = arm_pgd[start >> 20];
265 end_b = (start & ~0xfffff) + 0x100000;
266 for (; end_b < end; end_b += 0x100000)
267 if ((arm_pgd[end_b >> 20] ^ desc1) & 0xfffff)
272 sprintf(buf, "l1_fault");
277 tex_cb = ((desc1 >> 2) & 0x03) | ((desc1 >> 10) & 0x1c);
278 s = (desc1 >> 16) & 1;
279 xn = (desc1 >> 4) & 1;
280 ap = ((desc1 >> 10) & 3) | ((desc1 >> 13) & 4);
281 desc_type = (desc1 & (1 << 18)) ? 's' : 'h';
284 sprintf(buf, "reserved");
288 cpt = __va(desc1 & 0xfffffc00);
289 desc2 = cpt[(start >> 12) & 0xff];
293 for (end_b = start + 0x1000; end_b < end; end_b += 0x1000) {
294 if ((end_b & 0x000ff000) == 0) {
295 cpt_e = __va(arm_pgd[end_b >> 20] & 0xfffffc00);
296 if ((arm_pgd[end_b >> 20] ^ desc1) & 0x3ff)
300 // assume small pages
301 if ((cpt_e[(end_b >> 12) & 0xff] ^ desc2) & 0xfff)
307 sprintf(buf, "l2_fault");
310 tex_cb = ((desc2 >> 2) & 0x03) | ((desc2 >> 10) & 0x1c);
311 s = (desc2 >> 10) & 1;
312 xn = (desc2 >> 15) & 1;
313 ap = ((desc2 >> 4) & 3) | ((desc2 >> 7) & 4);
317 tex_cb = ((desc2 >> 2) & 0x03) | ((desc2 >> 4) & 0x1c);
318 s = (desc2 >> 10) & 1;
320 ap = ((desc2 >> 4) & 3) | ((desc2 >> 7) & 4);
325 if (control & (1 << 28)) { // TEX remap
326 // S (shareable) bit remapping
327 char s_normal[2] = { (prrr >> 18) & 1, (prrr >> 19) & 1 };
328 char s_device[2] = { (prrr >> 16) & 1, (prrr >> 17) & 1 };
332 type = (prrr >> tex_cb * 2) & 3;
335 sprintf(buf, "strongly-ordered");
338 sprintf(buf, "device");
342 sprintf(buf, "reserved/normal");
345 sprintf(buf + strlen(buf), "inner-%s-outer-%s",
346 cache_attrs4[(nmrr >> tex_cb * 2) & 3],
347 cache_attrs4[(nmrr >> (tex_cb * 2 + 16)) & 3]);
350 else if (tex_cb & 0x10) { // TEX[2] set
351 sprintf(buf, "inner-%s-outer-%s",
352 cache_attrs4[tex_cb & 3], cache_attrs4[(tex_cb >> 2) & 3]);
356 case 0x00: sprintf(buf, "strongly-ordered"); s = 1; break;
357 case 0x01: sprintf(buf, "shareable-device"); s = 1; break;
358 case 0x02: sprintf(buf, "inner-outer-WT-noWA"); break;
359 case 0x03: sprintf(buf, "inner-outer-WB-noWA"); break;
360 case 0x04: sprintf(buf, "inner-outer-non-cacheable"); break;
361 case 0x06: sprintf(buf, "implementation-defined"); break;
362 case 0x07: sprintf(buf, "inner-outer-WB-WA"); break;
363 case 0x08: sprintf(buf, "non-shareable-device"); s = 0; break;
364 default: sprintf(buf, "reserved"); break;
369 sprintf(buf + strlen(buf), "-shareable");
372 // use user permissions here
373 if (control & (1 << 29)) // AFE
374 sprintf(rw, "%c%c", (ap & 2) ? 'r' : '-',
375 ((ap & 2) && !(ap & 4)) ? 'w' : '-');
377 sprintf(rw, "%c%c", (ap & 2) ? 'r' : '-',
378 (ap == 3) ? 'w' : '-');
380 seq_printf(m, "%08lx-%08lx %s%c%c%c %-28s %n",
384 flags & VM_MAYSHARE ? 's' : 'p',
389 pad_len_spaces(m, len);
390 seq_path(m, &file->f_path, "\n");
392 pad_len_spaces(m, len);
402 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
404 struct mm_struct *mm = vma->vm_mm;
405 struct file *file = vma->vm_file;
406 vm_flags_t flags = vma->vm_flags;
