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 int show_arm_cache_attrs(struct seq_file *m, void *pgd,
214 unsigned long start, unsigned long end, vm_flags_t flags)
216 static const char *cache_attrs4[4] = { "noC", "WB-WA", "WT-noWA", "WB-noWA" };
217 u32 *arm_pgd = pgd, *cpt;
229 desc1 = arm_pgd[start >> 20];
233 sprintf(buf, "l1_fault");
238 tex_cb = ((desc1 >> 2) & 0x03) | ((desc1 >> 10) & 0x1c);
239 s = (desc1 >> 16) & 1;
240 xn = (desc1 >> 4) & 1;
241 ap = ((desc1 >> 10) & 3) | ((desc1 >> 13) & 4);
244 sprintf(buf, "reserved");
248 cpt = __va(desc1 & 0xfffffc00);
249 desc2 = cpt[(start >> 12) & 0xff];
253 sprintf(buf, "l2_fault");
256 tex_cb = ((desc2 >> 2) & 0x03) | ((desc2 >> 10) & 0x1c);
257 s = (desc2 >> 10) & 1;
258 xn = (desc2 >> 15) & 1;
259 ap = ((desc2 >> 4) & 3) | ((desc1 >> 7) & 4);
263 tex_cb = ((desc2 >> 2) & 0x03) | ((desc2 >> 4) & 0x1c);
264 s = (desc2 >> 10) & 1;
266 ap = ((desc2 >> 4) & 3) | ((desc1 >> 7) & 4);
271 asm ("mrc p15, 0, %0, c1, c0, 0" : "=r"(control));
272 asm ("mrc p15, 0, %0, c10, c2, 0" : "=r"(prrr)); // primary region RR
273 asm ("mrc p15, 0, %0, c10, c2, 1" : "=r"(nmrr)); // normal memory RR
275 if (control & (1 << 28)) { // TEX remap
276 // S (shareable) bit remapping
277 char s_normal[2] = { (prrr >> 18) & 1, (prrr >> 19) & 1 };
278 char s_device[2] = { (prrr >> 16) & 1, (prrr >> 17) & 1 };
282 type = (prrr >> tex_cb * 2) & 3;
285 sprintf(buf, "strongly-ordered");
288 sprintf(buf, "device");
292 sprintf(buf, "reserved/normal");
295 sprintf(buf + strlen(buf), "inner-%s-outer-%s",
296 cache_attrs4[(nmrr >> tex_cb * 2) & 3],
297 cache_attrs4[(nmrr >> (tex_cb * 2 + 16)) & 3]);
300 else if (tex_cb & 0x10) { // TEX[2] set
301 sprintf(buf, "inner-%s-outer-%s",
302 cache_attrs4[tex_cb & 3], cache_attrs4[(tex_cb >> 2) & 3]);
306 case 0x00: sprintf(buf, "strongly-ordered"); s = 1; break;
307 case 0x01: sprintf(buf, "shareable-device"); s = 1; break;
308 case 0x02: sprintf(buf, "inner-outer-WT-noWA"); break;
309 case 0x03: sprintf(buf, "inner-outer-WB-noWA"); break;
310 case 0x04: sprintf(buf, "inner-outer-non-cacheable"); break;
311 case 0x06: sprintf(buf, "implementation-defined"); break;
312 case 0x07: sprintf(buf, "inner-outer-WB-WA"); break;
313 case 0x08: sprintf(buf, "non-shareable-device"); s = 0; break;
314 default: sprintf(buf, "reserved"); break;
319 sprintf(buf + strlen(buf), "-shareable");
322 // use user permissions here
323 if (control & (1 << 29)) // AFE
324 sprintf(rw, "%c%c", (ap & 2) ? 'r' : '-',
325 ((ap & 2) && !(ap & 4)) ? 'w' : '-');
327 sprintf(rw, "%c%c", (ap & 2) ? 'r' : '-',
328 (ap == 3) ? 'w' : '-');
330 seq_printf(m, "%08lx-%08lx %s%c %-28s %n",
339 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma,
342 struct mm_struct *mm = vma->vm_mm;
343 struct file *file = vma->vm_file;
344 vm_flags_t flags = vma->vm_flags;
345 unsigned long ino = 0;
346 unsigned long long pgoff = 0;
347 unsigned long start, end;
