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>
14 #include <linux/mmu_notifier.h>
17 #include <asm/uaccess.h>
18 #include <asm/tlbflush.h>
21 void task_mem(struct seq_file
*m
, struct mm_struct
*mm
)
23 unsigned long data
, text
, lib
, swap
;
24 unsigned long hiwater_vm
, total_vm
, hiwater_rss
, total_rss
;
27 * Note: to minimize their overhead, mm maintains hiwater_vm and
28 * hiwater_rss only when about to *lower* total_vm or rss. Any
29 * collector of these hiwater stats must therefore get total_vm
30 * and rss too, which will usually be the higher. Barriers? not
31 * worth the effort, such snapshots can always be inconsistent.
33 hiwater_vm
= total_vm
= mm
->total_vm
;
34 if (hiwater_vm
< mm
->hiwater_vm
)
35 hiwater_vm
= mm
->hiwater_vm
;
36 hiwater_rss
= total_rss
= get_mm_rss(mm
);
37 if (hiwater_rss
< mm
->hiwater_rss
)
38 hiwater_rss
= mm
->hiwater_rss
;
40 data
= mm
->total_vm
- mm
->shared_vm
- mm
->stack_vm
;
41 text
= (PAGE_ALIGN(mm
->end_code
) - (mm
->start_code
& PAGE_MASK
)) >> 10;
42 lib
= (mm
->exec_vm
<< (PAGE_SHIFT
-10)) - text
;
43 swap
= get_mm_counter(mm
, MM_SWAPENTS
);
57 hiwater_vm
<< (PAGE_SHIFT
-10),
58 total_vm
<< (PAGE_SHIFT
-10),
59 mm
->locked_vm
<< (PAGE_SHIFT
-10),
60 mm
->pinned_vm
<< (PAGE_SHIFT
-10),
61 hiwater_rss
<< (PAGE_SHIFT
-10),
62 total_rss
<< (PAGE_SHIFT
-10),
63 data
<< (PAGE_SHIFT
-10),
64 mm
->stack_vm
<< (PAGE_SHIFT
-10), text
, lib
,
65 (PTRS_PER_PTE
* sizeof(pte_t
) *
66 atomic_long_read(&mm
->nr_ptes
)) >> 10,
67 swap
<< (PAGE_SHIFT
-10));
70 unsigned long task_vsize(struct mm_struct
*mm
)
72 return PAGE_SIZE
* mm
->total_vm
;
75 unsigned long task_statm(struct mm_struct
*mm
,
76 unsigned long *shared
, unsigned long *text
,
77 unsigned long *data
, unsigned long *resident
)
79 *shared
= get_mm_counter(mm
, MM_FILEPAGES
);
80 *text
= (PAGE_ALIGN(mm
->end_code
) - (mm
->start_code
& PAGE_MASK
))
82 *data
= mm
->total_vm
- mm
->shared_vm
;
83 *resident
= *shared
+ get_mm_counter(mm
, MM_ANONPAGES
);
87 static void pad_len_spaces(struct seq_file
*m
, int len
)
89 len
= 25 + sizeof(void*) * 6 - len
;
92 seq_printf(m
, "%*c", len
, ' ');
97 * These functions are for numa_maps but called in generic **maps seq_file
98 * ->start(), ->stop() ops.
100 * numa_maps scans all vmas under mmap_sem and checks their mempolicy.
101 * Each mempolicy object is controlled by reference counting. The problem here
102 * is how to avoid accessing dead mempolicy object.
104 * Because we're holding mmap_sem while reading seq_file, it's safe to access
105 * each vma's mempolicy, no vma objects will never drop refs to mempolicy.
107 * A task's mempolicy (task->mempolicy) has different behavior. task->mempolicy
108 * is set and replaced under mmap_sem but unrefed and cleared under task_lock().
109 * So, without task_lock(), we cannot trust get_vma_policy() because we cannot
110 * gurantee the task never exits under us. But taking task_lock() around
111 * get_vma_plicy() causes lock order problem.
113 * To access task->mempolicy without lock, we hold a reference count of an
114 * object pointed by task->mempolicy and remember it. This will guarantee
115 * that task->mempolicy points to an alive object or NULL in numa_maps accesses.
117 static void hold_task_mempolicy(struct proc_maps_private
*priv
)
119 struct task_struct
*task
= priv
->task
;
122 priv
->task_mempolicy
= task
->mempolicy
;
123 mpol_get(priv
->task_mempolicy
);
126 static void release_task_mempolicy(struct proc_maps_private
*priv
)
128 mpol_put(priv
->task_mempolicy
);
131 static void hold_task_mempolicy(struct proc_maps_private
*priv
)
134 static void release_task_mempolicy(struct proc_maps_private
*priv
)
139 static void vma_stop(struct proc_maps_private
*priv
, struct vm_area_struct
*vma
)
141 if (vma
&& vma
!= priv
->tail_vma
) {
142 struct mm_struct
*mm
= vma
->vm_mm
;
143 release_task_mempolicy(priv
);
144 up_read(&mm
->mmap_sem
);
149 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
151 struct proc_maps_private
*priv
= m
->private;
152 unsigned long last_addr
= m
->version
;
153 struct mm_struct
*mm
;
154 struct vm_area_struct
*vma
, *tail_vma
= NULL
;
157 /* Clear the per syscall fields in priv */
159 priv
->tail_vma
= NULL
;
162 * We remember last_addr rather than next_addr to hit with
163 * mmap_cache most of the time. We have zero last_addr at
164 * the beginning and also after lseek. We will have -1 last_addr
165 * after the end of the vmas.
