2 #include <linux/hugetlb.h>
3 #include <linux/huge_mm.h>
4 #include <linux/mount.h>
5 #include <linux/seq_file.h>
6 #include <linux/highmem.h>
7 #include <linux/ptrace.h>
8 #include <linux/slab.h>
9 #include <linux/pagemap.h>
10 #include <linux/mempolicy.h>
11 #include <linux/rmap.h>
12 #include <linux/swap.h>
13 #include <linux/swapops.h>
16 #include <asm/uaccess.h>
17 #include <asm/tlbflush.h>
20 void task_mem(struct seq_file
*m
, struct mm_struct
*mm
)
22 unsigned long data
, text
, lib
, swap
;
23 unsigned long hiwater_vm
, total_vm
, hiwater_rss
, total_rss
;
26 * Note: to minimize their overhead, mm maintains hiwater_vm and
27 * hiwater_rss only when about to *lower* total_vm or rss. Any
28 * collector of these hiwater stats must therefore get total_vm
29 * and rss too, which will usually be the higher. Barriers? not
30 * worth the effort, such snapshots can always be inconsistent.
32 hiwater_vm
= total_vm
= mm
->total_vm
;
33 if (hiwater_vm
< mm
->hiwater_vm
)
34 hiwater_vm
= mm
->hiwater_vm
;
35 hiwater_rss
= total_rss
= get_mm_rss(mm
);
36 if (hiwater_rss
< mm
->hiwater_rss
)
37 hiwater_rss
= mm
->hiwater_rss
;
39 data
= mm
->total_vm
- mm
->shared_vm
- mm
->stack_vm
;
40 text
= (PAGE_ALIGN(mm
->end_code
) - (mm
->start_code
& PAGE_MASK
)) >> 10;
41 lib
= (mm
->exec_vm
<< (PAGE_SHIFT
-10)) - text
;
42 swap
= get_mm_counter(mm
, MM_SWAPENTS
);
56 hiwater_vm
<< (PAGE_SHIFT
-10),
57 total_vm
<< (PAGE_SHIFT
-10),
58 mm
->locked_vm
<< (PAGE_SHIFT
-10),
59 mm
->pinned_vm
<< (PAGE_SHIFT
-10),
60 hiwater_rss
<< (PAGE_SHIFT
-10),
61 total_rss
<< (PAGE_SHIFT
-10),
62 data
<< (PAGE_SHIFT
-10),
63 mm
->stack_vm
<< (PAGE_SHIFT
-10), text
, lib
,
64 (PTRS_PER_PTE
*sizeof(pte_t
)*mm
->nr_ptes
) >> 10,
65 swap
<< (PAGE_SHIFT
-10));
68 unsigned long task_vsize(struct mm_struct
*mm
)
70 return PAGE_SIZE
* mm
->total_vm
;
73 unsigned long task_statm(struct mm_struct
*mm
,
74 unsigned long *shared
, unsigned long *text
,
75 unsigned long *data
, unsigned long *resident
)
77 *shared
= get_mm_counter(mm
, MM_FILEPAGES
);
78 *text
= (PAGE_ALIGN(mm
->end_code
) - (mm
->start_code
& PAGE_MASK
))
80 *data
= mm
->total_vm
- mm
->shared_vm
;
81 *resident
= *shared
+ get_mm_counter(mm
, MM_ANONPAGES
);
85 static void pad_len_spaces(struct seq_file
*m
, int len
)
87 len
= 25 + sizeof(void*) * 6 - len
;
90 seq_printf(m
, "%*c", len
, ' ');
95 * These functions are for numa_maps but called in generic **maps seq_file
96 * ->start(), ->stop() ops.
98 * numa_maps scans all vmas under mmap_sem and checks their mempolicy.
99 * Each mempolicy object is controlled by reference counting. The problem here
100 * is how to avoid accessing dead mempolicy object.
102 * Because we're holding mmap_sem while reading seq_file, it's safe to access
103 * each vma's mempolicy, no vma objects will never drop refs to mempolicy.
105 * A task's mempolicy (task->mempolicy) has different behavior. task->mempolicy
106 * is set and replaced under mmap_sem but unrefed and cleared under task_lock().
107 * So, without task_lock(), we cannot trust get_vma_policy() because we cannot
108 * gurantee the task never exits under us. But taking task_lock() around
109 * get_vma_plicy() causes lock order problem.
111 * To access task->mempolicy without lock, we hold a reference count of an
112 * object pointed by task->mempolicy and remember it. This will guarantee
113 * that task->mempolicy points to an alive object or NULL in numa_maps accesses.
