2 #include <linux/vmacache.h>
3 #include <linux/hugetlb.h>
4 #include <linux/huge_mm.h>
5 #include <linux/mount.h>
6 #include <linux/seq_file.h>
7 #include <linux/highmem.h>
8 #include <linux/ptrace.h>
9 #include <linux/slab.h>
10 #include <linux/pagemap.h>
11 #include <linux/mempolicy.h>
12 #include <linux/rmap.h>
13 #include <linux/swap.h>
14 #include <linux/sched/mm.h>
15 #include <linux/swapops.h>
16 #include <linux/mmu_notifier.h>
17 #include <linux/page_idle.h>
18 #include <linux/shmem_fs.h>
19 #include <linux/uaccess.h>
23 #include <asm/tlbflush.h>
26 void task_mem(struct seq_file
*m
, struct mm_struct
*mm
)
28 unsigned long text
, lib
, swap
, ptes
, pmds
, anon
, file
, shmem
;
29 unsigned long hiwater_vm
, total_vm
, hiwater_rss
, total_rss
;
31 anon
= get_mm_counter(mm
, MM_ANONPAGES
);
32 file
= get_mm_counter(mm
, MM_FILEPAGES
);
33 shmem
= get_mm_counter(mm
, MM_SHMEMPAGES
);
36 * Note: to minimize their overhead, mm maintains hiwater_vm and
37 * hiwater_rss only when about to *lower* total_vm or rss. Any
38 * collector of these hiwater stats must therefore get total_vm
39 * and rss too, which will usually be the higher. Barriers? not
40 * worth the effort, such snapshots can always be inconsistent.
42 hiwater_vm
= total_vm
= mm
->total_vm
;
43 if (hiwater_vm
< mm
->hiwater_vm
)
44 hiwater_vm
= mm
->hiwater_vm
;
45 hiwater_rss
= total_rss
= anon
+ file
+ shmem
;
46 if (hiwater_rss
< mm
->hiwater_rss
)
47 hiwater_rss
= mm
->hiwater_rss
;
49 text
= (PAGE_ALIGN(mm
->end_code
) - (mm
->start_code
& PAGE_MASK
)) >> 10;
50 lib
= (mm
->exec_vm
<< (PAGE_SHIFT
-10)) - text
;
51 swap
= get_mm_counter(mm
, MM_SWAPENTS
);
52 ptes
= PTRS_PER_PTE
* sizeof(pte_t
) * atomic_long_read(&mm
->nr_ptes
);
53 pmds
= PTRS_PER_PMD
* sizeof(pmd_t
) * mm_nr_pmds(mm
);
63 "RssShmem:\t%8lu kB\n"
71 hiwater_vm
<< (PAGE_SHIFT
-10),
72 total_vm
<< (PAGE_SHIFT
-10),
73 mm
->locked_vm
<< (PAGE_SHIFT
-10),
74 mm
->pinned_vm
<< (PAGE_SHIFT
-10),
75 hiwater_rss
<< (PAGE_SHIFT
-10),
76 total_rss
<< (PAGE_SHIFT
-10),
77 anon
<< (PAGE_SHIFT
-10),
78 file
<< (PAGE_SHIFT
-10),
79 shmem
<< (PAGE_SHIFT
-10),
80 mm
->data_vm
<< (PAGE_SHIFT
-10),
81 mm
->stack_vm
<< (PAGE_SHIFT
-10), text
, lib
,
84 swap
<< (PAGE_SHIFT
-10));
85 hugetlb_report_usage(m
, mm
);
88 unsigned long task_vsize(struct mm_struct
*mm
)
90 return PAGE_SIZE
* mm
->total_vm
;
93 unsigned long task_statm(struct mm_struct
*mm
,
94 unsigned long *shared
, unsigned long *text
,
95 unsigned long *data
, unsigned long *resident
)
97 *shared
= get_mm_counter(mm
, MM_FILEPAGES
) +
98 get_mm_counter(mm
, MM_SHMEMPAGES
);
99 *text
= (PAGE_ALIGN(mm
->end_code
) - (mm
->start_code
& PAGE_MASK
))
101 *data
= mm
->data_vm
+ mm
->stack_vm
;
102 *resident
= *shared
+ get_mm_counter(mm
, MM_ANONPAGES
);
108 * Save get_task_policy() for show_numa_map().
110 static void hold_task_mempolicy(struct proc_maps_private
*priv
)
112 struct task_struct
*task
= priv
->task
;
115 priv
->task_mempolicy
= get_task_policy(task
);
116 mpol_get(priv
->task_mempolicy
);
119 static void release_task_mempolicy(struct proc_maps_private
*priv
)
121 mpol_put(priv
->task_mempolicy
);
124 static void hold_task_mempolicy(struct proc_maps_private
*priv
)
127 static void release_task_mempolicy(struct proc_maps_private
*priv
)
132 static void vma_stop(struct proc_maps_private
*priv
)
134 struct mm_struct
*mm
= priv
->mm
;
136 release_task_mempolicy(priv
);
137 up_read(&mm
->mmap_sem
);
141 static struct vm_area_struct
*
142 m_next_vma(struct proc_maps_private
*priv
, struct vm_area_struct
*vma
)
144 if (vma
== priv
->tail_vma
)
146 return vma
->vm_next
?: priv
->tail_vma
;
149 static void m_cache_vma(struct seq_file
*m
, struct vm_area_struct
*vma
)
151 if (m
->count
< m
->size
) /* vma is copied successfully */
152 m
->version
= m_next_vma(m
->private, vma
) ? vma
->vm_end
: -1UL;
155 static void *m_start(struct seq_file
*m
, loff_t
*ppos
)
157 struct proc_maps_private
*priv
= m
->private;
158 unsigned long last_addr
= m
->version
;
159 struct mm_struct
*mm
;
160 struct vm_area_struct
*vma
;
161 unsigned int pos
= *ppos
;
163 /* See m_cache_vma(). Zero at the start or after lseek. */
164 if (last_addr
== -1UL)
167 priv
->task
= get_proc_task(priv
->inode
);
169 return ERR_PTR(-ESRCH
);
172 if (!mm
|| !mmget_not_zero(mm
))
175 down_read(&mm
->mmap_sem
);
176 hold_task_mempolicy(priv
);
177 priv
->tail_vma
= get_gate_vma(mm
);
180 vma
= find_vma(mm
, last_addr
- 1);
181 if (vma
&& vma
->vm_start
<= last_addr
)
182 vma
= m_next_vma(priv
, vma
);
188 if (pos
< mm
->map_count
) {
189 for (vma
= mm
->mmap
; pos
; pos
--) {
190 m
->version
= vma
->vm_start
;
196 /* we do not bother to update m->version in this case */
197 if (pos
== mm
->map_count
&& priv
->tail_vma
)
198 return priv
->tail_vma
;
204 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
206 struct proc_maps_private
*priv
= m
->private;
207 struct vm_area_struct
*next
;
210 next
= m_next_vma(priv
, v
);
216 static void m_stop(struct seq_file
*m
, void *v
)
218 struct proc_maps_private
*priv
= m
->private;
220 if (!IS_ERR_OR_NULL(v
))
223 put_task_struct(priv
->task
);
228 static int proc_maps_open(struct inode
*inode
, struct file
*file
,
229 const struct seq_operations
*ops
, int psize
)
231 struct proc_maps_private
*priv
= __seq_open_private(file
, ops
, psize
);
237 priv
->mm
= proc_mem_open(inode
, PTRACE_MODE_READ
);
238 if (IS_ERR(priv
->mm
)) {
239 int err
= PTR_ERR(priv
->mm
);
241 seq_release_private(inode
, file
);
248 static int proc_map_release(struct inode
*inode
, struct file
*file
)
250 struct seq_file
*seq
= file
->private_data
;
251 struct proc_maps_private
*priv
= seq
->private;
256 return seq_release_private(inode
, file
);
259 static int do_maps_open(struct inode
*inode
, struct file
*file
,
260 const struct seq_operations
*ops
)
262 return proc_maps_open(inode
, file
, ops
,
263 sizeof(struct proc_maps_private
));
267 * Indicate if the VMA is a stack for the given task; for
268 * /proc/PID/maps that is the stack of the main task.