168 if (last_addr
== -1UL)
171 priv
->task
= get_pid_task(priv
->pid
, PIDTYPE_PID
);
173 return ERR_PTR(-ESRCH
);
175 mm
= mm_access(priv
->task
, PTRACE_MODE_READ
);
176 if (!mm
|| IS_ERR(mm
))
178 down_read(&mm
->mmap_sem
);
180 tail_vma
= get_gate_vma(priv
->task
->mm
);
181 priv
->tail_vma
= tail_vma
;
182 hold_task_mempolicy(priv
);
183 /* Start with last addr hint */
184 vma
= find_vma(mm
, last_addr
);
185 if (last_addr
&& vma
) {
191 * Check the vma index is within the range and do
192 * sequential scan until m_index.
195 if ((unsigned long)l
< mm
->map_count
) {
202 if (l
!= mm
->map_count
)
203 tail_vma
= NULL
; /* After gate vma */
209 release_task_mempolicy(priv
);
210 /* End of vmas has been reached */
211 m
->version
= (tail_vma
!= NULL
)? 0: -1UL;
212 up_read(&mm
->mmap_sem
);
217 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
219 struct proc_maps_private
*priv
= m
->private;
220 struct vm_area_struct
*vma
= v
;
221 struct vm_area_struct
*tail_vma
= priv
->tail_vma
;
224 if (vma
&& (vma
!= tail_vma
) && vma
->vm_next
)
227 return (vma
!= tail_vma
)? tail_vma
: NULL
;
230 static void m_stop(struct seq_file
*m
, void *v
)
232 struct proc_maps_private
*priv
= m
->private;
233 struct vm_area_struct
*vma
= v
;
238 put_task_struct(priv
->task
);
241 static int do_maps_open(struct inode
*inode
, struct file
*file
,
242 const struct seq_operations
*ops
)
244 struct proc_maps_private
*priv
;
246 priv
= kzalloc(sizeof(*priv
), GFP_KERNEL
);
248 priv
->pid
= proc_pid(inode
);
249 ret
= seq_open(file
, ops
);
251 struct seq_file
*m
= file
->private_data
;
261 show_map_vma(struct seq_file
*m
, struct vm_area_struct
*vma
, int is_pid
)
263 struct mm_struct
*mm
= vma
->vm_mm
;
264 struct file
*file
= vma
->vm_file
;
265 struct proc_maps_private
*priv
= m
->private;
266 struct task_struct
*task
= priv
->task
;
267 vm_flags_t flags
= vma
->vm_flags
;
268 unsigned long ino
= 0;
269 unsigned long long pgoff
= 0;
270 unsigned long start
, end
;
273 const char *name
= NULL
;
276 struct inode
*inode
= file_inode(vma
->vm_file
);
277 dev
= inode
->i_sb
->s_dev
;
279 pgoff
= ((loff_t
)vma
->vm_pgoff
) << PAGE_SHIFT
;
282 /* We don't show the stack guard page in /proc/maps */
283 start
= vma
->vm_start
;
284 if (stack_guard_page_start(vma
, start
))
287 if (stack_guard_page_end(vma
, end
))
290 seq_printf(m
, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
293 flags
& VM_READ
? 'r' : '-',
294 flags
& VM_WRITE
? 'w' : '-',
295 flags
& VM_EXEC
? 'x' : '-',
296 flags
& VM_MAYSHARE
? 's' : 'p',
298 MAJOR(dev
), MINOR(dev
), ino
, &len
);
301 * Print the dentry name for named mappings, and a
302 * special [heap] marker for the heap:
305 pad_len_spaces(m
, len
);
306 seq_path(m
, &file
->f_path
, "\n");
310 name
= arch_vma_name(vma
);
319 if (vma
->vm_start
<= mm
->brk
&&
320 vma
->vm_end
>= mm
->start_brk
) {
325 tid
= vm_is_stack(task
, vma
, is_pid
);
329 * Thread stack in /proc/PID/task/TID/maps or
330 * the main process stack.
332 if (!is_pid
|| (vma
->vm_start
<= mm
->start_stack
&&
333 vma
->vm_end
>= mm
->start_stack
)) {
336 /* Thread stack in /proc/PID/maps */
337 pad_len_spaces(m
, len
);
338 seq_printf(m
, "[stack:%d]", tid
);
345 pad_len_spaces(m
, len
);
351 static int show_map(struct seq_file
*m
, void *v
, int is_pid
)
353 struct vm_area_struct
*vma
= v
;
354 struct proc_maps_private
*priv
= m
->private;
355 struct task_struct
*task
= priv
->task
;
357 show_map_vma(m
, vma
, is_pid
);
359 if (m
->count
< m
->size
) /* vma is copied successfully */
360 m
->version
= (vma
!= get_gate_vma(task
->mm
))
365 static int show_pid_map(struct seq_file
*m
, void *v
)
367 return show_map(m
, v
, 1);
370 static int show_tid_map(struct seq_file
*m
, void *v
)
372 return show_map(m
, v
, 0);
375 static const struct seq_operations proc_pid_maps_op
= {
382 static const struct seq_operations proc_tid_maps_op
= {
389 static int pid_maps_open(struct inode
*inode
, struct file
*file
)
391 return do_maps_open(inode
, file
, &proc_pid_maps_op
);
394 static int tid_maps_open(struct inode
*inode
, struct file
*file
)
396 return do_maps_open(inode
, file
, &proc_tid_maps_op
);
399 const struct file_operations proc_pid_maps_operations
= {
400 .open
= pid_maps_open
,
403 .release
= seq_release_private
,
406 const struct file_operations proc_tid_maps_operations
= {
407 .open
= tid_maps_open
,
410 .release
= seq_release_private
,
414 * Proportional Set Size(PSS): my share of RSS.
416 * PSS of a process is the count of pages it has in memory, where each
417 * page is divided by the number of processes sharing it. So if a
418 * process has 1000 pages all to itself, and 1000 shared with one other
419 * process, its PSS will be 1500.
421 * To keep (accumulated) division errors low, we adopt a 64bit
422 * fixed-point pss counter to minimize division errors. So (pss >>
423 * PSS_SHIFT) would be the real byte count.
