blob: 1030a716d155b4a19a91972b2ead074733b4add9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#include <linux/mm.h>
2#include <linux/hugetlb.h>
Dave Hansen22e057c2011-03-22 16:33:00 -07003#include <linux/huge_mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07004#include <linux/mount.h>
5#include <linux/seq_file.h>
Mauricio Line070ad42005-09-03 15:55:10 -07006#include <linux/highmem.h>
Kees Cook5096add2007-05-08 00:26:04 -07007#include <linux/ptrace.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09008#include <linux/slab.h>
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07009#include <linux/pagemap.h>
10#include <linux/mempolicy.h>
Dave Hansen22e057c2011-03-22 16:33:00 -070011#include <linux/rmap.h>
Matt Mackall85863e42008-02-04 22:29:04 -080012#include <linux/swap.h>
13#include <linux/swapops.h>
Mauricio Line070ad42005-09-03 15:55:10 -070014
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <asm/elf.h>
16#include <asm/uaccess.h>
Mauricio Line070ad42005-09-03 15:55:10 -070017#include <asm/tlbflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "internal.h"
19
Eric W. Biedermandf5f8312008-02-08 04:18:33 -080020void task_mem(struct seq_file *m, struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -070021{
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080022 unsigned long data, text, lib, swap;
Hugh Dickins365e9c872005-10-29 18:16:18 -070023 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
24
25 /*
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.
31 */
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
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;
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080042 swap = get_mm_counter(mm, MM_SWAPENTS);
Eric W. Biedermandf5f8312008-02-08 04:18:33 -080043 seq_printf(m,
Hugh Dickins365e9c872005-10-29 18:16:18 -070044 "VmPeak:\t%8lu kB\n"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 "VmSize:\t%8lu kB\n"
46 "VmLck:\t%8lu kB\n"
Christoph Lameterbc3e53f2011-10-31 17:07:30 -070047 "VmPin:\t%8lu kB\n"
Hugh Dickins365e9c872005-10-29 18:16:18 -070048 "VmHWM:\t%8lu kB\n"
Linus Torvalds1da177e2005-04-16 15:20:36 -070049 "VmRSS:\t%8lu kB\n"
50 "VmData:\t%8lu kB\n"
51 "VmStk:\t%8lu kB\n"
52 "VmExe:\t%8lu kB\n"
53 "VmLib:\t%8lu kB\n"
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080054 "VmPTE:\t%8lu kB\n"
55 "VmSwap:\t%8lu kB\n",
Hugh Dickins365e9c872005-10-29 18:16:18 -070056 hiwater_vm << (PAGE_SHIFT-10),
57 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 mm->locked_vm << (PAGE_SHIFT-10),
Christoph Lameterbc3e53f2011-10-31 17:07:30 -070059 mm->pinned_vm << (PAGE_SHIFT-10),
Hugh Dickins365e9c872005-10-29 18:16:18 -070060 hiwater_rss << (PAGE_SHIFT-10),
61 total_rss << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 data << (PAGE_SHIFT-10),
63 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080064 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
65 swap << (PAGE_SHIFT-10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070066}
67
68unsigned long task_vsize(struct mm_struct *mm)
69{
70 return PAGE_SIZE * mm->total_vm;
71}
72
Alexey Dobriyana2ade7b2011-01-12 17:00:32 -080073unsigned long task_statm(struct mm_struct *mm,
74 unsigned long *shared, unsigned long *text,
75 unsigned long *data, unsigned long *resident)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -080077 *shared = get_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
79 >> PAGE_SHIFT;
80 *data = mm->total_vm - mm->shared_vm;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -080081 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 return mm->total_vm;
83}
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085static void pad_len_spaces(struct seq_file *m, int len)
86{
87 len = 25 + sizeof(void*) * 6 - len;
88 if (len < 1)
89 len = 1;
90 seq_printf(m, "%*c", len, ' ');
91}
92
Matt Mackalla6198792008-02-04 22:29:03 -080093static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
Mauricio Line070ad42005-09-03 15:55:10 -070094{
Matt Mackalla6198792008-02-04 22:29:03 -080095 if (vma && vma != priv->tail_vma) {
96 struct mm_struct *mm = vma->vm_mm;
97 up_read(&mm->mmap_sem);
98 mmput(mm);
99 }
100}
101
102static void *m_start(struct seq_file *m, loff_t *pos)
103{
104 struct proc_maps_private *priv = m->private;
105 unsigned long last_addr = m->version;
106 struct mm_struct *mm;
107 struct vm_area_struct *vma, *tail_vma = NULL;
108 loff_t l = *pos;
109
110 /* Clear the per syscall fields in priv */
111 priv->task = NULL;
112 priv->tail_vma = NULL;
113
114 /*
115 * We remember last_addr rather than next_addr to hit with
116 * mmap_cache most of the time. We have zero last_addr at
117 * the beginning and also after lseek. We will have -1 last_addr
118 * after the end of the vmas.
119 */
120
121 if (last_addr == -1UL)
122 return NULL;
123
124 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
125 if (!priv->task)
Al Viroec6fd8a2011-02-15 22:22:54 -0500126 return ERR_PTR(-ESRCH);
Matt Mackalla6198792008-02-04 22:29:03 -0800127
128 mm = mm_for_maps(priv->task);
Al Viroec6fd8a2011-02-15 22:22:54 -0500129 if (!mm || IS_ERR(mm))
130 return mm;
Oleg Nesterov00f89d22009-07-10 03:27:38 +0200131 down_read(&mm->mmap_sem);
Matt Mackalla6198792008-02-04 22:29:03 -0800132
Stephen Wilson31db58b2011-03-13 15:49:15 -0400133 tail_vma = get_gate_vma(priv->task->mm);
Matt Mackalla6198792008-02-04 22:29:03 -0800134 priv->tail_vma = tail_vma;
135
136 /* Start with last addr hint */
137 vma = find_vma(mm, last_addr);
138 if (last_addr && vma) {
139 vma = vma->vm_next;
140 goto out;
141 }
142
143 /*
144 * Check the vma index is within the range and do
145 * sequential scan until m_index.
