1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/vmacache.h>
4 #include <linux/hugetlb.h>
5 #include <linux/huge_mm.h>
6 #include <linux/mount.h>
7 #include <linux/seq_file.h>
8 #include <linux/highmem.h>
9 #include <linux/ptrace.h>
10 #include <linux/slab.h>
11 #include <linux/pagemap.h>
12 #include <linux/mempolicy.h>
13 #include <linux/rmap.h>
14 #include <linux/swap.h>
15 #include <linux/sched/mm.h>
16 #include <linux/swapops.h>
17 #include <linux/mmu_notifier.h>
18 #include <linux/page_idle.h>
19 #include <linux/shmem_fs.h>
20 #include <linux/uaccess.h>
21 #include <linux/pkeys.h>
25 #include <asm/tlbflush.h>
28 #define SEQ_PUT_DEC(str, val) \
29 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
30 void task_mem(struct seq_file
*m
, struct mm_struct
*mm
)
32 unsigned long text
, lib
, swap
, anon
, file
, shmem
;
33 unsigned long hiwater_vm
, total_vm
, hiwater_rss
, total_rss
;
35 anon
= get_mm_counter(mm
, MM_ANONPAGES
);
36 file
= get_mm_counter(mm
, MM_FILEPAGES
);
37 shmem
= get_mm_counter(mm
, MM_SHMEMPAGES
);
40 * Note: to minimize their overhead, mm maintains hiwater_vm and
41 * hiwater_rss only when about to *lower* total_vm or rss. Any
42 * collector of these hiwater stats must therefore get total_vm
43 * and rss too, which will usually be the higher. Barriers? not
44 * worth the effort, such snapshots can always be inconsistent.
46 hiwater_vm
= total_vm
= mm
->total_vm
;
47 if (hiwater_vm
< mm
->hiwater_vm
)
48 hiwater_vm
= mm
->hiwater_vm
;
49 hiwater_rss
= total_rss
= anon
+ file
+ shmem
;
50 if (hiwater_rss
< mm
->hiwater_rss
)
51 hiwater_rss
= mm
->hiwater_rss
;
53 /* split executable areas between text and lib */
54 text
= PAGE_ALIGN(mm
->end_code
) - (mm
->start_code
& PAGE_MASK
);
55 text
= min(text
, mm
->exec_vm
<< PAGE_SHIFT
);
56 lib
= (mm
->exec_vm
<< PAGE_SHIFT
) - text
;
58 swap
= get_mm_counter(mm
, MM_SWAPENTS
);
59 SEQ_PUT_DEC("VmPeak:\t", hiwater_vm
);
60 SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm
);
61 SEQ_PUT_DEC(" kB\nVmLck:\t", mm
->locked_vm
);
62 SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm
->pinned_vm
));
63 SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss
);
64 SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss
);
65 SEQ_PUT_DEC(" kB\nRssAnon:\t", anon
);
66 SEQ_PUT_DEC(" kB\nRssFile:\t", file
);
67 SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem
);
68 SEQ_PUT_DEC(" kB\nVmData:\t", mm
->data_vm
);
69 SEQ_PUT_DEC(" kB\nVmStk:\t", mm
->stack_vm
);
70 seq_put_decimal_ull_width(m
,
71 " kB\nVmExe:\t", text
>> 10, 8);
72 seq_put_decimal_ull_width(m
,
73 " kB\nVmLib:\t", lib
>> 10, 8);
74 seq_put_decimal_ull_width(m
,
75 " kB\nVmPTE:\t", mm_pgtables_bytes(mm
) >> 10, 8);
76 SEQ_PUT_DEC(" kB\nVmSwap:\t", swap
);
78 hugetlb_report_usage(m
, mm
);
82 unsigned long task_vsize(struct mm_struct
*mm
)
84 return PAGE_SIZE
* mm
->total_vm
;
87 unsigned long task_statm(struct mm_struct
*mm
,
88 unsigned long *shared
, unsigned long *text
,
89 unsigned long *data
, unsigned long *resident
)
91 *shared
= get_mm_counter(mm
, MM_FILEPAGES
) +
92 get_mm_counter(mm
, MM_SHMEMPAGES
);
93 *text
= (PAGE_ALIGN(mm
->end_code
) - (mm
->start_code
& PAGE_MASK
))
95 *data
= mm
->data_vm
+ mm
->stack_vm
;
96 *resident
= *shared
+ get_mm_counter(mm
, MM_ANONPAGES
);
102 * Save get_task_policy() for show_numa_map().
104 static void hold_task_mempolicy(struct proc_maps_private
*priv
)
106 struct task_struct
*task
= priv
->task
;
109 priv
->task_mempolicy
= get_task_policy(task
);
110 mpol_get(priv
->task_mempolicy
);
113 static void release_task_mempolicy(struct proc_maps_private
*priv
)
115 mpol_put(priv
->task_mempolicy
);
118 static void hold_task_mempolicy(struct proc_maps_private
*priv
)
121 static void release_task_mempolicy(struct proc_maps_private
*priv
)
126 static void vma_stop(struct proc_maps_private
*priv
)
128 struct mm_struct
*mm
= priv
->mm
;
130 release_task_mempolicy(priv
);
131 up_read(&mm
->mmap_sem
);
135 static struct vm_area_struct
*
136 m_next_vma(struct proc_maps_private
*priv
, struct vm_area_struct
*vma
)
138 if (vma
== priv
->tail_vma
)
140 return vma
->vm_next
?: priv
->tail_vma
;
143 static void m_cache_vma(struct seq_file
*m
, struct vm_area_struct
*vma
)
145 if (m
->count
< m
->size
) /* vma is copied successfully */
146 m
->version
= m_next_vma(m
->private, vma
) ? vma
->vm_end
: -1UL;
149 static void *m_start(struct seq_file
*m
, loff_t
*ppos
)
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
;
155 unsigned int pos
= *ppos
;
157 /* See m_cache_vma(). Zero at the start or after lseek. */
158 if (last_addr
== -1UL)
161 priv
->task
= get_proc_task(priv
->inode
);
163 return ERR_PTR(-ESRCH
);
166 if (!mm
|| !mmget_not_zero(mm
))
169 down_read(&mm
->mmap_sem
);
170 hold_task_mempolicy(priv
);
171 priv
->tail_vma
= get_gate_vma(mm
);
174 vma
= find_vma(mm
, last_addr
- 1);
175 if (vma
&& vma
->vm_start
<= last_addr
)
176 vma
= m_next_vma(priv
, vma
);
182 if (pos
< mm
->map_count
) {
183 for (vma
= mm
->mmap
; pos
; pos
--) {
184 m
->version
= vma
->vm_start
;
190 /* we do not bother to update m->version in this case */
191 if (pos
== mm
->map_count
&& priv
->tail_vma
)
192 return priv
->tail_vma
;
198 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
200 struct proc_maps_private
*priv
= m
->private;
201 struct vm_area_struct
*next
;
204 next
= m_next_vma(priv
, v
);
210 static void m_stop(struct seq_file
*m
, void *v
)
212 struct proc_maps_private
*priv
= m
->private;
214 if (!IS_ERR_OR_NULL(v
))
217 put_task_struct(priv
->task
);
222 static int proc_maps_open(struct inode
*inode
, struct file
*file
,
223 const struct seq_operations
*ops
, int psize
)
225 struct proc_maps_private
*priv
= __seq_open_private(file
, ops
, psize
);
231 priv
->mm
= proc_mem_open(inode
, PTRACE_MODE_READ
);
232 if (IS_ERR(priv
->mm
)) {
233 int err
= PTR_ERR(priv
->mm
);
235 seq_release_private(inode
, file
);
242 static int proc_map_release(struct inode
*inode
, struct file
*file
)
244 struct seq_file
*seq
= file
->private_data
;
245 struct proc_maps_private
*priv
= seq
->private;
250 return seq_release_private(inode
, file
);
253 static int do_maps_open(struct inode
*inode
, struct file
*file
,
254 const struct seq_operations
*ops
)
256 return proc_maps_open(inode
, file
, ops
,
257 sizeof(struct proc_maps_private
));
261 * Indicate if the VMA is a stack for the given task; for
262 * /proc/PID/maps that is the stack of the main task.