425 * A shift of 12 before division means (assuming 4K page size):
426 * - 1M 3-user-pages add up to 8KB errors;
427 * - supports mapcount up to 2^24, or 16M;
428 * - supports PSS up to 2^52 bytes, or 4PB.
432 #ifdef CONFIG_PROC_PAGE_MONITOR
433 struct mem_size_stats
{
434 struct vm_area_struct
*vma
;
435 unsigned long resident
;
436 unsigned long shared_clean
;
437 unsigned long shared_dirty
;
438 unsigned long private_clean
;
439 unsigned long private_dirty
;
440 unsigned long referenced
;
441 unsigned long anonymous
;
442 unsigned long anonymous_thp
;
444 unsigned long nonlinear
;
449 static void smaps_pte_entry(pte_t ptent
, unsigned long addr
,
450 unsigned long ptent_size
, struct mm_walk
*walk
)
452 struct mem_size_stats
*mss
= walk
->private;
453 struct vm_area_struct
*vma
= mss
->vma
;
454 pgoff_t pgoff
= linear_page_index(vma
, addr
);
455 struct page
*page
= NULL
;
458 if (pte_present(ptent
)) {
459 page
= vm_normal_page(vma
, addr
, ptent
);
460 } else if (is_swap_pte(ptent
)) {
461 swp_entry_t swpent
= pte_to_swp_entry(ptent
);
463 if (!non_swap_entry(swpent
))
464 mss
->swap
+= ptent_size
;
465 else if (is_migration_entry(swpent
))
466 page
= migration_entry_to_page(swpent
);
467 } else if (pte_file(ptent
)) {
468 if (pte_to_pgoff(ptent
) != pgoff
)
469 mss
->nonlinear
+= ptent_size
;
476 mss
->anonymous
+= ptent_size
;
478 if (page
->index
!= pgoff
)
479 mss
->nonlinear
+= ptent_size
;
481 mss
->resident
+= ptent_size
;
482 /* Accumulate the size in pages that have been accessed. */
483 if (pte_young(ptent
) || PageReferenced(page
))
484 mss
->referenced
+= ptent_size
;
485 mapcount
= page_mapcount(page
);
487 if (pte_dirty(ptent
) || PageDirty(page
))
488 mss
->shared_dirty
+= ptent_size
;
490 mss
->shared_clean
+= ptent_size
;
491 mss
->pss
+= (ptent_size
<< PSS_SHIFT
) / mapcount
;
493 if (pte_dirty(ptent
) || PageDirty(page
))
494 mss
->private_dirty
+= ptent_size
;
496 mss
->private_clean
+= ptent_size
;
497 mss
->pss
+= (ptent_size
<< PSS_SHIFT
);
501 static int smaps_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
,
502 struct mm_walk
*walk
)
504 struct mem_size_stats
*mss
= walk
->private;
505 struct vm_area_struct
*vma
= mss
->vma
;
509 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
510 smaps_pte_entry(*(pte_t
*)pmd
, addr
, HPAGE_PMD_SIZE
, walk
);
511 spin_unlock(&walk
->mm
->page_table_lock
);
512 mss
->anonymous_thp
+= HPAGE_PMD_SIZE
;
516 if (pmd_trans_unstable(pmd
))
519 * The mmap_sem held all the way back in m_start() is what
520 * keeps khugepaged out of here and from collapsing things
523 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
524 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
525 smaps_pte_entry(*pte
, addr
, PAGE_SIZE
, walk
);
526 pte_unmap_unlock(pte
- 1, ptl
);
531 static void show_smap_vma_flags(struct seq_file
*m
, struct vm_area_struct
*vma
)
534 * Don't forget to update Documentation/ on changes.
536 static const char mnemonics
[BITS_PER_LONG
][2] = {
538 * In case if we meet a flag we don't know about.
540 [0 ... (BITS_PER_LONG
-1)] = "??",
542 [ilog2(VM_READ
)] = "rd",
543 [ilog2(VM_WRITE
)] = "wr",
544 [ilog2(VM_EXEC
)] = "ex",
545 [ilog2(VM_SHARED
)] = "sh",
546 [ilog2(VM_MAYREAD
)] = "mr",
547 [ilog2(VM_MAYWRITE
)] = "mw",
548 [ilog2(VM_MAYEXEC
)] = "me",
549 [ilog2(VM_MAYSHARE
)] = "ms",
550 [ilog2(VM_GROWSDOWN
)] = "gd",
551 [ilog2(VM_PFNMAP
)] = "pf",
552 [ilog2(VM_DENYWRITE
)] = "dw",
553 [ilog2(VM_LOCKED
)] = "lo",
554 [ilog2(VM_IO
)] = "io",
555 [ilog2(VM_SEQ_READ
)] = "sr",
556 [ilog2(VM_RAND_READ
)] = "rr",
557 [ilog2(VM_DONTCOPY
)] = "dc",
558 [ilog2(VM_DONTEXPAND
)] = "de",
559 [ilog2(VM_ACCOUNT
)] = "ac",
560 [ilog2(VM_NORESERVE
)] = "nr",
561 [ilog2(VM_HUGETLB
)] = "ht",
562 [ilog2(VM_NONLINEAR
)] = "nl",
563 [ilog2(VM_ARCH_1
)] = "ar",
564 [ilog2(VM_DONTDUMP
)] = "dd",
565 #ifdef CONFIG_MEM_SOFT_DIRTY
566 [ilog2(VM_SOFTDIRTY