146 */
147 vma = NULL;
148 if ((unsigned long)l < mm->map_count) {
149 vma = mm->mmap;
150 while (l-- && vma)
151 vma = vma->vm_next;
152 goto out;
153 }
154
155 if (l != mm->map_count)
156 tail_vma = NULL; /* After gate vma */
157
158out:
159 if (vma)
160 return vma;
161
162 /* End of vmas has been reached */
163 m->version = (tail_vma != NULL)? 0: -1UL;
164 up_read(&mm->mmap_sem);
165 mmput(mm);
166 return tail_vma;
167}
168
169static void *m_next(struct seq_file *m, void *v, loff_t *pos)
170{
171 struct proc_maps_private *priv = m->private;
172 struct vm_area_struct *vma = v;
173 struct vm_area_struct *tail_vma = priv->tail_vma;
174
175 (*pos)++;
176 if (vma && (vma != tail_vma) && vma->vm_next)
177 return vma->vm_next;
178 vma_stop(priv, vma);
179 return (vma != tail_vma)? tail_vma: NULL;
180}
181
182static void m_stop(struct seq_file *m, void *v)
183{
184 struct proc_maps_private *priv = m->private;
185 struct vm_area_struct *vma = v;
186
Linus Torvalds76597cd2011-03-27 19:09:29 -0700187 if (!IS_ERR(vma))
188 vma_stop(priv, vma);
Matt Mackalla6198792008-02-04 22:29:03 -0800189 if (priv->task)
190 put_task_struct(priv->task);
191}
192
193static int do_maps_open(struct inode *inode, struct file *file,
Jan Engelhardt03a44822008-02-08 04:21:19 -0800194 const struct seq_operations *ops)
Matt Mackalla6198792008-02-04 22:29:03 -0800195{
196 struct proc_maps_private *priv;
197 int ret = -ENOMEM;
198 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
199 if (priv) {
200 priv->pid = proc_pid(inode);
201 ret = seq_open(file, ops);
202 if (!ret) {
203 struct seq_file *m = file->private_data;
204 m->private = priv;
205 } else {
206 kfree(priv);
207 }
208 }
209 return ret;
210}
Mauricio Line070ad42005-09-03 15:55:10 -0700211
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700212static void
213show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
Mauricio Line070ad42005-09-03 15:55:10 -0700215 struct mm_struct *mm = vma->vm_mm;
216 struct file *file = vma->vm_file;
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700217 struct proc_maps_private *priv = m->private;
218 struct task_struct *task = priv->task;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900219 vm_flags_t flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 unsigned long ino = 0;
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700221 unsigned long long pgoff = 0;
Mikulas Patockaa09a79f2011-05-09 13:01:09 +0200222 unsigned long start, end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 dev_t dev = 0;
224 int len;
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700225 const char *name = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226
227 if (file) {
Josef "Jeff" Sipek2fddfee2006-12-08 02:36:36 -0800228 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 dev = inode->i_sb->s_dev;
230 ino = inode->i_ino;
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700231 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 }
233
Linus Torvaldsd7824372010-08-15 11:35:52 -0700234 /* We don't show the stack guard page in /proc/maps */
235 start = vma->vm_start;
Mikulas Patockaa09a79f2011-05-09 13:01:09 +0200236 if (stack_guard_page_start(vma, start))
237 start += PAGE_SIZE;
238 end = vma->vm_end;
239 if (stack_guard_page_end(vma, end))
240 end -= PAGE_SIZE;
Linus Torvaldsd7824372010-08-15 11:35:52 -0700241
Clement Calmels1804dc62008-08-20 14:09:00 -0700242 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
Linus Torvaldsd7824372010-08-15 11:35:52 -0700243 start,
Mikulas Patockaa09a79f2011-05-09 13:01:09 +0200244 end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 flags & VM_READ ? 'r' : '-',
246 flags & VM_WRITE ? 'w' : '-',
247 flags & VM_EXEC ? 'x' : '-',
248 flags & VM_MAYSHARE ? 's' : 'p',
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700249 pgoff,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 MAJOR(dev), MINOR(dev), ino, &len);
251
252 /*
253 * Print the dentry name for named mappings, and a
254 * special [heap] marker for the heap:
255 */
Mauricio Line070ad42005-09-03 15:55:10 -0700256 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 pad_len_spaces(m, len);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800258 seq_path(m, &file->f_path, "\n");
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700259 goto done;
260 }
261
262 name = arch_vma_name(vma);
263 if (!name) {
264 pid_t tid;
265
266 if (!mm) {
267 name = "[vdso]";
268 goto done;
269 }
270
271 if (vma->vm_start <= mm->brk &&
272 vma->vm_end >= mm->start_brk) {
273 name = "[heap]";
274 goto done;
275 }
276
277 tid = vm_is_stack(task, vma, is_pid);
278
279 if (tid != 0) {
280 /*
281 * Thread stack in /proc/PID/task/TID/maps or
282 * the main process stack.
283 */
284 if (!is_pid || (vma->vm_start <= mm->start_stack &&
285 vma->vm_end >= mm->start_stack)) {
286 name = "[stack]";
Ingo Molnare6e54942006-06-27 02:53:50 -0700287 } else {
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700288 /* Thread stack in /proc/PID/maps */
289 pad_len_spaces(m, len);
290 seq_printf(m, "[stack:%d]", tid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700292 }
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700293 }
294
295done:
296 if (name) {
297 pad_len_spaces(m, len);
298 seq_puts(m, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 }
300 seq_putc(m, '\n');
Joe Korty7c88db02008-10-16 15:27:09 +0400301}
302
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700303static int show_map(struct seq_file *m, void *v, int is_pid)
Joe Korty7c88db02008-10-16 15:27:09 +0400304{
305 struct vm_area_struct *vma = v;
306 struct proc_maps_private *priv = m->private;
307 struct task_struct *task = priv->task;
308
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700309 show_map_vma(m, vma, is_pid);
Mauricio Line070ad42005-09-03 15:55:10 -0700310
Mauricio Line070ad42005-09-03 15:55:10 -0700311 if (m->count < m->size) /* vma is copied successfully */
Stephen Wilson31db58b2011-03-13 15:49:15 -0400312 m->version = (vma != get_gate_vma(task->mm))
313 ? vma->vm_start : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 return 0;
315}
316
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700317static int show_pid_map(struct seq_file *m, void *v)
318{
319 return show_map(m, v, 1);
320}
321
322static int show_tid_map(struct seq_file *m, void *v)
323{
324 return show_map(m, v, 0);
325}
326
Jan Engelhardt03a44822008-02-08 04:21:19 -0800327static const struct seq_operations proc_pid_maps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800328 .start = m_start,
329 .next = m_next,
330 .stop = m_stop,
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700331 .show = show_pid_map
Matt Mackalla6198792008-02-04 22:29:03 -0800332};
333
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700334static const struct seq_operations proc_tid_maps_op = {
335 .start = m_start,
336 .next = m_next,
337 .stop = m_stop,
338 .show = show_tid_map
339};
340
341static int pid_maps_open(struct inode *inode, struct file *file)
Matt Mackalla6198792008-02-04 22:29:03 -0800342{
343 return do_maps_open(inode, file, &proc_pid_maps_op);
344}
345
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700346static int tid_maps_open(struct inode *inode, struct file *file)
347{
348 return do_maps_open(inode, file, &proc_tid_maps_op);
349}
350
351const struct file_operations proc_pid_maps_operations = {
352 .open = pid_maps_open,
353 .read = seq_read,
354 .llseek = seq_lseek,
355 .release = seq_release_private,
356};
357
358const struct file_operations proc_tid_maps_operations = {
359 .open = tid_maps_open,
Matt Mackalla6198792008-02-04 22:29:03 -0800360 .read = seq_read,
361 .llseek = seq_lseek,
362 .release = seq_release_private,
363};
364
365/*
366 * Proportional Set Size(PSS): my share of RSS.