264 static int is_stack(struct vm_area_struct
*vma
)
267 * We make no effort to guess what a given thread considers to be
268 * its "stack". It's not even well-defined for programs written
271 return vma
->vm_start
<= vma
->vm_mm
->start_stack
&&
272 vma
->vm_end
>= vma
->vm_mm
->start_stack
;
275 static void show_vma_header_prefix(struct seq_file
*m
,
276 unsigned long start
, unsigned long end
,
277 vm_flags_t flags
, unsigned long long pgoff
,
278 dev_t dev
, unsigned long ino
)
280 seq_setwidth(m
, 25 + sizeof(void *) * 6 - 1);
281 seq_put_hex_ll(m
, NULL
, start
, 8);
282 seq_put_hex_ll(m
, "-", end
, 8);
284 seq_putc(m
, flags
& VM_READ
? 'r' : '-');
285 seq_putc(m
, flags
& VM_WRITE
? 'w' : '-');
286 seq_putc(m
, flags
& VM_EXEC
? 'x' : '-');
287 seq_putc(m
, flags
& VM_MAYSHARE
? 's' : 'p');
288 seq_put_hex_ll(m
, " ", pgoff
, 8);
289 seq_put_hex_ll(m
, " ", MAJOR(dev
), 2);
290 seq_put_hex_ll(m
, ":", MINOR(dev
), 2);
291 seq_put_decimal_ull(m
, " ", ino
);
296 show_map_vma(struct seq_file
*m
, struct vm_area_struct
*vma
)
298 struct mm_struct
*mm
= vma
->vm_mm
;
299 struct file
*file
= vma
->vm_file
;
300 vm_flags_t flags
= vma
->vm_flags
;
301 unsigned long ino
= 0;
302 unsigned long long pgoff
= 0;
303 unsigned long start
, end
;
305 const char *name
= NULL
;
308 struct inode
*inode
= file_inode(vma
->vm_file
);
309 dev
= inode
->i_sb
->s_dev
;
311 pgoff
= ((loff_t
)vma
->vm_pgoff
) << PAGE_SHIFT
;
314 start
= vma
->vm_start
;
316 show_vma_header_prefix(m
, start
, end
, flags
, pgoff
, dev
, ino
);
319 * Print the dentry name for named mappings, and a
320 * special [heap] marker for the heap:
324 seq_file_path(m
, file
, "\n");
328 if (vma
->vm_ops
&& vma
->vm_ops
->name
) {
329 name
= vma
->vm_ops
->name(vma
);
334 name
= arch_vma_name(vma
);
341 if (vma
->vm_start
<= mm
->brk
&&
342 vma
->vm_end
>= mm
->start_brk
) {
359 static int show_map(struct seq_file
*m
, void *v
)
366 static const struct seq_operations proc_pid_maps_op
= {
373 static int pid_maps_open(struct inode
*inode
, struct file
*file
)
375 return do_maps_open(inode
, file
, &proc_pid_maps_op
);
378 const struct file_operations proc_pid_maps_operations
= {
379 .open
= pid_maps_open
,
382 .release
= proc_map_release
,
386 * Proportional Set Size(PSS): my share of RSS.
388 * PSS of a process is the count of pages it has in memory, where each
389 * page is divided by the number of processes sharing it. So if a
390 * process has 1000 pages all to itself, and 1000 shared with one other
391 * process, its PSS will be 1500.
393 * To keep (accumulated) division errors low, we adopt a 64bit
394 * fixed-point pss counter to minimize division errors. So (pss >>
395 * PSS_SHIFT) would be the real byte count.
397 * A shift of 12 before division means (assuming 4K page size):
398 * - 1M 3-user-pages add up to 8KB errors;
399 * - supports mapcount up to 2^24, or 16M;
400 * - supports PSS up to 2^52 bytes, or 4PB.
404 #ifdef CONFIG_PROC_PAGE_MONITOR
405 struct mem_size_stats
{
406 unsigned long resident
;
407 unsigned long shared_clean
;
408 unsigned long shared_dirty
;
409 unsigned long private_clean
;
410 unsigned long private_dirty
;
411 unsigned long referenced
;
412 unsigned long anonymous
;
413 unsigned long lazyfree
;
414 unsigned long anonymous_thp
;
415 unsigned long shmem_thp
;
417 unsigned long shared_hugetlb
;
418 unsigned long private_hugetlb
;
422 bool check_shmem_swap
;
425 static void smaps_account(struct mem_size_stats
*mss
, struct page
*page
,
426 bool compound
, bool young
, bool dirty
, bool locked
)
428 int i
, nr
= compound
? 1 << compound_order(page
) : 1;
429 unsigned long size
= nr
* PAGE_SIZE
;
431 if (PageAnon(page
)) {
432 mss
->anonymous
+= size
;
433 if (!PageSwapBacked(page
) && !dirty
&& !PageDirty(page
))
434 mss
->lazyfree
+= size
;
437 mss
->resident
+= size
;
438 /* Accumulate the size in pages that have been accessed. */
439 if (young
|| page_is_young(page
) || PageReferenced(page
))
440 mss
->referenced
+= size
;
443 * page_count(page) == 1 guarantees the page is mapped exactly once.