)] = "sd",
568 [ilog2(VM_MIXEDMAP
)] = "mm",
569 [ilog2(VM_HUGEPAGE
)] = "hg",
570 [ilog2(VM_NOHUGEPAGE
)] = "nh",
571 [ilog2(VM_MERGEABLE
)] = "mg",
575 seq_puts(m
, "VmFlags: ");
576 for (i
= 0; i
< BITS_PER_LONG
; i
++) {
577 if (vma
->vm_flags
& (1UL << i
)) {
578 seq_printf(m
, "%c%c ",
579 mnemonics
[i
][0], mnemonics
[i
][1]);
585 static int show_smap(struct seq_file
*m
, void *v
, int is_pid
)
587 struct proc_maps_private
*priv
= m
->private;
588 struct task_struct
*task
= priv
->task
;
589 struct vm_area_struct
*vma
= v
;
590 struct mem_size_stats mss
;
591 struct mm_walk smaps_walk
= {
592 .pmd_entry
= smaps_pte_range
,
597 memset(&mss
, 0, sizeof mss
);
599 /* mmap_sem is held in m_start */
600 if (vma
->vm_mm
&& !is_vm_hugetlb_page(vma
))
601 walk_page_range(vma
->vm_start
, vma
->vm_end
, &smaps_walk
);
603 show_map_vma(m
, vma
, is_pid
);
609 "Shared_Clean: %8lu kB\n"
610 "Shared_Dirty: %8lu kB\n"
611 "Private_Clean: %8lu kB\n"
612 "Private_Dirty: %8lu kB\n"
613 "Referenced: %8lu kB\n"
614 "Anonymous: %8lu kB\n"
615 "AnonHugePages: %8lu kB\n"
617 "KernelPageSize: %8lu kB\n"
618 "MMUPageSize: %8lu kB\n"
620 (vma
->vm_end
- vma
->vm_start
) >> 10,
622 (unsigned long)(mss
.pss
>> (10 + PSS_SHIFT
)),
623 mss
.shared_clean
>> 10,
624 mss
.shared_dirty
>> 10,
625 mss
.private_clean
>> 10,
626 mss
.private_dirty
>> 10,
627 mss
.referenced
>> 10,
629 mss
.anonymous_thp
>> 10,
631 vma_kernel_pagesize(vma
) >> 10,
632 vma_mmu_pagesize(vma
) >> 10,
633 (vma
->vm_flags
& VM_LOCKED
) ?
634 (unsigned long)(mss
.pss
>> (10 + PSS_SHIFT
)) : 0);
636 if (vma
->vm_flags
& VM_NONLINEAR
)
637 seq_printf(m
, "Nonlinear: %8lu kB\n",
638 mss
.nonlinear
>> 10);
640 show_smap_vma_flags(m
, vma
);
642 if (m
->count
< m
->size
) /* vma is copied successfully */
643 m
->version
= (vma
!= get_gate_vma(task
->mm
))
648 static int show_pid_smap(struct seq_file
*m
, void *v
)
650 return show_smap(m
, v
, 1);
653 static int show_tid_smap(struct seq_file
*m
, void *v
)
655 return show_smap(m
, v
, 0);
658 static const struct seq_operations proc_pid_smaps_op
= {
662 .show
= show_pid_smap
665 static const struct seq_operations proc_tid_smaps_op
= {
669 .show
= show_tid_smap
672 static int pid_smaps_open(struct inode
*inode
, struct file
*file
)
674 return do_maps_open(inode
, file
, &proc_pid_smaps_op
);
677 static int tid_smaps_open(struct inode
*inode
, struct file
*file
)
679 return do_maps_open(inode
, file
, &proc_tid_smaps_op
);
682 const struct file_operations proc_pid_smaps_operations
= {
683 .open
= pid_smaps_open
,
686 .release
= seq_release_private
,
689 const struct file_operations proc_tid_smaps_operations
= {
690 .open
= tid_smaps_open
,
693 .release
= seq_release_private
,
697 * We do not want to have constant page-shift bits sitting in
698 * pagemap entries and are about to reuse them some time soon.
700 * Here's the "migration strategy":
701 * 1. when the system boots these bits remain what they are,
702 * but a warning about future change is printed in log;
703 * 2. once anyone clears soft-dirty bits via clear_refs file,
704 * these flag is set to denote, that user is aware of the
705 * new API and those page-shift bits change their meaning.
706 * The respective warning is printed in dmesg;
707 * 3. In a couple of releases we will remove all the mentions
708 * of page-shift in pagemap entries.
711 static bool soft_dirty_cleared __read_mostly
;
713 enum clear_refs_types
{
717 CLEAR_REFS_SOFT_DIRTY
,
721 struct clear_refs_private
{
722 struct vm_area_struct
*vma
;
723 enum clear_refs_types type
;
726 static inline void clear_soft_dirty(struct vm_area_struct
*vma
,
727 unsigned long addr
, pte_t
*pte
)
729 #ifdef CONFIG_MEM_SOFT_DIRTY
731 * The soft-dirty tracker uses #PF-s to catch writes
732 * to pages, so write-protect the pte as well. See the
733 * Documentation/vm/soft-dirty.txt for full description
734 * of how soft-dirty works.