367 *
368 * PSS of a process is the count of pages it has in memory, where each
369 * page is divided by the number of processes sharing it. So if a
370 * process has 1000 pages all to itself, and 1000 shared with one other
371 * process, its PSS will be 1500.
372 *
373 * To keep (accumulated) division errors low, we adopt a 64bit
374 * fixed-point pss counter to minimize division errors. So (pss >>
375 * PSS_SHIFT) would be the real byte count.
376 *
377 * A shift of 12 before division means (assuming 4K page size):
378 * - 1M 3-user-pages add up to 8KB errors;
379 * - supports mapcount up to 2^24, or 16M;
380 * - supports PSS up to 2^52 bytes, or 4PB.
381 */
382#define PSS_SHIFT 12
383
Matt Mackall1e883282008-02-04 22:29:07 -0800384#ifdef CONFIG_PROC_PAGE_MONITOR
Peter Zijlstra214e4712008-04-28 02:12:55 -0700385struct mem_size_stats {
Matt Mackalla6198792008-02-04 22:29:03 -0800386 struct vm_area_struct *vma;
387 unsigned long resident;
388 unsigned long shared_clean;
389 unsigned long shared_dirty;
390 unsigned long private_clean;
391 unsigned long private_dirty;
392 unsigned long referenced;
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700393 unsigned long anonymous;
Dave Hansen4031a212011-03-22 16:33:01 -0700394 unsigned long anonymous_thp;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700395 unsigned long swap;
Matt Mackalla6198792008-02-04 22:29:03 -0800396 u64 pss;
397};
398
Dave Hansenae11c4d2011-03-22 16:32:58 -0700399
400static void smaps_pte_entry(pte_t ptent, unsigned long addr,
Dave Hansen3c9acc72011-03-22 16:32:59 -0700401 unsigned long ptent_size, struct mm_walk *walk)
Dave Hansenae11c4d2011-03-22 16:32:58 -0700402{
403 struct mem_size_stats *mss = walk->private;
404 struct vm_area_struct *vma = mss->vma;
405 struct page *page;
406 int mapcount;
407
408 if (is_swap_pte(ptent)) {
Dave Hansen3c9acc72011-03-22 16:32:59 -0700409 mss->swap += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700410 return;
411 }
412
413 if (!pte_present(ptent))
414 return;
415
416 page = vm_normal_page(vma, addr, ptent);
417 if (!page)
418 return;
419
420 if (PageAnon(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700421 mss->anonymous += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700422
Dave Hansen3c9acc72011-03-22 16:32:59 -0700423 mss->resident += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700424 /* Accumulate the size in pages that have been accessed. */
425 if (pte_young(ptent) || PageReferenced(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700426 mss->referenced += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700427 mapcount = page_mapcount(page);
428 if (mapcount >= 2) {
429 if (pte_dirty(ptent) || PageDirty(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700430 mss->shared_dirty += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700431 else
Dave Hansen3c9acc72011-03-22 16:32:59 -0700432 mss->shared_clean += ptent_size;
433 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700434 } else {
435 if (pte_dirty(ptent) || PageDirty(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700436 mss->private_dirty += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700437 else
Dave Hansen3c9acc72011-03-22 16:32:59 -0700438 mss->private_clean += ptent_size;
439 mss->pss += (ptent_size << PSS_SHIFT);
Dave Hansenae11c4d2011-03-22 16:32:58 -0700440 }
441}
442
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800443static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700444 struct mm_walk *walk)
Mauricio Line070ad42005-09-03 15:55:10 -0700445{
Dave Hansen21650092008-06-12 15:21:47 -0700446 struct mem_size_stats *mss = walk->private;
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800447 struct vm_area_struct *vma = mss->vma;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700448 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700449 spinlock_t *ptl;
Mauricio Line070ad42005-09-03 15:55:10 -0700450
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700451 if (pmd_trans_huge_lock(pmd, vma) == 1) {
452 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
Dave Hansen22e057c2011-03-22 16:33:00 -0700453 spin_unlock(&walk->mm->page_table_lock);
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700454 mss->anonymous_thp += HPAGE_PMD_SIZE;
455 return 0;
Dave Hansen22e057c2011-03-22 16:33:00 -0700456 }
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700457
458 if (pmd_trans_unstable(pmd))
459 return 0;
Dave Hansen22e057c2011-03-22 16:33:00 -0700460 /*
461 * The mmap_sem held all the way back in m_start() is what
462 * keeps khugepaged out of here and from collapsing things
463 * in here.
