blob: c7e3a163295cb8169163cdaa2468d5d8f54a4a6a [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
Joe Korty7c88db02008-10-16 15:27:09 +0400212static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
Mauricio Line070ad42005-09-03 15:55:10 -0700214 struct mm_struct *mm = vma->vm_mm;
215 struct file *file = vma->vm_file;
KOSAKI Motohiroca16d142011-05-26 19:16:19 +0900216 vm_flags_t flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 unsigned long ino = 0;
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700218 unsigned long long pgoff = 0;
Mikulas Patockaa09a79f2011-05-09 13:01:09 +0200219 unsigned long start, end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 dev_t dev = 0;
221 int len;
222
223 if (file) {
Josef "Jeff" Sipek2fddfee2006-12-08 02:36:36 -0800224 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 dev = inode->i_sb->s_dev;
226 ino = inode->i_ino;
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700227 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 }
229
Linus Torvaldsd7824372010-08-15 11:35:52 -0700230 /* We don't show the stack guard page in /proc/maps */
231 start = vma->vm_start;
Mikulas Patockaa09a79f2011-05-09 13:01:09 +0200232 if (stack_guard_page_start(vma, start))
233 start += PAGE_SIZE;
234 end = vma->vm_end;
235 if (stack_guard_page_end(vma, end))
236 end -= PAGE_SIZE;
Linus Torvaldsd7824372010-08-15 11:35:52 -0700237
Clement Calmels1804dc62008-08-20 14:09:00 -0700238 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
Linus Torvaldsd7824372010-08-15 11:35:52 -0700239 start,
Mikulas Patockaa09a79f2011-05-09 13:01:09 +0200240 end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 flags & VM_READ ? 'r' : '-',
242 flags & VM_WRITE ? 'w' : '-',
243 flags & VM_EXEC ? 'x' : '-',
244 flags & VM_MAYSHARE ? 's' : 'p',
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700245 pgoff,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 MAJOR(dev), MINOR(dev), ino, &len);
247
248 /*
249 * Print the dentry name for named mappings, and a
250 * special [heap] marker for the heap:
251 */
Mauricio Line070ad42005-09-03 15:55:10 -0700252 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 pad_len_spaces(m, len);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800254 seq_path(m, &file->f_path, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 } else {
Ingo Molnare6e54942006-06-27 02:53:50 -0700256 const char *name = arch_vma_name(vma);
257 if (!name) {
258 if (mm) {
Aaro Koskinen0db0c012011-03-23 16:42:50 -0700259 if (vma->vm_start <= mm->brk &&
260 vma->vm_end >= mm->start_brk) {
Ingo Molnare6e54942006-06-27 02:53:50 -0700261 name = "[heap]";
262 } else if (vma->vm_start <= mm->start_stack &&
263 vma->vm_end >= mm->start_stack) {
264 name = "[stack]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700266 } else {
267 name = "[vdso]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700269 }
270 if (name) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 pad_len_spaces(m, len);
Ingo Molnare6e54942006-06-27 02:53:50 -0700272 seq_puts(m, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 }
274 }
275 seq_putc(m, '\n');
Joe Korty7c88db02008-10-16 15:27:09 +0400276}
277
278static int show_map(struct seq_file *m, void *v)
279{
280 struct vm_area_struct *vma = v;
281 struct proc_maps_private *priv = m->private;
282 struct task_struct *task = priv->task;
283
284 show_map_vma(m, vma);
Mauricio Line070ad42005-09-03 15:55:10 -0700285
Mauricio Line070ad42005-09-03 15:55:10 -0700286 if (m->count < m->size) /* vma is copied successfully */
Stephen Wilson31db58b2011-03-13 15:49:15 -0400287 m->version = (vma != get_gate_vma(task->mm))
288 ? vma->vm_start : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 return 0;
290}
291
Jan Engelhardt03a44822008-02-08 04:21:19 -0800292static const struct seq_operations proc_pid_maps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800293 .start = m_start,
294 .next = m_next,
295 .stop = m_stop,
296 .show = show_map
297};
298
299static int maps_open(struct inode *inode, struct file *file)
300{
301 return do_maps_open(inode, file, &proc_pid_maps_op);
302}
303
304const struct file_operations proc_maps_operations = {
305 .open = maps_open,
306 .read = seq_read,
307 .llseek = seq_lseek,
308 .release = seq_release_private,
309};
310
311/*
312 * Proportional Set Size(PSS): my share of RSS.
313 *
314 * PSS of a process is the count of pages it has in memory, where each
315 * page is divided by the number of processes sharing it. So if a
316 * process has 1000 pages all to itself, and 1000 shared with one other
317 * process, its PSS will be 1500.
318 *
319 * To keep (accumulated) division errors low, we adopt a 64bit
320 * fixed-point pss counter to minimize division errors. So (pss >>
321 * PSS_SHIFT) would be the real byte count.
322 *
323 * A shift of 12 before division means (assuming 4K page size):
324 * - 1M 3-user-pages add up to 8KB errors;
325 * - supports mapcount up to 2^24, or 16M;
326 * - supports PSS up to 2^52 bytes, or 4PB.