407 unsigned long ino = 0;
408 unsigned long long pgoff = 0;
409 unsigned long start, end;
414 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
415 dev = inode->i_sb->s_dev;
417 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
420 /* We don't show the stack guard page in /proc/maps */
421 start = vma->vm_start;
422 if (stack_guard_page_start(vma, start))
425 if (stack_guard_page_end(vma, end))
428 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
431 flags & VM_READ ? 'r' : '-',
432 flags & VM_WRITE ? 'w' : '-',
433 flags & VM_EXEC ? 'x' : '-',
434 flags & VM_MAYSHARE ? 's' : 'p',
436 MAJOR(dev), MINOR(dev), ino, &len);
439 * Print the dentry name for named mappings, and a
440 * special [heap] marker for the heap:
443 pad_len_spaces(m, len);
444 seq_path(m, &file->f_path, "\n");
446 const char *name = arch_vma_name(vma);
449 if (vma->vm_start <= mm->brk &&
450 vma->vm_end >= mm->start_brk) {
452 } else if (vma->vm_start <= mm->start_stack &&
453 vma->vm_end >= mm->start_stack) {
461 pad_len_spaces(m, len);
468 static int show_map(struct seq_file *m, void *v)
470 struct vm_area_struct *vma = v;
471 struct proc_maps_private *priv = m->private;
472 struct task_struct *task = priv->task;
474 show_map_vma(m, vma);
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_maps_op = {
489 static int maps_open(struct inode *inode, struct file *file)
491 return do_maps_open(inode, file, &proc_pid_maps_op);
494 const struct file_operations proc_maps_operations = {
498 .release = seq_release_private,
502 static int show_armv7_map(struct seq_file *m, void *v)
504 struct vm_area_struct *vma = v;
505 struct proc_maps_private *priv = m->private;
506 struct task_struct *task = priv->task;
508 show_arm_map_vma(m, vma);
510 if (m->count < m->size) /* vma is copied successfully */
511 m->version = (vma != get_gate_vma(task->mm))
516 static const struct seq_operations proc_pid_armv7_maps_op = {
520 .show = show_armv7_map
524 static int armv7_maps_open(struct inode *inode, struct file *file)
526 return do_maps_open(inode, file, &proc_pid_armv7_maps_op);
529 const struct file_operations proc_armv7_maps_operations = {
530 .open = armv7_maps_open,
533 .release = seq_release_private,
537 * Proportional Set Size(PSS): my share of RSS.
539 * PSS of a process is the count of pages it has in memory, where each
540 * page is divided by the number of processes sharing it. So if a
541 * process has 1000 pages all to itself, and 1000 shared with one other
542 * process, its PSS will be 1500.
544 * To keep (accumulated) division errors low, we adopt a 64bit
545 * fixed-point pss counter to minimize division errors. So (pss >>
546 * PSS_SHIFT) would be the real byte count.
548 * A shift of 12 before division means (assuming 4K page size):
549 * - 1M 3-user-pages add up to 8KB errors;
550 * - supports mapcount up to 2^24, or 16M;
551 * - supports PSS up to 2^52 bytes, or 4PB.