738 if (pte_present(ptent
)) {
739 ptent
= pte_wrprotect(ptent
);
740 ptent
= pte_clear_flags(ptent
, _PAGE_SOFT_DIRTY
);
741 } else if (is_swap_pte(ptent
)) {
742 ptent
= pte_swp_clear_soft_dirty(ptent
);
743 } else if (pte_file(ptent
)) {
744 ptent
= pte_file_clear_soft_dirty(ptent
);
747 if (vma
->vm_flags
& VM_SOFTDIRTY
)
748 vma
->vm_flags
&= ~VM_SOFTDIRTY
;
750 set_pte_at(vma
->vm_mm
, addr
, pte
, ptent
);
754 static int clear_refs_pte_range(pmd_t
*pmd
, unsigned long addr
,
755 unsigned long end
, struct mm_walk
*walk
)
757 struct clear_refs_private
*cp
= walk
->private;
758 struct vm_area_struct
*vma
= cp
->vma
;
763 split_huge_page_pmd(vma
, addr
, pmd
);
764 if (pmd_trans_unstable(pmd
))
767 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
768 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
) {
771 if (cp
->type
== CLEAR_REFS_SOFT_DIRTY
) {
772 clear_soft_dirty(vma
, addr
, pte
);
776 if (!pte_present(ptent
))
779 page
= vm_normal_page(vma
, addr
, ptent
);
783 /* Clear accessed and referenced bits. */
784 ptep_test_and_clear_young(vma
, addr
, pte
);
785 ClearPageReferenced(page
);
787 pte_unmap_unlock(pte
- 1, ptl
);
792 static ssize_t
clear_refs_write(struct file
*file
, const char __user
*buf
,
793 size_t count
, loff_t
*ppos
)
795 struct task_struct
*task
;
796 char buffer
[PROC_NUMBUF
];
797 struct mm_struct
*mm
;
798 struct vm_area_struct
*vma
;
799 enum clear_refs_types type
;
803 memset(buffer
, 0, sizeof(buffer
));
804 if (count
> sizeof(buffer
) - 1)
805 count
= sizeof(buffer
) - 1;
806 if (copy_from_user(buffer
, buf
, count
))
808 rv
= kstrtoint(strstrip(buffer
), 10, &itype
);
811 type
= (enum clear_refs_types
)itype
;
812 if (type
< CLEAR_REFS_ALL
|| type
>= CLEAR_REFS_LAST
)
815 if (type
== CLEAR_REFS_SOFT_DIRTY
) {
816 soft_dirty_cleared
= true;
817 pr_warn_once("The pagemap bits 55-60 has changed their meaning! "
818 "See the linux/Documentation/vm/pagemap.txt for details.\n");
821 task
= get_proc_task(file_inode(file
));
824 mm
= get_task_mm(task
);
826 struct clear_refs_private cp
= {
829 struct mm_walk clear_refs_walk
= {
830 .pmd_entry
= clear_refs_pte_range
,
834 down_read(&mm
->mmap_sem
);
835 if (type
== CLEAR_REFS_SOFT_DIRTY
)
836 mmu_notifier_invalidate_range_start(mm
, 0, -1);
837 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
839 if (is_vm_hugetlb_page(vma
))
842 * Writing 1 to /proc/pid/clear_refs affects all pages.
844 * Writing 2 to /proc/pid/clear_refs only affects
847 * Writing 3 to /proc/pid/clear_refs only affects file
850 if (type
== CLEAR_REFS_ANON
&& vma
->vm_file
)
852 if (type
== CLEAR_REFS_MAPPED
&& !vma
->vm_file
)
854 walk_page_range(vma
->vm_start
, vma
->vm_end
,
857 if (type
== CLEAR_REFS_SOFT_DIRTY
)
858 mmu_notifier_invalidate_range_end(mm
, 0, -1);
860 up_read(&mm
->mmap_sem
);
863 put_task_struct(task
);
868 const struct file_operations proc_clear_refs_operations
= {
869 .write
= clear_refs_write
,
870 .llseek
= noop_llseek
,
878 int pos
, len
; /* units: PM_ENTRY_BYTES, not bytes */
879 pagemap_entry_t
*buffer
;
883 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
884 #define PAGEMAP_WALK_MASK (PMD_MASK)
886 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
887 #define PM_STATUS_BITS 3
888 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
889 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
890 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
891 #define PM_PSHIFT_BITS 6
892 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
893 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
894 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
895 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
896 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
897 /* in "new" pagemap pshift bits are occupied with more status bits */
898 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT))
900 #define __PM_SOFT_DIRTY (1LL)
901 #define PM_PRESENT PM_STATUS(4LL)
902 #define PM_SWAP PM_STATUS(2LL)
903 #define PM_FILE PM_STATUS(1LL)
904 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0)
905 #define PM_END_OF_BUFFER 1
907 static inline pagemap_entry_t
make_pme(u64 val
)
909 return (pagemap_entry_t
) { .pme
= val
};
912 static int add_to_pagemap(unsigned long addr
, pagemap_entry_t
*pme
,
913 struct pagemapread
*pm
)
915 pm
->buffer
[pm
->pos
++] = *pme
;
916 if (pm
->pos
>= pm
->len
)
917 return PM_END_OF_BUFFER
;
921 static int pagemap_pte_hole(unsigned long start
, unsigned long end
,
922 struct mm_walk
*walk
)
924 struct pagemapread
*pm
= walk
->private;
927 pagemap_entry_t pme
= make_pme(PM_NOT_PRESENT(pm
->v2
));
929 for (addr
= start
; addr
< end
; addr
+= PAGE_SIZE
) {
930 err
= add_to_pagemap(addr
, &pme
, pm
);
937 static void pte_to_pagemap_entry(pagemap_entry_t
*pme
, struct pagemapread
*pm
,
938 struct vm_area_struct
*vma
, unsigned long addr
, pte_t pte
)
941 struct page
*page
= NULL
;
944 if (pte_present(pte
)) {
945 frame
= pte_pfn(pte
);
947 page
= vm_normal_page(vma
, addr
, pte
);
948 if (pte_soft_dirty(pte
))
949 flags2
|= __PM_SOFT_DIRTY
;
950 } else if (is_swap_pte(pte
)) {
952 if (pte_swp_soft_dirty(pte
))
953 flags2
|= __PM_SOFT_DIRTY
;
954 entry
= pte_to_swp_entry(pte
);
955 frame
= swp_type(entry
) |
956 (swp_offset(entry
) << MAX_SWAPFILES_SHIFT
);
958 if (is_migration_entry(entry
))
959 page
= migration_entry_to_page(entry
);
961 if (vma
->vm_flags
& VM_SOFTDIRTY
)
962 flags2
|= __PM_SOFT_DIRTY
;
963 *pme
= make_pme(PM_NOT_PRESENT(pm
->v2
) | PM_STATUS2(pm
->v2
, flags2
));
967 if (page
&& !PageAnon(page
))
969 if ((vma
->vm_flags
& VM_SOFTDIRTY
))
970 flags2
|= __PM_SOFT_DIRTY
;
972 *pme
= make_pme(PM_PFRAME(frame
) | PM_STATUS2(pm
->v2
, flags2
) | flags
);
975 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
976 static void thp_pmd_to_pagemap_entry(pagemap_entry_t
*pme
, struct pagemapread
*pm
,
977 pmd_t pmd
, int offset
, int pmd_flags2
)
980 * Currently pmd for thp is always present because thp can not be
981 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
982 * This if-check is just to prepare for future implementation.