327 */
328#define PSS_SHIFT 12
329
Matt Mackall1e883282008-02-04 22:29:07 -0800330#ifdef CONFIG_PROC_PAGE_MONITOR
Peter Zijlstra214e4712008-04-28 02:12:55 -0700331struct mem_size_stats {
Matt Mackalla6198792008-02-04 22:29:03 -0800332 struct vm_area_struct *vma;
333 unsigned long resident;
334 unsigned long shared_clean;
335 unsigned long shared_dirty;
336 unsigned long private_clean;
337 unsigned long private_dirty;
338 unsigned long referenced;
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700339 unsigned long anonymous;
Dave Hansen4031a212011-03-22 16:33:01 -0700340 unsigned long anonymous_thp;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700341 unsigned long swap;
Matt Mackalla6198792008-02-04 22:29:03 -0800342 u64 pss;
343};
344
Dave Hansenae11c4d2011-03-22 16:32:58 -0700345
346static void smaps_pte_entry(pte_t ptent, unsigned long addr,
Dave Hansen3c9acc72011-03-22 16:32:59 -0700347 unsigned long ptent_size, struct mm_walk *walk)
Dave Hansenae11c4d2011-03-22 16:32:58 -0700348{
349 struct mem_size_stats *mss = walk->private;
350 struct vm_area_struct *vma = mss->vma;
351 struct page *page;
352 int mapcount;
353
354 if (is_swap_pte(ptent)) {
Dave Hansen3c9acc72011-03-22 16:32:59 -0700355 mss->swap += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700356 return;
357 }
358
359 if (!pte_present(ptent))
360 return;
361
362 page = vm_normal_page(vma, addr, ptent);
363 if (!page)
364 return;
365
366 if (PageAnon(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700367 mss->anonymous += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700368
Dave Hansen3c9acc72011-03-22 16:32:59 -0700369 mss->resident += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700370 /* Accumulate the size in pages that have been accessed. */
371 if (pte_young(ptent) || PageReferenced(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700372 mss->referenced += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700373 mapcount = page_mapcount(page);
374 if (mapcount >= 2) {
375 if (pte_dirty(ptent) || PageDirty(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700376 mss->shared_dirty += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700377 else
Dave Hansen3c9acc72011-03-22 16:32:59 -0700378 mss->shared_clean += ptent_size;
379 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700380 } else {
381 if (pte_dirty(ptent) || PageDirty(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700382 mss->private_dirty += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700383 else
Dave Hansen3c9acc72011-03-22 16:32:59 -0700384 mss->private_clean += ptent_size;
385 mss->pss += (ptent_size << PSS_SHIFT);
Dave Hansenae11c4d2011-03-22 16:32:58 -0700386 }
387}
388
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800389static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700390 struct mm_walk *walk)
Mauricio Line070ad42005-09-03 15:55:10 -0700391{
Dave Hansen21650092008-06-12 15:21:47 -0700392 struct mem_size_stats *mss = walk->private;
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800393 struct vm_area_struct *vma = mss->vma;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700394 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700395 spinlock_t *ptl;
Mauricio Line070ad42005-09-03 15:55:10 -0700396
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700397 if (pmd_trans_huge_lock(pmd, vma) == 1) {
398 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
Dave Hansen22e057c2011-03-22 16:33:00 -0700399 spin_unlock(&walk->mm->page_table_lock);
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700400 mss->anonymous_thp += HPAGE_PMD_SIZE;
401 return 0;
Dave Hansen22e057c2011-03-22 16:33:00 -0700402 }
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700403
404 if (pmd_trans_unstable(pmd))
405 return 0;
Dave Hansen22e057c2011-03-22 16:33:00 -0700406 /*
407 * The mmap_sem held all the way back in m_start() is what
408 * keeps khugepaged out of here and from collapsing things
409 * in here.