555 #ifdef CONFIG_PROC_PAGE_MONITOR
556 struct mem_size_stats {
557 struct vm_area_struct *vma;
558 unsigned long resident;
559 unsigned long shared_clean;
560 unsigned long shared_dirty;
561 unsigned long private_clean;
562 unsigned long private_dirty;
563 unsigned long referenced;
564 unsigned long anonymous;
565 unsigned long anonymous_thp;
571 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
572 unsigned long ptent_size, struct mm_walk *walk)
574 struct mem_size_stats *mss = walk->private;
575 struct vm_area_struct *vma = mss->vma;
579 if (is_swap_pte(ptent)) {
580 mss->swap += ptent_size;
584 if (!pte_present(ptent))
587 page = vm_normal_page(vma, addr, ptent);
592 mss->anonymous += ptent_size;
594 mss->resident += ptent_size;
595 /* Accumulate the size in pages that have been accessed. */
596 if (pte_young(ptent) || PageReferenced(page))
597 mss->referenced += ptent_size;
598 mapcount = page_mapcount(page);
600 if (pte_dirty(ptent) || PageDirty(page))
601 mss->shared_dirty += ptent_size;
603 mss->shared_clean += ptent_size;
604 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
606 if (pte_dirty(ptent) || PageDirty(page))
607 mss->private_dirty += ptent_size;
609 mss->private_clean += ptent_size;
610 mss->pss += (ptent_size << PSS_SHIFT);
614 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
615 struct mm_walk *walk)
617 struct mem_size_stats *mss = walk->private;
618 struct vm_area_struct *vma = mss->vma;
622 spin_lock(&walk->mm->page_table_lock);
623 if (pmd_trans_huge(*pmd)) {
624 if (pmd_trans_splitting(*pmd)) {
625 spin_unlock(&walk->mm->page_table_lock);
626 wait_split_huge_page(vma->anon_vma, pmd);
628 smaps_pte_entry(*(pte_t *)pmd, addr,
629 HPAGE_PMD_SIZE, walk);
630 spin_unlock(&walk->mm->page_table_lock);
631 mss->anonymous_thp += HPAGE_PMD_SIZE;
635 spin_unlock(&walk->mm->page_table_lock);
638 if (pmd_trans_unstable(pmd))
641 * The mmap_sem held all the way back in m_start() is what
642 * keeps khugepaged out of here and from collapsing things
645 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
646 for (; addr != end; pte++, addr += PAGE_SIZE)
647 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
648 pte_unmap_unlock(pte - 1, ptl);
653 static int show_smap(struct seq_file *m, void *v)
655 struct proc_maps_private *priv = m->private;
656 struct task_struct *task = priv->task;
657 struct vm_area_struct *vma = v;
658 struct mem_size_stats mss;
659 struct mm_walk smaps_walk = {
660 .pmd_entry = smaps_pte_range,
665 memset(&mss, 0, sizeof mss);
667 /* mmap_sem is held in m_start */
668 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
669 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
671 show_map_vma(m, vma);
677 "Shared_Clean: %8lu kB\n"
678 "Shared_Dirty: %8lu kB\n"
679 "Private_Clean: %8lu kB\n"
680 "Private_Dirty: %8lu kB\n"
681 "Referenced: %8lu kB\n"
682 "Anonymous: %8lu kB\n"
683 "AnonHugePages: %8lu kB\n"
685 "KernelPageSize: %8lu kB\n"
686 "MMUPageSize: %8lu kB\n"
688 (vma->vm_end - vma->vm_start) >> 10,
690 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
691 mss.shared_clean >> 10,
692 mss.shared_dirty >> 10,
693 mss.private_clean >> 10,
694 mss.private_dirty >> 10,
695 mss.referenced >> 10,
697 mss.anonymous_thp >> 10,
699 vma_kernel_pagesize(vma) >> 10,
700 vma_mmu_pagesize(vma) >> 10,
701 (vma->vm_flags & VM_LOCKED) ?
702 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
704 if (m->count < m->size) /* vma is copied successfully */
705 m->version = (vma != get_gate_vma(task->mm))
710 static const struct seq_operations proc_pid_smaps_op = {
717 static int smaps_open(struct inode *inode, struct file *file)
719 return do_maps_open(inode, file, &proc_pid_smaps_op);
722 const struct file_operations proc_smaps_operations = {
726 .release = seq_release_private,
729 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
730 unsigned long end, struct mm_walk *walk)
732 struct vm_area_struct *vma = walk->private;
737 split_huge_page_pmd(walk->mm, pmd);
738 if (pmd_trans_unstable(pmd))
741 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
742 for (; addr != end; pte++, addr += PAGE_SIZE) {
744 if (!pte_present(ptent))
747 page = vm_normal_page(vma, addr, ptent);
751 if (PageReserved(page))
754 /* Clear accessed and referenced bits. */
755 ptep_test_and_clear_young(vma, addr, pte);
756 ClearPageReferenced(page);
758 pte_unmap_unlock(pte - 1, ptl);
763 #define CLEAR_REFS_ALL 1
764 #define CLEAR_REFS_ANON 2
765 #define CLEAR_REFS_MAPPED 3
767 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
768 size_t count, loff_t *ppos)
770 struct task_struct *task;
771 char buffer[PROC_NUMBUF];
772 struct mm_struct *mm;
773 struct vm_area_struct *vma;
777 memset(buffer, 0, sizeof(buffer));
778 if (count > sizeof(buffer) - 1)
779 count = sizeof(buffer) - 1;
780 if (copy_from_user(buffer, buf, count))
782 rv = kstrtoint(strstrip(buffer), 10, &type);
785 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
787 task = get_proc_task(file->f_path.dentry->d_inode);
790 mm = get_task_mm(task);
792 struct mm_walk clear_refs_walk = {
793 .pmd_entry = clear_refs_pte_range,
796 down_read(&mm->mmap_sem);
797 for (vma = mm->mmap; vma; vma = vma->vm_next) {
798 clear_refs_walk.private = vma;
799 if (is_vm_hugetlb_page(vma))
802 * Writing 1 to /proc/pid/clear_refs affects all pages.