115 static void hold_task_mempolicy(struct proc_maps_private
*priv
)
117 struct task_struct
*task
= priv
->task
;
120 priv
->task_mempolicy
= task
->mempolicy
;
121 mpol_get(priv
->task_mempolicy
);
124 static void release_task_mempolicy(struct proc_maps_private
*priv
)
126 mpol_put(priv
->task_mempolicy
);
129 static void hold_task_mempolicy(struct proc_maps_private
*priv
)
132 static void release_task_mempolicy(struct proc_maps_private
*priv
)
137 static void vma_stop(struct proc_maps_private
*priv
, struct vm_area_struct
*vma
)
139 if (vma
&& vma
!= priv
->tail_vma
) {
140 struct mm_struct
*mm
= vma
->vm_mm
;
141 release_task_mempolicy(priv
);
142 up_read(&mm
->mmap_sem
);
147 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
149 struct proc_maps_private
*priv
= m
->private;
150 unsigned long last_addr
= m
->version
;
151 struct mm_struct
*mm
;
152 struct vm_area_struct
*vma
, *tail_vma
= NULL
;
155 /* Clear the per syscall fields in priv */
157 priv
->tail_vma
= NULL
;
160 * We remember last_addr rather than next_addr to hit with
161 * mmap_cache most of the time. We have zero last_addr at
162 * the beginning and also after lseek. We will have -1 last_addr
163 * after the end of the vmas.
166 if (last_addr
== -1UL)
169 priv
->task
= get_pid_task(priv
->pid
, PIDTYPE_PID
);
171 return ERR_PTR(-ESRCH
);
173 mm
= mm_access(priv
->task
, PTRACE_MODE_READ
);
174 if (!mm
|| IS_ERR(mm
))
176 down_read(&mm
->mmap_sem
);
178 tail_vma
= get_gate_vma(priv
->task
->mm
);
179 priv
->tail_vma
= tail_vma
;
180 hold_task_mempolicy(priv
);
181 /* Start with last addr hint */
182 vma
= find_vma(mm
, last_addr
);
183 if (last_addr
&& vma
) {
189 * Check the vma index is within the range and do
190 * sequential scan until m_index.
193 if ((unsigned long)l
< mm
->map_count
) {
200 if (l
!= mm
->map_count
)
201 tail_vma
= NULL
; /* After gate vma */
207 release_task_mempolicy(priv
);
208 /* End of vmas has been reached */
209 m
->version
= (tail_vma
!= NULL
)? 0: -1UL;
210 up_read(&mm
->mmap_sem
);
215 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
217 struct proc_maps_private
*priv
= m
->private;
218 struct vm_area_struct
*vma
= v
;
219 struct vm_area_struct
*tail_vma
= priv
->tail_vma
;
222 if (vma
&& (vma
!= tail_vma
) && vma
->vm_next
)
225 return (vma
!= tail_vma
)? tail_vma
: NULL
;
228 static void m_stop(struct seq_file
*m
, void *v
)
230 struct proc_maps_private
*priv
= m
->private;
231 struct vm_area_struct
*vma
= v
;
236 put_task_struct(priv
->task
);
239 static int do_maps_open(struct inode
*inode
, struct file
*file
,
240 const struct seq_operations
*ops
)
242 struct proc_maps_private
*priv
;
244 priv
= kzalloc(sizeof(*priv
), GFP_KERNEL
);
246 priv
->pid
= proc_pid(inode
);
247 ret
= seq_open(file
, ops
);
249 struct seq_file
*m
= file
->private_data
;
259 show_map_vma(struct seq_file
*m
, struct vm_area_struct
*vma
, int is_pid
)
261 struct mm_struct
*mm
= vma
->vm_mm
;
262 struct file
*file
= vma
->vm_file
;
263 struct proc_maps_private
*priv
= m
->private;
264 struct task_struct
*task
= priv
->task
;
265 vm_flags_t flags
= vma
->vm_flags
;
266 unsigned long ino
= 0;
267 unsigned long long pgoff
= 0;
268 unsigned long start
, end
;
271 const char *name
= NULL
;
274 struct inode
*inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
275 dev
= inode
->i_sb
->s_dev
;
277 pgoff
= ((loff_t
)vma
->vm_pgoff
) << PAGE_SHIFT
;
280 /* We don't show the stack guard page in /proc/maps */
281 start
= vma
->vm_start
;
282 if (stack_guard_page_start(vma
, start
))
285 if (stack_guard_page_end(vma
, end
))
288 seq_printf(m
, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
291 flags
& VM_READ
? 'r' : '-',
292 flags
& VM_WRITE
? 'w' : '-',
293 flags
& VM_EXEC
? 'x' : '-',
294 flags
& VM_MAYSHARE
? 's' : 'p',
296 MAJOR(dev
), MINOR(dev
), ino
, &len
);
299 * Print the dentry name for named mappings, and a
300 * special [heap] marker for the heap:
303 pad_len_spaces(m
, len
);
304 seq_path(m
, &file
->f_path
, "\n");
308 name
= arch_vma_name(vma
);
317 if (vma
->vm_start
<= mm
->brk
&&
318 vma
->vm_end
>= mm
->start_brk
) {
323 tid
= vm_is_stack(task
, vma
, is_pid
);
327 * Thread stack in /proc/PID/task/TID/maps or
328 * the main process stack.