410 */
Hugh Dickins705e87c2005-10-29 18:16:27 -0700411 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Dave Hansenae11c4d2011-03-22 16:32:58 -0700412 for (; addr != end; pte++, addr += PAGE_SIZE)
Dave Hansen3c9acc72011-03-22 16:32:59 -0700413 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700414 pte_unmap_unlock(pte - 1, ptl);
415 cond_resched();
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800416 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700417}
418
Mauricio Line070ad42005-09-03 15:55:10 -0700419static int show_smap(struct seq_file *m, void *v)
420{
Joe Korty7c88db02008-10-16 15:27:09 +0400421 struct proc_maps_private *priv = m->private;
422 struct task_struct *task = priv->task;
Mauricio Line070ad42005-09-03 15:55:10 -0700423 struct vm_area_struct *vma = v;
Mauricio Line070ad42005-09-03 15:55:10 -0700424 struct mem_size_stats mss;
Dave Hansen21650092008-06-12 15:21:47 -0700425 struct mm_walk smaps_walk = {
426 .pmd_entry = smaps_pte_range,
427 .mm = vma->vm_mm,
428 .private = &mss,
429 };
Mauricio Line070ad42005-09-03 15:55:10 -0700430
431 memset(&mss, 0, sizeof mss);
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800432 mss.vma = vma;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900433 /* mmap_sem is held in m_start */
Nick Piggin5ddfae12006-03-06 15:42:57 -0800434 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700435 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
Matt Mackall4752c362008-02-04 22:29:02 -0800436
Joe Korty7c88db02008-10-16 15:27:09 +0400437 show_map_vma(m, vma);
Matt Mackall4752c362008-02-04 22:29:02 -0800438
439 seq_printf(m,
440 "Size: %8lu kB\n"
441 "Rss: %8lu kB\n"
442 "Pss: %8lu kB\n"
443 "Shared_Clean: %8lu kB\n"
444 "Shared_Dirty: %8lu kB\n"
445 "Private_Clean: %8lu kB\n"
446 "Private_Dirty: %8lu kB\n"
Peter Zijlstra214e4712008-04-28 02:12:55 -0700447 "Referenced: %8lu kB\n"
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700448 "Anonymous: %8lu kB\n"
Dave Hansen4031a212011-03-22 16:33:01 -0700449 "AnonHugePages: %8lu kB\n"
Mel Gorman08fba692009-01-06 14:38:53 -0800450 "Swap: %8lu kB\n"
Mel Gorman33402892009-01-06 14:38:54 -0800451 "KernelPageSize: %8lu kB\n"
Nikanth Karthikesan2d905082011-01-13 15:45:53 -0800452 "MMUPageSize: %8lu kB\n"
453 "Locked: %8lu kB\n",
Matt Mackall4752c362008-02-04 22:29:02 -0800454 (vma->vm_end - vma->vm_start) >> 10,
455 mss.resident >> 10,
456 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
457 mss.shared_clean >> 10,
458 mss.shared_dirty >> 10,
459 mss.private_clean >> 10,
460 mss.private_dirty >> 10,
Peter Zijlstra214e4712008-04-28 02:12:55 -0700461 mss.referenced >> 10,
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700462 mss.anonymous >> 10,
Dave Hansen4031a212011-03-22 16:33:01 -0700463 mss.anonymous_thp >> 10,
Mel Gorman08fba692009-01-06 14:38:53 -0800464 mss.swap >> 10,
Mel Gorman33402892009-01-06 14:38:54 -0800465 vma_kernel_pagesize(vma) >> 10,
Nikanth Karthikesan2d905082011-01-13 15:45:53 -0800466 vma_mmu_pagesize(vma) >> 10,
467 (vma->vm_flags & VM_LOCKED) ?
468 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
Matt Mackall4752c362008-02-04 22:29:02 -0800469
Joe Korty7c88db02008-10-16 15:27:09 +0400470 if (m->count < m->size) /* vma is copied successfully */
Stephen Wilson31db58b2011-03-13 15:49:15 -0400471 m->version = (vma != get_gate_vma(task->mm))
472 ? vma->vm_start : 0;
Joe Korty7c88db02008-10-16 15:27:09 +0400473 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700474}
475
Jan Engelhardt03a44822008-02-08 04:21:19 -0800476static const struct seq_operations proc_pid_smaps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800477 .start = m_start,
478 .next = m_next,
479 .stop = m_stop,
480 .show = show_smap
481};
482
483static int smaps_open(struct inode *inode, struct file *file)
484{
485 return do_maps_open(inode, file, &proc_pid_smaps_op);
486}
487
488const struct file_operations proc_smaps_operations = {
489 .open = smaps_open,
490 .read = seq_read,
491 .llseek = seq_lseek,
492 .release = seq_release_private,
493};
494
495static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
Dave Hansen21650092008-06-12 15:21:47 -0700496 unsigned long end, struct mm_walk *walk)
Matt Mackalla6198792008-02-04 22:29:03 -0800497{
Dave Hansen21650092008-06-12 15:21:47 -0700498 struct vm_area_struct *vma = walk->private;
Matt Mackalla6198792008-02-04 22:29:03 -0800499 pte_t *pte, ptent;
500 spinlock_t *ptl;
501 struct page *page;
502
Dave Hansen03319322011-03-22 16:32:56 -0700503 split_huge_page_pmd(walk->mm, pmd);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700504 if (pmd_trans_unstable(pmd))
505 return 0;
Dave Hansen03319322011-03-22 16:32:56 -0700506
Matt Mackalla6198792008-02-04 22:29:03 -0800507 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
508 for (; addr != end; pte++, addr += PAGE_SIZE) {
509 ptent = *pte;
510 if (!pte_present(ptent))
511 continue;
512
513 page = vm_normal_page(vma, addr, ptent);
514 if (!page)
515 continue;
516
Will Deacon85e72aa2012-01-20 14:34:09 -0800517 if (PageReserved(page))
518 continue;
519
Matt Mackalla6198792008-02-04 22:29:03 -0800520 /* Clear accessed and referenced bits. */
521 ptep_test_and_clear_young(vma, addr, pte);
522 ClearPageReferenced(page);
523 }
524 pte_unmap_unlock(pte - 1, ptl);
525 cond_resched();
526 return 0;
527}
528
Moussa A. Ba398499d2009-09-21 17:02:29 -0700529#define CLEAR_REFS_ALL 1
