blob: 7dcd2a250495d9a1777e6d7992637fb0a4897a81 [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
Dave Hansen22e057c2011-03-22 16:33:00 -0700397 spin_lock(&walk->mm->page_table_lock);
398 if (pmd_trans_huge(*pmd)) {
399 if (pmd_trans_splitting(*pmd)) {
400 spin_unlock(&walk->mm->page_table_lock);
401 wait_split_huge_page(vma->anon_vma, pmd);
402 } else {
403 smaps_pte_entry(*(pte_t *)pmd, addr,
404 HPAGE_PMD_SIZE, walk);
405 spin_unlock(&walk->mm->page_table_lock);
Dave Hansen4031a212011-03-22 16:33:01 -0700406 mss->anonymous_thp += HPAGE_PMD_SIZE;
Dave Hansen22e057c2011-03-22 16:33:00 -0700407 return 0;
408 }
409 } else {
410 spin_unlock(&walk->mm->page_table_lock);
411 }
412 /*
413 * The mmap_sem held all the way back in m_start() is what
414 * keeps khugepaged out of here and from collapsing things
415 * in here.
416 */
Hugh Dickins705e87c2005-10-29 18:16:27 -0700417 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Dave Hansenae11c4d2011-03-22 16:32:58 -0700418 for (; addr != end; pte++, addr += PAGE_SIZE)
Dave Hansen3c9acc72011-03-22 16:32:59 -0700419 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700420 pte_unmap_unlock(pte - 1, ptl);
421 cond_resched();
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800422 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700423}
424
Mauricio Line070ad42005-09-03 15:55:10 -0700425static int show_smap(struct seq_file *m, void *v)
426{
Joe Korty7c88db02008-10-16 15:27:09 +0400427 struct proc_maps_private *priv = m->private;
428 struct task_struct *task = priv->task;
Mauricio Line070ad42005-09-03 15:55:10 -0700429 struct vm_area_struct *vma = v;
Mauricio Line070ad42005-09-03 15:55:10 -0700430 struct mem_size_stats mss;
Dave Hansen21650092008-06-12 15:21:47 -0700431 struct mm_walk smaps_walk = {
432 .pmd_entry = smaps_pte_range,
433 .mm = vma->vm_mm,
434 .private = &mss,
435 };
Mauricio Line070ad42005-09-03 15:55:10 -0700436
437 memset(&mss, 0, sizeof mss);
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800438 mss.vma = vma;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900439 /* mmap_sem is held in m_start */
Nick Piggin5ddfae12006-03-06 15:42:57 -0800440 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700441 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
Matt Mackall4752c362008-02-04 22:29:02 -0800442
Joe Korty7c88db02008-10-16 15:27:09 +0400443 show_map_vma(m, vma);
Matt Mackall4752c362008-02-04 22:29:02 -0800444
445 seq_printf(m,
446 "Size: %8lu kB\n"
447 "Rss: %8lu kB\n"
448 "Pss: %8lu kB\n"
449 "Shared_Clean: %8lu kB\n"
450 "Shared_Dirty: %8lu kB\n"
451 "Private_Clean: %8lu kB\n"
452 "Private_Dirty: %8lu kB\n"
Peter Zijlstra214e4712008-04-28 02:12:55 -0700453 "Referenced: %8lu kB\n"
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700454 "Anonymous: %8lu kB\n"
Dave Hansen4031a212011-03-22 16:33:01 -0700455 "AnonHugePages: %8lu kB\n"
Mel Gorman08fba692009-01-06 14:38:53 -0800456 "Swap: %8lu kB\n"
Mel Gorman33402892009-01-06 14:38:54 -0800457 "KernelPageSize: %8lu kB\n"
Nikanth Karthikesan2d905082011-01-13 15:45:53 -0800458 "MMUPageSize: %8lu kB\n"
459 "Locked: %8lu kB\n",
Matt Mackall4752c362008-02-04 22:29:02 -0800460 (vma->vm_end - vma->vm_start) >> 10,
461 mss.resident >> 10,
462 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
463 mss.shared_clean >> 10,
464 mss.shared_dirty >> 10,
465 mss.private_clean >> 10,
466 mss.private_dirty >> 10,
Peter Zijlstra214e4712008-04-28 02:12:55 -0700467 mss.referenced >> 10,
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700468 mss.anonymous >> 10,
Dave Hansen4031a212011-03-22 16:33:01 -0700469 mss.anonymous_thp >> 10,
Mel Gorman08fba692009-01-06 14:38:53 -0800470 mss.swap >> 10,
Mel Gorman33402892009-01-06 14:38:54 -0800471 vma_kernel_pagesize(vma) >> 10,
Nikanth Karthikesan2d905082011-01-13 15:45:53 -0800472 vma_mmu_pagesize(vma) >> 10,
473 (vma->vm_flags & VM_LOCKED) ?