352 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
353 dev = inode->i_sb->s_dev;
355 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
358 /* We don't show the stack guard page in /proc/maps */
359 start = vma->vm_start;
360 if (stack_guard_page_start(vma, start))
363 if (stack_guard_page_end(vma, end))
368 len = show_arm_cache_attrs(m, mm->pgd, start, end, flags);
371 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
374 flags & VM_READ ? 'r' : '-',
375 flags & VM_WRITE ? 'w' : '-',
376 flags & VM_EXEC ? 'x' : '-',
377 flags & VM_MAYSHARE ? 's' : 'p',
379 MAJOR(dev), MINOR(dev), ino, &len);
382 * Print the dentry name for named mappings, and a
383 * special [heap] marker for the heap:
386 pad_len_spaces(m, len);
387 seq_path(m, &file->f_path, "\n");
389 const char *name = arch_vma_name(vma);
392 if (vma->vm_start <= mm->brk &&
393 vma->vm_end >= mm->start_brk) {
395 } else if (vma->vm_start <= mm->start_stack &&
396 vma->vm_end >= mm->start_stack) {
404 pad_len_spaces(m, len);
411 static int show_map(struct seq_file *m, void *v)
413 struct vm_area_struct *vma = v;
414 struct proc_maps_private *priv = m->private;
415 struct task_struct *task = priv->task;
417 show_map_vma(m, vma, 0);
419 if (m->count < m->size) /* vma is copied successfully */
420 m->version = (vma != get_gate_vma(task->mm))
425 static const struct seq_operations proc_pid_maps_op = {
432 static int maps_open(struct inode *inode, struct file *file)
434 return do_maps_open(inode, file, &proc_pid_maps_op);
437 const struct file_operations proc_maps_operations = {
441 .release = seq_release_private,
444 static int show_armv7_map(struct seq_file *m, void *v)
446 struct vm_area_struct *vma = v;
447 struct proc_maps_private *priv = m->private;
448 struct task_struct *task = priv->task;
450 show_map_vma(m, vma, 1);
452 if (m->count < m->size) /* vma is copied successfully */
453 m->version = (vma != get_gate_vma(task->mm))
458 static const struct seq_operations proc_pid_armv7_maps_op = {
462 .show = show_armv7_map
465 static int armv7_maps_open(struct inode *inode, struct file *file)
467 return do_maps_open(inode, file, &proc_pid_armv7_maps_op);
470 const struct file_operations proc_armv7_maps_operations = {
471 .open = armv7_maps_open,
474 .release = seq_release_private,
478 * Proportional Set Size(PSS): my share of RSS.
480 * PSS of a process is the count of pages it has in memory, where each
481 * page is divided by the number of processes sharing it. So if a
482 * process has 1000 pages all to itself, and 1000 shared with one other
483 * process, its PSS will be 1500.
485 * To keep (accumulated) division errors low, we adopt a 64bit
486 * fixed-point pss counter to minimize division errors. So (pss >>
487 * PSS_SHIFT) would be the real byte count.
489 * A shift of 12 before division means (assuming 4K page size):
490 * - 1M 3-user-pages add up to 8KB errors;
491 * - supports mapcount up to 2^24, or 16M;
492 * - supports PSS up to 2^52 bytes, or 4PB.