530#define CLEAR_REFS_ANON 2
531#define CLEAR_REFS_MAPPED 3
532
Matt Mackallf248dcb2008-02-04 22:29:03 -0800533static ssize_t clear_refs_write(struct file *file, const char __user *buf,
534 size_t count, loff_t *ppos)
David Rientjesb813e932007-05-06 14:49:24 -0700535{
Matt Mackallf248dcb2008-02-04 22:29:03 -0800536 struct task_struct *task;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700537 char buffer[PROC_NUMBUF];
Matt Mackallf248dcb2008-02-04 22:29:03 -0800538 struct mm_struct *mm;
David Rientjesb813e932007-05-06 14:49:24 -0700539 struct vm_area_struct *vma;
Alexey Dobriyan0a8cb8e2011-05-26 16:25:50 -0700540 int type;
541 int rv;
David Rientjesb813e932007-05-06 14:49:24 -0700542
Matt Mackallf248dcb2008-02-04 22:29:03 -0800543 memset(buffer, 0, sizeof(buffer));
544 if (count > sizeof(buffer) - 1)
545 count = sizeof(buffer) - 1;
546 if (copy_from_user(buffer, buf, count))
547 return -EFAULT;
Alexey Dobriyan0a8cb8e2011-05-26 16:25:50 -0700548 rv = kstrtoint(strstrip(buffer), 10, &type);
549 if (rv < 0)
550 return rv;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700551 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
Matt Mackallf248dcb2008-02-04 22:29:03 -0800552 return -EINVAL;
Matt Mackallf248dcb2008-02-04 22:29:03 -0800553 task = get_proc_task(file->f_path.dentry->d_inode);
554 if (!task)
555 return -ESRCH;
556 mm = get_task_mm(task);
557 if (mm) {
Andrew Morton20cbc972008-07-05 12:29:05 -0700558 struct mm_walk clear_refs_walk = {
559 .pmd_entry = clear_refs_pte_range,
560 .mm = mm,
561 };
Matt Mackallf248dcb2008-02-04 22:29:03 -0800562 down_read(&mm->mmap_sem);
Dave Hansen21650092008-06-12 15:21:47 -0700563 for (vma = mm->mmap; vma; vma = vma->vm_next) {
564 clear_refs_walk.private = vma;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700565 if (is_vm_hugetlb_page(vma))
566 continue;
567 /*
568 * Writing 1 to /proc/pid/clear_refs affects all pages.
569 *
570 * Writing 2 to /proc/pid/clear_refs only affects
571 * Anonymous pages.
572 *
573 * Writing 3 to /proc/pid/clear_refs only affects file
574 * mapped pages.
575 */
576 if (type == CLEAR_REFS_ANON && vma->vm_file)
577 continue;
578 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
579 continue;
580 walk_page_range(vma->vm_start, vma->vm_end,
581 &clear_refs_walk);
Dave Hansen21650092008-06-12 15:21:47 -0700582 }
Matt Mackallf248dcb2008-02-04 22:29:03 -0800583 flush_tlb_mm(mm);
584 up_read(&mm->mmap_sem);
585 mmput(mm);
586 }
587 put_task_struct(task);
Vincent Lifb92a4b2009-09-22 16:45:36 -0700588
589 return count;
David Rientjesb813e932007-05-06 14:49:24 -0700590}
591
Matt Mackallf248dcb2008-02-04 22:29:03 -0800592const struct file_operations proc_clear_refs_operations = {
593 .write = clear_refs_write,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200594 .llseek = noop_llseek,
Matt Mackallf248dcb2008-02-04 22:29:03 -0800595};
596
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700597typedef struct {
598 u64 pme;
599} pagemap_entry_t;
600
Matt Mackall85863e42008-02-04 22:29:04 -0800601struct pagemapread {
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900602 int pos, len;
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700603 pagemap_entry_t *buffer;
Matt Mackall85863e42008-02-04 22:29:04 -0800604};
605
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700606#define PAGEMAP_WALK_SIZE (PMD_SIZE)
607#define PAGEMAP_WALK_MASK (PMD_MASK)
608
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500609#define PM_ENTRY_BYTES sizeof(u64)
610#define PM_STATUS_BITS 3
611#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
612#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
613#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
614#define PM_PSHIFT_BITS 6
615#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
616#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
617#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
618#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
619#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
620
621#define PM_PRESENT PM_STATUS(4LL)
622#define PM_SWAP PM_STATUS(2LL)
623#define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
Matt Mackall85863e42008-02-04 22:29:04 -0800624#define PM_END_OF_BUFFER 1
625
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700626static inline pagemap_entry_t make_pme(u64 val)
627{
628 return (pagemap_entry_t) { .pme = val };
629}
630
631static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
Matt Mackall85863e42008-02-04 22:29:04 -0800632 struct pagemapread *pm)
633{
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700634 pm->buffer[pm->pos++] = *pme;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900635 if (pm->pos >= pm->len)
Thomas Tuttleaae86792008-06-05 22:46:31 -0700636 return PM_END_OF_BUFFER;
Matt Mackall85863e42008-02-04 22:29:04 -0800637 return 0;
638}
639
640static int pagemap_pte_hole(unsigned long start, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700641 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800642{
Dave Hansen21650092008-06-12 15:21:47 -0700643 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800644 unsigned long addr;
645 int err = 0;
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700646 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
647
Matt Mackall85863e42008-02-04 22:29:04 -0800648 for (addr = start; addr < end; addr += PAGE_SIZE) {