270 static int is_stack(struct proc_maps_private
*priv
,
271 struct vm_area_struct
*vma
)
274 * We make no effort to guess what a given thread considers to be
275 * its "stack". It's not even well-defined for programs written
278 return vma
->vm_start
<= vma
->vm_mm
->start_stack
&&
279 vma
->vm_end
>= vma
->vm_mm
->start_stack
;
283 show_map_vma(struct seq_file
*m
, struct vm_area_struct
*vma
, int is_pid
)
285 struct mm_struct
*mm
= vma
->vm_mm
;
286 struct file
*file
= vma
->vm_file
;
287 struct proc_maps_private
*priv
= m
->private;
288 vm_flags_t flags
= vma
->vm_flags
;
289 unsigned long ino
= 0;
290 unsigned long long pgoff
= 0;
291 unsigned long start
, end
;
293 const char *name
= NULL
;
298 file
= vma_pr_or_file(vma
);
299 inode
= file_inode(file
);
300 dev
= inode
->i_sb
->s_dev
;
302 pgoff
= ((loff_t
)vma
->vm_pgoff
) << PAGE_SHIFT
;
305 start
= vma
->vm_start
;
308 seq_setwidth(m
, 25 + sizeof(void *) * 6 - 1);
309 seq_printf(m
, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
312 flags
& VM_READ
? 'r' : '-',
313 flags
& VM_WRITE
? 'w' : '-',
314 flags
& VM_EXEC
? 'x' : '-',
315 flags
& VM_MAYSHARE
? 's' : 'p',
317 MAJOR(dev
), MINOR(dev
), ino
);
320 * Print the dentry name for named mappings, and a
321 * special [heap] marker for the heap:
325 seq_file_path(m
, file
, "\n");
329 if (vma
->vm_ops
&& vma
->vm_ops
->name
) {
330 name
= vma
->vm_ops
->name(vma
);
335 name
= arch_vma_name(vma
);
342 if (vma
->vm_start
<= mm
->brk
&&
343 vma
->vm_end
>= mm
->start_brk
) {
348 if (is_stack(priv
, vma
))
360 static int show_map(struct seq_file
*m
, void *v
, int is_pid
)
362 show_map_vma(m
, v
, is_pid
);
367 static int show_pid_map(struct seq_file
*m
, void *v
)
369 return show_map(m
, v
, 1);
372 static int show_tid_map(struct seq_file
*m
, void *v
)
374 return show_map(m
, v
, 0);
377 static const struct seq_operations proc_pid_maps_op
= {
384 static const struct seq_operations proc_tid_maps_op
= {
391 static int pid_maps_open(struct inode
*inode
, struct file
*file
)
393 return do_maps_open(inode
, file
, &proc_pid_maps_op
);
396 static int tid_maps_open(struct inode
*inode
, struct file
*file
)
398 return do_maps_open(inode
, file
, &proc_tid_maps_op
);
401 const struct file_operations proc_pid_maps_operations
= {
402 .open
= pid_maps_open
,
405 .release
= proc_map_release
,
408 const struct file_operations proc_tid_maps_operations
= {
409 .open
= tid_maps_open
,
412 .release
= proc_map_release
,
416 * Proportional Set Size(PSS): my share of RSS.
418 * PSS of a process is the count of pages it has in memory, where each
419 * page is divided by the number of processes sharing it. So if a
420 * process has 1000 pages all to itself, and 1000 shared with one other
421 * process, its PSS will be 1500.
423 * To keep (accumulated) division errors low, we adopt a 64bit
424 * fixed-point pss counter to minimize division errors. So (pss >>
425 * PSS_SHIFT) would be the real byte count.
427 * A shift of 12 before division means (assuming 4K page size):
428 * - 1M 3-user-pages add up to 8KB errors;
429 * - supports mapcount up to 2^24, or 16M;
430 * - supports PSS up to 2^52 bytes, or 4PB.
434 #ifdef CONFIG_PROC_PAGE_MONITOR
435 struct mem_size_stats
{
436 unsigned long resident
;
437 unsigned long shared_clean
;
438 unsigned long shared_dirty
;
439 unsigned long private_clean
;
440 unsigned long private_dirty
;
441 unsigned long referenced
;
442 unsigned long anonymous
;
443 unsigned long lazyfree
;
444 unsigned long anonymous_thp
;
445 unsigned long shmem_thp
;
447 unsigned long shared_hugetlb
;
448 unsigned long private_hugetlb
;
451 bool check_shmem_swap
;
454 static void smaps_account(struct mem_size_stats
*mss
, struct page
*page
,
455 bool compound
, bool young
, bool dirty
)
457 int i
, nr
= compound
? 1 << compound_order(page
) : 1;
458 unsigned long size
= nr
* PAGE_SIZE
;
460 if (PageAnon(page
)) {
461 mss
->anonymous
+= size
;
462 if (!PageSwapBacked(page
) && !dirty
&& !PageDirty(page
))
463 mss
->lazyfree
+= size
;
466 mss
->resident
+= size
;
467 /* Accumulate the size in pages that have been accessed. */
468 if (young
|| page_is_young(page
) || PageReferenced(page
))
469 mss
->referenced
+= size
;
472 * page_count(page) == 1 guarantees the page is mapped exactly once.