444 * If any subpage of the compound page mapped with PTE it would elevate
447 if (page_count(page
) == 1) {
448 if (dirty
|| PageDirty(page
))
449 mss
->private_dirty
+= size
;
451 mss
->private_clean
+= size
;
452 mss
->pss
+= (u64
)size
<< PSS_SHIFT
;
454 mss
->pss_locked
+= (u64
)size
<< PSS_SHIFT
;
458 for (i
= 0; i
< nr
; i
++, page
++) {
459 int mapcount
= page_mapcount(page
);
460 unsigned long pss
= (PAGE_SIZE
<< PSS_SHIFT
);
463 if (dirty
|| PageDirty(page
))
464 mss
->shared_dirty
+= PAGE_SIZE
;
466 mss
->shared_clean
+= PAGE_SIZE
;
467 mss
->pss
+= pss
/ mapcount
;
469 mss
->pss_locked
+= pss
/ mapcount
;
471 if (dirty
|| PageDirty(page
))
472 mss
->private_dirty
+= PAGE_SIZE
;
474 mss
->private_clean
+= PAGE_SIZE
;
477 mss
->pss_locked
+= pss
;
483 static int smaps_pte_hole(unsigned long addr
, unsigned long end
,
484 struct mm_walk
*walk
)
486 struct mem_size_stats
*mss
= walk
->private;
488 mss
->swap
+= shmem_partial_swap_usage(
489 walk
->vma
->vm_file
->f_mapping
, addr
, end
);
495 static void smaps_pte_entry(pte_t
*pte
, unsigned long addr
,
496 struct mm_walk
*walk
)
498 struct mem_size_stats
*mss
= walk
->private;
499 struct vm_area_struct
*vma
= walk
->vma
;
500 bool locked
= !!(vma
->vm_flags
& VM_LOCKED
);
501 struct page
*page
= NULL
;
503 if (pte_present(*pte
)) {
504 page
= vm_normal_page(vma
, addr
, *pte
);
505 } else if (is_swap_pte(*pte
)) {
506 swp_entry_t swpent
= pte_to_swp_entry(*pte
);
508 if (!non_swap_entry(swpent
)) {
511 mss
->swap
+= PAGE_SIZE
;
512 mapcount
= swp_swapcount(swpent
);
514 u64 pss_delta
= (u64
)PAGE_SIZE
<< PSS_SHIFT
;
516 do_div(pss_delta
, mapcount
);
517 mss
->swap_pss
+= pss_delta
;
519 mss
->swap_pss
+= (u64
)PAGE_SIZE
<< PSS_SHIFT
;
521 } else if (is_migration_entry(swpent
))
522 page
= migration_entry_to_page(swpent
);
523 else if (is_device_private_entry(swpent
))
524 page
= device_private_entry_to_page(swpent
);
525 } else if (unlikely(IS_ENABLED(CONFIG_SHMEM
) && mss
->check_shmem_swap
526 && pte_none(*pte
))) {
527 page
= find_get_entry(vma
->vm_file
->f_mapping
,
528 linear_page_index(vma
, addr
));
532 if (xa_is_value(page
))
533 mss
->swap
+= PAGE_SIZE
;
543 smaps_account(mss
, page
, false, pte_young(*pte
), pte_dirty(*pte
), locked
);
546 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
547 static void smaps_pmd_entry(pmd_t
*pmd
, unsigned long addr
,
548 struct mm_walk
*walk
)
550 struct mem_size_stats
*mss
= walk
->private;
551 struct vm_area_struct
*vma
= walk
->vma
;
552 bool locked
= !!(vma
->vm_flags
& VM_LOCKED
);
555 /* FOLL_DUMP will return -EFAULT on huge zero page */
556 page
= follow_trans_huge_pmd(vma
, addr
, pmd
, FOLL_DUMP
);
557 if (IS_ERR_OR_NULL(page
))
560 mss
->anonymous_thp
+= HPAGE_PMD_SIZE
;
561 else if (PageSwapBacked(page
))
562 mss
->shmem_thp
+= HPAGE_PMD_SIZE
;
563 else if (is_zone_device_page(page
))
566 VM_BUG_ON_PAGE(1, page
);
567 smaps_account(mss
, page
, true, pmd_young(*pmd
), pmd_dirty(*pmd
), locked
);
570 static void smaps_pmd_entry(pmd_t
*pmd
, unsigned long addr
,
571 struct mm_walk
*walk
)
576 static int smaps_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
,
577 struct mm_walk
*walk
)
579 struct vm_area_struct
*vma
= walk
->vma
;
583 ptl
= pmd_trans_huge_lock(pmd
, vma
);
585 if (pmd_present(*pmd
))
586 smaps_pmd_entry(pmd
, addr
, walk
);
591 if (pmd_trans_unstable(pmd
))
594 * The mmap_sem held all the way back in m_start() is what
595 * keeps khugepaged out of here and from collapsing things
598 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
599 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
600 smaps_pte_entry(pte
, addr
, walk
);
601 pte_unmap_unlock(pte
- 1, ptl
);
607 static void show_smap_vma_flags(struct seq_file
*m
, struct vm_area_struct
*vma
)
610 * Don't forget to update Documentation/ on changes.
612 static const char mnemonics
[BITS_PER_LONG
][2] = {
614 * In case if we meet a flag we don't know about.
616 [0 ... (BITS_PER_LONG
-1)] = "??",
618 [ilog2(VM_READ
)] = "rd",
619 [ilog2(VM_WRITE
)] = "wr",
620 [ilog2(VM_EXEC
)] = "ex",
621 [ilog2(VM_SHARED
)] = "sh",
622 [ilog2(VM_MAYREAD
)] = "mr",
623 [ilog2(VM_MAYWRITE
)] = "mw",
624 [ilog2(VM_MAYEXEC
)] = "me",
625 [ilog2(VM_MAYSHARE
)] = "ms",
626 [ilog2(VM_GROWSDOWN
)] = "gd",
627 [ilog2(VM_PFNMAP
)] = "pf",
628 [ilog2(VM_DENYWRITE
)] = "dw",
629 #ifdef CONFIG_X86_INTEL_MPX
630 [ilog2(VM_MPX
)] = "mp",
632 [ilog2(VM_LOCKED
)] = "lo",
633 [ilog2(VM_IO
)] = "io",
634 [ilog2(VM_SEQ_READ
)] = "sr",
635 [ilog2(VM_RAND_READ
)] = "rr",
636 [ilog2(VM_DONTCOPY
)] = "dc",
637 [ilog2(VM_DONTEXPAND
)] = "de",
638 [ilog2(VM_ACCOUNT
)] = "ac",
639 [ilog2(VM_NORESERVE
)] = "nr",
640 [ilog2(VM_HUGETLB
)] = "ht",
641 [ilog2(VM_SYNC
)] = "sf",
642 [ilog2(VM_ARCH_1
)] = "ar",
643 [ilog2(VM_WIPEONFORK
)] = "wf",
644 [ilog2(VM_DONTDUMP
)] = "dd",
645 #ifdef CONFIG_MEM_SOFT_DIRTY
646 [ilog2(VM_SOFTDIRTY
)] = "sd",
648 [ilog2(VM_MIXEDMAP
)] = "mm",
649 [ilog2(VM_HUGEPAGE
)] = "hg",
650 [ilog2(VM_NOHUGEPAGE
)] = "nh",
651 [ilog2(VM_MERGEABLE
)] = "mg",
652 [ilog2(VM_UFFD_MISSING
)]= "um",
653 [ilog2(VM_UFFD_WP
)] = "uw",
654 #ifdef CONFIG_ARCH_HAS_PKEYS
655 /* These come out via ProtectionKey: */
656 [ilog2(VM_PKEY_BIT0
)] = "",
657 [ilog2(VM_PKEY_BIT1
)] = "",
658 [ilog2(VM_PKEY_BIT2
)] = "",
659 [ilog2(VM_PKEY_BIT3
)] = "",
661 [ilog2(VM_PKEY_BIT4
)] = "",
663 #endif /* CONFIG_ARCH_HAS_PKEYS */
667 seq_puts(m
, "VmFlags: ");
668 for (i
= 0; i
< BITS_PER_LONG
; i
++) {
669 if (!mnemonics
[i
][0])
671 if (vma
->vm_flags
& (1UL << i
)) {
672 seq_putc(m
, mnemonics
[i
][0]);
673 seq_putc(m
, mnemonics
[i
][1]);
680 #ifdef CONFIG_HUGETLB_PAGE
681 static int smaps_hugetlb_range(pte_t
*pte
, unsigned long hmask
,
682 unsigned long addr
, unsigned long end
,
683 struct mm_walk
*walk
)
685 struct mem_size_stats
*mss
= walk
->private;
686 struct vm_area_struct
*vma
= walk
->vma
;
687 struct page
*page
= NULL
;
689 if (pte_present(*pte
)) {
690 page
= vm_normal_page(vma
, addr
, *pte
);
691 } else if (is_swap_pte(*pte
)) {
692 swp_entry_t swpent