330 if (!is_pid
|| (vma
->vm_start
<= mm
->start_stack
&&
331 vma
->vm_end
>= mm
->start_stack
)) {
334 /* Thread stack in /proc/PID/maps */
335 pad_len_spaces(m
, len
);
336 seq_printf(m
, "[stack:%d]", tid
);
343 pad_len_spaces(m
, len
);
349 static int show_map(struct seq_file
*m
, void *v
, int is_pid
)
351 struct vm_area_struct
*vma
= v
;
352 struct proc_maps_private
*priv
= m
->private;
353 struct task_struct
*task
= priv
->task
;
355 show_map_vma(m
, vma
, is_pid
);
357 if (m
->count
< m
->size
) /* vma is copied successfully */
358 m
->version
= (vma
!= get_gate_vma(task
->mm
))
363 static int show_pid_map(struct seq_file
*m
, void *v
)
365 return show_map(m
, v
, 1);
368 static int show_tid_map(struct seq_file
*m
, void *v
)
370 return show_map(m
, v
, 0);
373 static const struct seq_operations proc_pid_maps_op
= {
380 static const struct seq_operations proc_tid_maps_op
= {
387 static int pid_maps_open(struct inode
*inode
, struct file
*file
)
389 return do_maps_open(inode
, file
, &proc_pid_maps_op
);
392 static int tid_maps_open(struct inode
*inode
, struct file
*file
)
394 return do_maps_open(inode
, file
, &proc_tid_maps_op
);
397 const struct file_operations proc_pid_maps_operations
= {
398 .open
= pid_maps_open
,
401 .release
= seq_release_private
,
404 const struct file_operations proc_tid_maps_operations
= {
405 .open
= tid_maps_open
,
408 .release
= seq_release_private
,
412 * Proportional Set Size(PSS): my share of RSS.
414 * PSS of a process is the count of pages it has in memory, where each
415 * page is divided by the number of processes sharing it. So if a
416 * process has 1000 pages all to itself, and 1000 shared with one other
417 * process, its PSS will be 1500.
419 * To keep (accumulated) division errors low, we adopt a 64bit
420 * fixed-point pss counter to minimize division errors. So (pss >>
421 * PSS_SHIFT) would be the real byte count.
423 * A shift of 12 before division means (assuming 4K page size):
424 * - 1M 3-user-pages add up to 8KB errors;
425 * - supports mapcount up to 2^24, or 16M;
426 * - supports PSS up to 2^52 bytes, or 4PB.
430 #ifdef CONFIG_PROC_PAGE_MONITOR
431 struct mem_size_stats
{
432 struct vm_area_struct
*vma
;
433 unsigned long resident
;
434 unsigned long shared_clean
;
435 unsigned long shared_dirty
;
436 unsigned long private_clean
;
437 unsigned long private_dirty
;
438 unsigned long referenced
;
439 unsigned long anonymous
;
440 unsigned long anonymous_thp
;
442 unsigned long nonlinear
;
447 static void smaps_pte_entry(pte_t ptent
, unsigned long addr
,
448 unsigned long ptent_size
, struct mm_walk
*walk
)
450 struct mem_size_stats
*mss
= walk
->private;
451 struct vm_area_struct
*vma
= mss
->vma
;
452 pgoff_t pgoff
= linear_page_index(vma
, addr
);
453 struct page
*page
= NULL
;
456 if (pte_present(ptent
)) {
457 page
= vm_normal_page(vma
, addr
, ptent
);
458 } else if (is_swap_pte(ptent
)) {
459 swp_entry_t swpent
= pte_to_swp_entry(ptent
);
461 if (!non_swap_entry(swpent
))
462 mss
->swap
+= ptent_size
;
463 else if (is_migration_entry(swpent
))
464 page
= migration_entry_to_page(swpent
);
465 } else if (pte_file(ptent
)) {
466 if (pte_to_pgoff(ptent
) != pgoff
)
467 mss
->nonlinear
+= ptent_size
;
474 mss
->anonymous
+= ptent_size
;
476 if (page
->index
!= pgoff
)
477 mss
->nonlinear
+= ptent_size
;
479 mss
->resident
+= ptent_size
;
480 /* Accumulate the size in pages that have been accessed. */
481 if (pte_young(ptent
) || PageReferenced(page
))
482 mss
->referenced
+= ptent_size
;
483 mapcount
= page_mapcount(page
);
485 if (pte_dirty(ptent
) || PageDirty(page
))
486 mss
->shared_dirty
+= ptent_size
;
488 mss
->shared_clean
+= ptent_size
;
489 mss
->pss
+= (ptent_size
<< PSS_SHIFT
) / mapcount
;
491 if (pte_dirty(ptent
) || PageDirty(page
))
492 mss
->private_dirty
+= ptent_size
;
494 mss
->private_clean
+= ptent_size
;
495 mss
->pss
+= (ptent_size
<< PSS_SHIFT
);
499 static int smaps_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
,
500 struct mm_walk
*walk
)
502 struct mem_size_stats
*mss
= walk
->private;
503 struct vm_area_struct
*vma
= mss
->vma
;
507 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
508 smaps_pte_entry(*(pte_t
*)pmd
, addr
, HPAGE_PMD_SIZE
, walk
);
509 spin_unlock(&walk
->mm
->page_table_lock
);
510 mss
->anonymous_thp
+= HPAGE_PMD_SIZE