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700649 err = add_to_pagemap(addr, &pme, pm);
Matt Mackall85863e42008-02-04 22:29:04 -0800650 if (err)
651 break;
652 }
653 return err;
654}
655
Adrian Bunk9d02dbc2008-04-28 02:12:11 -0700656static u64 swap_pte_to_pagemap_entry(pte_t pte)
Matt Mackall85863e42008-02-04 22:29:04 -0800657{
658 swp_entry_t e = pte_to_swp_entry(pte);
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500659 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
Matt Mackall85863e42008-02-04 22:29:04 -0800660}
661
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700662static void pte_to_pagemap_entry(pagemap_entry_t *pme, pte_t pte)
Dave Hansenbcf80392008-06-12 15:21:48 -0700663{
Dave Hansenbcf80392008-06-12 15:21:48 -0700664 if (is_swap_pte(pte))
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700665 *pme = make_pme(PM_PFRAME(swap_pte_to_pagemap_entry(pte))
666 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP);
Dave Hansenbcf80392008-06-12 15:21:48 -0700667 else if (pte_present(pte))
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700668 *pme = make_pme(PM_PFRAME(pte_pfn(pte))
669 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
Dave Hansenbcf80392008-06-12 15:21:48 -0700670}
671
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700672#ifdef CONFIG_TRANSPARENT_HUGEPAGE
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700673static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
674 pmd_t pmd, int offset)
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700675{
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700676 /*
677 * Currently pmd for thp is always present because thp can not be
678 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
679 * This if-check is just to prepare for future implementation.
680 */
681 if (pmd_present(pmd))
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700682 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
683 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700684}
685#else
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700686static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme,
687 pmd_t pmd, int offset)
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700688{
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700689}
690#endif
691
Matt Mackall85863e42008-02-04 22:29:04 -0800692static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700693 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800694{
Dave Hansenbcf80392008-06-12 15:21:48 -0700695 struct vm_area_struct *vma;
Dave Hansen21650092008-06-12 15:21:47 -0700696 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800697 pte_t *pte;
698 int err = 0;
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700699 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
Matt Mackall85863e42008-02-04 22:29:04 -0800700
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700701 if (pmd_trans_unstable(pmd))
702 return 0;
Dave Hansen03319322011-03-22 16:32:56 -0700703
Dave Hansenbcf80392008-06-12 15:21:48 -0700704 /* find the first VMA at or above 'addr' */
705 vma = find_vma(walk->mm, addr);
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700706 spin_lock(&walk->mm->page_table_lock);
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700707 if (pmd_trans_huge_lock(pmd, vma) == 1) {
708 for (; addr != end; addr += PAGE_SIZE) {
709 unsigned long offset;
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700710
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700711 offset = (addr & ~PAGEMAP_WALK_MASK) >>
712 PAGE_SHIFT;
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700713 thp_pmd_to_pagemap_entry(&pme, *pmd, offset);
714 err = add_to_pagemap(addr, &pme, pm);
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700715 if (err)
716 break;
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700717 }
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700718 spin_unlock(&walk->mm->page_table_lock);
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700719 return err;
Naoya Horiguchi5aaabe82012-03-21 16:33:57 -0700720 }
721
Matt Mackall85863e42008-02-04 22:29:04 -0800722 for (; addr != end; addr += PAGE_SIZE) {
Dave Hansenbcf80392008-06-12 15:21:48 -0700723
724 /* check to see if we've left 'vma' behind
725 * and need a new, higher one */
726 if (vma && (addr >= vma->vm_end))
727 vma = find_vma(walk->mm, addr);
728
729 /* check that 'vma' actually covers this address,
730 * and that it isn't a huge page vma */
731 if (vma && (vma->vm_start <= addr) &&
732 !is_vm_hugetlb_page(vma)) {
733 pte = pte_offset_map(pmd, addr);
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700734 pte_to_pagemap_entry(&pme, *pte);
Dave Hansenbcf80392008-06-12 15:21:48 -0700735 /* unmap before userspace copy */
736 pte_unmap(pte);
737 }
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700738 err = add_to_pagemap(addr, &pme, pm);
Matt Mackall85863e42008-02-04 22:29:04 -0800739 if (err)
740 return err;
741 }
742
743 cond_resched();
744
745 return err;
746}
747
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700748#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700749static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme,
750 pte_t pte, int offset)
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800751{
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800752 if (pte_present(pte))