= pte_to_swp_entry(*pte
);
694 if (is_migration_entry(swpent
))
695 page
= migration_entry_to_page(swpent
);
696 else if (is_device_private_entry(swpent
))
697 page
= device_private_entry_to_page(swpent
);
700 int mapcount
= page_mapcount(page
);
703 mss
->shared_hugetlb
+= huge_page_size(hstate_vma(vma
));
705 mss
->private_hugetlb
+= huge_page_size(hstate_vma(vma
));
709 #endif /* HUGETLB_PAGE */
711 static void smap_gather_stats(struct vm_area_struct
*vma
,
712 struct mem_size_stats
*mss
)
714 struct mm_walk smaps_walk
= {
715 .pmd_entry
= smaps_pte_range
,
716 #ifdef CONFIG_HUGETLB_PAGE
717 .hugetlb_entry
= smaps_hugetlb_range
,
722 smaps_walk
.private = mss
;
725 /* In case of smaps_rollup, reset the value from previous vma */
726 mss
->check_shmem_swap
= false;
727 if (vma
->vm_file
&& shmem_mapping(vma
->vm_file
->f_mapping
)) {
729 * For shared or readonly shmem mappings we know that all
730 * swapped out pages belong to the shmem object, and we can
731 * obtain the swap value much more efficiently. For private
732 * writable mappings, we might have COW pages that are
733 * not affected by the parent swapped out pages of the shmem
734 * object, so we have to distinguish them during the page walk.
735 * Unless we know that the shmem object (or the part mapped by
736 * our VMA) has no swapped out pages at all.
738 unsigned long shmem_swapped
= shmem_swap_usage(vma
);
740 if (!shmem_swapped
|| (vma
->vm_flags
& VM_SHARED
) ||
741 !(vma
->vm_flags
& VM_WRITE
)) {
742 mss
->swap
+= shmem_swapped
;
744 mss
->check_shmem_swap
= true;
745 smaps_walk
.pte_hole
= smaps_pte_hole
;
749 /* mmap_sem is held in m_start */
750 walk_page_vma(vma
, &smaps_walk
);
753 #define SEQ_PUT_DEC(str, val) \
754 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
756 /* Show the contents common for smaps and smaps_rollup */
757 static void __show_smap(struct seq_file
*m
, const struct mem_size_stats
*mss
)
759 SEQ_PUT_DEC("Rss: ", mss
->resident
);
760 SEQ_PUT_DEC(" kB\nPss: ", mss
->pss
>> PSS_SHIFT
);
761 SEQ_PUT_DEC(" kB\nShared_Clean: ", mss
->shared_clean
);
762 SEQ_PUT_DEC(" kB\nShared_Dirty: ", mss
->shared_dirty
);
763 SEQ_PUT_DEC(" kB\nPrivate_Clean: ", mss
->private_clean
);
764 SEQ_PUT_DEC(" kB\nPrivate_Dirty: ", mss
->private_dirty
);
765 SEQ_PUT_DEC(" kB\nReferenced: ", mss
->referenced
);
766 SEQ_PUT_DEC(" kB\nAnonymous: ", mss
->anonymous
);
767 SEQ_PUT_DEC(" kB\nLazyFree: ", mss
->lazyfree
);
768 SEQ_PUT_DEC(" kB\nAnonHugePages: ", mss
->anonymous_thp
);
769 SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss
->shmem_thp
);
770 SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss
->shared_hugetlb
);
771 seq_put_decimal_ull_width(m
, " kB\nPrivate_Hugetlb: ",
772 mss
->private_hugetlb
>> 10, 7);
773 SEQ_PUT_DEC(" kB\nSwap: ", mss
->swap
);
774 SEQ_PUT_DEC(" kB\nSwapPss: ",
775 mss
->swap_pss
>> PSS_SHIFT
);
776 SEQ_PUT_DEC(" kB\nLocked: ",
777 mss
->pss_locked
>> PSS_SHIFT
);
778 seq_puts(m
, " kB\n");
781 static int show_smap(struct seq_file
*m
, void *v
)
783 struct vm_area_struct
*vma
= v
;
784 struct mem_size_stats mss
;
786 memset(&mss
, 0, sizeof(mss
));
788 smap_gather_stats(vma
, &mss
);
790 show_map_vma(m
, vma
);
792 SEQ_PUT_DEC("Size: ", vma
->vm_end
- vma
->vm_start
);
793 SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma
));
794 SEQ_PUT_DEC(" kB\nMMUPageSize: ", vma_mmu_pagesize(vma
));
795 seq_puts(m
, " kB\n");
797 __show_smap(m
, &mss
);
799 seq_printf(m
, "THPeligible: %d\n", transparent_hugepage_enabled(vma
));
801 if (arch_pkeys_enabled())
802 seq_printf(m
, "ProtectionKey: %8u\n", vma_pkey(vma
));
803 show_smap_vma_flags(m
, vma
);
810 static int show_smaps_rollup(struct seq_file
*m
, void *v
)
812 struct proc_maps_private
*priv
= m
->private;
813 struct mem_size_stats mss
;
814 struct mm_struct
*mm
;
815 struct vm_area_struct
*vma
;
816 unsigned long last_vma_end
= 0;
819 priv
->task
= get_proc_task(priv
->inode
);
824 if (!mm
|| !mmget_not_zero(mm
)) {
829 memset(&mss
, 0, sizeof(mss
));
831 down_read(&mm
->mmap_sem
);
832 hold_task_mempolicy(priv
);
834 for (vma
= priv
->mm
->mmap
; vma
; vma
= vma
->vm_next
) {
835 smap_gather_stats(vma
, &mss
);
836 last_vma_end
= vma
->vm_end
;
839 show_vma_header_prefix(m
, priv
->mm
->mmap
->vm_start
,
840 last_vma_end
, 0, 0, 0, 0);
842 seq_puts(m
, "[rollup]\n");
844 __show_smap(m
, &mss
);
846 release_task_mempolicy(priv
);
847 up_read(&mm
->mmap_sem
);
851 put_task_struct(priv
->task
);
858 static const struct seq_operations proc_pid_smaps_op
= {
865 static int pid_smaps_open(struct inode
*inode
, struct file
*file
)
867 return do_maps_open(inode
, file
, &proc_pid_smaps_op
);
870 static int smaps_rollup_open(struct inode
*inode
, struct file
*file
)
873 struct proc_maps_private
*priv
;
875 priv
= kzalloc(sizeof(*priv
), GFP_KERNEL_ACCOUNT
);
879 ret
= single_open(file
, show_smaps_rollup
, priv
);
884 priv
->mm
= proc_mem_open(inode
, PTRACE_MODE_READ
);
885 if (IS_ERR(priv
->mm
)) {
886 ret
= PTR_ERR(priv
->mm
);
888 single_release(inode
, file
);
899 static int smaps_rollup_release(struct inode
*inode
, struct file
*file
)
901 struct seq_file
*seq
= file
->private_data
;
902 struct proc_maps_private
*priv
= seq
->private;
908 return single_release(inode
, file
);
911 const struct file_operations proc_pid_smaps_operations
= {
912 .open
= pid_smaps_open
,
915 .release
= proc_map_release
,
918 const struct file_operations proc_pid_smaps_rollup_operations
= {
919 .open
= smaps_rollup_open
,
922 .release
= smaps_rollup_release
,
925 enum clear_refs_types
{
929 CLEAR_REFS_SOFT_DIRTY
,
930 CLEAR_REFS_MM_HIWATER_RSS
,
934 struct clear_refs_private
{
935 enum clear_refs_types type
;
938 #ifdef CONFIG_MEM_SOFT_DIRTY
939 static inline void clear_soft_dirty(struct vm_area_struct
*vma
,
940 unsigned long addr
, pte_t
*pte
)
943 * The soft-dirty tracker uses #PF-s to catch writes
944 * to pages, so write-protect the pte as well. See the
945 * Documentation/admin-guide/mm/soft-dirty.rst for full description
946 * of how soft-dirty works.