;
514 if (pmd_trans_unstable(pmd
))
517 * The mmap_sem held all the way back in m_start() is what
518 * keeps khugepaged out of here and from collapsing things
521 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
522 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
523 smaps_pte_entry(*pte
, addr
, PAGE_SIZE
, walk
);
524 pte_unmap_unlock(pte
- 1, ptl
);
529 static int show_smap(struct seq_file
*m
, void *v
, int is_pid
)
531 struct proc_maps_private
*priv
= m
->private;
532 struct task_struct
*task
= priv
->task
;
533 struct vm_area_struct
*vma
= v
;
534 struct mem_size_stats mss
;
535 struct mm_walk smaps_walk
= {
536 .pmd_entry
= smaps_pte_range
,
541 memset(&mss
, 0, sizeof mss
);
543 /* mmap_sem is held in m_start */
544 if (vma
->vm_mm
&& !is_vm_hugetlb_page(vma
))
545 walk_page_range(vma
->vm_start
, vma
->vm_end
, &smaps_walk
);
547 show_map_vma(m
, vma
, is_pid
);
553 "Shared_Clean: %8lu kB\n"
554 "Shared_Dirty: %8lu kB\n"
555 "Private_Clean: %8lu kB\n"
556 "Private_Dirty: %8lu kB\n"
557 "Referenced: %8lu kB\n"
558 "Anonymous: %8lu kB\n"
559 "AnonHugePages: %8lu kB\n"
561 "KernelPageSize: %8lu kB\n"
562 "MMUPageSize: %8lu kB\n"
564 (vma
->vm_end
- vma
->vm_start
) >> 10,
566 (unsigned long)(mss
.pss
>> (10 + PSS_SHIFT
)),
567 mss
.shared_clean
>> 10,
568 mss
.shared_dirty
>> 10,
569 mss
.private_clean
>> 10,
570 mss
.private_dirty
>> 10,
571 mss
.referenced
>> 10,
573 mss
.anonymous_thp
>> 10,
575 vma_kernel_pagesize(vma
) >> 10,
576 vma_mmu_pagesize(vma
) >> 10,
577 (vma
->vm_flags
& VM_LOCKED
) ?
578 (unsigned long)(mss
.pss
>> (10 + PSS_SHIFT
)) : 0);
580 if (vma
->vm_flags
& VM_NONLINEAR
)
581 seq_printf(m
, "Nonlinear: %8lu kB\n",
582 mss
.nonlinear
>> 10);
584 if (m
->count
< m
->size
) /* vma is copied successfully */
585 m
->version
= (vma
!= get_gate_vma(task
->mm
))
590 static int show_pid_smap(struct seq_file
*m
, void *v
)
592 return show_smap(m
, v
, 1);
595 static int show_tid_smap(struct seq_file
*m
, void *v
)
597 return show_smap(m
, v
, 0);
600 static const struct seq_operations proc_pid_smaps_op
= {
604 .show
= show_pid_smap
607 static const struct seq_operations proc_tid_smaps_op
= {
611 .show
= show_tid_smap
614 static int pid_smaps_open(struct inode
*inode
, struct file
*file
)
616 return do_maps_open(inode
, file
, &proc_pid_smaps_op
);
619 static int tid_smaps_open(struct inode
*inode
, struct file
*file
)
621 return do_maps_open(inode
, file
, &proc_tid_smaps_op
);
624 const struct file_operations proc_pid_smaps_operations
= {
625 .open
= pid_smaps_open
,
628 .release
= seq_release_private
,
631 const struct file_operations proc_tid_smaps_operations
= {
632 .open
= tid_smaps_open
,
635 .release
= seq_release_private
,
638 static int clear_refs_pte_range(pmd_t
*pmd
, unsigned long addr
,
639 unsigned long end
, struct mm_walk
*walk
)
641 struct vm_area_struct
*vma
= walk
->private;
646 split_huge_page_pmd(walk
->mm
, pmd
);
647 if (pmd_trans_unstable(pmd
))
650 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
651 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
) {
653 if (!pte_present(ptent
))
656 page
= vm_normal_page(vma
, addr
, ptent
);
660 /* Clear accessed and referenced bits. */
661 ptep_test_and_clear_young(vma
, addr
, pte
);
662 ClearPageReferenced(page
);
664 pte_unmap_unlock(pte
- 1, ptl
);
669 #define CLEAR_REFS_ALL 1
670 #define CLEAR_REFS_ANON 2
671 #define CLEAR_REFS_MAPPED 3
673 static ssize_t
clear_refs_write(struct file
*file
, const char __user
*buf
,
674 size_t count
, loff_t
*ppos
)
676 struct task_struct
*task
;
677 char buffer
[PROC_NUMBUF
];
678 struct mm_struct
*mm
;
679 struct vm_area_struct
*vma
;
683 memset(buffer
, 0, sizeof(buffer
));
684 if (count
> sizeof(buffer
) - 1)
685 count
= sizeof(buffer
) - 1;
686 if (copy_from_user(buffer
, buf
, count
))
688 rv
= kstrtoint(strstrip(buffer
), 10, &type
);
691 if (type
< CLEAR_REFS_ALL
|| type
> CLEAR_REFS_MAPPED
)
693 task
= get_proc_task(file
->f_path
.dentry
->d_inode
);
696 mm
= get_task_mm(task
);
698 struct mm_walk clear_refs_walk
= {
699 .pmd_entry
= clear_refs_pte_range
,
702 down_read(&mm
->mmap_sem
);
703 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
704 clear_refs_walk
.private = vma
;
705 if (is_vm_hugetlb_page(vma
))
708 * Writing 1 to /proc/pid/clear_refs affects all pages.