464 */
Hugh Dickins705e87c2005-10-29 18:16:27 -0700465 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Dave Hansenae11c4d2011-03-22 16:32:58 -0700466 for (; addr != end; pte++, addr += PAGE_SIZE)
Dave Hansen3c9acc72011-03-22 16:32:59 -0700467 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700468 pte_unmap_unlock(pte - 1, ptl);
469 cond_resched();
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800470 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700471}
472
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700473static int show_smap(struct seq_file *m, void *v, int is_pid)
Mauricio Line070ad42005-09-03 15:55:10 -0700474{
Joe Korty7c88db02008-10-16 15:27:09 +0400475 struct proc_maps_private *priv = m->private;
476 struct task_struct *task = priv->task;
Mauricio Line070ad42005-09-03 15:55:10 -0700477 struct vm_area_struct *vma = v;
Mauricio Line070ad42005-09-03 15:55:10 -0700478 struct mem_size_stats mss;
Dave Hansen21650092008-06-12 15:21:47 -0700479 struct mm_walk smaps_walk = {
480 .pmd_entry = smaps_pte_range,
481 .mm = vma->vm_mm,
482 .private = &mss,
483 };
Mauricio Line070ad42005-09-03 15:55:10 -0700484
485 memset(&mss, 0, sizeof mss);
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800486 mss.vma = vma;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900487 /* mmap_sem is held in m_start */
Nick Piggin5ddfae12006-03-06 15:42:57 -0800488 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700489 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
Matt Mackall4752c362008-02-04 22:29:02 -0800490
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700491 show_map_vma(m, vma, is_pid);
Matt Mackall4752c362008-02-04 22:29:02 -0800492
493 seq_printf(m,
494 "Size: %8lu kB\n"
495 "Rss: %8lu kB\n"
496 "Pss: %8lu kB\n"
497 "Shared_Clean: %8lu kB\n"
498 "Shared_Dirty: %8lu kB\n"
499 "Private_Clean: %8lu kB\n"
500 "Private_Dirty: %8lu kB\n"
Peter Zijlstra214e4712008-04-28 02:12:55 -0700501 "Referenced: %8lu kB\n"
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700502 "Anonymous: %8lu kB\n"
Dave Hansen4031a212011-03-22 16:33:01 -0700503 "AnonHugePages: %8lu kB\n"
Mel Gorman08fba692009-01-06 14:38:53 -0800504 "Swap: %8lu kB\n"
Mel Gorman33402892009-01-06 14:38:54 -0800505 "KernelPageSize: %8lu kB\n"
Nikanth Karthikesan2d905082011-01-13 15:45:53 -0800506 "MMUPageSize: %8lu kB\n"
507 "Locked: %8lu kB\n",
Matt Mackall4752c362008-02-04 22:29:02 -0800508 (vma->vm_end - vma->vm_start) >> 10,
509 mss.resident >> 10,
510 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
511 mss.shared_clean >> 10,
512 mss.shared_dirty >> 10,
513 mss.private_clean >> 10,
514 mss.private_dirty >> 10,
Peter Zijlstra214e4712008-04-28 02:12:55 -0700515 mss.referenced >> 10,
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700516 mss.anonymous >> 10,
Dave Hansen4031a212011-03-22 16:33:01 -0700517 mss.anonymous_thp >> 10,
Mel Gorman08fba692009-01-06 14:38:53 -0800518 mss.swap >> 10,
Mel Gorman33402892009-01-06 14:38:54 -0800519 vma_kernel_pagesize(vma) >> 10,
Nikanth Karthikesan2d905082011-01-13 15:45:53 -0800520 vma_mmu_pagesize(vma) >> 10,
521 (vma->vm_flags & VM_LOCKED) ?
522 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
Matt Mackall4752c362008-02-04 22:29:02 -0800523
Joe Korty7c88db02008-10-16 15:27:09 +0400524 if (m->count < m->size) /* vma is copied successfully */
Stephen Wilson31db58b2011-03-13 15:49:15 -0400525 m->version = (vma != get_gate_vma(task->mm))
526 ? vma->vm_start : 0;
Joe Korty7c88db02008-10-16 15:27:09 +0400527 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700528}
529
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700530static int show_pid_smap(struct seq_file *m, void *v)
531{
532 return show_smap(m, v, 1);
533}
534
535static int show_tid_smap(struct seq_file *m, void *v)
536{
537 return show_smap(m, v, 0);
538}
539
Jan Engelhardt03a44822008-02-08 04:21:19 -0800540static const struct seq_operations proc_pid_smaps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800541 .start = m_start,
542 .next = m_next,
543 .stop = m_stop,
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700544 .show = show_pid_smap
Matt Mackalla6198792008-02-04 22:29:03 -0800545};
546
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700547static const struct seq_operations proc_tid_smaps_op = {
548 .start = m_start,
549 .next = m_next,
550 .stop = m_stop,
551 .show = show_tid_smap
552};
553
554static int pid_smaps_open(struct inode *inode, struct file *file)
Matt Mackalla6198792008-02-04 22:29:03 -0800555{
556 return do_maps_open(inode, file, &proc_pid_smaps_op);
557}
558
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -0700559static int tid_smaps_open(struct inode *inode, struct file *file)
560{
561 return do_maps_open(inode, file, &proc_tid_smaps_op);
562}
563
564const struct file_operations proc_pid_smaps_operations = {
565 .open = pid_smaps_open,
566 .read = seq_read,
567 .llseek = seq_lseek,
568 .release = seq_release_private,
569};
570
571const struct file_operations proc_tid_smaps_operations = {
572 .open = tid_smaps_open,
Matt Mackalla6198792008-02-04 22:29:03 -0800573 .read = seq_read,
574 .llseek = seq_lseek,
575 .release = seq_release_private,
576};
577
578static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
Dave Hansen21650092008-06-12 15:21:47 -0700579 unsigned long end, struct mm_walk *walk)
Matt Mackalla6198792008-02-04 22:29:03 -0800580{
Dave Hansen21650092008-06-12 15:21:47 -0700581 struct vm_area_struct *vma = walk->private;
Matt Mackalla6198792008-02-04 22:29:03 -0800582 pte_t *pte, ptent;
583 spinlock_t *ptl;
584 struct page *page;
585
Dave Hansen03319322011-03-22 16:32:56 -0700586 split_huge_page_pmd(walk->mm, pmd);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700587 if (pmd_trans_unstable(pmd))
588 return 0;
Dave Hansen03319322011-03-22 16:32:56 -0700589
Matt Mackalla6198792008-02-04 22:29:03 -0800590 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
591 for (; addr != end; pte++, addr += PAGE_SIZE) {
592 ptent = *pte;
593 if (!pte_present(ptent))
594 continue;
595
596 page = vm_normal_page(vma, addr, ptent);
597 if (!page)
598 continue;
599
600 /* Clear accessed and referenced bits. */
601 ptep_test_and_clear_young(vma, addr, pte);
602 ClearPageReferenced(page);
603 }
604 pte_unmap_unlock(pte - 1, ptl);
605 cond_resched();
606 return 0;
607}
608
Moussa A. Ba398499d2009-09-21 17:02:29 -0700609#define CLEAR_REFS_ALL 1
610#define CLEAR_REFS_ANON 2
611#define CLEAR_REFS_MAPPED 3
612
Matt Mackallf248dcb2008-02-04 22:29:03 -0800613static ssize_t clear_refs_write(struct file *file, const char __user *buf,
614 size_t count, loff_t *ppos)
David Rientjesb813e932007-05-06 14:49:24 -0700615{
Matt Mackallf248dcb2008-02-04 22:29:03 -0800616 struct task_struct *task;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700617 char buffer[PROC_NUMBUF];
Matt Mackallf248dcb2008-02-04 22:29:03 -0800618 struct mm_struct *mm;
David Rientjesb813e932007-05-06 14:49:24 -0700619 struct vm_area_struct *vma;
Alexey Dobriyan0a8cb8e2011-05-26 16:25:50 -0700620 int type;
621 int rv;
David Rientjesb813e932007-05-06 14:49:24 -0700622
Matt Mackallf248dcb2008-02-04 22:29:03 -0800623 memset(buffer, 0, sizeof(buffer));
624 if (count > sizeof(buffer) - 1)
625 count = sizeof(buffer) - 1;
626 if (copy_from_user(buffer, buf, count))
627 return -EFAULT;
Alexey Dobriyan0a8cb8e2011-05-26 16:25:50 -0700628 rv = kstrtoint(strstrip(buffer), 10, &type);
629 if (rv < 0)
630 return rv;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700631 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
Matt Mackallf248dcb2008-02-04 22:29:03 -0800632 return -EINVAL;
Matt Mackallf248dcb2008-02-04 22:29:03 -0800633 task = get_proc_task(file->f_path.dentry->d_inode);
634 if (!task)
635 return -ESRCH;
636 mm = get_task_mm(task);
637 if (mm) {
Andrew Morton20cbc972008-07-05 12:29:05 -0700638 struct mm_walk clear_refs_walk = {
639 .pmd_entry = clear_refs_pte_range,
640 .mm = mm,
641 };
Matt Mackallf248dcb2008-02-04 22:29:03 -0800642 down_read(&mm->mmap_sem);
Dave Hansen21650092008-06-12 15:21:47 -0700643 for (vma = mm->mmap; vma; vma = vma->vm_next) {
644 clear_refs_walk.private = vma;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700645 if (is_vm_hugetlb_page(vma))
646 continue;
647 /*
648 * Writing 1 to /proc/pid/clear_refs affects all pages.