473 * If any subpage of the compound page mapped with PTE it would elevate
476 if (page_count(page
) == 1) {
477 if (dirty
|| PageDirty(page
))
478 mss
->private_dirty
+= size
;
480 mss
->private_clean
+= size
;
481 mss
->pss
+= (u64
)size
<< PSS_SHIFT
;
485 for (i
= 0; i
< nr
; i
++, page
++) {
486 int mapcount
= page_mapcount(page
);
489 if (dirty
|| PageDirty(page
))
490 mss
->shared_dirty
+= PAGE_SIZE
;
492 mss
->shared_clean
+= PAGE_SIZE
;
493 mss
->pss
+= (PAGE_SIZE
<< PSS_SHIFT
) / mapcount
;
495 if (dirty
|| PageDirty(page
))
496 mss
->private_dirty
+= PAGE_SIZE
;
498 mss
->private_clean
+= PAGE_SIZE
;
499 mss
->pss
+= PAGE_SIZE
<< PSS_SHIFT
;
505 static int smaps_pte_hole(unsigned long addr
, unsigned long end
,
506 struct mm_walk
*walk
)
508 struct mem_size_stats
*mss
= walk
->private;
510 mss
->swap
+= shmem_partial_swap_usage(
511 walk
->vma
->vm_file
->f_mapping
, addr
, end
);
517 static void smaps_pte_entry(pte_t
*pte
, unsigned long addr
,
518 struct mm_walk
*walk
)
520 struct mem_size_stats
*mss
= walk
->private;
521 struct vm_area_struct
*vma
= walk
->vma
;
522 struct page
*page
= NULL
;
524 if (pte_present(*pte
)) {
525 page
= vm_normal_page(vma
, addr
, *pte
);
526 } else if (is_swap_pte(*pte
)) {
527 swp_entry_t swpent
= pte_to_swp_entry(*pte
);
529 if (!non_swap_entry(swpent
)) {
532 mss
->swap
+= PAGE_SIZE
;
533 mapcount
= swp_swapcount(swpent
);
535 u64 pss_delta
= (u64
)PAGE_SIZE
<< PSS_SHIFT
;
537 do_div(pss_delta
, mapcount
);
538 mss
->swap_pss
+= pss_delta
;
540 mss
->swap_pss
+= (u64
)PAGE_SIZE
<< PSS_SHIFT
;
542 } else if (is_migration_entry(swpent
))
543 page
= migration_entry_to_page(swpent
);
544 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM
) && mss
->check_shmem_swap
545 && pte_none(*pte
))) {
546 page
= find_get_entry(vma
->vm_file
->f_mapping
,
547 linear_page_index(vma
, addr
));
551 if (radix_tree_exceptional_entry(page
))
552 mss
->swap
+= PAGE_SIZE
;
562 smaps_account(mss
, page
, false, pte_young(*pte
), pte_dirty(*pte
));
565 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
566 static void smaps_pmd_entry(pmd_t
*pmd
, unsigned long addr
,
567 struct mm_walk
*walk
)
569 struct mem_size_stats
*mss
= walk
->private;
570 struct vm_area_struct
*vma
= walk
->vma
;
573 /* FOLL_DUMP will return -EFAULT on huge zero page */
574 page
= follow_trans_huge_pmd(vma
, addr
, pmd
, FOLL_DUMP
);
575 if (IS_ERR_OR_NULL(page
))
578 mss
->anonymous_thp
+= HPAGE_PMD_SIZE
;
579 else if (PageSwapBacked(page
))
580 mss
->shmem_thp
+= HPAGE_PMD_SIZE
;
581 else if (is_zone_device_page(page
))
584 VM_BUG_ON_PAGE(1, page
);
585 smaps_account(mss
, page
, true, pmd_young(*pmd
), pmd_dirty(*pmd
));
588 static void smaps_pmd_entry(pmd_t
*pmd
, unsigned long addr
,
589 struct mm_walk
*walk
)
594 static int smaps_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
,
595 struct mm_walk
*walk
)
597 struct vm_area_struct
*vma
= walk
->vma
;
601 ptl
= pmd_trans_huge_lock(pmd
, vma
);
603 smaps_pmd_entry(pmd
, addr
, walk
);
608 if (pmd_trans_unstable(pmd
))
611 * The mmap_sem held all the way back in m_start() is what
612 * keeps khugepaged out of here and from collapsing things
615 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
616 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
617 smaps_pte_entry(pte
, addr
, walk
);
618 pte_unmap_unlock(pte
- 1, ptl
);
623 static void show_smap_vma_flags(struct seq_file
*m
, struct vm_area_struct
*vma
)
626 * Don't forget to update Documentation/ on changes.
628 static const char mnemonics
[BITS_PER_LONG
][2] = {
630 * In case if we meet a flag we don't know about.