710 * Writing 2 to /proc/pid/clear_refs only affects
713 * Writing 3 to /proc/pid/clear_refs only affects file
716 if (type
== CLEAR_REFS_ANON
&& vma
->vm_file
)
718 if (type
== CLEAR_REFS_MAPPED
&& !vma
->vm_file
)
720 walk_page_range(vma
->vm_start
, vma
->vm_end
,
724 up_read(&mm
->mmap_sem
);
727 put_task_struct(task
);
732 const struct file_operations proc_clear_refs_operations
= {
733 .write
= clear_refs_write
,
734 .llseek
= noop_llseek
,
743 pagemap_entry_t
*buffer
;
746 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
747 #define PAGEMAP_WALK_MASK (PMD_MASK)
749 #define PM_ENTRY_BYTES sizeof(u64)
750 #define PM_STATUS_BITS 3
751 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
752 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
753 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
754 #define PM_PSHIFT_BITS 6
755 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
756 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
757 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
758 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
759 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
761 #define PM_PRESENT PM_STATUS(4LL)
762 #define PM_SWAP PM_STATUS(2LL)
763 #define PM_FILE PM_STATUS(1LL)
764 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
765 #define PM_END_OF_BUFFER 1
767 static inline pagemap_entry_t
make_pme(u64 val
)
769 return (pagemap_entry_t
) { .pme
= val
};
772 static int add_to_pagemap(unsigned long addr
, pagemap_entry_t
*pme
,
773 struct pagemapread
*pm
)
775 pm
->buffer
[pm
->pos
++] = *pme
;
776 if (pm
->pos
>= pm
->len
)
777 return PM_END_OF_BUFFER
;
781 static int pagemap_pte_hole(unsigned long start
, unsigned long end
,
782 struct mm_walk
*walk
)
784 struct pagemapread
*pm
= walk
->private;
787 pagemap_entry_t pme
= make_pme(PM_NOT_PRESENT
);
789 for (addr
= start
; addr
< end
; addr
+= PAGE_SIZE
) {
790 err
= add_to_pagemap(addr
, &pme
, pm
);
797 static void pte_to_pagemap_entry(pagemap_entry_t
*pme
,
798 struct vm_area_struct
*vma
, unsigned long addr
, pte_t pte
)
801 struct page
*page
= NULL
;
803 if (pte_present(pte
)) {
804 frame
= pte_pfn(pte
);
806 page
= vm_normal_page(vma
, addr
, pte
);
807 } else if (is_swap_pte(pte
)) {
808 swp_entry_t entry
= pte_to_swp_entry(pte
);
810 frame
= swp_type(entry
) |
811 (swp_offset(entry
) << MAX_SWAPFILES_SHIFT
);
813 if (is_migration_entry(entry
))
814 page
= migration_entry_to_page(entry
);
816 *pme
= make_pme(PM_NOT_PRESENT
);
820 if (page
&& !PageAnon(page
))
823 *pme
= make_pme(PM_PFRAME(frame
) | PM_PSHIFT(PAGE_SHIFT
) | flags
);
826 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
827 static void thp_pmd_to_pagemap_entry(pagemap_entry_t
*pme
,
828 pmd_t pmd
, int offset
)
831 * Currently pmd for thp is always present because thp can not be
832 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
833 * This if-check is just to prepare for future implementation.