649 *
650 * Writing 2 to /proc/pid/clear_refs only affects
651 * Anonymous pages.
652 *
653 * Writing 3 to /proc/pid/clear_refs only affects file
654 * mapped pages.
655 */
656 if (type == CLEAR_REFS_ANON && vma->vm_file)
657 continue;
658 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
659 continue;
660 walk_page_range(vma->vm_start, vma->vm_end,
661 &clear_refs_walk);
Dave Hansen21650092008-06-12 15:21:47 -0700662 }
Matt Mackallf248dcb2008-02-04 22:29:03 -0800663 flush_tlb_mm(mm);
664 up_read(&mm->mmap_sem);
665 mmput(mm);
666 }
667 put_task_struct(task);
Vincent Lifb92a4b2009-09-22 16:45:36 -0700668
669 return count;
David Rientjesb813e932007-05-06 14:49:24 -0700670}
671
Matt Mackallf248dcb2008-02-04 22:29:03 -0800672const struct file_operations proc_clear_refs_operations = {
673 .write = clear_refs_write,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200674 .llseek = noop_llseek,
Matt Mackallf248dcb2008-02-04 22:29:03 -0800675};
676
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700677typedef struct {
678 u64 pme;
679} pagemap_entry_t;
680
Matt Mackall85863e42008-02-04 22:29:04 -0800681struct pagemapread {
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900682 int pos, len;
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700683 pagemap_entry_t *buffer;
Matt Mackall85863e42008-02-04 22:29:04 -0800684};
685
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700686#define PAGEMAP_WALK_SIZE (PMD_SIZE)
687#define PAGEMAP_WALK_MASK (PMD_MASK)
688
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500689#define PM_ENTRY_BYTES sizeof(u64)
690#define PM_STATUS_BITS 3
691#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
692#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
693#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
694#define PM_PSHIFT_BITS 6
695#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
696#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
697#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
698#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
699#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
700
701#define PM_PRESENT PM_STATUS(4LL)
702#define PM_SWAP PM_STATUS(2LL)
703#define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
Matt Mackall85863e42008-02-04 22:29:04 -0800704#define PM_END_OF_BUFFER 1
705
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700706static inline pagemap_entry_t make_pme(u64 val)
707{
708 return (pagemap_entry_t) { .pme = val };
709}
710
711static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
Matt Mackall85863e42008-02-04 22:29:04 -0800712 struct pagemapread *pm)
713{
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700714 pm->buffer[pm->pos++] = *pme;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900715 if (pm->pos >= pm->len)
Thomas Tuttleaae86792008-06-05 22:46:31 -0700716 return PM_END_OF_BUFFER;
Matt Mackall85863e42008-02-04 22:29:04 -0800717 return 0;
718}
719
720static int pagemap_pte_hole(unsigned long start, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700721 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800722{
Dave Hansen21650092008-06-12 15:21:47 -0700723 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800724 unsigned long addr;
725 int err = 0;
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700726 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
727
Matt Mackall85863e42008-02-04 22:29:04 -0800728 for (addr = start; addr < end; addr += PAGE_SIZE) {
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700729 err = add_to_pagemap(addr, &pme, pm);
Matt Mackall85863e42008-02-04 22:29:04 -0800730 if (err)
731 break;
732 }
733 return err;
734}
735
Adrian Bunk9d02dbc2008-04-28 02:12:11 -0700736static u64 swap_pte_to_pagemap_entry(pte_t pte)
Matt Mackall85863e42008-02-04 22:29:04 -0800737{
738 swp_entry_t e = pte_to_swp_entry(pte);
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500739 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
Matt Mackall85863e42008-02-04 22:29:04 -0800740}
741
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700742static void pte_to_pagemap_entry(pagemap_entry_t *pme, pte_t pte)
Dave Hansenbcf80392008-06-12 15:21:48 -0700743{
Dave Hansenbcf80392008-06-12 15:21:48 -0700744 if (is_swap_pte(pte))
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700745 *pme = make_pme(PM_PFRAME(swap_pte_to_pagemap_entry(pte))
746 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP);
Dave Hansenbcf80392008-06-12 15:21:48 -0700747 else if (pte_present(pte))
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700748 *pme = make_pme(PM_PFRAME(pte_pfn(pte))
749 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
Konstantin Khlebnikov16fbdce2012-05-10 13:01:43 -0700750 else
751 *pme = make_pme(PM_NOT_PRESENT);
Dave Hansenbcf80392008-06-12 15:21:48 -0700752}
753
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700754#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700755static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
756 pmd_t pmd, int offset)
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700757{
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700758 /*
759 * Currently pmd for thp is always present because thp can not be
760 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
761 * This if-check is just to prepare for future implementation.