632 [0 ... (BITS_PER_LONG
-1)] = "??",
634 [ilog2(VM_READ
)] = "rd",
635 [ilog2(VM_WRITE
)] = "wr",
636 [ilog2(VM_EXEC
)] = "ex",
637 [ilog2(VM_SHARED
)] = "sh",
638 [ilog2(VM_MAYREAD
)] = "mr",
639 [ilog2(VM_MAYWRITE
)] = "mw",
640 [ilog2(VM_MAYEXEC
)] = "me",
641 [ilog2(VM_MAYSHARE
)] = "ms",
642 [ilog2(VM_GROWSDOWN
)] = "gd",
643 [ilog2(VM_PFNMAP
)] = "pf",
644 [ilog2(VM_DENYWRITE
)] = "dw",
645 #ifdef CONFIG_X86_INTEL_MPX
646 [ilog2(VM_MPX
)] = "mp",
648 [ilog2(VM_LOCKED
)] = "lo",
649 [ilog2(VM_IO
)] = "io",
650 [ilog2(VM_SEQ_READ
)] = "sr",
651 [ilog2(VM_RAND_READ
)] = "rr",
652 [ilog2(VM_DONTCOPY
)] = "dc",
653 [ilog2(VM_DONTEXPAND
)] = "de",
654 [ilog2(VM_ACCOUNT
)] = "ac",
655 [ilog2(VM_NORESERVE
)] = "nr",
656 [ilog2(VM_HUGETLB
)] = "ht",
657 [ilog2(VM_ARCH_1
)] = "ar",
658 [ilog2(VM_DONTDUMP
)] = "dd",
659 #ifdef CONFIG_MEM_SOFT_DIRTY
660 [ilog2(VM_SOFTDIRTY
)] = "sd",
662 [ilog2(VM_MIXEDMAP
)] = "mm",
663 [ilog2(VM_HUGEPAGE
)] = "hg",
664 [ilog2(VM_NOHUGEPAGE
)] = "nh",
665 [ilog2(VM_MERGEABLE
)] = "mg",
666 [ilog2(VM_UFFD_MISSING
)]= "um",
667 [ilog2(VM_UFFD_WP
)] = "uw",
668 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
669 /* These come out via ProtectionKey: */
670 [ilog2(VM_PKEY_BIT0
)] = "",
671 [ilog2(VM_PKEY_BIT1
)] = "",
672 [ilog2(VM_PKEY_BIT2
)] = "",
673 [ilog2(VM_PKEY_BIT3
)] = "",
678 seq_puts(m
, "VmFlags: ");
679 for (i
= 0; i
< BITS_PER_LONG
; i
++) {
680 if (!mnemonics
[i
][0])
682 if (vma
->vm_flags
& (1UL << i
)) {
683 seq_printf(m
, "%c%c ",
684 mnemonics
[i
][0], mnemonics
[i
][1]);
690 #ifdef CONFIG_HUGETLB_PAGE
691 static int smaps_hugetlb_range(pte_t
*pte
, unsigned long hmask
,
692 unsigned long addr
, unsigned long end
,
693 struct mm_walk
*walk
)
695 struct mem_size_stats
*mss
= walk
->private;
696 struct vm_area_struct
*vma
= walk
->vma
;
697 struct page
*page
= NULL
;
699 if (pte_present(*pte
)) {
700 page
= vm_normal_page(vma
, addr
, *pte
);
701 } else if (is_swap_pte(*pte
)) {
702 swp_entry_t swpent
= pte_to_swp_entry(*pte
);
704 if (is_migration_entry(swpent
))
705 page
= migration_entry_to_page(swpent
);
708 int mapcount
= page_mapcount(page
);
711 mss
->shared_hugetlb
+= huge_page_size(hstate_vma(vma
));
713 mss
->private_hugetlb
+= huge_page_size(hstate_vma(vma
));
717 #endif /* HUGETLB_PAGE */
719 void __weak
arch_show_smap(struct seq_file
*m
, struct vm_area_struct
*vma
)
723 static int show_smap(struct seq_file
*m
, void *v
, int is_pid
)
725 struct vm_area_struct
*vma
= v
;
726 struct mem_size_stats mss
;
727 struct mm_walk smaps_walk
= {
728 .pmd_entry
= smaps_pte_range
,
729 #ifdef CONFIG_HUGETLB_PAGE
730 .hugetlb_entry
= smaps_hugetlb_range
,
736 memset(&mss
, 0, sizeof mss
);
739 if (vma
->vm_file
&& shmem_mapping(vma
->vm_file
->f_mapping
)) {
741 * For shared or readonly shmem mappings we know that all
742 * swapped out pages belong to the shmem object, and we can
743 * obtain the swap value much more efficiently. For private
744 * writable mappings, we might have COW pages that are
745 * not affected by the parent swapped out pages of the shmem
746 * object, so we have to distinguish them during the page walk.
747 * Unless we know that the shmem object (or the part mapped by
748 * our VMA) has no swapped out pages at all.
750 unsigned long shmem_swapped
= shmem_swap_usage(vma
);
752 if (!shmem_swapped
|| (vma
->vm_flags
& VM_SHARED
) ||
753 !(vma
->vm_flags
& VM_WRITE
)) {
754 mss
.swap
= shmem_swapped
;
756 mss
.check_shmem_swap
= true;
757 smaps_walk
.pte_hole
= smaps_pte_hole
;
762 /* mmap_sem is held in m_start */
763 walk_page_vma(vma
, &smaps_walk
);
765 show_map_vma(m
, vma
, is_pid
);
771 "Shared_Clean: %8lu kB\n"
772 "Shared_Dirty: %8lu kB\n"
773 "Private_Clean: %8lu kB\n"
774 "Private_Dirty: %8lu kB\n"
775 "Referenced: %8lu kB\n"
776 "Anonymous: %8lu kB\n"
777 "LazyFree: %8lu kB\n"
778 "AnonHugePages: %8lu kB\n"
779 "ShmemPmdMapped: %8lu kB\n"
780 "Shared_Hugetlb: %8lu kB\n"
781 "Private_Hugetlb: %7lu kB\n"
784 "KernelPageSize: %8lu kB\n"
785 "MMUPageSize: %8lu kB\n"
787 (vma
->vm_end
- vma
->vm_start
) >> 10,
789 (unsigned long)(mss
.pss
>> (10 + PSS_SHIFT
)),
790 mss
.shared_clean
>> 10,
791 mss
.shared_dirty
>> 10,
792 mss
.private_clean
>> 10,
793 mss
.private_dirty
>> 10,
794 mss
.referenced
>> 10,
797 mss
.anonymous_thp
>> 10,
799 mss
.shared_hugetlb
>> 10,
800 mss
.private_hugetlb
>> 10,
802 (unsigned long)(mss
.swap_pss
>> (10 + PSS_SHIFT
)),
803 vma_kernel_pagesize(vma
) >> 10,
804 vma_mmu_pagesize(vma
) >> 10,
805 (vma
->vm_flags
& VM_LOCKED
) ?
806 (unsigned long)(mss
.pss
>> (10 + PSS_SHIFT
)) : 0);
808 arch_show_smap(m
, vma
);
809 show_smap_vma_flags(m
, vma
);
814 static int show_pid_smap(struct seq_file
*m
, void *v
)
816 return show_smap(m
, v
, 1);
819 static int show_tid_smap(struct seq_file
*m
, void *v
)
821 return show_smap(m
, v
, 0);
824 static const struct seq_operations proc_pid_smaps_op
= {
828 .show
= show_pid_smap
831 static const struct seq_operations proc_tid_smaps_op
= {
835 .show
= show_tid_smap
838 static int pid_smaps_open(struct inode
*inode
, struct file
*file
)
840 return do_maps_open(inode
, file
, &proc_pid_smaps_op
);
843 static int tid_smaps_open(struct inode
*inode
, struct file
*file
)
845 return do_maps_open(inode
, file
, &proc_tid_smaps_op
);
848 const struct file_operations proc_pid_smaps_operations
= {
849 .open
= pid_smaps_open
,
852 .release
= proc_map_release
,
855 const struct file_operations proc_tid_smaps_operations
= {
856 .open
= tid_smaps_open
,
859 .release
= proc_map_release
,
862 enum clear_refs_types
{
866 CLEAR_REFS_SOFT_DIRTY
,
867 CLEAR_REFS_MM_HIWATER_RSS
,
871 struct clear_refs_private
{
872 enum clear_refs_types type
;
875 #ifdef CONFIG_MEM_SOFT_DIRTY
876 static inline void clear_soft_dirty(struct vm_area_struct
*vma
,
877 unsigned long addr
, pte_t
*pte
)
880 * The soft-dirty tracker uses #PF-s to catch writes
881 * to pages, so write-protect the pte as well. See the
882 * Documentation/vm/soft-dirty.txt for full description
883 * of how soft-dirty works.