835 if (pmd_present(pmd
))
836 *pme
= make_pme(PM_PFRAME(pmd_pfn(pmd
) + offset
)
837 | PM_PSHIFT(PAGE_SHIFT
) | PM_PRESENT
);
839 *pme
= make_pme(PM_NOT_PRESENT
);
842 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t
*pme
,
843 pmd_t pmd
, int offset
)
848 static int pagemap_pte_range(pmd_t
*pmd
, unsigned long addr
, unsigned long end
,
849 struct mm_walk
*walk
)
851 struct vm_area_struct
*vma
;
852 struct pagemapread
*pm
= walk
->private;
855 pagemap_entry_t pme
= make_pme(PM_NOT_PRESENT
);
857 /* find the first VMA at or above 'addr' */
858 vma
= find_vma(walk
->mm
, addr
);
859 if (vma
&& pmd_trans_huge_lock(pmd
, vma
) == 1) {
860 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
861 unsigned long offset
;
863 offset
= (addr
& ~PAGEMAP_WALK_MASK
) >>
865 thp_pmd_to_pagemap_entry(&pme
, *pmd
, offset
);
866 err
= add_to_pagemap(addr
, &pme
, pm
);
870 spin_unlock(&walk
->mm
->page_table_lock
);
874 if (pmd_trans_unstable(pmd
))
876 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
878 /* check to see if we've left 'vma' behind
879 * and need a new, higher one */
880 if (vma
&& (addr
>= vma
->vm_end
)) {
881 vma
= find_vma(walk
->mm
, addr
);
882 pme
= make_pme(PM_NOT_PRESENT
);
885 /* check that 'vma' actually covers this address,
886 * and that it isn't a huge page vma */
887 if (vma
&& (vma
->vm_start
<= addr
) &&
888 !is_vm_hugetlb_page(vma
)) {
889 pte
= pte_offset_map(pmd
, addr
);
890 pte_to_pagemap_entry(&pme
, vma
, addr
, *pte
);
891 /* unmap before userspace copy */
894 err
= add_to_pagemap(addr
, &pme
, pm
);
904 #ifdef CONFIG_HUGETLB_PAGE
905 static void huge_pte_to_pagemap_entry(pagemap_entry_t
*pme
,
906 pte_t pte
, int offset
)
908 if (pte_present(pte
))
909 *pme
= make_pme(PM_PFRAME(pte_pfn(pte
) + offset
)
910 | PM_PSHIFT(PAGE_SHIFT
) | PM_PRESENT
);
912 *pme
= make_pme(PM_NOT_PRESENT
);
915 /* This function walks within one hugetlb entry in the single call */
916 static int pagemap_hugetlb_range(pte_t
*pte
, unsigned long hmask
,
917 unsigned long addr
, unsigned long end
,
918 struct mm_walk
*walk
)
920 struct pagemapread
*pm
= walk
->private;
924 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
925 int offset
= (addr
& ~hmask
) >> PAGE_SHIFT
;
926 huge_pte_to_pagemap_entry(&pme
, *pte
, offset
);
927 err
= add_to_pagemap(addr
, &pme
, pm
);
936 #endif /* HUGETLB_PAGE */
939 * /proc/pid/pagemap - an array mapping virtual pages to pfns
941 * For each page in the address space, this file contains one 64-bit entry
942 * consisting of the following:
944 * Bits 0-54 page frame number (PFN) if present
945 * Bits 0-4 swap type if swapped
946 * Bits 5-54 swap offset if swapped
947 * Bits 55-60 page shift (page size = 1<<page shift)
948 * Bit 61 page is file-page or shared-anon
949 * Bit 62 page swapped
950 * Bit 63 page present
952 * If the page is not present but in swap, then the PFN contains an
953 * encoding of the swap file number and the page's offset into the
954 * swap. Unmapped pages return a null PFN. This allows determining
955 * precisely which pages are mapped (or in swap) and comparing mapped
956 * pages between processes.
958 * Efficient users of this interface will use /proc/pid/maps to
959 * determine which areas of memory are actually mapped and llseek to
960 * skip over unmapped regions.
962 static ssize_t
pagemap_read(struct file
*file
, char __user
*buf
,
963 size_t count
, loff_t
*ppos
)
965 struct task_struct
*task
= get_proc_task(file
->f_path
.dentry
->d_inode
);
966 struct mm_struct
*mm
;
967 struct pagemapread pm
;
969 struct mm_walk pagemap_walk
= {};
972 unsigned long start_vaddr
;
973 unsigned long end_vaddr
;
980 /* file position must be aligned */
981 if ((*ppos
% PM_ENTRY_BYTES
) || (count
% PM_ENTRY_BYTES
))
988 pm
.len
= PM_ENTRY_BYTES
* (PAGEMAP_WALK_SIZE
>> PAGE_SHIFT
);
989 pm
.buffer
= kmalloc(pm
.len
, GFP_TEMPORARY
);
994 mm
= mm_access(task
, PTRACE_MODE_READ
);
996 if (!mm
|| IS_ERR(mm
))
999 pagemap_walk
.pmd_entry
= pagemap_pte_range
;
1000 pagemap_walk
.pte_hole
= pagemap_pte_hole
;
1001 #ifdef CONFIG_HUGETLB_PAGE
1002 pagemap_walk
.hugetlb_entry
= pagemap_hugetlb_range
;
1004 pagemap_walk
.mm
= mm
;
1005 pagemap_walk
.private = &pm
;
1008 svpfn
= src
/ PM_ENTRY_BYTES
;
1009 start_vaddr
= svpfn
<< PAGE_SHIFT
;
1010 end_vaddr
= TASK_SIZE_OF(task
);
1012 /* watch out for wraparound */
1013 if (svpfn
> TASK_SIZE_OF(task
) >> PAGE_SHIFT
)
1014 start_vaddr
= end_vaddr
;
1017 * The odds are that this will stop walking way
1018 * before end_vaddr, because the length of the
1019 * user buffer is tracked in "pm", and the walk
1020 * will stop when we hit the end of the buffer.