762 */
763 if (pmd_present(pmd))
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700764 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
765 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
Konstantin Khlebnikov16fbdce2012-05-10 13:01:43 -0700766 else
767 *pme = make_pme(PM_NOT_PRESENT);
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700768}
769#else
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700770static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
771 pmd_t pmd, int offset)
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700772{
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700773}
774#endif
775
Matt Mackall85863e42008-02-04 22:29:04 -0800776static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700777 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800778{
Dave Hansenbcf80392008-06-12 15:21:48 -0700779 struct vm_area_struct *vma;
Dave Hansen21650092008-06-12 15:21:47 -0700780 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800781 pte_t *pte;
782 int err = 0;
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700783 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
Matt Mackall85863e42008-02-04 22:29:04 -0800784
Dave Hansenbcf80392008-06-12 15:21:48 -0700785 /* find the first VMA at or above 'addr' */
786 vma = find_vma(walk->mm, addr);
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700787 if (pmd_trans_huge_lock(pmd, vma) == 1) {
788 for (; addr != end; addr += PAGE_SIZE) {
789 unsigned long offset;
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700790
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700791 offset = (addr & ~PAGEMAP_WALK_MASK) >>
792 PAGE_SHIFT;
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700793 thp_pmd_to_pagemap_entry(&pme, *pmd, offset);
794 err = add_to_pagemap(addr, &pme, pm);
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700795 if (err)
796 break;
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700797 }
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700798 spin_unlock(&walk->mm->page_table_lock);
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700799 return err;
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700800 }
801
Andrea Arcangeli45f83ce2012-03-28 14:42:40 -0700802 if (pmd_trans_unstable(pmd))
803 return 0;
Matt Mackall85863e42008-02-04 22:29:04 -0800804 for (; addr != end; addr += PAGE_SIZE) {
Dave Hansenbcf80392008-06-12 15:21:48 -0700805
806 /* check to see if we've left 'vma' behind
807 * and need a new, higher one */
Konstantin Khlebnikov16fbdce2012-05-10 13:01:43 -0700808 if (vma && (addr >= vma->vm_end)) {
Dave Hansenbcf80392008-06-12 15:21:48 -0700809 vma = find_vma(walk->mm, addr);
Konstantin Khlebnikov16fbdce2012-05-10 13:01:43 -0700810 pme = make_pme(PM_NOT_PRESENT);
811 }
Dave Hansenbcf80392008-06-12 15:21:48 -0700812
813 /* check that 'vma' actually covers this address,
814 * and that it isn't a huge page vma */
815 if (vma && (vma->vm_start <= addr) &&
816 !is_vm_hugetlb_page(vma)) {
817 pte = pte_offset_map(pmd, addr);
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700818 pte_to_pagemap_entry(&pme, *pte);
Dave Hansenbcf80392008-06-12 15:21:48 -0700819 /* unmap before userspace copy */
820 pte_unmap(pte);
821 }
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700822 err = add_to_pagemap(addr, &pme, pm);
Matt Mackall85863e42008-02-04 22:29:04 -0800823 if (err)
824 return err;
825 }
826
827 cond_resched();
828
829 return err;
830}
831
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700832#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700833static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme,
834 pte_t pte, int offset)
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800835{
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800836 if (pte_present(pte))
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700837 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset)
838 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
Konstantin Khlebnikov16fbdce2012-05-10 13:01:43 -0700839 else
840 *pme = make_pme(PM_NOT_PRESENT);
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800841}
842
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700843/* This function walks within one hugetlb entry in the single call */
844static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
845 unsigned long addr, unsigned long end,
846 struct mm_walk *walk)
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800847{
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800848 struct pagemapread *pm = walk->private;
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800849 int err = 0;
Konstantin Khlebnikov16fbdce2012-05-10 13:01:43 -0700850 pagemap_entry_t pme;
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800851
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800852 for (; addr != end; addr += PAGE_SIZE) {
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700853 int offset = (addr & ~hmask) >> PAGE_SHIFT;
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700854 huge_pte_to_pagemap_entry(&pme, *pte, offset);
855 err = add_to_pagemap(addr, &pme, pm);
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800856 if (err)
857 return err;
858 }
859
860 cond_resched();
861
862 return err;
863}
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700864#endif /* HUGETLB_PAGE */
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800865
Matt Mackall85863e42008-02-04 22:29:04 -0800866/*
867 * /proc/pid/pagemap - an array mapping virtual pages to pfns
868 *
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500869 * For each page in the address space, this file contains one 64-bit entry
870 * consisting of the following:
871 *
872 * Bits 0-55 page frame number (PFN) if present
873 * Bits 0-4 swap type if swapped
874 * Bits 5-55 swap offset if swapped
875 * Bits 55-60 page shift (page size = 1<<page shift)
876 * Bit 61 reserved for future use
877 * Bit 62 page swapped
878 * Bit 63 page present
879 *
880 * If the page is not present but in swap, then the PFN contains an
881 * encoding of the swap file number and the page's offset into the
882 * swap. Unmapped pages return a null PFN. This allows determining
Matt Mackall85863e42008-02-04 22:29:04 -0800883 * precisely which pages are mapped (or in swap) and comparing mapped
884 * pages between processes.
885 *
886 * Efficient users of this interface will use /proc/pid/maps to
887 * determine which areas of memory are actually mapped and llseek to
888 * skip over unmapped regions.
889 */
890static ssize_t pagemap_read(struct file *file, char __user *buf,
891 size_t count, loff_t *ppos)
892{
893 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
Matt Mackall85863e42008-02-04 22:29:04 -0800894 struct mm_struct *mm;
895 struct pagemapread pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800896 int ret = -ESRCH;
Alexey Dobriyanee1e6ab2008-07-21 14:21:36 -0700897 struct mm_walk pagemap_walk = {};
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700898 unsigned long src;
899 unsigned long svpfn;
900 unsigned long start_vaddr;
901 unsigned long end_vaddr;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900902 int copied = 0;
Matt Mackall85863e42008-02-04 22:29:04 -0800903
904 if (!task)
905 goto out;
906
Matt Mackall85863e42008-02-04 22:29:04 -0800907 ret = -EINVAL;
908 /* file position must be aligned */
Thomas Tuttleaae86792008-06-05 22:46:31 -0700909 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700910 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800911
912 ret = 0;
Vitaly Mayatskikh08161782009-04-30 15:08:18 -0700913 if (!count)
914 goto out_task;
915
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900916 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
917 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700918 ret = -ENOMEM;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900919 if (!pm.buffer)
KOSAKI Motohiro98bc93e2011-05-26 16:25:53 -0700920 goto out_task;
921
922 mm = mm_for_maps(task);
923 ret = PTR_ERR(mm);
924 if (!mm || IS_ERR(mm))
925 goto out_free;
Matt Mackall85863e42008-02-04 22:29:04 -0800926
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700927 pagemap_walk.pmd_entry = pagemap_pte_range;
928 pagemap_walk.pte_hole = pagemap_pte_hole;
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700929#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800930 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700931#endif
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700932 pagemap_walk.mm = mm;
933 pagemap_walk.private = &pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800934
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700935 src = *ppos;
936 svpfn = src / PM_ENTRY_BYTES;
937 start_vaddr = svpfn << PAGE_SHIFT;
938 end_vaddr = TASK_SIZE_OF(task);
Matt Mackall85863e42008-02-04 22:29:04 -0800939
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700940 /* watch out for wraparound */
941 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
942 start_vaddr = end_vaddr;
943
944 /*
945 * The odds are that this will stop walking way
946 * before end_vaddr, because the length of the
947 * user buffer is tracked in "pm", and the walk
948 * will stop when we hit the end of the buffer.