887 if (pte_present(ptent
)) {
888 ptent
= ptep_modify_prot_start(vma
->vm_mm
, addr
, pte
);
889 ptent
= pte_wrprotect(ptent
);
890 ptent
= pte_clear_soft_dirty(ptent
);
891 ptep_modify_prot_commit(vma
->vm_mm
, addr
, pte
, ptent
);
892 } else if (is_swap_pte(ptent
)) {
893 ptent
= pte_swp_clear_soft_dirty(ptent
);
894 set_pte_at(vma
->vm_mm
, addr
, pte
, ptent
);
898 static inline void clear_soft_dirty(struct vm_area_struct
*vma
,
899 unsigned long addr
, pte_t
*pte
)
904 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
905 static inline void clear_soft_dirty_pmd(struct vm_area_struct
*vma
,
906 unsigned long addr
, pmd_t
*pmdp
)
910 /* See comment in change_huge_pmd() */
911 pmdp_invalidate(vma
, addr
, pmdp
);
912 if (pmd_dirty(*pmdp
))
913 pmd
= pmd_mkdirty(pmd
);
914 if (pmd_young(*pmdp
))
915 pmd
= pmd_mkyoung(pmd
);
917 pmd
= pmd_wrprotect(pmd
);
918 pmd
= pmd_clear_soft_dirty(pmd
);
920 set_pmd_at(vma
->vm_mm
, addr
, pmdp
, pmd
);
923 static inline void clear_soft_dirty_pmd(struct vm_area_struct
*vma
,
924 unsigned long addr
, pmd_t
*pmdp
)
929 static int clear_refs_pte_range(pmd_t
*pmd
, unsigned long addr
,
930 unsigned long end
, struct mm_walk
*walk
)
932 struct clear_refs_private
*cp
= walk
->private;
933 struct vm_area_struct
*vma
= walk
->vma
;
938 ptl
= pmd_trans_huge_lock(pmd
, vma
);
940 if (cp
->type
== CLEAR_REFS_SOFT_DIRTY
) {
941 clear_soft_dirty_pmd(vma
, addr
, pmd
);
945 page
= pmd_page(*pmd
);
947 /* Clear accessed and referenced bits. */
948 pmdp_test_and_clear_young(vma
, addr
, pmd
);
949 test_and_clear_page_young(page
);
950 ClearPageReferenced(page
);
956 if (pmd_trans_unstable(pmd
))
959 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
960 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
) {
963 if (cp
->type
== CLEAR_REFS_SOFT_DIRTY
) {
964 clear_soft_dirty(vma
, addr
, pte
);
968 if (!pte_present(ptent
))
971 page
= vm_normal_page(vma
, addr
, ptent
);
975 /* Clear accessed and referenced bits. */
976 ptep_test_and_clear_young(vma
, addr
, pte
);
977 test_and_clear_page_young(page
);
978 ClearPageReferenced(page
);
980 pte_unmap_unlock(pte
- 1, ptl
);
985 static int clear_refs_test_walk(unsigned long start
, unsigned long end
,
986 struct mm_walk
*walk
)
988 struct clear_refs_private
*cp
= walk
->private;
989 struct vm_area_struct
*vma
= walk
->vma
;
991 if (vma
->vm_flags
& VM_PFNMAP
)
995 * Writing 1 to /proc/pid/clear_refs affects all pages.
996 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
997 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
998 * Writing 4 to /proc/pid/clear_refs affects all pages.
1000 if (cp
->type
== CLEAR_REFS_ANON
&& vma
->vm_file
)
1002 if (cp
->type
== CLEAR_REFS_MAPPED
&& !vma
->vm_file
)
1007 static ssize_t
clear_refs_write(struct file
*file
, const char __user
*buf
,
1008 size_t count
, loff_t
*ppos
)
1010 struct task_struct
*task
;
1011 char buffer
[PROC_NUMBUF
];
1012 struct mm_struct
*mm
;
1013 struct vm_area_struct
*vma
;
1014 enum clear_refs_types type
;
1015 struct mmu_gather tlb
;
1019 memset(buffer
, 0, sizeof(buffer
));
1020 if (count
> sizeof(buffer
) - 1)
1021 count
= sizeof(buffer
) - 1;
1022 if (copy_from_user(buffer
, buf
, count
))
1024 rv
= kstrtoint(strstrip(buffer
), 10, &itype
);
1027 type
= (enum clear_refs_types
)itype
;
1028 if (type
< CLEAR_REFS_ALL
|| type
>= CLEAR_REFS_LAST
)
1031 task
= get_proc_task(file_inode(file
));
1034 mm
= get_task_mm(task
);
1036 struct clear_refs_private cp
= {
1039 struct mm_walk clear_refs_walk
= {
1040 .pmd_entry
= clear_refs_pte_range
,
1041 .test_walk
= clear_refs_test_walk
,
1046 if (type
== CLEAR_REFS_MM_HIWATER_RSS
) {
1047 if (down_write_killable(&mm
->mmap_sem
)) {
1053 * Writing 5 to /proc/pid/clear_refs resets the peak
1054 * resident set size to this mm's current rss value.