984 if (pmd_present(pmd
))
985 *pme
= make_pme(PM_PFRAME(pmd_pfn(pmd
) + offset
)
986 | PM_STATUS2(pm
->v2
, pmd_flags2
) | PM_PRESENT
);
988 *pme
= make_pme(PM_NOT_PRESENT(pm
->v2
) | PM_STATUS2(pm
->v2
, pmd_flags2
));
991 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t
*pme
, struct pagemapread
*pm
,
992 pmd_t pmd
, int offset
, int pmd_flags2
)
997 static int pagemap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
,
998 struct mm_walk
*walk
)
1000 struct vm_area_struct
*vma
;
1001 struct pagemapread
*pm
= walk
->private;
1004 pagemap_entry_t pme
= make_pme(PM_NOT_PRESENT(pm
->v2
));
1006 /* find the first VMA at or above 'addr' */
1007 vma
= find_vma(walk
->mm
, addr
);
1008 if (vma
&& pmd_trans_huge_lock(pmd
, vma
) == 1) {
1011 if ((vma
->vm_flags
& VM_SOFTDIRTY
) || pmd_soft_dirty(*pmd
))
1012 pmd_flags2
= __PM_SOFT_DIRTY
;
1016 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
1017 unsigned long offset
;
1019 offset
= (addr
& ~PAGEMAP_WALK_MASK
) >>
1021 thp_pmd_to_pagemap_entry(&pme
, pm
, *pmd
, offset
, pmd_flags2
);
1022 err
= add_to_pagemap(addr
, &pme
, pm
);
1026 spin_unlock(&walk
->mm
->page_table_lock
);
1030 if (pmd_trans_unstable(pmd
))
1032 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
1035 /* check to see if we've left 'vma' behind
1036 * and need a new, higher one */
1037 if (vma
&& (addr
>= vma
->vm_end
)) {
1038 vma
= find_vma(walk
->mm
, addr
);
1039 if (vma
&& (vma
->vm_flags
& VM_SOFTDIRTY
))
1040 flags2
= __PM_SOFT_DIRTY
;
1043 pme
= make_pme(PM_NOT_PRESENT(pm
->v2
) | PM_STATUS2(pm
->v2
, flags2
));
1046 /* check that 'vma' actually covers this address,
1047 * and that it isn't a huge page vma */
1048 if (vma
&& (vma
->vm_start
<= addr
) &&
1049 !is_vm_hugetlb_page(vma
)) {
1050 pte
= pte_offset_map(pmd
, addr
);
1051 pte_to_pagemap_entry(&pme
, pm
, vma
, addr
, *pte
);
1052 /* unmap before userspace copy */
1055 err
= add_to_pagemap(addr
, &pme
, pm
);
1065 #ifdef CONFIG_HUGETLB_PAGE
1066 static void huge_pte_to_pagemap_entry(pagemap_entry_t
*pme
, struct pagemapread
*pm
,
1067 pte_t pte
, int offset
, int flags2
)
1069 if (pte_present(pte
))
1070 *pme
= make_pme(PM_PFRAME(pte_pfn(pte
) + offset
) |
1071 PM_STATUS2(pm
->v2
, flags2
) |
1074 *pme
= make_pme(PM_NOT_PRESENT(pm
->v2
) |
1075 PM_STATUS2(pm
->v2
, flags2
));
1078 /* This function walks within one hugetlb entry in the single call */
1079 static int pagemap_hugetlb_range(pte_t
*pte
, unsigned long hmask
,
1080 unsigned long addr
, unsigned long end
,
1081 struct mm_walk
*walk
)
1083 struct pagemapread
*pm
= walk
->private;
1084 struct vm_area_struct
*vma
;
1087 pagemap_entry_t pme
;
1089 vma
= find_vma(walk
->mm
, addr
);
1092 if (vma
&& (vma
->vm_flags
& VM_SOFTDIRTY
))
1093 flags2
= __PM_SOFT_DIRTY
;
1097 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
1098 int offset
= (addr
& ~hmask
) >> PAGE_SHIFT
;
1099 huge_pte_to_pagemap_entry(&pme
, pm
, *pte
, offset
, flags2
);
1100 err
= add_to_pagemap(addr
, &pme
, pm
);
1109 #endif /* HUGETLB_PAGE */
1112 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1114 * For each page in the address space, this file contains one 64-bit entry
1115 * consisting of the following:
1117 * Bits 0-54 page frame number (PFN) if present
1118 * Bits 0-4 swap type if swapped
1119 * Bits 5-54 swap offset if swapped
1120 * Bits 55-60 page shift (page size = 1<<page shift)
1121 * Bit 61 page is file-page or shared-anon
1122 * Bit 62 page swapped
1123 * Bit 63 page present
1125 * If the page is not present but in swap, then the PFN contains an
1126 * encoding of the swap file number and the page's offset into the
1127 * swap. Unmapped pages return a null PFN. This allows determining
1128 * precisely which pages are mapped (or in swap) and comparing mapped
1129 * pages between processes.
1131 * Efficient users of this interface will use /proc/pid/maps to
1132 * determine which areas of memory are actually mapped and llseek to
1133 * skip over unmapped regions.