950 if (pte_present(ptent
)) {
953 old_pte
= ptep_modify_prot_start(vma
, addr
, pte
);
954 ptent
= pte_wrprotect(old_pte
);
955 ptent
= pte_clear_soft_dirty(ptent
);
956 ptep_modify_prot_commit(vma
, addr
, pte
, old_pte
, ptent
);
957 } else if (is_swap_pte(ptent
)) {
958 ptent
= pte_swp_clear_soft_dirty(ptent
);
959 set_pte_at(vma
->vm_mm
, addr
, pte
, ptent
);
963 static inline void clear_soft_dirty(struct vm_area_struct
*vma
,
964 unsigned long addr
, pte_t
*pte
)
969 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
970 static inline void clear_soft_dirty_pmd(struct vm_area_struct
*vma
,
971 unsigned long addr
, pmd_t
*pmdp
)
973 pmd_t old
, pmd
= *pmdp
;
975 if (pmd_present(pmd
)) {
976 /* See comment in change_huge_pmd() */
977 old
= pmdp_invalidate(vma
, addr
, pmdp
);
979 pmd
= pmd_mkdirty(pmd
);
981 pmd
= pmd_mkyoung(pmd
);
983 pmd
= pmd_wrprotect(pmd
);
984 pmd
= pmd_clear_soft_dirty(pmd
);
986 set_pmd_at(vma
->vm_mm
, addr
, pmdp
, pmd
);
987 } else if (is_migration_entry(pmd_to_swp_entry(pmd
))) {
988 pmd
= pmd_swp_clear_soft_dirty(pmd
);
989 set_pmd_at(vma
->vm_mm
, addr
, pmdp
, pmd
);
993 static inline void clear_soft_dirty_pmd(struct vm_area_struct
*vma
,
994 unsigned long addr
, pmd_t
*pmdp
)
999 static int clear_refs_pte_range(pmd_t
*pmd
, unsigned long addr
,
1000 unsigned long end
, struct mm_walk
*walk
)
1002 struct clear_refs_private
*cp
= walk
->private;
1003 struct vm_area_struct
*vma
= walk
->vma
;
1008 ptl
= pmd_trans_huge_lock(pmd
, vma
);
1010 if (cp
->type
== CLEAR_REFS_SOFT_DIRTY
) {
1011 clear_soft_dirty_pmd(vma
, addr
, pmd
);
1015 if (!pmd_present(*pmd
))
1018 page
= pmd_page(*pmd
);
1020 /* Clear accessed and referenced bits. */
1021 pmdp_test_and_clear_young(vma
, addr
, pmd
);
1022 test_and_clear_page_young(page
);
1023 ClearPageReferenced(page
);
1029 if (pmd_trans_unstable(pmd
))
1032 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1033 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
) {
1036 if (cp
->type
== CLEAR_REFS_SOFT_DIRTY
) {
1037 clear_soft_dirty(vma
, addr
, pte
);
1041 if (!pte_present(ptent
))
1044 page
= vm_normal_page(vma
, addr
, ptent
);
1048 /* Clear accessed and referenced bits. */
1049 ptep_test_and_clear_young(vma
, addr
, pte
);
1050 test_and_clear_page_young(page
);
1051 ClearPageReferenced(page
);
1053 pte_unmap_unlock(pte
- 1, ptl
);
1058 static int clear_refs_test_walk(unsigned long start
, unsigned long end
,
1059 struct mm_walk
*walk
)
1061 struct clear_refs_private
*cp
= walk
->private;
1062 struct vm_area_struct
*vma
= walk
->vma
;
1064 if (vma
->vm_flags
& VM_PFNMAP
)
1068 * Writing 1 to /proc/pid/clear_refs affects all pages.
1069 * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1070 * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1071 * Writing 4 to /proc/pid/clear_refs affects all pages.
1073 if (cp
->type
== CLEAR_REFS_ANON
&& vma
->vm_file
)
1075 if (cp
->type
== CLEAR_REFS_MAPPED
&& !vma
->vm_file
)
1080 static ssize_t
clear_refs_write(struct file
*file
, const char __user
*buf
,
1081 size_t count
, loff_t
*ppos
)
1083 struct task_struct
*task
;
1084 char buffer
[PROC_NUMBUF
];
1085 struct mm_struct
*mm
;
1086 struct vm_area_struct
*vma
;
1087 enum clear_refs_types type
;
1088 struct mmu_gather tlb
;
1092 memset(buffer
, 0, sizeof(buffer
));
1093 if (count
> sizeof(buffer
) - 1)
1094 count
= sizeof(buffer
) - 1;
1095 if (copy_from_user(buffer
, buf
, count
))
1097 rv
= kstrtoint(strstrip(buffer
), 10, &itype
);
1100 type
= (enum clear_refs_types
)itype
;
1101 if (type
< CLEAR_REFS_ALL
|| type
>= CLEAR_REFS_LAST
)
1104 task
= get_proc_task(file_inode(file
));
1107 mm
= get_task_mm(task
);
1109 struct mmu_notifier_range range
;
1110 struct clear_refs_private cp
= {
1113 struct mm_walk clear_refs_walk
= {
1114 .pmd_entry
= clear_refs_pte_range
,
1115 .test_walk
= clear_refs_test_walk
,
1120 if (type
== CLEAR_REFS_MM_HIWATER_RSS
) {
1121 if (down_write_killable(&mm
->mmap_sem
)) {
1127 * Writing 5 to /proc/pid/clear_refs resets the peak
1128 * resident set size to this mm's current rss value.