1056 reset_mm_hiwater_rss(mm
);
1057 up_write(&mm
->mmap_sem
);
1061 down_read(&mm
->mmap_sem
);
1062 tlb_gather_mmu(&tlb
, mm
, 0, -1);
1063 if (type
== CLEAR_REFS_SOFT_DIRTY
) {
1064 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1065 if (!(vma
->vm_flags
& VM_SOFTDIRTY
))
1067 up_read(&mm
->mmap_sem
);
1068 if (down_write_killable(&mm
->mmap_sem
)) {
1072 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1073 vma
->vm_flags
&= ~VM_SOFTDIRTY
;
1074 vma_set_page_prot(vma
);
1076 downgrade_write(&mm
->mmap_sem
);
1079 mmu_notifier_invalidate_range_start(mm
, 0, -1);
1081 walk_page_range(0, mm
->highest_vm_end
, &clear_refs_walk
);
1082 if (type
== CLEAR_REFS_SOFT_DIRTY
)
1083 mmu_notifier_invalidate_range_end(mm
, 0, -1);
1084 tlb_finish_mmu(&tlb
, 0, -1);
1085 up_read(&mm
->mmap_sem
);
1089 put_task_struct(task
);
1094 const struct file_operations proc_clear_refs_operations
= {
1095 .write
= clear_refs_write
,
1096 .llseek
= noop_llseek
,
1103 struct pagemapread
{
1104 int pos
, len
; /* units: PM_ENTRY_BYTES, not bytes */
1105 pagemap_entry_t
*buffer
;
1109 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1110 #define PAGEMAP_WALK_MASK (PMD_MASK)
1112 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1113 #define PM_PFRAME_BITS 55
1114 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1115 #define PM_SOFT_DIRTY BIT_ULL(55)
1116 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1117 #define PM_FILE BIT_ULL(61)
1118 #define PM_SWAP BIT_ULL(62)
1119 #define PM_PRESENT BIT_ULL(63)
1121 #define PM_END_OF_BUFFER 1
1123 static inline pagemap_entry_t
make_pme(u64 frame
, u64 flags
)
1125 return (pagemap_entry_t
) { .pme
= (frame
& PM_PFRAME_MASK
) | flags
};
1128 static int add_to_pagemap(unsigned long addr
, pagemap_entry_t
*pme
,
1129 struct pagemapread
*pm
)
1131 pm
->buffer
[pm
->pos
++] = *pme
;
1132 if (pm
->pos
>= pm
->len
)
1133 return PM_END_OF_BUFFER
;
1137 static int pagemap_pte_hole(unsigned long start
, unsigned long end
,
1138 struct mm_walk
*walk
)
1140 struct pagemapread
*pm
= walk
->private;
1141 unsigned long addr
= start
;
1144 while (addr
< end
) {
1145 struct vm_area_struct
*vma
= find_vma(walk
->mm
, addr
);
1146 pagemap_entry_t pme
= make_pme(0, 0);
1147 /* End of address space hole, which we mark as non-present. */
1148 unsigned long hole_end
;
1151 hole_end
= min(end
, vma
->vm_start
);
1155 for (; addr
< hole_end
; addr
+= PAGE_SIZE
) {
1156 err
= add_to_pagemap(addr
, &pme
, pm
);
1164 /* Addresses in the VMA. */
1165 if (vma
->vm_flags
& VM_SOFTDIRTY
)
1166 pme
= make_pme(0, PM_SOFT_DIRTY
);
1167 for (; addr
< min(end
, vma
->vm_end
); addr
+= PAGE_SIZE
) {
1168 err
= add_to_pagemap(addr
, &pme
, pm
);
1177 static pagemap_entry_t
pte_to_pagemap_entry(struct pagemapread
*pm
,
1178 struct vm_area_struct
*vma
, unsigned long addr
, pte_t pte
)
1180 u64 frame
= 0, flags
= 0;
1181 struct page
*page
= NULL
;
1183 if (pte_present(pte
)) {
1185 frame
= pte_pfn(pte
);
1186 flags
|= PM_PRESENT
;
1187 page
= vm_normal_page(vma
, addr
, pte
);
1188 if (pte_soft_dirty(pte
))
1189 flags
|= PM_SOFT_DIRTY
;
1190 } else if (is_swap_pte(pte
)) {
1192 if (pte_swp_soft_dirty(pte
))
1193 flags
|= PM_SOFT_DIRTY
;
1194 entry
= pte_to_swp_entry(pte
);
1195 frame
= swp_type(entry
) |
1196 (swp_offset(entry
) << MAX_SWAPFILES_SHIFT
);
1198 if (is_migration_entry(entry
))
1199 page
= migration_entry_to_page(entry
);
1202 if (page
&& !PageAnon(page
))
1204 if (page
&& page_mapcount(page
) == 1)
1205 flags
|= PM_MMAP_EXCLUSIVE
;
1206 if (vma
->vm_flags
& VM_SOFTDIRTY
)
1207 flags
|= PM_SOFT_DIRTY
;
1209 return make_pme(frame
, flags
);
1212 static int pagemap_pmd_range(pmd_t
*pmdp
, unsigned long addr
, unsigned long end
,
1213 struct mm_walk
*walk
)
1215 struct vm_area_struct
*vma
= walk
->vma
;
1216 struct pagemapread
*pm
= walk
->private;
1218 pte_t
*pte
, *orig_pte
;
1221 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1222 ptl
= pmd_trans_huge_lock(pmdp
, vma
);
1224 u64 flags
= 0, frame
= 0;
1227 if ((vma
->vm_flags
& VM_SOFTDIRTY
) || pmd_soft_dirty(pmd
))
1228 flags
|= PM_SOFT_DIRTY
;
1231 * Currently pmd for thp is always present because thp
1232 * can not be swapped-out, migrated, or HWPOISONed
1233 * (split in such cases instead.)
1234 * This if-check is just to prepare for future implementation.
1236 if (pmd_present(pmd
)) {
1237 struct page
*page
= pmd_page(pmd
);
1239 if (page_mapcount(page
) == 1)
1240 flags
|= PM_MMAP_EXCLUSIVE
;
1242 flags
|= PM_PRESENT
;
1244 frame
= pmd_pfn(pmd
) +
1245 ((addr
& ~PMD_MASK
) >> PAGE_SHIFT
);
1248 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
1249 pagemap_entry_t pme
= make_pme(frame
, flags
);
1251 err
= add_to_pagemap(addr
, &pme
, pm
);
1254 if (pm
->show_pfn
&& (flags
& PM_PRESENT
))
1261 if (pmd_trans_unstable(pmdp
))
1263 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1266 * We can assume that @vma always points to a valid one and @end never
1267 * goes beyond vma->vm_end.