1135 static ssize_t
pagemap_read(struct file
*file
, char __user
*buf
,
1136 size_t count
, loff_t
*ppos
)
1138 struct task_struct
*task
= get_proc_task(file_inode(file
));
1139 struct mm_struct
*mm
;
1140 struct pagemapread pm
;
1142 struct mm_walk pagemap_walk
= {};
1144 unsigned long svpfn
;
1145 unsigned long start_vaddr
;
1146 unsigned long end_vaddr
;
1153 /* file position must be aligned */
1154 if ((*ppos
% PM_ENTRY_BYTES
) || (count
% PM_ENTRY_BYTES
))
1161 pm
.v2
= soft_dirty_cleared
;
1162 pm
.len
= (PAGEMAP_WALK_SIZE
>> PAGE_SHIFT
);
1163 pm
.buffer
= kmalloc(pm
.len
* PM_ENTRY_BYTES
, GFP_TEMPORARY
);
1168 mm
= mm_access(task
, PTRACE_MODE_READ
);
1170 if (!mm
|| IS_ERR(mm
))
1173 pagemap_walk
.pmd_entry
= pagemap_pte_range
;
1174 pagemap_walk
.pte_hole
= pagemap_pte_hole
;
1175 #ifdef CONFIG_HUGETLB_PAGE
1176 pagemap_walk
.hugetlb_entry
= pagemap_hugetlb_range
;
1178 pagemap_walk
.mm
= mm
;
1179 pagemap_walk
.private = &pm
;
1182 svpfn
= src
/ PM_ENTRY_BYTES
;
1183 start_vaddr
= svpfn
<< PAGE_SHIFT
;
1184 end_vaddr
= TASK_SIZE_OF(task
);
1186 /* watch out for wraparound */
1187 if (svpfn
> TASK_SIZE_OF(task
) >> PAGE_SHIFT
)
1188 start_vaddr
= end_vaddr
;
1191 * The odds are that this will stop walking way
1192 * before end_vaddr, because the length of the
1193 * user buffer is tracked in "pm", and the walk
1194 * will stop when we hit the end of the buffer.
1197 while (count
&& (start_vaddr
< end_vaddr
)) {
1202 end
= (start_vaddr
+ PAGEMAP_WALK_SIZE
) & PAGEMAP_WALK_MASK
;
1204 if (end
< start_vaddr
|| end
> end_vaddr
)
1206 down_read(&mm
->mmap_sem
);
1207 ret
= walk_page_range(start_vaddr
, end
, &pagemap_walk
);
1208 up_read(&mm
->mmap_sem
);
1211 len
= min(count
, PM_ENTRY_BYTES
* pm
.pos
);
1212 if (copy_to_user(buf
, pm
.buffer
, len
)) {
1221 if (!ret
|| ret
== PM_END_OF_BUFFER
)
1229 put_task_struct(task
);
1234 static int pagemap_open(struct inode
*inode
, struct file
*file
)
1236 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about "
1237 "to stop being page-shift some time soon. See the "
1238 "linux/Documentation/vm/pagemap.txt for details.\n");
1242 const struct file_operations proc_pagemap_operations
= {
1243 .llseek
= mem_lseek
, /* borrow this */
1244 .read
= pagemap_read
,
1245 .open
= pagemap_open
,
1247 #endif /* CONFIG_PROC_PAGE_MONITOR */
1252 struct vm_area_struct
*vma
;
1253 unsigned long pages
;
1255 unsigned long active
;
1256 unsigned long writeback
;
1257 unsigned long mapcount_max
;
1258 unsigned long dirty
;
1259 unsigned long swapcache
;
1260 unsigned long node
[MAX_NUMNODES
];
1263 struct numa_maps_private
{
1264 struct proc_maps_private proc_maps
;
1265 struct numa_maps md
;
1268 static void gather_stats(struct page
*page
, struct numa_maps
*md
, int pte_dirty
,
1269 unsigned long nr_pages
)
1271 int count
= page_mapcount(page
);
1273 md
->pages
+= nr_pages
;
1274 if (pte_dirty
|| PageDirty(page
))
1275 md
->dirty
+= nr_pages
;
1277 if (PageSwapCache(page
))
1278 md
->swapcache
+= nr_pages
;
1280 if (PageActive(page
) || PageUnevictable(page
))
1281 md
->active
+= nr_pages
;
1283 if (PageWriteback(page
))
1284 md
->writeback
+= nr_pages
;
1287 md
->anon
+= nr_pages
;
1289 if (count
> md
->mapcount_max
)
1290 md
->mapcount_max
= count
;
1292 md
->node
[page_to_nid(page
)] += nr_pages
;
1295 static struct page
*can_gather_numa_stats(pte_t pte
, struct vm_area_struct
*vma
,
1301 if (!pte_present(pte
))
1304 page
= vm_normal_page(vma
, addr
, pte
);
1308 if (PageReserved(page
))
1311 nid
= page_to_nid(page
);
1312 if (!node_isset(nid
, node_states
[N_MEMORY
]))
1318 static int gather_pte_stats(pmd_t
*pmd
, unsigned long addr
,
1319 unsigned long end
, struct mm_walk
*walk
)
1321 struct numa_maps
*md
;
1328 if (pmd_trans_huge_lock(pmd
, md
->vma
) == 1) {
1329 pte_t huge_pte
= *(pte_t
*)pmd
;
1332 page
= can_gather_numa_stats(huge_pte
, md
->vma
, addr
);
1334 gather_stats(page
, md
, pte_dirty(huge_pte
),
1335 HPAGE_PMD_SIZE
/PAGE_SIZE
);
1336 spin_unlock(&walk
->mm
->page_table_lock
);
1340 if (pmd_trans_unstable(pmd
))
1342 orig_pte
= pte
= pte_offset_map_lock(walk
->mm
, pmd
, addr
, &ptl
);
1344 struct page
*page
= can_gather_numa_stats(*pte
, md
->vma
, addr
);
1347 gather_stats(page
, md
, pte_dirty(*pte
), 1);
1349 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1350 pte_unmap_unlock(orig_pte
, ptl
);
1353 #ifdef CONFIG_HUGETLB_PAGE
1354 static int gather_hugetbl_stats(pte_t
*pte
, unsigned long hmask
,
1355 unsigned long addr
, unsigned long end
, struct mm_walk
*walk
)
1357 struct numa_maps
*md
;
1363 page
= pte_page(*pte
);
1368 gather_stats(page
, md
, pte_dirty(*pte
), 1);
1373 static int gather_hugetbl_stats(pte_t
*pte
, unsigned long hmask
,
1374 unsigned long addr
, unsigned long end
, struct mm_walk
*walk
)
1381 * Display pages allocated per node and memory policy via /proc.