1023 while (count
&& (start_vaddr
< end_vaddr
)) {
1028 end
= (start_vaddr
+ PAGEMAP_WALK_SIZE
) & PAGEMAP_WALK_MASK
;
1030 if (end
< start_vaddr
|| end
> end_vaddr
)
1032 down_read(&mm
->mmap_sem
);
1033 ret
= walk_page_range(start_vaddr
, end
, &pagemap_walk
);
1034 up_read(&mm
->mmap_sem
);
1037 len
= min(count
, PM_ENTRY_BYTES
* pm
.pos
);
1038 if (copy_to_user(buf
, pm
.buffer
, len
)) {
1047 if (!ret
|| ret
== PM_END_OF_BUFFER
)
1055 put_task_struct(task
);
1060 const struct file_operations proc_pagemap_operations
= {
1061 .llseek
= mem_lseek
, /* borrow this */
1062 .read
= pagemap_read
,
1064 #endif /* CONFIG_PROC_PAGE_MONITOR */
1069 struct vm_area_struct
*vma
;
1070 unsigned long pages
;
1072 unsigned long active
;
1073 unsigned long writeback
;
1074 unsigned long mapcount_max
;
1075 unsigned long dirty
;
1076 unsigned long swapcache
;
1077 unsigned long node
[MAX_NUMNODES
];
1080 struct numa_maps_private
{
1081 struct proc_maps_private proc_maps
;
1082 struct numa_maps md
;
1085 static void gather_stats(struct page
*page
, struct numa_maps
*md
, int pte_dirty
,
1086 unsigned long nr_pages
)
1088 int count
= page_mapcount(page
);
1090 md
->pages
+= nr_pages
;
1091 if (pte_dirty
|| PageDirty(page
))
1092 md
->dirty
+= nr_pages
;
1094 if (PageSwapCache(page
))
1095 md
->swapcache
+= nr_pages
;
1097 if (PageActive(page
) || PageUnevictable(page
))
1098 md
->active
+= nr_pages
;
1100 if (PageWriteback(page
))
1101 md
->writeback
+= nr_pages
;
1104 md
->anon
+= nr_pages
;
1106 if (count
> md
->mapcount_max
)
1107 md
->mapcount_max
= count
;
1109 md
->node
[page_to_nid(page
)] += nr_pages
;
1112 static struct page
*can_gather_numa_stats(pte_t pte
, struct vm_area_struct
*vma
,
1118 if (!pte_present(pte
))
1121 page
= vm_normal_page(vma
, addr
, pte
);
1125 if (PageReserved(page
))
1128 nid
= page_to_nid(page
);
1129 if (!node_isset(nid
, node_states
[N_HIGH_MEMORY
]))
1135 static int gather_pte_stats(pmd_t
*pmd
, unsigned long addr
,
1136 unsigned long end
, struct mm_walk
*walk
)
1138 struct numa_maps
*md
;
1145 if (pmd_trans_huge_lock(pmd
, md
->vma
) == 1) {
1146 pte_t huge_pte
= *(pte_t
*)pmd
;
1149 page
= can_gather_numa_stats(huge_pte
, md
->vma
, addr
);
1151 gather_stats(page
, md
, pte_dirty(huge_pte
),
1152 HPAGE_PMD_SIZE
/PAGE_SIZE
);
1153 spin_unlock(&walk
->mm
->page_table_lock
);
1157 if (pmd_trans_unstable(pmd
))
1159 orig_pte
= pte
= pte_offset_map_lock(walk
->mm
, pmd
, addr
, &ptl
);
1161 struct page
*page
= can_gather_numa_stats(*pte
, md
->vma
, addr
);
1164 gather_stats(page
, md
, pte_dirty(*pte
), 1);
1166 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1167 pte_unmap_unlock(orig_pte
, ptl
);
1170 #ifdef CONFIG_HUGETLB_PAGE
1171 static int gather_hugetbl_stats(pte_t
*pte
, unsigned long hmask
,
1172 unsigned long addr
, unsigned long end
, struct mm_walk
*walk
)
1174 struct numa_maps
*md
;
1180 page
= pte_page(*pte
);
1185 gather_stats(page
, md
, pte_dirty(*pte
), 1);
1190 static int gather_hugetbl_stats(pte_t
*pte
, unsigned long hmask
,
1191 unsigned long addr
, unsigned long end
, struct mm_walk
*walk
)
1198 * Display pages allocated per node and memory policy via /proc.