496 #ifdef CONFIG_PROC_PAGE_MONITOR
497 struct mem_size_stats {
498 struct vm_area_struct *vma;
499 unsigned long resident;
500 unsigned long shared_clean;
501 unsigned long shared_dirty;
502 unsigned long private_clean;
503 unsigned long private_dirty;
504 unsigned long referenced;
505 unsigned long anonymous;
506 unsigned long anonymous_thp;
512 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
513 unsigned long ptent_size, struct mm_walk *walk)
515 struct mem_size_stats *mss = walk->private;
516 struct vm_area_struct *vma = mss->vma;
520 if (is_swap_pte(ptent)) {
521 mss->swap += ptent_size;
525 if (!pte_present(ptent))
528 page = vm_normal_page(vma, addr, ptent);
533 mss->anonymous += ptent_size;
535 mss->resident += ptent_size;
536 /* Accumulate the size in pages that have been accessed. */
537 if (pte_young(ptent) || PageReferenced(page))
538 mss->referenced += ptent_size;
539 mapcount = page_mapcount(page);
541 if (pte_dirty(ptent) || PageDirty(page))
542 mss->shared_dirty += ptent_size;
544 mss->shared_clean += ptent_size;
545 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
547 if (pte_dirty(ptent) || PageDirty(page))
548 mss->private_dirty += ptent_size;
550 mss->private_clean += ptent_size;
551 mss->pss += (ptent_size << PSS_SHIFT);
555 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
556 struct mm_walk *walk)
558 struct mem_size_stats *mss = walk->private;
559 struct vm_area_struct *vma = mss->vma;
563 spin_lock(&walk->mm->page_table_lock);
564 if (pmd_trans_huge(*pmd)) {
565 if (pmd_trans_splitting(*pmd)) {
566 spin_unlock(&walk->mm->page_table_lock);
567 wait_split_huge_page(vma->anon_vma, pmd);
569 smaps_pte_entry(*(pte_t *)pmd, addr,
570 HPAGE_PMD_SIZE, walk);
571 spin_unlock(&walk->mm->page_table_lock);
572 mss->anonymous_thp += HPAGE_PMD_SIZE;
576 spin_unlock(&walk->mm->page_table_lock);
579 if (pmd_trans_unstable(pmd))
582 * The mmap_sem held all the way back in m_start() is what
583 * keeps khugepaged out of here and from collapsing things
586 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
587 for (; addr != end; pte++, addr += PAGE_SIZE)
588 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
589 pte_unmap_unlock(pte - 1, ptl);
594 static int show_smap(struct seq_file *m, void *v)
596 struct proc_maps_private *priv = m->private;
597 struct task_struct *task = priv->task;
598 struct vm_area_struct *vma = v;
599 struct mem_size_stats mss;
600 struct mm_walk smaps_walk = {
601 .pmd_entry = smaps_pte_range,
606 memset(&mss, 0, sizeof mss);
608 /* mmap_sem is held in m_start */
609 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
610 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
612 show_map_vma(m, vma, 0);
618 "Shared_Clean: %8lu kB\n"
619 "Shared_Dirty: %8lu kB\n"
620 "Private_Clean: %8lu kB\n"
621 "Private_Dirty: %8lu kB\n"
622 "Referenced: %8lu kB\n"
623 "Anonymous: %8lu kB\n"
624 "AnonHugePages: %8lu kB\n"
626 "KernelPageSize: %8lu kB\n"
627 "MMUPageSize: %8lu kB\n"
629 (vma->vm_end - vma->vm_start) >> 10,
631 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
632 mss.shared_clean >> 10,
633 mss.shared_dirty >> 10,
634 mss.private_clean >> 10,
635 mss.private_dirty >> 10,
636 mss.referenced >> 10,
638 mss.anonymous_thp >> 10,
640 vma_kernel_pagesize(vma) >> 10,
641 vma_mmu_pagesize(vma) >> 10,
642 (vma->vm_flags & VM_LOCKED) ?
643 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
645 if (m->count < m->size) /* vma is copied successfully */
646 m->version = (vma != get_gate_vma(task->mm))
651 static const struct seq_operations proc_pid_smaps_op = {
658 static int smaps_open(struct inode *inode, struct file *file)
660 return do_maps_open(inode, file, &proc_pid_smaps_op);
663 const struct file_operations proc_smaps_operations = {
667 .release = seq_release_private,
670 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
671 unsigned long end, struct mm_walk *walk)
673 struct vm_area_struct *vma = walk->private;
678 split_huge_page_pmd(walk->mm, pmd);
679 if (pmd_trans_unstable(pmd))
682 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
683 for (; addr != end; pte++, addr += PAGE_SIZE) {
685 if (!pte_present(ptent))
688 page = vm_normal_page(vma, addr, ptent);
692 if (PageReserved(page))
695 /* Clear accessed and referenced bits. */
696 ptep_test_and_clear_young(vma, addr, pte);
697 ClearPageReferenced(page);
699 pte_unmap_unlock(pte - 1, ptl);
704 #define CLEAR_REFS_ALL 1
705 #define CLEAR_REFS_ANON 2
706 #define CLEAR_REFS_MAPPED 3
708 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
709 size_t count, loff_t *ppos)
711 struct task_struct *task;
712 char buffer[PROC_NUMBUF];
713 struct mm_struct *mm;
714 struct vm_area_struct *vma;
718 memset(buffer, 0, sizeof(buffer));
719 if (count > sizeof(buffer) - 1)
720 count = sizeof(buffer) - 1;
721 if (copy_from_user(buffer, buf, count))
723 rv = kstrtoint(strstrip(buffer), 10, &type);
726 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
728 task = get_proc_task(file->f_path.dentry->d_inode);
731 mm = get_task_mm(task);
733 struct mm_walk clear_refs_walk = {
734 .pmd_entry = clear_refs_pte_range,
737 down_read(&mm->mmap_sem);
738 for (vma = mm->mmap; vma; vma = vma->vm_next) {
739 clear_refs_walk.private = vma;
740 if (is_vm_hugetlb_page(vma))
743 * Writing 1 to /proc/pid/clear_refs affects all pages.