1383 static int show_numa_map(struct seq_file
*m
, void *v
, int is_pid
)
1385 struct numa_maps_private
*numa_priv
= m
->private;
1386 struct proc_maps_private
*proc_priv
= &numa_priv
->proc_maps
;
1387 struct vm_area_struct
*vma
= v
;
1388 struct numa_maps
*md
= &numa_priv
->md
;
1389 struct file
*file
= vma
->vm_file
;
1390 struct task_struct
*task
= proc_priv
->task
;
1391 struct mm_struct
*mm
= vma
->vm_mm
;
1392 struct mm_walk walk
= {};
1393 struct mempolicy
*pol
;
1400 /* Ensure we start with an empty set of numa_maps statistics. */
1401 memset(md
, 0, sizeof(*md
));
1405 walk
.hugetlb_entry
= gather_hugetbl_stats
;
1406 walk
.pmd_entry
= gather_pte_stats
;
1410 pol
= get_vma_policy(task
, vma
, vma
->vm_start
);
1411 mpol_to_str(buffer
, sizeof(buffer
), pol
);
1414 seq_printf(m
, "%08lx %s", vma
->vm_start
, buffer
);
1417 seq_printf(m
, " file=");
1418 seq_path(m
, &file
->f_path
, "\n\t= ");
1419 } else if (vma
->vm_start
<= mm
->brk
&& vma
->vm_end
>= mm
->start_brk
) {
1420 seq_printf(m
, " heap");
1422 pid_t tid
= vm_is_stack(task
, vma
, is_pid
);
1425 * Thread stack in /proc/PID/task/TID/maps or
1426 * the main process stack.
1428 if (!is_pid
|| (vma
->vm_start
<= mm
->start_stack
&&
1429 vma
->vm_end
>= mm
->start_stack
))
1430 seq_printf(m
, " stack");
1432 seq_printf(m
, " stack:%d", tid
);
1436 if (is_vm_hugetlb_page(vma
))
1437 seq_printf(m
, " huge");
1439 walk_page_range(vma
->vm_start
, vma
->vm_end
, &walk
);
1445 seq_printf(m
, " anon=%lu", md
->anon
);
1448 seq_printf(m
, " dirty=%lu", md
->dirty
);
1450 if (md
->pages
!= md
->anon
&& md
->pages
!= md
->dirty
)
1451 seq_printf(m
, " mapped=%lu", md
->pages
);
1453 if (md
->mapcount_max
> 1)
1454 seq_printf(m
, " mapmax=%lu", md
->mapcount_max
);
1457 seq_printf(m
, " swapcache=%lu", md
->swapcache
);
1459 if (md
->active
< md
->pages
&& !is_vm_hugetlb_page(vma
))
1460 seq_printf(m
, " active=%lu", md
->active
);
1463 seq_printf(m
, " writeback=%lu", md
->writeback
);
1465 for_each_node_state(nid
, N_MEMORY
)
1467 seq_printf(m
, " N%d=%lu", nid
, md
->node
[nid
]);
1471 if (m
->count
< m
->size
)
1472 m
->version
= (vma
!= proc_priv
->tail_vma
) ? vma
->vm_start
: 0;
1476 static int show_pid_numa_map(struct seq_file
*m
, void *v
)
1478 return show_numa_map(m
, v
, 1);
1481 static int show_tid_numa_map(struct seq_file
*m
, void *v
)
1483 return show_numa_map(m
, v
, 0);
1486 static const struct seq_operations proc_pid_numa_maps_op
= {
1490 .show
= show_pid_numa_map
,
1493 static const struct seq_operations proc_tid_numa_maps_op
= {
1497 .show
= show_tid_numa_map
,
1500 static int numa_maps_open(struct inode
*inode
, struct file
*file
,
1501 const struct seq_operations
*ops
)
1503 struct numa_maps_private
*priv
;
1505 priv
= kzalloc(sizeof(*priv
), GFP_KERNEL
);
1507 priv
->proc_maps
.pid
= proc_pid(inode
);
1508 ret
= seq_open(file
, ops
);
1510 struct seq_file
*m
= file
->private_data
;
1519 static int pid_numa_maps_open(struct inode
*inode
, struct file
*file
)
1521 return numa_maps_open(inode
, file
, &proc_pid_numa_maps_op
);
1524 static int tid_numa_maps_open(struct inode
*inode
, struct file
*file
)
1526 return numa_maps_open(inode
, file
, &proc_tid_numa_maps_op
);
1529 const struct file_operations proc_pid_numa_maps_operations
= {
1530 .open
= pid_numa_maps_open
,
1532 .llseek
= seq_lseek
,
1533 .release
= seq_release_private
,
1536 const struct file_operations proc_tid_numa_maps_operations
= {
1537 .open
= tid_numa_maps_open
,
1539 .llseek
= seq_lseek
,
1540 .release
= seq_release_private
,
1542 #endif /* CONFIG_NUMA */