1200 static int show_numa_map(struct seq_file
*m
, void *v
, int is_pid
)
1202 struct numa_maps_private
*numa_priv
= m
->private;
1203 struct proc_maps_private
*proc_priv
= &numa_priv
->proc_maps
;
1204 struct vm_area_struct
*vma
= v
;
1205 struct numa_maps
*md
= &numa_priv
->md
;
1206 struct file
*file
= vma
->vm_file
;
1207 struct task_struct
*task
= proc_priv
->task
;
1208 struct mm_struct
*mm
= vma
->vm_mm
;
1209 struct mm_walk walk
= {};
1210 struct mempolicy
*pol
;
1217 /* Ensure we start with an empty set of numa_maps statistics. */
1218 memset(md
, 0, sizeof(*md
));
1222 walk
.hugetlb_entry
= gather_hugetbl_stats
;
1223 walk
.pmd_entry
= gather_pte_stats
;
1227 pol
= get_vma_policy(task
, vma
, vma
->vm_start
);
1228 mpol_to_str(buffer
, sizeof(buffer
), pol
, 0);
1231 seq_printf(m
, "%08lx %s", vma
->vm_start
, buffer
);
1234 seq_printf(m
, " file=");
1235 seq_path(m
, &file
->f_path
, "\n\t= ");
1236 } else if (vma
->vm_start
<= mm
->brk
&& vma
->vm_end
>= mm
->start_brk
) {
1237 seq_printf(m
, " heap");
1239 pid_t tid
= vm_is_stack(task
, vma
, is_pid
);
1242 * Thread stack in /proc/PID/task/TID/maps or
1243 * the main process stack.
1245 if (!is_pid
|| (vma
->vm_start
<= mm
->start_stack
&&
1246 vma
->vm_end
>= mm
->start_stack
))
1247 seq_printf(m
, " stack");
1249 seq_printf(m
, " stack:%d", tid
);
1253 if (is_vm_hugetlb_page(vma
))
1254 seq_printf(m
, " huge");
1256 walk_page_range(vma
->vm_start
, vma
->vm_end
, &walk
);
1262 seq_printf(m
, " anon=%lu", md
->anon
);
1265 seq_printf(m
, " dirty=%lu", md
->dirty
);
1267 if (md
->pages
!= md
->anon
&& md
->pages
!= md
->dirty
)
1268 seq_printf(m
, " mapped=%lu", md
->pages
);
1270 if (md
->mapcount_max
> 1)
1271 seq_printf(m
, " mapmax=%lu", md
->mapcount_max
);
1274 seq_printf(m
, " swapcache=%lu", md
->swapcache
);
1276 if (md
->active
< md
->pages
&& !is_vm_hugetlb_page(vma
))
1277 seq_printf(m
, " active=%lu", md
->active
);
1280 seq_printf(m
, " writeback=%lu", md
->writeback
);
1282 for_each_node_state(n
, N_HIGH_MEMORY
)
1284 seq_printf(m
, " N%d=%lu", n
, md
->node
[n
]);
1288 if (m
->count
< m
->size
)
1289 m
->version
= (vma
!= proc_priv
->tail_vma
) ? vma
->vm_start
: 0;
1293 static int show_pid_numa_map(struct seq_file
*m
, void *v
)
1295 return show_numa_map(m
, v
, 1);
1298 static int show_tid_numa_map(struct seq_file
*m
, void *v
)
1300 return show_numa_map(m
, v
, 0);
1303 static const struct seq_operations proc_pid_numa_maps_op
= {
1307 .show
= show_pid_numa_map
,
1310 static const struct seq_operations proc_tid_numa_maps_op
= {
1314 .show
= show_tid_numa_map
,
1317 static int numa_maps_open(struct inode
*inode
, struct file
*file
,
1318 const struct seq_operations
*ops
)
1320 struct numa_maps_private
*priv
;
1322 priv
= kzalloc(sizeof(*priv
), GFP_KERNEL
);
1324 priv
->proc_maps
.pid
= proc_pid(inode
);
1325 ret
= seq_open(file
, ops
);
1327 struct seq_file
*m
= file
->private_data
;
1336 static int pid_numa_maps_open(struct inode
*inode
, struct file
*file
)
1338 return numa_maps_open(inode
, file
, &proc_pid_numa_maps_op
);
1341 static int tid_numa_maps_open(struct inode
*inode
, struct file
*file
)
1343 return numa_maps_open(inode
, file
, &proc_tid_numa_maps_op
);
1346 const struct file_operations proc_pid_numa_maps_operations
= {
1347 .open
= pid_numa_maps_open
,
1349 .llseek
= seq_lseek
,
1350 .release
= seq_release_private
,
1353 const struct file_operations proc_tid_numa_maps_operations
= {
1354 .open
= tid_numa_maps_open
,
1356 .llseek
= seq_lseek
,
1357 .release
= seq_release_private
,
1359 #endif /* CONFIG_NUMA */