474 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
Matt Mackall4752c362008-02-04 22:29:02 -0800475
Joe Korty7c88db02008-10-16 15:27:09 +0400476 if (m->count < m->size) /* vma is copied successfully */
Stephen Wilson31db58b2011-03-13 15:49:15 -0400477 m->version = (vma != get_gate_vma(task->mm))
478 ? vma->vm_start : 0;
Joe Korty7c88db02008-10-16 15:27:09 +0400479 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700480}
481
Jan Engelhardt03a44822008-02-08 04:21:19 -0800482static const struct seq_operations proc_pid_smaps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800483 .start = m_start,
484 .next = m_next,
485 .stop = m_stop,
486 .show = show_smap
487};
488
489static int smaps_open(struct inode *inode, struct file *file)
490{
491 return do_maps_open(inode, file, &proc_pid_smaps_op);
492}
493
494const struct file_operations proc_smaps_operations = {
495 .open = smaps_open,
496 .read = seq_read,
497 .llseek = seq_lseek,
498 .release = seq_release_private,
499};
500
501static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
Dave Hansen21650092008-06-12 15:21:47 -0700502 unsigned long end, struct mm_walk *walk)
Matt Mackalla6198792008-02-04 22:29:03 -0800503{
Dave Hansen21650092008-06-12 15:21:47 -0700504 struct vm_area_struct *vma = walk->private;
Matt Mackalla6198792008-02-04 22:29:03 -0800505 pte_t *pte, ptent;
506 spinlock_t *ptl;
507 struct page *page;
508
Dave Hansen03319322011-03-22 16:32:56 -0700509 split_huge_page_pmd(walk->mm, pmd);
510
Matt Mackalla6198792008-02-04 22:29:03 -0800511 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
512 for (; addr != end; pte++, addr += PAGE_SIZE) {
513 ptent = *pte;
514 if (!pte_present(ptent))
515 continue;
516
517 page = vm_normal_page(vma, addr, ptent);
518 if (!page)
519 continue;
520
Will Deacon85e72aa2012-01-20 14:34:09 -0800521 if (PageReserved(page))
522 continue;
523
Matt Mackalla6198792008-02-04 22:29:03 -0800524 /* Clear accessed and referenced bits. */
525 ptep_test_and_clear_young(vma, addr, pte);
526 ClearPageReferenced(page);
527 }
528 pte_unmap_unlock(pte - 1, ptl);
529 cond_resched();
530 return 0;
531}
532
Moussa A. Ba398499d2009-09-21 17:02:29 -0700533#define CLEAR_REFS_ALL 1
534#define CLEAR_REFS_ANON 2
535#define CLEAR_REFS_MAPPED 3
536
Matt Mackallf248dcb2008-02-04 22:29:03 -0800537static ssize_t clear_refs_write(struct file *file, const char __user *buf,
538 size_t count, loff_t *ppos)
David Rientjesb813e932007-05-06 14:49:24 -0700539{
Matt Mackallf248dcb2008-02-04 22:29:03 -0800540 struct task_struct *task;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700541 char buffer[PROC_NUMBUF];
Matt Mackallf248dcb2008-02-04 22:29:03 -0800542 struct mm_struct *mm;
David Rientjesb813e932007-05-06 14:49:24 -0700543 struct vm_area_struct *vma;
Alexey Dobriyan0a8cb8e2011-05-26 16:25:50 -0700544 int type;
545 int rv;
David Rientjesb813e932007-05-06 14:49:24 -0700546
Matt Mackallf248dcb2008-02-04 22:29:03 -0800547 memset(buffer, 0, sizeof(buffer));
548 if (count > sizeof(buffer) - 1)
549 count = sizeof(buffer) - 1;
550 if (copy_from_user(buffer, buf, count))
551 return -EFAULT;
Alexey Dobriyan0a8cb8e2011-05-26 16:25:50 -0700552 rv = kstrtoint(strstrip(buffer), 10, &type);
553 if (rv < 0)
554 return rv;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700555 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
Matt Mackallf248dcb2008-02-04 22:29:03 -0800556 return -EINVAL;
Matt Mackallf248dcb2008-02-04 22:29:03 -0800557 task = get_proc_task(file->f_path.dentry->d_inode);
558 if (!task)
559 return -ESRCH;
560 mm = get_task_mm(task);
561 if (mm) {
Andrew Morton20cbc972008-07-05 12:29:05 -0700562 struct mm_walk clear_refs_walk = {
563 .pmd_entry = clear_refs_pte_range,
564 .mm = mm,
565 };
Matt Mackallf248dcb2008-02-04 22:29:03 -0800566 down_read(&mm->mmap_sem);
Dave Hansen21650092008-06-12 15:21:47 -0700567 for (vma = mm->mmap; vma; vma = vma->vm_next) {
568 clear_refs_walk.private = vma;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700569 if (is_vm_hugetlb_page(vma))
570 continue;
571 /*
572 * Writing 1 to /proc/pid/clear_refs affects all pages.
573 *
574 * Writing 2 to /proc/pid/clear_refs only affects
575 * Anonymous pages.
576 *
577 * Writing 3 to /proc/pid/clear_refs only affects file
578 * mapped pages.