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700753 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset)
754 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT);
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800755}
756
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700757/* This function walks within one hugetlb entry in the single call */
758static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
759 unsigned long addr, unsigned long end,
760 struct mm_walk *walk)
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800761{
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800762 struct pagemapread *pm = walk->private;
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800763 int err = 0;
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700764 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT);
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800765
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800766 for (; addr != end; addr += PAGE_SIZE) {
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700767 int offset = (addr & ~hmask) >> PAGE_SHIFT;
Naoya Horiguchi092b50b2012-03-21 16:33:59 -0700768 huge_pte_to_pagemap_entry(&pme, *pte, offset);
769 err = add_to_pagemap(addr, &pme, pm);
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800770 if (err)
771 return err;
772 }
773
774 cond_resched();
775
776 return err;
777}
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700778#endif /* HUGETLB_PAGE */
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800779
Matt Mackall85863e42008-02-04 22:29:04 -0800780/*
781 * /proc/pid/pagemap - an array mapping virtual pages to pfns
782 *
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500783 * For each page in the address space, this file contains one 64-bit entry
784 * consisting of the following:
785 *
786 * Bits 0-55 page frame number (PFN) if present
787 * Bits 0-4 swap type if swapped
788 * Bits 5-55 swap offset if swapped
789 * Bits 55-60 page shift (page size = 1<<page shift)
790 * Bit 61 reserved for future use
791 * Bit 62 page swapped
792 * Bit 63 page present
793 *
794 * If the page is not present but in swap, then the PFN contains an
795 * encoding of the swap file number and the page's offset into the
796 * swap. Unmapped pages return a null PFN. This allows determining
Matt Mackall85863e42008-02-04 22:29:04 -0800797 * precisely which pages are mapped (or in swap) and comparing mapped
798 * pages between processes.
799 *
800 * Efficient users of this interface will use /proc/pid/maps to
801 * determine which areas of memory are actually mapped and llseek to
802 * skip over unmapped regions.
803 */
804static ssize_t pagemap_read(struct file *file, char __user *buf,
805 size_t count, loff_t *ppos)
806{
807 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
Matt Mackall85863e42008-02-04 22:29:04 -0800808 struct mm_struct *mm;
809 struct pagemapread pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800810 int ret = -ESRCH;
Alexey Dobriyanee1e6ab2008-07-21 14:21:36 -0700811 struct mm_walk pagemap_walk = {};
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700812 unsigned long src;
813 unsigned long svpfn;
814 unsigned long start_vaddr;
815 unsigned long end_vaddr;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900816 int copied = 0;
Matt Mackall85863e42008-02-04 22:29:04 -0800817
818 if (!task)
819 goto out;
820
Matt Mackall85863e42008-02-04 22:29:04 -0800821 ret = -EINVAL;
822 /* file position must be aligned */
Thomas Tuttleaae86792008-06-05 22:46:31 -0700823 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700824 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800825
826 ret = 0;
Vitaly Mayatskikh08161782009-04-30 15:08:18 -0700827 if (!count)
828 goto out_task;
829
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900830 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
831 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700832 ret = -ENOMEM;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900833 if (!pm.buffer)
KOSAKI Motohiro98bc93e2011-05-26 16:25:53 -0700834 goto out_task;
835
836 mm = mm_for_maps(task);
837 ret = PTR_ERR(mm);
838 if (!mm || IS_ERR(mm))
839 goto out_free;
Matt Mackall85863e42008-02-04 22:29:04 -0800840
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700841 pagemap_walk.pmd_entry = pagemap_pte_range;
842 pagemap_walk.pte_hole = pagemap_pte_hole;
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700843#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800844 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700845#endif
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700846 pagemap_walk.mm = mm;
847 pagemap_walk.private = &pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800848
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700849 src = *ppos;
850 svpfn = src / PM_ENTRY_BYTES;
851 start_vaddr = svpfn << PAGE_SHIFT;
852 end_vaddr = TASK_SIZE_OF(task);
Matt Mackall85863e42008-02-04 22:29:04 -0800853
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700854 /* watch out for wraparound */
855 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
856 start_vaddr = end_vaddr;
857
858 /*
859 * The odds are that this will stop walking way
860 * before end_vaddr, because the length of the
861 * user buffer is tracked in "pm", and the walk
862 * will stop when we hit the end of the buffer.