1269 orig_pte
= pte
= pte_offset_map_lock(walk
->mm
, pmdp
, addr
, &ptl
);
1270 for (; addr
< end
; pte
++, addr
+= PAGE_SIZE
) {
1271 pagemap_entry_t pme
;
1273 pme
= pte_to_pagemap_entry(pm
, vma
, addr
, *pte
);
1274 err
= add_to_pagemap(addr
, &pme
, pm
);
1278 pte_unmap_unlock(orig_pte
, ptl
);
1285 #ifdef CONFIG_HUGETLB_PAGE
1286 /* This function walks within one hugetlb entry in the single call */
1287 static int pagemap_hugetlb_range(pte_t
*ptep
, unsigned long hmask
,
1288 unsigned long addr
, unsigned long end
,
1289 struct mm_walk
*walk
)
1291 struct pagemapread
*pm
= walk
->private;
1292 struct vm_area_struct
*vma
= walk
->vma
;
1293 u64 flags
= 0, frame
= 0;
1297 if (vma
->vm_flags
& VM_SOFTDIRTY
)
1298 flags
|= PM_SOFT_DIRTY
;
1300 pte
= huge_ptep_get(ptep
);
1301 if (pte_present(pte
)) {
1302 struct page
*page
= pte_page(pte
);
1304 if (!PageAnon(page
))
1307 if (page_mapcount(page
) == 1)
1308 flags
|= PM_MMAP_EXCLUSIVE
;
1310 flags
|= PM_PRESENT
;
1312 frame
= pte_pfn(pte
) +
1313 ((addr
& ~hmask
) >> PAGE_SHIFT
);
1316 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
1317 pagemap_entry_t pme
= make_pme(frame
, flags
);
1319 err
= add_to_pagemap(addr
, &pme
, pm
);
1322 if (pm
->show_pfn
&& (flags
& PM_PRESENT
))
1330 #endif /* HUGETLB_PAGE */
1333 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1335 * For each page in the address space, this file contains one 64-bit entry
1336 * consisting of the following:
1338 * Bits 0-54 page frame number (PFN) if present
1339 * Bits 0-4 swap type if swapped
1340 * Bits 5-54 swap offset if swapped
1341 * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt)
1342 * Bit 56 page exclusively mapped
1344 * Bit 61 page is file-page or shared-anon
1345 * Bit 62 page swapped
1346 * Bit 63 page present
1348 * If the page is not present but in swap, then the PFN contains an
1349 * encoding of the swap file number and the page's offset into the
1350 * swap. Unmapped pages return a null PFN. This allows determining
1351 * precisely which pages are mapped (or in swap) and comparing mapped
1352 * pages between processes.
1354 * Efficient users of this interface will use /proc/pid/maps to
1355 * determine which areas of memory are actually mapped and llseek to
1356 * skip over unmapped regions.
1358 static ssize_t
pagemap_read(struct file
*file
, char __user
*buf
,
1359 size_t count
, loff_t
*ppos
)
1361 struct mm_struct
*mm
= file
->private_data
;
1362 struct pagemapread pm
;
1363 struct mm_walk pagemap_walk
= {};
1365 unsigned long svpfn
;
1366 unsigned long start_vaddr
;
1367 unsigned long end_vaddr
;
1368 int ret
= 0, copied
= 0;
1370 if (!mm
|| !mmget_not_zero(mm
))
1374 /* file position must be aligned */
1375 if ((*ppos
% PM_ENTRY_BYTES
) || (count
% PM_ENTRY_BYTES
))
1382 /* do not disclose physical addresses: attack vector */
1383 pm
.show_pfn
= file_ns_capable(file
, &init_user_ns
, CAP_SYS_ADMIN
);
1385 pm
.len
= (PAGEMAP_WALK_SIZE
>> PAGE_SHIFT
);
1386 pm
.buffer
= kmalloc(pm
.len
* PM_ENTRY_BYTES
, GFP_TEMPORARY
);
1391 pagemap_walk
.pmd_entry
= pagemap_pmd_range
;
1392 pagemap_walk
.pte_hole
= pagemap_pte_hole
;
1393 #ifdef CONFIG_HUGETLB_PAGE
1394 pagemap_walk
.hugetlb_entry
= pagemap_hugetlb_range
;
1396 pagemap_walk
.mm
= mm
;
1397 pagemap_walk
.private = &pm
;
1400 svpfn
= src
/ PM_ENTRY_BYTES
;
1401 start_vaddr
= svpfn
<< PAGE_SHIFT
;
1402 end_vaddr
= mm
->task_size
;
1404 /* watch out for wraparound */
1405 if (svpfn
> mm
->task_size
>> PAGE_SHIFT
)
1406 start_vaddr
= end_vaddr
;
1409 * The odds are that this will stop walking way
1410 * before end_vaddr, because the length of the
1411 * user buffer is tracked in "pm", and the walk
1412 * will stop when we hit the end of the buffer.
1415 while (count
&& (start_vaddr
< end_vaddr
)) {
1420 end
= (start_vaddr
+ PAGEMAP_WALK_SIZE
) & PAGEMAP_WALK_MASK
;
1422 if (end
< start_vaddr
|| end
> end_vaddr
)
1424 down_read(&mm
->mmap_sem
);
1425 ret
= walk_page_range(start_vaddr
, end
, &pagemap_walk
);
1426 up_read(&mm
->mmap_sem
);
1429 len
= min(count
, PM_ENTRY_BYTES
* pm
.pos
);
1430 if (copy_to_user(buf
, pm
.buffer
, len
)) {
1439 if (!ret
|| ret
== PM_END_OF_BUFFER
)
1450 static int pagemap_open(struct inode
*inode
, struct file
*file
)
1452 struct mm_struct
*mm
;
1454 mm
= proc_mem_open(inode
, PTRACE_MODE_READ
);
1457 file
->private_data
= mm
;
1461 static int pagemap_release(struct inode
*inode
, struct file
*file
)
1463 struct mm_struct
*mm
= file
->private_data
;
1470 const struct file_operations proc_pagemap_operations
= {
1471 .llseek
= mem_lseek
, /* borrow this */
1472 .read
= pagemap_read
,
1473 .open
= pagemap_open
,
1474 .release
= pagemap_release
,
1476 #endif /* CONFIG_PROC_PAGE_MONITOR */
1481 unsigned long pages
;
1483 unsigned long active
;
1484 unsigned long writeback
;
1485 unsigned long mapcount_max
;
1486 unsigned long dirty
;
1487 unsigned long swapcache
;
1488 unsigned long node
[MAX_NUMNODES
];
1491 struct numa_maps_private
{
1492 struct proc_maps_private proc_maps
;
1493 struct numa_maps md
;
1496 static void gather_stats(struct page
*page
, struct numa_maps
*md
, int pte_dirty
,
1497 unsigned long nr_pages
)
1499 int count
= page_mapcount(page
);
1501 md
->pages
+= nr_pages
;
1502 if (pte_dirty
|| PageDirty(page
))
1503 md
->dirty
+= nr_pages
;
1505 if (PageSwapCache(page
))
1506 md
->swapcache
+= nr_pages
;
1508 if (PageActive(page
) || PageUnevictable(page
))
1509 md
->active