1130 reset_mm_hiwater_rss(mm
);
1131 up_write(&mm
->mmap_sem
);
1135 down_read(&mm
->mmap_sem
);
1136 tlb_gather_mmu(&tlb
, mm
, 0, -1);
1137 if (type
== CLEAR_REFS_SOFT_DIRTY
) {
1138 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1139 if (!(vma
->vm_flags
& VM_SOFTDIRTY
))
1141 up_read(&mm
->mmap_sem
);
1142 if (down_write_killable(&mm
->mmap_sem
)) {
1147 * Avoid to modify vma->vm_flags
1148 * without locked ops while the
1149 * coredump reads the vm_flags.
1151 if (!mmget_still_valid(mm
)) {
1153 * Silently return "count"
1154 * like if get_task_mm()
1155 * failed. FIXME: should this
1156 * function have returned
1157 * -ESRCH if get_task_mm()
1159 * get_proc_task() fails?
1161 up_write(&mm
->mmap_sem
);
1164 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1165 vma
->vm_flags
&= ~VM_SOFTDIRTY
;
1166 vma_set_page_prot(vma
);
1168 downgrade_write(&mm
->mmap_sem
);
1172 mmu_notifier_range_init(&range
, MMU_NOTIFY_SOFT_DIRTY
,
1173 0, NULL
, mm
, 0, -1UL);
1174 mmu_notifier_invalidate_range_start(&range
);
1176 walk_page_range(0, mm
->highest_vm_end
, &clear_refs_walk
);
1177 if (type
== CLEAR_REFS_SOFT_DIRTY
)
1178 mmu_notifier_invalidate_range_end(&range
);
1179 tlb_finish_mmu(&tlb
, 0, -1);
1180 up_read(&mm
->mmap_sem
);
1184 put_task_struct(task
);
1189 const struct file_operations proc_clear_refs_operations
= {
1190 .write
= clear_refs_write
,
1191 .llseek
= noop_llseek
,
1198 struct pagemapread
{
1199 int pos
, len
; /* units: PM_ENTRY_BYTES, not bytes */
1200 pagemap_entry_t
*buffer
;
1204 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
1205 #define PAGEMAP_WALK_MASK (PMD_MASK)
1207 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t)
1208 #define PM_PFRAME_BITS 55
1209 #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1210 #define PM_SOFT_DIRTY BIT_ULL(55)
1211 #define PM_MMAP_EXCLUSIVE BIT_ULL(56)
1212 #define PM_FILE BIT_ULL(61)
1213 #define PM_SWAP BIT_ULL(62)
1214 #define PM_PRESENT BIT_ULL(63)
1216 #define PM_END_OF_BUFFER 1
1218 static inline pagemap_entry_t
make_pme(u64 frame
, u64 flags
)
1220 return (pagemap_entry_t
) { .pme
= (frame
& PM_PFRAME_MASK
) | flags
};
1223 static int add_to_pagemap(unsigned long addr
, pagemap_entry_t
*pme
,
1224 struct pagemapread
*pm
)
1226 pm
->buffer
[pm
->pos
++] = *pme
;
1227 if (pm
->pos
>= pm
->len
)
1228 return PM_END_OF_BUFFER
;
1232 static int pagemap_pte_hole(unsigned long start
, unsigned long end
,
1233 struct mm_walk
*walk
)
1235 struct pagemapread
*pm
= walk
->private;
1236 unsigned long addr
= start
;
1239 while (addr
< end
) {
1240 struct vm_area_struct
*vma
= find_vma(walk
->mm
, addr
);
1241 pagemap_entry_t pme
= make_pme(0, 0);
1242 /* End of address space hole, which we mark as non-present. */
1243 unsigned long hole_end
;
1246 hole_end
= min(end
, vma
->vm_start
);
1250 for (; addr
< hole_end
; addr
+= PAGE_SIZE
) {
1251 err
= add_to_pagemap(addr
, &pme
, pm
);
1259 /* Addresses in the VMA. */
1260 if (vma
->vm_flags
& VM_SOFTDIRTY
)
1261 pme
= make_pme(0, PM_SOFT_DIRTY
);
1262 for (; addr
< min(end
, vma
->vm_end
); addr
+= PAGE_SIZE
) {
1263 err
= add_to_pagemap(addr
, &pme
, pm
);
1272 static pagemap_entry_t
pte_to_pagemap_entry(struct pagemapread
*pm
,
1273 struct vm_area_struct
*vma
, unsigned long addr
, pte_t pte
)
1275 u64 frame
= 0, flags
= 0;
1276 struct page
*page
= NULL
;
1278 if (pte_present(pte
)) {
1280 frame
= pte_pfn(pte
);
1281 flags
|= PM_PRESENT
;
1282 page
= _vm_normal_page(vma
, addr
, pte
, true);
1283 if (pte_soft_dirty(pte
))
1284 flags
|= PM_SOFT_DIRTY
;
1285 } else if (is_swap_pte(pte
)) {
1287 if (pte_swp_soft_dirty(pte
))
1288 flags
|= PM_SOFT_DIRTY
;
1289 entry
= pte_to_swp_entry(pte
);
1291 frame
= swp_type(entry
) |
1292 (swp_offset(entry
) << MAX_SWAPFILES_SHIFT
);
1294 if (is_migration_entry(entry
))
1295 page
= migration_entry_to_page(entry
);
1297 if (is_device_private_entry(entry
))
1298 page
= device_private_entry_to_page(entry
);
1301 if (page
&& !PageAnon(page
))
1303 if (page
&& page_mapcount(page
) == 1)
1304 flags
|= PM_MMAP_EXCLUSIVE
;
1305 if (vma
->vm_flags
& VM_SOFTDIRTY
)
1306 flags
|= PM_SOFT_DIRTY
;
1308 return make_pme(frame
, flags
);
1311 static int pagemap_pmd_range(pmd_t
*pmdp
, unsigned long addr
, unsigned long end
,
1312 struct mm_walk
*walk
)
1314 struct vm_area_struct
*vma
= walk
->vma
;
1315 struct pagemapread
*pm
= walk
->private;
1317 pte_t
*pte
, *orig_pte
;
1320 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1321 ptl
= pmd_trans_huge_lock(pmdp
, vma
);
1323 u64 flags
= 0, frame
= 0;
1325 struct page
*page
= NULL
;
1327 if (vma
->vm_flags
& VM_SOFTDIRTY
)
1328 flags
|= PM_SOFT_DIRTY
;
1330 if (pmd_present(pmd
)) {
1331 page
= pmd_page(pmd
);
1333 flags
|= PM_PRESENT
;
1334 if (pmd_soft_dirty(pmd
))
1335 flags
|= PM_SOFT_DIRTY
;
1337 frame
= pmd_pfn(pmd
) +
1338 ((addr
& ~PMD_MASK
) >> PAGE_SHIFT
);
1340 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1341 else if (is_swap_pmd(pmd
)) {
1342 swp_entry_t entry
= pmd_to_swp_entry(pmd
);
1343 unsigned long offset
;
1346 offset
= swp_offset(entry
) +
1347 ((addr
& ~PMD_MASK
) >> PAGE_SHIFT
);
1348 frame
= swp_type(entry
) |
1349 (offset
<< MAX_SWAPFILES_SHIFT
);
1352 if (pmd_swp_soft_dirty(pmd
))
1353 flags
|= PM_SOFT_DIRTY
;
1354 VM_BUG_ON(!is_pmd_migration_entry(pmd
));
1355 page
= migration_entry_to_page(entry
);
1359 if (page
&& page_mapcount(page
) == 1)
1360 flags
|= PM_MMAP_EXCLUSIVE
;
1362 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
1363 pagemap_entry_t pme
= make_pme(frame
, flags
);
1365 err
= add_to_pagemap(addr
, &pme
, pm
);
1369 if (flags
& PM_PRESENT
)
1371 else if (flags
& PM_SWAP
)
1372 frame
+= (1 << MAX_SWAPFILES_SHIFT
);
1379 if (pmd_trans_unstable(pmdp
))
1381 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1384 * We can assume that @vma always points to a valid one and @end never
1385 * goes beyond vma->vm_end.