804 * Writing 2 to /proc/pid/clear_refs only affects
807 * Writing 3 to /proc/pid/clear_refs only affects file
810 if (type == CLEAR_REFS_ANON && vma->vm_file)
812 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
814 walk_page_range(vma->vm_start, vma->vm_end,
818 up_read(&mm->mmap_sem);
821 put_task_struct(task);
826 const struct file_operations proc_clear_refs_operations = {
827 .write = clear_refs_write,
828 .llseek = noop_llseek,
832 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
836 #define PM_ENTRY_BYTES sizeof(u64)
837 #define PM_STATUS_BITS 3
838 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
839 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
840 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
841 #define PM_PSHIFT_BITS 6
842 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
843 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
844 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
845 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
846 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
848 #define PM_PRESENT PM_STATUS(4LL)
849 #define PM_SWAP PM_STATUS(2LL)
850 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
851 #define PM_END_OF_BUFFER 1
853 static int add_to_pagemap(unsigned long addr, u64 pfn,
854 struct pagemapread *pm)
856 pm->buffer[pm->pos++] = pfn;
857 if (pm->pos >= pm->len)
858 return PM_END_OF_BUFFER;
862 static int pagemap_pte_hole(unsigned long start, unsigned long end,
863 struct mm_walk *walk)
865 struct pagemapread *pm = walk->private;
868 for (addr = start; addr < end; addr += PAGE_SIZE) {
869 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
876 static u64 swap_pte_to_pagemap_entry(pte_t pte)
878 swp_entry_t e = pte_to_swp_entry(pte);
879 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
882 static u64 pte_to_pagemap_entry(pte_t pte)
885 if (is_swap_pte(pte))
886 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
887 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
888 else if (pte_present(pte))
889 pme = PM_PFRAME(pte_pfn(pte))
890 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
894 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
895 struct mm_walk *walk)
897 struct vm_area_struct *vma;
898 struct pagemapread *pm = walk->private;
902 split_huge_page_pmd(walk->mm, pmd);
903 if (pmd_trans_unstable(pmd))
906 /* find the first VMA at or above 'addr' */
907 vma = find_vma(walk->mm, addr);
908 for (; addr != end; addr += PAGE_SIZE) {
909 u64 pfn = PM_NOT_PRESENT;
911 /* check to see if we've left 'vma' behind
912 * and need a new, higher one */
913 if (vma && (addr >= vma->vm_end))
914 vma = find_vma(walk->mm, addr);
916 /* check that 'vma' actually covers this address,
917 * and that it isn't a huge page vma */
918 if (vma && (vma->vm_start <= addr) &&
919 !is_vm_hugetlb_page(vma)) {
920 pte = pte_offset_map(pmd, addr);
921 pfn = pte_to_pagemap_entry(*pte);
922 /* unmap before userspace copy */
925 err = add_to_pagemap(addr, pfn, pm);
935 #ifdef CONFIG_HUGETLB_PAGE
936 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
939 if (pte_present(pte))
940 pme = PM_PFRAME(pte_pfn(pte) + offset)
941 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
945 /* This function walks within one hugetlb entry in the single call */
946 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
947 unsigned long addr, unsigned long end,
948 struct mm_walk *walk)
950 struct pagemapread *pm = walk->private;
954 for (; addr != end; addr += PAGE_SIZE) {
955 int offset = (addr & ~hmask) >> PAGE_SHIFT;
956 pfn = huge_pte_to_pagemap_entry(*pte, offset);
957 err = add_to_pagemap(addr, pfn, pm);
966 #endif /* HUGETLB_PAGE */
969 * /proc/pid/pagemap - an array mapping virtual pages to pfns
971 * For each page in the address space, this file contains one 64-bit entry
972 * consisting of the following:
974 * Bits 0-55 page frame number (PFN) if present
975 * Bits 0-4 swap type if swapped
976 * Bits 5-55 swap offset if swapped
977 * Bits 55-60 page shift (page size = 1<<page shift)
978 * Bit 61 reserved for future use
979 * Bit 62 page swapped
980 * Bit 63 page present
982 * If the page is not present but in swap, then the PFN contains an
983 * encoding of the swap file number and the page's offset into the
984 * swap. Unmapped pages return a null PFN. This allows determining
985 * precisely which pages are mapped (or in swap) and comparing mapped
986 * pages between processes.