745 * Writing 2 to /proc/pid/clear_refs only affects
748 * Writing 3 to /proc/pid/clear_refs only affects file
751 if (type == CLEAR_REFS_ANON && vma->vm_file)
753 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
755 walk_page_range(vma->vm_start, vma->vm_end,
759 up_read(&mm->mmap_sem);
762 put_task_struct(task);
767 const struct file_operations proc_clear_refs_operations = {
768 .write = clear_refs_write,
769 .llseek = noop_llseek,
773 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */
777 #define PM_ENTRY_BYTES sizeof(u64)
778 #define PM_STATUS_BITS 3
779 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
780 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
781 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
782 #define PM_PSHIFT_BITS 6
783 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
784 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
785 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
786 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
787 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
789 #define PM_PRESENT PM_STATUS(4LL)
790 #define PM_SWAP PM_STATUS(2LL)
791 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
792 #define PM_END_OF_BUFFER 1
794 static int add_to_pagemap(unsigned long addr, u64 pfn,
795 struct pagemapread *pm)
797 pm->buffer[pm->pos++] = pfn;
798 if (pm->pos >= pm->len)
799 return PM_END_OF_BUFFER;
803 static int pagemap_pte_hole(unsigned long start, unsigned long end,
804 struct mm_walk *walk)
806 struct pagemapread *pm = walk->private;
809 for (addr = start; addr < end; addr += PAGE_SIZE) {
810 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
817 static u64 swap_pte_to_pagemap_entry(pte_t pte)
819 swp_entry_t e = pte_to_swp_entry(pte);
820 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
823 static u64 pte_to_pagemap_entry(pte_t pte)
826 if (is_swap_pte(pte))
827 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
828 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
829 else if (pte_present(pte))
830 pme = PM_PFRAME(pte_pfn(pte))
831 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
835 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
836 struct mm_walk *walk)
838 struct vm_area_struct *vma;
839 struct pagemapread *pm = walk->private;
843 split_huge_page_pmd(walk->mm, pmd);
844 if (pmd_trans_unstable(pmd))
847 /* find the first VMA at or above 'addr' */
848 vma = find_vma(walk->mm, addr);
849 for (; addr != end; addr += PAGE_SIZE) {
850 u64 pfn = PM_NOT_PRESENT;
852 /* check to see if we've left 'vma' behind
853 * and need a new, higher one */
854 if (vma && (addr >= vma->vm_end))
855 vma = find_vma(walk->mm, addr);
857 /* check that 'vma' actually covers this address,
858 * and that it isn't a huge page vma */
859 if (vma && (vma->vm_start <= addr) &&
860 !is_vm_hugetlb_page(vma)) {
861 pte = pte_offset_map(pmd, addr);
862 pfn = pte_to_pagemap_entry(*pte);
863 /* unmap before userspace copy */
866 err = add_to_pagemap(addr, pfn, pm);
876 #ifdef CONFIG_HUGETLB_PAGE
877 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
880 if (pte_present(pte))
881 pme = PM_PFRAME(pte_pfn(pte) + offset)
882 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
886 /* This function walks within one hugetlb entry in the single call */
887 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
888 unsigned long addr, unsigned long end,
889 struct mm_walk *walk)
891 struct pagemapread *pm = walk->private;
895 for (; addr != end; addr += PAGE_SIZE) {
896 int offset = (addr & ~hmask) >> PAGE_SHIFT;
897 pfn = huge_pte_to_pagemap_entry(*pte, offset);
898 err = add_to_pagemap(addr, pfn, pm);
907 #endif /* HUGETLB_PAGE */
910 * /proc/pid/pagemap - an array mapping virtual pages to pfns
912 * For each page in the address space, this file contains one 64-bit entry
913 * consisting of the following:
915 * Bits 0-55 page frame number (PFN) if present
916 * Bits 0-4 swap type if swapped
917 * Bits 5-55 swap offset if swapped
918 * Bits 55-60 page shift (page size = 1<<page shift)
919 * Bit 61 reserved for future use
920 * Bit 62 page swapped
921 * Bit 63 page present
923 * If the page is not present but in swap, then the PFN contains an
924 * encoding of the swap file number and the page's offset into the
925 * swap. Unmapped pages return a null PFN. This allows determining
926 * precisely which pages are mapped (or in swap) and comparing mapped
927 * pages between processes.