1387 orig_pte
= pte
= pte_offset_map_lock(walk
->mm
, pmdp
, addr
, &ptl
);
1388 for (; addr
< end
; pte
++, addr
+= PAGE_SIZE
) {
1389 pagemap_entry_t pme
;
1391 pme
= pte_to_pagemap_entry(pm
, vma
, addr
, *pte
);
1392 err
= add_to_pagemap(addr
, &pme
, pm
);
1396 pte_unmap_unlock(orig_pte
, ptl
);
1403 #ifdef CONFIG_HUGETLB_PAGE
1404 /* This function walks within one hugetlb entry in the single call */
1405 static int pagemap_hugetlb_range(pte_t
*ptep
, unsigned long hmask
,
1406 unsigned long addr
, unsigned long end
,
1407 struct mm_walk
*walk
)
1409 struct pagemapread
*pm
= walk
->private;
1410 struct vm_area_struct
*vma
= walk
->vma
;
1411 u64 flags
= 0, frame
= 0;
1415 if (vma
->vm_flags
& VM_SOFTDIRTY
)
1416 flags
|= PM_SOFT_DIRTY
;
1418 pte
= huge_ptep_get(ptep
);
1419 if (pte_present(pte
)) {
1420 struct page
*page
= pte_page(pte
);
1422 if (!PageAnon(page
))
1425 if (page_mapcount(page
) == 1)
1426 flags
|= PM_MMAP_EXCLUSIVE
;
1428 flags
|= PM_PRESENT
;
1430 frame
= pte_pfn(pte
) +
1431 ((addr
& ~hmask
) >> PAGE_SHIFT
);
1434 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
1435 pagemap_entry_t pme
= make_pme(frame
, flags
);
1437 err
= add_to_pagemap(addr
, &pme
, pm
);
1440 if (pm
->show_pfn
&& (flags
& PM_PRESENT
))
1448 #endif /* HUGETLB_PAGE */
1451 * /proc/pid/pagemap - an array mapping virtual pages to pfns
1453 * For each page in the address space, this file contains one 64-bit entry
1454 * consisting of the following:
1456 * Bits 0-54 page frame number (PFN) if present
1457 * Bits 0-4 swap type if swapped
1458 * Bits 5-54 swap offset if swapped
1459 * Bit 55 pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1460 * Bit 56 page exclusively mapped
1462 * Bit 61 page is file-page or shared-anon
1463 * Bit 62 page swapped
1464 * Bit 63 page present
1466 * If the page is not present but in swap, then the PFN contains an
1467 * encoding of the swap file number and the page's offset into the
1468 * swap. Unmapped pages return a null PFN. This allows determining
1469 * precisely which pages are mapped (or in swap) and comparing mapped
1470 * pages between processes.
1472 * Efficient users of this interface will use /proc/pid/maps to
1473 * determine which areas of memory are actually mapped and llseek to
1474 * skip over unmapped regions.
1476 static ssize_t
pagemap_read(struct file
*file
, char __user
*buf
,
1477 size_t count
, loff_t
*ppos
)
1479 struct mm_struct
*mm
= file
->private_data
;
1480 struct pagemapread pm
;
1481 struct mm_walk pagemap_walk
= {};
1483 unsigned long svpfn
;
1484 unsigned long start_vaddr
;
1485 unsigned long end_vaddr
;
1486 int ret
= 0, copied
= 0;
1488 if (!mm
|| !mmget_not_zero(mm
))
1492 /* file position must be aligned */
1493 if ((*ppos
% PM_ENTRY_BYTES
) || (count
% PM_ENTRY_BYTES
))
1500 /* do not disclose physical addresses: attack vector */
1501 pm
.show_pfn
= file_ns_capable(file
, &init_user_ns
, CAP_SYS_ADMIN
);
1503 pm
.len
= (PAGEMAP_WALK_SIZE
>> PAGE_SHIFT
);
1504 pm
.buffer
= kmalloc_array(pm
.len
, PM_ENTRY_BYTES
, GFP_KERNEL
);
1509 pagemap_walk
.pmd_entry
= pagemap_pmd_range
;
1510 pagemap_walk
.pte_hole
= pagemap_pte_hole
;
1511 #ifdef CONFIG_HUGETLB_PAGE
1512 pagemap_walk
.hugetlb_entry
= pagemap_hugetlb_range
;
1514 pagemap_walk
.mm
= mm
;
1515 pagemap_walk
.private = &pm
;
1518 svpfn
= src
/ PM_ENTRY_BYTES
;
1519 start_vaddr
= svpfn
<< PAGE_SHIFT
;
1520 end_vaddr
= mm
->task_size
;
1522 /* watch out for wraparound */
1523 if (svpfn
> mm
->task_size
>> PAGE_SHIFT
)
1524 start_vaddr
= end_vaddr
;
1527 * The odds are that this will stop walking way
1528 * before end_vaddr, because the length of the
1529 * user buffer is tracked in "pm", and the walk
1530 * will stop when we hit the end of the buffer.