863 */
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900864 ret = 0;
865 while (count && (start_vaddr < end_vaddr)) {
866 int len;
867 unsigned long end;
Matt Mackall85863e42008-02-04 22:29:04 -0800868
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900869 pm.pos = 0;
Naoya Horiguchiea251c12010-11-24 12:57:13 -0800870 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900871 /* overflow ? */
872 if (end < start_vaddr || end > end_vaddr)
873 end = end_vaddr;
874 down_read(&mm->mmap_sem);
875 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
876 up_read(&mm->mmap_sem);
877 start_vaddr = end;
878
879 len = min(count, PM_ENTRY_BYTES * pm.pos);
Dan Carpenter309361e02010-04-06 13:45:39 +0300880 if (copy_to_user(buf, pm.buffer, len)) {
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900881 ret = -EFAULT;
KOSAKI Motohiro98bc93e2011-05-26 16:25:53 -0700882 goto out_mm;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900883 }
884 copied += len;
885 buf += len;
886 count -= len;
Matt Mackall85863e42008-02-04 22:29:04 -0800887 }
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900888 *ppos += copied;
889 if (!ret || ret == PM_END_OF_BUFFER)
890 ret = copied;
891
Marcelo Tosattifb393802008-03-13 12:32:35 -0700892out_mm:
893 mmput(mm);
KOSAKI Motohiro98bc93e2011-05-26 16:25:53 -0700894out_free:
895 kfree(pm.buffer);
Matt Mackall85863e42008-02-04 22:29:04 -0800896out_task:
897 put_task_struct(task);
898out:
899 return ret;
900}
901
902const struct file_operations proc_pagemap_operations = {
903 .llseek = mem_lseek, /* borrow this */
904 .read = pagemap_read,
905};
Matt Mackall1e883282008-02-04 22:29:07 -0800906#endif /* CONFIG_PROC_PAGE_MONITOR */
Matt Mackall85863e42008-02-04 22:29:04 -0800907
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700908#ifdef CONFIG_NUMA
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700909
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700910struct numa_maps {
911 struct vm_area_struct *vma;
912 unsigned long pages;
913 unsigned long anon;
914 unsigned long active;
915 unsigned long writeback;
916 unsigned long mapcount_max;
917 unsigned long dirty;
918 unsigned long swapcache;
919 unsigned long node[MAX_NUMNODES];
920};
921
Stephen Wilson5b52fc82011-05-24 17:12:49 -0700922struct numa_maps_private {
923 struct proc_maps_private proc_maps;
924 struct numa_maps md;
925};
926
Dave Hanseneb4866d2011-09-20 15:19:38 -0700927static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
928 unsigned long nr_pages)
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700929{
930 int count = page_mapcount(page);
931
Dave Hanseneb4866d2011-09-20 15:19:38 -0700932 md->pages += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700933 if (pte_dirty || PageDirty(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -0700934 md->dirty += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700935
936 if (PageSwapCache(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -0700937 md->swapcache += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700938
939 if (PageActive(page) || PageUnevictable(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -0700940 md->active += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700941
942 if (PageWriteback(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -0700943 md->writeback += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700944
945 if (PageAnon(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -0700946 md->anon += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700947
948 if (count > md->mapcount_max)
949 md->mapcount_max = count;
950
Dave Hanseneb4866d2011-09-20 15:19:38 -0700951 md->node[page_to_nid(page)] += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700952}
953
Dave Hansen3200a8a2011-09-20 15:19:39 -0700954static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
955 unsigned long addr)
956{
957 struct page *page;
958 int nid;
959
960 if (!pte_present(pte))
961 return NULL;
962
963 page = vm_normal_page(vma, addr, pte);
964 if (!page)
965 return NULL;
966
967 if (PageReserved(page))
968 return NULL;
969
970 nid = page_to_nid(page);
971 if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
972 return NULL;
973
974 return page;
975}
976
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700977static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
978 unsigned long end, struct mm_walk *walk)
979{
980 struct numa_maps *md;
981 spinlock_t *ptl;
982 pte_t *orig_pte;
983 pte_t *pte;
984
985 md = walk->private;
Dave Hansen32ef4382011-09-20 15:19:41 -0700986
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700987 if (pmd_trans_huge_lock(pmd, md->vma) == 1) {
988 pte_t huge_pte = *(pte_t *)pmd;
989 struct page *page;
990
991 page = can_gather_numa_stats(huge_pte, md->vma, addr);
992 if (page)
993 gather_stats(page, md, pte_dirty(huge_pte),
994 HPAGE_PMD_SIZE/PAGE_SIZE);
Dave Hansen32ef4382011-09-20 15:19:41 -0700995 spin_unlock(&walk->mm->page_table_lock);
Naoya Horiguchi025c5b22012-03-21 16:33:57 -0700996 return 0;
Dave Hansen32ef4382011-09-20 15:19:41 -0700997 }
998
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700999 if (pmd_trans_unstable(pmd))
1000 return 0;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001001 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1002 do {
Dave Hansen3200a8a2011-09-20 15:19:39 -07001003 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001004 if (!page)
1005 continue;
Dave Hanseneb4866d2011-09-20 15:19:38 -07001006 gather_stats(page, md, pte_dirty(*pte), 1);
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001007
1008 } while (pte++, addr += PAGE_SIZE, addr != end);
1009 pte_unmap_unlock(orig_pte, ptl);
1010 return 0;
1011}
1012#ifdef CONFIG_HUGETLB_PAGE
1013static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1014 unsigned long addr, unsigned long end, struct mm_walk *walk)
1015{
1016 struct numa_maps *md;
1017 struct page *page;
1018
1019 if (pte_none(*pte))
1020 return 0;
1021
1022 page = pte_page(*pte);
1023 if (!page)
1024 return 0;
1025
1026 md = walk->private;
Dave Hanseneb4866d2011-09-20 15:19:38 -07001027 gather_stats(page, md, pte_dirty(*pte), 1);
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001028 return 0;
1029}
1030
1031#else
1032static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1033 unsigned long addr, unsigned long end, struct mm_walk *walk)
1034{
1035 return 0;
1036}
1037#endif
1038
1039/*
1040 * Display pages allocated per node and memory policy via /proc.