949 */
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900950 ret = 0;
951 while (count && (start_vaddr < end_vaddr)) {
952 int len;
953 unsigned long end;
Matt Mackall85863e42008-02-04 22:29:04 -0800954
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900955 pm.pos = 0;
Naoya Horiguchiea251c12010-11-24 12:57:13 -0800956 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900957 /* overflow ? */
958 if (end < start_vaddr || end > end_vaddr)
959 end = end_vaddr;
960 down_read(&mm->mmap_sem);
961 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
962 up_read(&mm->mmap_sem);
963 start_vaddr = end;
964
965 len = min(count, PM_ENTRY_BYTES * pm.pos);
Dan Carpenter309361e2010-04-06 13:45:39 +0300966 if (copy_to_user(buf, pm.buffer, len)) {
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900967 ret = -EFAULT;
KOSAKI Motohiro98bc93e2011-05-26 16:25:53 -0700968 goto out_mm;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900969 }
970 copied += len;
971 buf += len;
972 count -= len;
Matt Mackall85863e42008-02-04 22:29:04 -0800973 }
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900974 *ppos += copied;
975 if (!ret || ret == PM_END_OF_BUFFER)
976 ret = copied;
977
Marcelo Tosattifb393802008-03-13 12:32:35 -0700978out_mm:
979 mmput(mm);
KOSAKI Motohiro98bc93e2011-05-26 16:25:53 -0700980out_free:
981 kfree(pm.buffer);
Matt Mackall85863e42008-02-04 22:29:04 -0800982out_task:
983 put_task_struct(task);
984out:
985 return ret;
986}
987
988const struct file_operations proc_pagemap_operations = {
989 .llseek = mem_lseek, /* borrow this */
990 .read = pagemap_read,
991};
Matt Mackall1e883282008-02-04 22:29:07 -0800992#endif /* CONFIG_PROC_PAGE_MONITOR */
Matt Mackall85863e42008-02-04 22:29:04 -0800993
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700994#ifdef CONFIG_NUMA
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700995
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700996struct numa_maps {
997 struct vm_area_struct *vma;
998 unsigned long pages;
999 unsigned long anon;
1000 unsigned long active;
1001 unsigned long writeback;
1002 unsigned long mapcount_max;
1003 unsigned long dirty;
1004 unsigned long swapcache;
1005 unsigned long node[MAX_NUMNODES];
1006};
1007
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001008struct numa_maps_private {
1009 struct proc_maps_private proc_maps;
1010 struct numa_maps md;
1011};
1012
Dave Hanseneb4866d2011-09-20 15:19:38 -07001013static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1014 unsigned long nr_pages)
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001015{
1016 int count = page_mapcount(page);
1017
Dave Hanseneb4866d2011-09-20 15:19:38 -07001018 md->pages += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001019 if (pte_dirty || PageDirty(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -07001020 md->dirty += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001021
1022 if (PageSwapCache(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -07001023 md->swapcache += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001024
1025 if (PageActive(page) || PageUnevictable(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -07001026 md->active += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001027
1028 if (PageWriteback(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -07001029 md->writeback += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001030
1031 if (PageAnon(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -07001032 md->anon += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001033
1034 if (count > md->mapcount_max)
1035 md->mapcount_max = count;
1036
Dave Hanseneb4866d2011-09-20 15:19:38 -07001037 md->node[page_to_nid(page)] += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001038}
1039
Dave Hansen3200a8a2011-09-20 15:19:39 -07001040static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1041 unsigned long addr)
1042{
1043 struct page *page;
1044 int nid;
1045
1046 if (!pte_present(pte))
1047 return NULL;
1048
1049 page = vm_normal_page(vma, addr, pte);
1050 if (!page)
1051 return NULL;
1052
1053 if (PageReserved(page))
1054 return NULL;
1055
1056 nid = page_to_nid(page);
1057 if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
1058 return NULL;
1059
1060 return page;
1061}
1062
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001063static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1064 unsigned long end, struct mm_walk *walk)
1065{
1066 struct numa_maps *md;
1067 spinlock_t *ptl;
1068 pte_t *orig_pte;
1069 pte_t *pte;
1070
1071 md = walk->private;
Dave Hansen32ef4382011-09-20 15:19:41 -07001072
Naoya Horiguchi025c5b22012-03-21 16:33:57 -07001073 if (pmd_trans_huge_lock(pmd, md->vma) == 1) {
1074 pte_t huge_pte = *(pte_t *)pmd;
1075 struct page *page;
1076
1077 page = can_gather_numa_stats(huge_pte, md->vma, addr);
1078 if (page)
1079 gather_stats(page, md, pte_dirty(huge_pte),
1080 HPAGE_PMD_SIZE/PAGE_SIZE);
Dave Hansen32ef4382011-09-20 15:19:41 -07001081 spin_unlock(&walk->mm->page_table_lock);
Naoya Horiguchi025c5b22012-03-21 16:33:57 -07001082 return 0;
Dave Hansen32ef4382011-09-20 15:19:41 -07001083 }
1084
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -07001085 if (pmd_trans_unstable(pmd))
1086 return 0;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001087 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1088 do {
Dave Hansen3200a8a2011-09-20 15:19:39 -07001089 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001090 if (!page)
1091 continue;
Dave Hanseneb4866d2011-09-20 15:19:38 -07001092 gather_stats(page, md, pte_dirty(*pte), 1);
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001093
1094 } while (pte++, addr += PAGE_SIZE, addr != end);
1095 pte_unmap_unlock(orig_pte, ptl);
1096 return 0;
1097}
1098#ifdef CONFIG_HUGETLB_PAGE
1099static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1100 unsigned long addr, unsigned long end, struct mm_walk *walk)
1101{
1102 struct numa_maps *md;
1103 struct page *page;
1104
1105 if (pte_none(*pte))
1106 return 0;
1107
1108 page = pte_page(*pte);
1109 if (!page)
1110 return 0;
1111
1112 md = walk->private;
Dave Hanseneb4866d2011-09-20 15:19:38 -07001113 gather_stats(page, md, pte_dirty(*pte), 1);
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001114 return 0;
1115}
1116
1117#else
1118static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1119 unsigned long addr, unsigned long end, struct mm_walk *walk)
1120{
1121 return 0;
1122}
1123#endif
1124
1125/*
1126 * Display pages allocated per node and memory policy via /proc.