1533 while (count
&& (start_vaddr
< end_vaddr
)) {
1538 end
= (start_vaddr
+ PAGEMAP_WALK_SIZE
) & PAGEMAP_WALK_MASK
;
1540 if (end
< start_vaddr
|| end
> end_vaddr
)
1542 down_read(&mm
->mmap_sem
);
1543 ret
= walk_page_range(start_vaddr
, end
, &pagemap_walk
);
1544 up_read(&mm
->mmap_sem
);
1547 len
= min(count
, PM_ENTRY_BYTES
* pm
.pos
);
1548 if (copy_to_user(buf
, pm
.buffer
, len
)) {
1557 if (!ret
|| ret
== PM_END_OF_BUFFER
)
1568 static int pagemap_open(struct inode
*inode
, struct file
*file
)
1570 struct mm_struct
*mm
;
1572 mm
= proc_mem_open(inode
, PTRACE_MODE_READ
);
1575 file
->private_data
= mm
;
1579 static int pagemap_release(struct inode
*inode
, struct file
*file
)
1581 struct mm_struct
*mm
= file
->private_data
;
1588 const struct file_operations proc_pagemap_operations
= {
1589 .llseek
= mem_lseek
, /* borrow this */
1590 .read
= pagemap_read
,
1591 .open
= pagemap_open
,
1592 .release
= pagemap_release
,
1594 #endif /* CONFIG_PROC_PAGE_MONITOR */
1599 unsigned long pages
;
1601 unsigned long active
;
1602 unsigned long writeback
;
1603 unsigned long mapcount_max
;
1604 unsigned long dirty
;
1605 unsigned long swapcache
;
1606 unsigned long node
[MAX_NUMNODES
];
1609 struct numa_maps_private
{
1610 struct proc_maps_private proc_maps
;
1611 struct numa_maps md
;
1614 static void gather_stats(struct page
*page
, struct numa_maps
*md
, int pte_dirty
,
1615 unsigned long nr_pages
)
1617 int count
= page_mapcount(page
);
1619 md
->pages
+= nr_pages
;
1620 if (pte_dirty
|| PageDirty(page
))
1621 md
->dirty
+= nr_pages
;
1623 if (PageSwapCache(page
))
1624 md
->swapcache
+= nr_pages
;
1626 if (PageActive(page
) || PageUnevictable(page
))
1627 md
->active
+= nr_pages
;
1629 if (PageWriteback(page
))
1630 md
->writeback
+= nr_pages
;
1633 md
->anon
+= nr_pages
;
1635 if (count
> md
->mapcount_max
)
1636 md
->mapcount_max
= count
;
1638 md
->node
[page_to_nid(page
)] += nr_pages
;
1641 static struct page
*can_gather_numa_stats(pte_t pte
, struct vm_area_struct
*vma
,
1647 if (!pte_present(pte
))
1650 page
= vm_normal_page(vma
, addr
, pte
);
1654 if (PageReserved(page
))
1657 nid
= page_to_nid(page
);
1658 if (!node_isset(nid
, node_states
[N_MEMORY
]))
1664 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1665 static struct page
*can_gather_numa_stats_pmd(pmd_t pmd
,
1666 struct vm_area_struct
*vma
,
1672 if (!pmd_present(pmd
))
1675 page
= vm_normal_page_pmd(vma
, addr
, pmd
);
1679 if (PageReserved(page
))
1682 nid
= page_to_nid(page
);
1683 if (!node_isset(nid
, node_states
[N_MEMORY
]))
1690 static int gather_pte_stats(pmd_t
*pmd
, unsigned long addr
,
1691 unsigned long end
, struct mm_walk
*walk
)
1693 struct numa_maps
*md
= walk
->private;
1694 struct vm_area_struct
*vma
= walk
->vma
;
1699 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1700 ptl
= pmd_trans_huge_lock(pmd
, vma
);
1704 page
= can_gather_numa_stats_pmd(*pmd
, vma
, addr
);
1706 gather_stats(page
, md
, pmd_dirty(*pmd
),
1707 HPAGE_PMD_SIZE
/PAGE_SIZE
);
1712 if (pmd_trans_unstable(pmd
))
1715 orig_pte
= pte
= pte_offset_map_lock(walk
->mm
, pmd
, addr
, &ptl
);
1717 struct page
*page
= can_gather_numa_stats(*pte
, vma
, addr
);
1720 gather_stats(page
, md
, pte_dirty(*pte
), 1);
1722 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1723 pte_unmap_unlock(orig_pte
, ptl
);
1727 #ifdef CONFIG_HUGETLB_PAGE
1728 static int gather_hugetlb_stats(pte_t
*pte
, unsigned long hmask
,
1729 unsigned long addr
, unsigned long end
, struct mm_walk
*walk
)
1731 pte_t huge_pte
= huge_ptep_get(pte
);
1732 struct numa_maps
*md
;
1735 if (!pte_present(huge_pte
))
1738 page
= pte_page(huge_pte
);
1743 gather_stats(page
, md
, pte_dirty(huge_pte
), 1);
1748 static int gather_hugetlb_stats(pte_t
*pte
, unsigned long hmask
,
1749 unsigned long addr
, unsigned long end
, struct mm_walk
*walk
)
1756 * Display pages allocated per node and memory policy via /proc.
1758 static int show_numa_map(struct seq_file
*m
, void *v
)
1760 struct numa_maps_private
*numa_priv
= m
->private;
1761 struct proc_maps_private
*proc_priv
= &numa_priv
->proc_maps
;
1762 struct vm_area_struct
*vma
= v
;
1763 struct numa_maps
*md
= &numa_priv
->md
;
1764 struct file
*file
= vma
->vm_file
;
1765 struct mm_struct
*mm
= vma
->vm_mm
;
1766 struct mm_walk walk
= {
1767 .hugetlb_entry
= gather_hugetlb_stats
,
1768 .pmd_entry
= gather_pte_stats
,
1772 struct mempolicy
*pol
;
1779 /* Ensure we start with an empty set of numa_maps statistics. */
1780 memset(md
, 0, sizeof(*md
));
1782 pol
= __get_vma_policy(vma
, vma
->vm_start
);
1784 mpol_to_str(buffer
, sizeof(buffer
), pol
);
1787 mpol_to_str(buffer
, sizeof(buffer
), proc_priv
->task_mempolicy
);
1790 seq_printf(m
, "%08lx %s", vma
->vm_start
, buffer
);
1793 seq_puts(m
, " file=");
1794 seq_file_path(m
, file
, "\n\t= ");
1795 } else if (vma
->vm_start
<= mm
->brk
&& vma
->vm_end
>= mm
->start_brk
) {
1796 seq_puts(m
, " heap");
1797 } else if (is_stack(vma
)) {
1798 seq_puts(m
, " stack");
1801 if (is_vm_hugetlb_page(vma
))
1802 seq_puts(m
, " huge");
1804 /* mmap_sem is held by m_start */
1805 walk_page_vma(vma
, &walk
);
1811 seq_printf(m
, " anon=%lu", md
->anon
);
1814 seq_printf(m
, " dirty=%lu", md
->dirty
);
1816 if (md
->pages
!= md
->anon
&& md
->pages
!= md
->dirty
)
1817 seq_printf(m
, " mapped=%lu", md
->pages
);
1819 if (md
->mapcount_max
> 1)
1820 seq_printf(m
, " mapmax=%lu", md
->mapcount_max
);
1823 seq_printf(m
, " swapcache=%lu", md
->swapcache
);
1825 if (md
->active
< md
->pages
&& !is_vm_hugetlb_page(vma
))
1826 seq_printf(m
, " active=%lu", md
->active
);
1829 seq_printf(m
, " writeback=%lu", md
->writeback
);
1831 for_each_node_state(nid
, N_MEMORY
)
1833 seq_printf(m
, " N%d=%lu", nid
, md
->node
[nid
]);
1835 seq_printf(m
, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma
) >> 10);
1838 m_cache_vma(m
, vma
);
1842 static const struct seq_operations proc_pid_numa_maps_op
= {
1846 .show
= show_numa_map
,
1849 static int pid_numa_maps_open(struct inode
*inode
, struct file
*file
)
1851 return proc_maps_open(inode
, file
, &proc_pid_numa_maps_op
,
1852 sizeof(struct numa_maps_private
));
1855 const struct file_operations proc_pid_numa_maps_operations
= {
1856 .open
= pid_numa_maps_open
,
1858 .llseek
= seq_lseek
,
1859 .release
= proc_map_release
,
1862 #endif /* CONFIG_NUMA */