929 * Efficient users of this interface will use /proc/pid/maps to
930 * determine which areas of memory are actually mapped and llseek to
931 * skip over unmapped regions.
933 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
934 #define PAGEMAP_WALK_MASK (PMD_MASK)
935 static ssize_t pagemap_read(struct file *file, char __user *buf,
936 size_t count, loff_t *ppos)
938 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
939 struct mm_struct *mm;
940 struct pagemapread pm;
942 struct mm_walk pagemap_walk = {};
945 unsigned long start_vaddr;
946 unsigned long end_vaddr;
953 /* file position must be aligned */
954 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
961 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
962 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
967 mm = mm_for_maps(task);
969 if (!mm || IS_ERR(mm))
972 pagemap_walk.pmd_entry = pagemap_pte_range;
973 pagemap_walk.pte_hole = pagemap_pte_hole;
974 #ifdef CONFIG_HUGETLB_PAGE
975 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
977 pagemap_walk.mm = mm;
978 pagemap_walk.private = ±
981 svpfn = src / PM_ENTRY_BYTES;
982 start_vaddr = svpfn << PAGE_SHIFT;
983 end_vaddr = TASK_SIZE_OF(task);
985 /* watch out for wraparound */
986 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
987 start_vaddr = end_vaddr;
990 * The odds are that this will stop walking way
991 * before end_vaddr, because the length of the
992 * user buffer is tracked in "pm", and the walk
993 * will stop when we hit the end of the buffer.
996 while (count && (start_vaddr < end_vaddr)) {
1001 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1003 if (end < start_vaddr || end > end_vaddr)
1005 down_read(&mm->mmap_sem);
1006 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1007 up_read(&mm->mmap_sem);
1010 len = min(count, PM_ENTRY_BYTES * pm.pos);
1011 if (copy_to_user(buf, pm.buffer, len)) {
1020 if (!ret || ret == PM_END_OF_BUFFER)
1028 put_task_struct(task);
1033 const struct file_operations proc_pagemap_operations = {
1034 .llseek = mem_lseek, /* borrow this */
1035 .read = pagemap_read,
1037 #endif /* CONFIG_PROC_PAGE_MONITOR */
1042 struct vm_area_struct *vma;
1043 unsigned long pages;
1045 unsigned long active;
1046 unsigned long writeback;
1047 unsigned long mapcount_max;
1048 unsigned long dirty;
1049 unsigned long swapcache;
1050 unsigned long node[MAX_NUMNODES];
1053 struct numa_maps_private {
1054 struct proc_maps_private proc_maps;
1055 struct numa_maps md;
1058 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1059 unsigned long nr_pages)
1061 int count = page_mapcount(page);
1063 md->pages += nr_pages;
1064 if (pte_dirty || PageDirty(page))
1065 md->dirty += nr_pages;
1067 if (PageSwapCache(page))
1068 md->swapcache += nr_pages;
1070 if (PageActive(page) || PageUnevictable(page))
1071 md->active += nr_pages;
1073 if (PageWriteback(page))
1074 md->writeback += nr_pages;
1077 md->anon += nr_pages;
1079 if (count > md->mapcount_max)
1080 md->mapcount_max = count;
1082 md->node[page_to_nid(page)] += nr_pages;
1085 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1091 if (!pte_present(pte))
1094 page = vm_normal_page(vma, addr, pte);
1098 if (PageReserved(page))
1101 nid = page_to_nid(page);
1102 if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
1108 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1109 unsigned long end, struct mm_walk *walk)
1111 struct numa_maps *md;
1117 spin_lock(&walk->mm->page_table_lock);
1118 if (pmd_trans_huge(*pmd)) {
1119 if (pmd_trans_splitting(*pmd)) {