988 * Efficient users of this interface will use /proc/pid/maps to
989 * determine which areas of memory are actually mapped and llseek to
990 * skip over unmapped regions.
992 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
993 #define PAGEMAP_WALK_MASK (PMD_MASK)
994 static ssize_t pagemap_read(struct file *file, char __user *buf,
995 size_t count, loff_t *ppos)
997 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
998 struct mm_struct *mm;
999 struct pagemapread pm;
1001 struct mm_walk pagemap_walk = {};
1003 unsigned long svpfn;
1004 unsigned long start_vaddr;
1005 unsigned long end_vaddr;
1012 /* file position must be aligned */
1013 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1020 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1021 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1026 mm = mm_for_maps(task);
1028 if (!mm || IS_ERR(mm))
1031 pagemap_walk.pmd_entry = pagemap_pte_range;
1032 pagemap_walk.pte_hole = pagemap_pte_hole;
1033 #ifdef CONFIG_HUGETLB_PAGE
1034 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1036 pagemap_walk.mm = mm;
1037 pagemap_walk.private = ±
1040 svpfn = src / PM_ENTRY_BYTES;
1041 start_vaddr = svpfn << PAGE_SHIFT;
1042 end_vaddr = TASK_SIZE_OF(task);
1044 /* watch out for wraparound */
1045 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1046 start_vaddr = end_vaddr;
1049 * The odds are that this will stop walking way
1050 * before end_vaddr, because the length of the
1051 * user buffer is tracked in "pm", and the walk
1052 * will stop when we hit the end of the buffer.
1055 while (count && (start_vaddr < end_vaddr)) {
1060 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1062 if (end < start_vaddr || end > end_vaddr)
1064 down_read(&mm->mmap_sem);
1065 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1066 up_read(&mm->mmap_sem);
1069 len = min(count, PM_ENTRY_BYTES * pm.pos);
1070 if (copy_to_user(buf, pm.buffer, len)) {
1079 if (!ret || ret == PM_END_OF_BUFFER)
1087 put_task_struct(task);
1092 const struct file_operations proc_pagemap_operations = {
1093 .llseek = mem_lseek, /* borrow this */
1094 .read = pagemap_read,
1096 #endif /* CONFIG_PROC_PAGE_MONITOR */
1101 struct vm_area_struct *vma;
1102 unsigned long pages;
1104 unsigned long active;
1105 unsigned long writeback;
1106 unsigned long mapcount_max;
1107 unsigned long dirty;
1108 unsigned long swapcache;
1109 unsigned long node[MAX_NUMNODES];
1112 struct numa_maps_private {
1113 struct proc_maps_private proc_maps;
1114 struct numa_maps md;
1117 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1118 unsigned long nr_pages)
1120 int count = page_mapcount(page);
1122 md->pages += nr_pages;
1123 if (pte_dirty || PageDirty(page))
1124 md->dirty += nr_pages;
1126 if (PageSwapCache(page))
1127 md->swapcache += nr_pages;
1129 if (PageActive(page) || PageUnevictable(page))
1130 md->active += nr_pages;
1132 if (PageWriteback(page))
1133 md->writeback += nr_pages;
1136 md->anon += nr_pages;
1138 if (count > md->mapcount_max)
1139 md->mapcount_max = count;
1141 md->node[page_to_nid(page)] += nr_pages;
1144 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1150 if (!pte_present(pte))
1153 page = vm_normal_page(vma, addr, pte);
1157 if (PageReserved(page))
1160 nid = page_to_nid(page);
1161 if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
1167 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1168 unsigned long end, struct mm_walk *walk)
1170 struct numa_maps *md;
1176 spin_lock(&walk->mm->page_table_lock);
1177 if (pmd_trans_huge(*pmd)) {