1127 */
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -07001128static int show_numa_map(struct seq_file *m, void *v, int is_pid)
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001129{
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001130 struct numa_maps_private *numa_priv = m->private;
1131 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001132 struct vm_area_struct *vma = v;
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001133 struct numa_maps *md = &numa_priv->md;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001134 struct file *file = vma->vm_file;
1135 struct mm_struct *mm = vma->vm_mm;
1136 struct mm_walk walk = {};
1137 struct mempolicy *pol;
1138 int n;
1139 char buffer[50];
1140
1141 if (!mm)
1142 return 0;
1143
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001144 /* Ensure we start with an empty set of numa_maps statistics. */
1145 memset(md, 0, sizeof(*md));
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001146
1147 md->vma = vma;
1148
1149 walk.hugetlb_entry = gather_hugetbl_stats;
1150 walk.pmd_entry = gather_pte_stats;
1151 walk.private = md;
1152 walk.mm = mm;
1153
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001154 pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001155 mpol_to_str(buffer, sizeof(buffer), pol, 0);
1156 mpol_cond_put(pol);
1157
1158 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1159
1160 if (file) {
1161 seq_printf(m, " file=");
1162 seq_path(m, &file->f_path, "\n\t= ");
1163 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1164 seq_printf(m, " heap");
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -07001165 } else {
1166 pid_t tid = vm_is_stack(proc_priv->task, vma, is_pid);
1167 if (tid != 0) {
1168 /*
1169 * Thread stack in /proc/PID/task/TID/maps or
1170 * the main process stack.
1171 */
1172 if (!is_pid || (vma->vm_start <= mm->start_stack &&
1173 vma->vm_end >= mm->start_stack))
1174 seq_printf(m, " stack");
1175 else
1176 seq_printf(m, " stack:%d", tid);
1177 }
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001178 }
1179
Andrew Mortonfc360bd2011-10-31 17:06:32 -07001180 if (is_vm_hugetlb_page(vma))
1181 seq_printf(m, " huge");
1182
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001183 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1184
1185 if (!md->pages)
1186 goto out;
1187
1188 if (md->anon)
1189 seq_printf(m, " anon=%lu", md->anon);
1190
1191 if (md->dirty)
1192 seq_printf(m, " dirty=%lu", md->dirty);
1193
1194 if (md->pages != md->anon && md->pages != md->dirty)
1195 seq_printf(m, " mapped=%lu", md->pages);
1196
1197 if (md->mapcount_max > 1)
1198 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1199
1200 if (md->swapcache)
1201 seq_printf(m, " swapcache=%lu", md->swapcache);
1202
1203 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1204 seq_printf(m, " active=%lu", md->active);
1205
1206 if (md->writeback)
1207 seq_printf(m, " writeback=%lu", md->writeback);
1208
1209 for_each_node_state(n, N_HIGH_MEMORY)
1210 if (md->node[n])
1211 seq_printf(m, " N%d=%lu", n, md->node[n]);
1212out:
1213 seq_putc(m, '\n');
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001214
1215 if (m->count < m->size)
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001216 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001217 return 0;
1218}
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001219
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -07001220static int show_pid_numa_map(struct seq_file *m, void *v)
1221{
1222 return show_numa_map(m, v, 1);
1223}
1224
1225static int show_tid_numa_map(struct seq_file *m, void *v)
1226{
1227 return show_numa_map(m, v, 0);
1228}
1229
Jan Engelhardt03a44822008-02-08 04:21:19 -08001230static const struct seq_operations proc_pid_numa_maps_op = {
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -07001231 .start = m_start,
1232 .next = m_next,
1233 .stop = m_stop,
1234 .show = show_pid_numa_map,
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001235};
Eric W. Biederman662795d2006-06-26 00:25:48 -07001236
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -07001237static const struct seq_operations proc_tid_numa_maps_op = {
1238 .start = m_start,
1239 .next = m_next,
1240 .stop = m_stop,
1241 .show = show_tid_numa_map,
1242};
1243
1244static int numa_maps_open(struct inode *inode, struct file *file,
1245 const struct seq_operations *ops)
Eric W. Biederman662795d2006-06-26 00:25:48 -07001246{
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001247 struct numa_maps_private *priv;
1248 int ret = -ENOMEM;
1249 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1250 if (priv) {
1251 priv->proc_maps.pid = proc_pid(inode);
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -07001252 ret = seq_open(file, ops);
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001253 if (!ret) {
1254 struct seq_file *m = file->private_data;
1255 m->private = priv;
1256 } else {
1257 kfree(priv);
1258 }
1259 }
1260 return ret;
Eric W. Biederman662795d2006-06-26 00:25:48 -07001261}
1262
Siddhesh Poyarekarb7643752012-03-21 16:34:04 -07001263static int pid_numa_maps_open(struct inode *inode, struct file *file)
1264{
1265 return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1266}
1267
1268static int tid_numa_maps_open(struct inode *inode, struct file *file)
1269{
1270 return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1271}
1272
1273const struct file_operations proc_pid_numa_maps_operations = {
1274 .open = pid_numa_maps_open,
1275 .read = seq_read,
1276 .llseek = seq_lseek,
1277 .release = seq_release_private,
1278};
1279
1280const struct file_operations proc_tid_numa_maps_operations = {
1281 .open = tid_numa_maps_open,
Eric W. Biederman662795d2006-06-26 00:25:48 -07001282 .read = seq_read,
1283 .llseek = seq_lseek,
Eric W. Biederman99f89552006-06-26 00:25:55 -07001284 .release = seq_release_private,
Eric W. Biederman662795d2006-06-26 00:25:48 -07001285};
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001286#endif /* CONFIG_NUMA */