1120 spin_unlock(&walk->mm->page_table_lock);
1121 wait_split_huge_page(md->vma->anon_vma, pmd);
1123 pte_t huge_pte = *(pte_t *)pmd;
1126 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1128 gather_stats(page, md, pte_dirty(huge_pte),
1129 HPAGE_PMD_SIZE/PAGE_SIZE);
1130 spin_unlock(&walk->mm->page_table_lock);
1134 spin_unlock(&walk->mm->page_table_lock);
1137 if (pmd_trans_unstable(pmd))
1139 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1141 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1144 gather_stats(page, md, pte_dirty(*pte), 1);
1146 } while (pte++, addr += PAGE_SIZE, addr != end);
1147 pte_unmap_unlock(orig_pte, ptl);
1150 #ifdef CONFIG_HUGETLB_PAGE
1151 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1152 unsigned long addr, unsigned long end, struct mm_walk *walk)
1154 struct numa_maps *md;
1160 page = pte_page(*pte);
1165 gather_stats(page, md, pte_dirty(*pte), 1);
1170 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1171 unsigned long addr, unsigned long end, struct mm_walk *walk)
1178 * Display pages allocated per node and memory policy via /proc.
1180 static int show_numa_map(struct seq_file *m, void *v)
1182 struct numa_maps_private *numa_priv = m->private;
1183 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1184 struct vm_area_struct *vma = v;
1185 struct numa_maps *md = &numa_priv->md;
1186 struct file *file = vma->vm_file;
1187 struct mm_struct *mm = vma->vm_mm;
1188 struct mm_walk walk = {};
1189 struct mempolicy *pol;
1196 /* Ensure we start with an empty set of numa_maps statistics. */
1197 memset(md, 0, sizeof(*md));
1201 walk.hugetlb_entry = gather_hugetbl_stats;
1202 walk.pmd_entry = gather_pte_stats;
1206 pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
1207 mpol_to_str(buffer, sizeof(buffer), pol, 0);
1210 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1213 seq_printf(m, " file=");
1214 seq_path(m, &file->f_path, "\n\t= ");
1215 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1216 seq_printf(m, " heap");
1217 } else if (vma->vm_start <= mm->start_stack &&
1218 vma->vm_end >= mm->start_stack) {
1219 seq_printf(m, " stack");
1222 if (is_vm_hugetlb_page(vma))
1223 seq_printf(m, " huge");
1225 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1231 seq_printf(m, " anon=%lu", md->anon);
1234 seq_printf(m, " dirty=%lu", md->dirty);
1236 if (md->pages != md->anon && md->pages != md->dirty)
1237 seq_printf(m, " mapped=%lu", md->pages);
1239 if (md->mapcount_max > 1)
1240 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1243 seq_printf(m, " swapcache=%lu", md->swapcache);
1245 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1246 seq_printf(m, " active=%lu", md->active);
1249 seq_printf(m, " writeback=%lu", md->writeback);
1251 for_each_node_state(n, N_HIGH_MEMORY)
1253 seq_printf(m, " N%d=%lu", n, md->node[n]);
1257 if (m->count < m->size)
1258 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1262 static const struct seq_operations proc_pid_numa_maps_op = {
1266 .show = show_numa_map,
1269 static int numa_maps_open(struct inode *inode, struct file *file)
1271 struct numa_maps_private *priv;
1273 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1275 priv->proc_maps.pid = proc_pid(inode);
1276 ret = seq_open(file, &proc_pid_numa_maps_op);
1278 struct seq_file *m = file->private_data;
1287 const struct file_operations proc_numa_maps_operations = {
1288 .open = numa_maps_open,
1290 .llseek = seq_lseek,
1291 .release = seq_release_private,
1293 #endif /* CONFIG_NUMA */