1178 if (pmd_trans_splitting(*pmd)) {
1179 spin_unlock(&walk->mm->page_table_lock);
1180 wait_split_huge_page(md->vma->anon_vma, pmd);
1182 pte_t huge_pte = *(pte_t *)pmd;
1185 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1187 gather_stats(page, md, pte_dirty(huge_pte),
1188 HPAGE_PMD_SIZE/PAGE_SIZE);
1189 spin_unlock(&walk->mm->page_table_lock);
1193 spin_unlock(&walk->mm->page_table_lock);
1196 if (pmd_trans_unstable(pmd))
1198 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1200 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1203 gather_stats(page, md, pte_dirty(*pte), 1);
1205 } while (pte++, addr += PAGE_SIZE, addr != end);
1206 pte_unmap_unlock(orig_pte, ptl);
1209 #ifdef CONFIG_HUGETLB_PAGE
1210 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1211 unsigned long addr, unsigned long end, struct mm_walk *walk)
1213 struct numa_maps *md;
1219 page = pte_page(*pte);
1224 gather_stats(page, md, pte_dirty(*pte), 1);
1229 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1230 unsigned long addr, unsigned long end, struct mm_walk *walk)
1237 * Display pages allocated per node and memory policy via /proc.
1239 static int show_numa_map(struct seq_file *m, void *v)
1241 struct numa_maps_private *numa_priv = m->private;
1242 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1243 struct vm_area_struct *vma = v;
1244 struct numa_maps *md = &numa_priv->md;
1245 struct file *file = vma->vm_file;
1246 struct mm_struct *mm = vma->vm_mm;
1247 struct mm_walk walk = {};
1248 struct mempolicy *pol;
1255 /* Ensure we start with an empty set of numa_maps statistics. */
1256 memset(md, 0, sizeof(*md));
1260 walk.hugetlb_entry = gather_hugetbl_stats;
1261 walk.pmd_entry = gather_pte_stats;
1265 pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
1266 mpol_to_str(buffer, sizeof(buffer), pol, 0);
1269 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1272 seq_printf(m, " file=");
1273 seq_path(m, &file->f_path, "\n\t= ");
1274 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1275 seq_printf(m, " heap");
1276 } else if (vma->vm_start <= mm->start_stack &&
1277 vma->vm_end >= mm->start_stack) {
1278 seq_printf(m, " stack");
1281 if (is_vm_hugetlb_page(vma))
1282 seq_printf(m, " huge");
1284 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1290 seq_printf(m, " anon=%lu", md->anon);
1293 seq_printf(m, " dirty=%lu", md->dirty);
1295 if (md->pages != md->anon && md->pages != md->dirty)
1296 seq_printf(m, " mapped=%lu", md->pages);
1298 if (md->mapcount_max > 1)
1299 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1302 seq_printf(m, " swapcache=%lu", md->swapcache);
1304 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1305 seq_printf(m, " active=%lu", md->active);
1308 seq_printf(m, " writeback=%lu", md->writeback);
1310 for_each_node_state(n, N_HIGH_MEMORY)
1312 seq_printf(m, " N%d=%lu", n, md->node[n]);
1316 if (m->count < m->size)
1317 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1321 static const struct seq_operations proc_pid_numa_maps_op = {
1325 .show = show_numa_map,
1328 static int numa_maps_open(struct inode *inode, struct file *file)
1330 struct numa_maps_private *priv;
1332 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1334 priv->proc_maps.pid = proc_pid(inode);
1335 ret = seq_open(file, &proc_pid_numa_maps_op);
1337 struct seq_file *m = file->private_data;
1346 const struct file_operations proc_numa_maps_operations = {
1347 .open = numa_maps_open,
1349 .llseek = seq_lseek,
1350 .release = seq_release_private,
1352 #endif /* CONFIG_NUMA */