1041 */
1042static int show_numa_map(struct seq_file *m, void *v)
1043{
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001044 struct numa_maps_private *numa_priv = m->private;
1045 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001046 struct vm_area_struct *vma = v;
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001047 struct numa_maps *md = &numa_priv->md;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001048 struct file *file = vma->vm_file;
1049 struct mm_struct *mm = vma->vm_mm;
1050 struct mm_walk walk = {};
1051 struct mempolicy *pol;
1052 int n;
1053 char buffer[50];
1054
1055 if (!mm)
1056 return 0;
1057
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001058 /* Ensure we start with an empty set of numa_maps statistics. */
1059 memset(md, 0, sizeof(*md));
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001060
1061 md->vma = vma;
1062
1063 walk.hugetlb_entry = gather_hugetbl_stats;
1064 walk.pmd_entry = gather_pte_stats;
1065 walk.private = md;
1066 walk.mm = mm;
1067
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001068 pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001069 mpol_to_str(buffer, sizeof(buffer), pol, 0);
1070 mpol_cond_put(pol);
1071
1072 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1073
1074 if (file) {
1075 seq_printf(m, " file=");
1076 seq_path(m, &file->f_path, "\n\t= ");
1077 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1078 seq_printf(m, " heap");
1079 } else if (vma->vm_start <= mm->start_stack &&
1080 vma->vm_end >= mm->start_stack) {
1081 seq_printf(m, " stack");
1082 }
1083
Andrew Mortonfc360bd2011-10-31 17:06:32 -07001084 if (is_vm_hugetlb_page(vma))
1085 seq_printf(m, " huge");
1086
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001087 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1088
1089 if (!md->pages)
1090 goto out;
1091
1092 if (md->anon)
1093 seq_printf(m, " anon=%lu", md->anon);
1094
1095 if (md->dirty)
1096 seq_printf(m, " dirty=%lu", md->dirty);
1097
1098 if (md->pages != md->anon && md->pages != md->dirty)
1099 seq_printf(m, " mapped=%lu", md->pages);
1100
1101 if (md->mapcount_max > 1)
1102 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1103
1104 if (md->swapcache)
1105 seq_printf(m, " swapcache=%lu", md->swapcache);
1106
1107 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1108 seq_printf(m, " active=%lu", md->active);
1109
1110 if (md->writeback)
1111 seq_printf(m, " writeback=%lu", md->writeback);
1112
1113 for_each_node_state(n, N_HIGH_MEMORY)
1114 if (md->node[n])
1115 seq_printf(m, " N%d=%lu", n, md->node[n]);
1116out:
1117 seq_putc(m, '\n');
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001118
1119 if (m->count < m->size)
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001120 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001121 return 0;
1122}
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001123
Jan Engelhardt03a44822008-02-08 04:21:19 -08001124static const struct seq_operations proc_pid_numa_maps_op = {
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08001125 .start = m_start,
1126 .next = m_next,
1127 .stop = m_stop,
Alexey Dobriyan3bbfe052008-10-10 03:27:16 +04001128 .show = show_numa_map,
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001129};
Eric W. Biederman662795d2006-06-26 00:25:48 -07001130
1131static int numa_maps_open(struct inode *inode, struct file *file)
1132{
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001133 struct numa_maps_private *priv;
1134 int ret = -ENOMEM;
1135 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1136 if (priv) {
1137 priv->proc_maps.pid = proc_pid(inode);
1138 ret = seq_open(file, &proc_pid_numa_maps_op);
1139 if (!ret) {
1140 struct seq_file *m = file->private_data;
1141 m->private = priv;
1142 } else {
1143 kfree(priv);
1144 }
1145 }
1146 return ret;
Eric W. Biederman662795d2006-06-26 00:25:48 -07001147}
1148
Arjan van de Ven00977a52007-02-12 00:55:34 -08001149const struct file_operations proc_numa_maps_operations = {
Eric W. Biederman662795d2006-06-26 00:25:48 -07001150 .open = numa_maps_open,
1151 .read = seq_read,
1152 .llseek = seq_lseek,
Eric W. Biederman99f89552006-06-26 00:25:55 -07001153 .release = seq_release_private,
Eric W. Biederman662795d2006-06-26 00:25:48 -07001154};
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001155#endif /* CONFIG_NUMA */