+= nr_pages
;
1511 if (PageWriteback(page
))
1512 md
->writeback
+= nr_pages
;
1515 md
->anon
+= nr_pages
;
1517 if (count
> md
->mapcount_max
)
1518 md
->mapcount_max
= count
;
1520 md
->node
[page_to_nid(page
)] += nr_pages
;
1523 static struct page
*can_gather_numa_stats(pte_t pte
, struct vm_area_struct
*vma
,
1529 if (!pte_present(pte
))
1532 page
= vm_normal_page(vma
, addr
, pte
);
1536 if (PageReserved(page
))
1539 nid
= page_to_nid(page
);
1540 if (!node_isset(nid
, node_states
[N_MEMORY
]))
1546 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1547 static struct page
*can_gather_numa_stats_pmd(pmd_t pmd
,
1548 struct vm_area_struct
*vma
,
1554 if (!pmd_present(pmd
))
1557 page
= vm_normal_page_pmd(vma
, addr
, pmd
);
1561 if (PageReserved(page
))
1564 nid
= page_to_nid(page
);
1565 if (!node_isset(nid
, node_states
[N_MEMORY
]))
1572 static int gather_pte_stats(pmd_t
*pmd
, unsigned long addr
,
1573 unsigned long end
, struct mm_walk
*walk
)
1575 struct numa_maps
*md
= walk
->private;
1576 struct vm_area_struct
*vma
= walk
->vma
;
1581 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1582 ptl
= pmd_trans_huge_lock(pmd
, vma
);
1586 page
= can_gather_numa_stats_pmd(*pmd
, vma
, addr
);
1588 gather_stats(page
, md
, pmd_dirty(*pmd
),
1589 HPAGE_PMD_SIZE
/PAGE_SIZE
);
1594 if (pmd_trans_unstable(pmd
))
1597 orig_pte
= pte
= pte_offset_map_lock(walk
->mm
, pmd
, addr
, &ptl
);
1599 struct page
*page
= can_gather_numa_stats(*pte
, vma
, addr
);
1602 gather_stats(page
, md
, pte_dirty(*pte
), 1);
1604 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1605 pte_unmap_unlock(orig_pte
, ptl
);
1609 #ifdef CONFIG_HUGETLB_PAGE
1610 static int gather_hugetlb_stats(pte_t
*pte
, unsigned long hmask
,
1611 unsigned long addr
, unsigned long end
, struct mm_walk
*walk
)
1613 pte_t huge_pte
= huge_ptep_get(pte
);
1614 struct numa_maps
*md
;
1617 if (!pte_present(huge_pte
))
1620 page
= pte_page(huge_pte
);
1625 gather_stats(page
, md
, pte_dirty(huge_pte
), 1);
1630 static int gather_hugetlb_stats(pte_t
*pte
, unsigned long hmask
,
1631 unsigned long addr
, unsigned long end
, struct mm_walk
*walk
)
1638 * Display pages allocated per node and memory policy via /proc.
1640 static int show_numa_map(struct seq_file
*m
, void *v
, int is_pid
)
1642 struct numa_maps_private
*numa_priv
= m
->private;
1643 struct proc_maps_private
*proc_priv
= &numa_priv
->proc_maps
;
1644 struct vm_area_struct
*vma
= v
;
1645 struct numa_maps
*md
= &numa_priv
->md
;
1646 struct file
*file
= vma_pr_or_file(vma
);
1647 struct mm_struct
*mm
= vma
->vm_mm
;
1648 struct mm_walk walk
= {
1649 .hugetlb_entry
= gather_hugetlb_stats
,
1650 .pmd_entry
= gather_pte_stats
,
1654 struct mempolicy
*pol
;
1661 /* Ensure we start with an empty set of numa_maps statistics. */
1662 memset(md
, 0, sizeof(*md
));
1664 pol
= __get_vma_policy(vma
, vma
->vm_start
);
1666 mpol_to_str(buffer
, sizeof(buffer
), pol
);
1669 mpol_to_str(buffer
, sizeof(buffer
), proc_priv
->task_mempolicy
);
1672 seq_printf(m
, "%08lx %s", vma
->vm_start
, buffer
);
1675 seq_puts(m
, " file=");
1676 seq_file_path(m
, file
, "\n\t= ");
1677 } else if (vma
->vm_start
<= mm
->brk
&& vma
->vm_end
>= mm
->start_brk
) {
1678 seq_puts(m
, " heap");
1679 } else if (is_stack(proc_priv
, vma
)) {
1680 seq_puts(m
, " stack");
1683 if (is_vm_hugetlb_page(vma
))
1684 seq_puts(m
, " huge");
1686 /* mmap_sem is held by m_start */
1687 walk_page_vma(vma
, &walk
);
1693 seq_printf(m
, " anon=%lu", md
->anon
);
1696 seq_printf(m
, " dirty=%lu", md
->dirty
);
1698 if (md
->pages
!= md
->anon
&& md
->pages
!= md
->dirty
)
1699 seq_printf(m
, " mapped=%lu", md
->pages
);
1701 if (md
->mapcount_max
> 1)
1702 seq_printf(m
, " mapmax=%lu", md
->mapcount_max
);
1705 seq_printf(m
, " swapcache=%lu", md
->swapcache
);
1707 if (md
->active
< md
->pages
&& !is_vm_hugetlb_page(vma
))
1708 seq_printf(m
, " active=%lu", md
->active
);
1711 seq_printf(m
, " writeback=%lu", md
->writeback
);
1713 for_each_node_state(nid
, N_MEMORY
)
1715 seq_printf(m
, " N%d=%lu", nid
, md
->node
[nid
]);
1717 seq_printf(m
, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma
) >> 10);
1720 m_cache_vma(m
, vma
);
1724 static int show_pid_numa_map(struct seq_file
*m
, void *v
)
1726 return show_numa_map(m
, v
, 1);
1729 static int show_tid_numa_map(struct seq_file
*m
, void *v
)
1731 return show_numa_map(m
, v
, 0);
1734 static const struct seq_operations proc_pid_numa_maps_op
= {
1738 .show
= show_pid_numa_map
,
1741 static const struct seq_operations proc_tid_numa_maps_op
= {
1745 .show
= show_tid_numa_map
,
1748 static int numa_maps_open(struct inode
*inode
, struct file
*file
,
1749 const struct seq_operations
*ops
)
1751 return proc_maps_open(inode
, file
, ops
,
1752 sizeof(struct numa_maps_private
));
1755 static int pid_numa_maps_open(struct inode
*inode
, struct file
*file
)
1757 return numa_maps_open(inode
, file
, &proc_pid_numa_maps_op
);
1760 static int tid_numa_maps_open(struct inode
*inode
, struct file
*file
)
1762 return numa_maps_open(inode
, file
, &proc_tid_numa_maps_op
);
1765 const struct file_operations proc_pid_numa_maps_operations
= {
1766 .open
= pid_numa_maps_open
,
1768 .llseek
= seq_lseek
,
1769 .release
= proc_map_release
,
1772 const struct file_operations proc_tid_numa_maps_operations
= {
1773 .open
= tid_numa_maps_open
,
1775 .llseek
= seq_lseek
,
1776 .release
= proc_map_release
,
1778 #endif /* CONFIG_NUMA */