579 */
580 if (type == CLEAR_REFS_ANON && vma->vm_file)
581 continue;
582 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
583 continue;
584 walk_page_range(vma->vm_start, vma->vm_end,
585 &clear_refs_walk);
Dave Hansen21650092008-06-12 15:21:47 -0700586 }
Matt Mackallf248dcb2008-02-04 22:29:03 -0800587 flush_tlb_mm(mm);
588 up_read(&mm->mmap_sem);
589 mmput(mm);
590 }
591 put_task_struct(task);
Vincent Lifb92a4b2009-09-22 16:45:36 -0700592
593 return count;
David Rientjesb813e932007-05-06 14:49:24 -0700594}
595
Matt Mackallf248dcb2008-02-04 22:29:03 -0800596const struct file_operations proc_clear_refs_operations = {
597 .write = clear_refs_write,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200598 .llseek = noop_llseek,
Matt Mackallf248dcb2008-02-04 22:29:03 -0800599};
600
Matt Mackall85863e42008-02-04 22:29:04 -0800601struct pagemapread {
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900602 int pos, len;
603 u64 *buffer;
Matt Mackall85863e42008-02-04 22:29:04 -0800604};
605
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500606#define PM_ENTRY_BYTES sizeof(u64)
607#define PM_STATUS_BITS 3
608#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
609#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
610#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
611#define PM_PSHIFT_BITS 6
612#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
613#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
614#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
615#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
616#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
617
618#define PM_PRESENT PM_STATUS(4LL)
619#define PM_SWAP PM_STATUS(2LL)
620#define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
Matt Mackall85863e42008-02-04 22:29:04 -0800621#define PM_END_OF_BUFFER 1
622
623static int add_to_pagemap(unsigned long addr, u64 pfn,
624 struct pagemapread *pm)
625{
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900626 pm->buffer[pm->pos++] = pfn;
627 if (pm->pos >= pm->len)
Thomas Tuttleaae86792008-06-05 22:46:31 -0700628 return PM_END_OF_BUFFER;
Matt Mackall85863e42008-02-04 22:29:04 -0800629 return 0;
630}
631
632static int pagemap_pte_hole(unsigned long start, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700633 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800634{
Dave Hansen21650092008-06-12 15:21:47 -0700635 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800636 unsigned long addr;
637 int err = 0;
638 for (addr = start; addr < end; addr += PAGE_SIZE) {
639 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
640 if (err)
641 break;
642 }
643 return err;
644}
645
Adrian Bunk9d02dbc2008-04-28 02:12:11 -0700646static u64 swap_pte_to_pagemap_entry(pte_t pte)
Matt Mackall85863e42008-02-04 22:29:04 -0800647{
648 swp_entry_t e = pte_to_swp_entry(pte);
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500649 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
Matt Mackall85863e42008-02-04 22:29:04 -0800650}
651
Matt Mackall49c50342008-12-09 13:14:21 -0800652static u64 pte_to_pagemap_entry(pte_t pte)
Dave Hansenbcf80392008-06-12 15:21:48 -0700653{
Matt Mackall49c50342008-12-09 13:14:21 -0800654 u64 pme = 0;
Dave Hansenbcf80392008-06-12 15:21:48 -0700655 if (is_swap_pte(pte))
656 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
657 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
658 else if (pte_present(pte))
659 pme = PM_PFRAME(pte_pfn(pte))
660 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
661 return pme;
662}
663
Matt Mackall85863e42008-02-04 22:29:04 -0800664static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700665 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800666{
Dave Hansenbcf80392008-06-12 15:21:48 -0700667 struct vm_area_struct *vma;
Dave Hansen21650092008-06-12 15:21:47 -0700668 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800669 pte_t *pte;
670 int err = 0;
671
Dave Hansen03319322011-03-22 16:32:56 -0700672 split_huge_page_pmd(walk->mm, pmd);
673
Dave Hansenbcf80392008-06-12 15:21:48 -0700674 /* find the first VMA at or above 'addr' */
675 vma = find_vma(walk->mm, addr);
Matt Mackall85863e42008-02-04 22:29:04 -0800676 for (; addr != end; addr += PAGE_SIZE) {
677 u64 pfn = PM_NOT_PRESENT;
Dave Hansenbcf80392008-06-12 15:21:48 -0700678
679 /* check to see if we've left 'vma' behind
680 * and need a new, higher one */
681 if (vma && (addr >= vma->vm_end))
682 vma = find_vma(walk->mm, addr);
683
684 /* check that 'vma' actually covers this address,
685 * and that it isn't a huge page vma */
686 if (vma && (vma->vm_start <= addr) &&
687 !is_vm_hugetlb_page(vma)) {
688 pte = pte_offset_map(pmd, addr);
689 pfn = pte_to_pagemap_entry(*pte);
690 /* unmap before userspace copy */
691 pte_unmap(pte);
692 }
Matt Mackall85863e42008-02-04 22:29:04 -0800693 err = add_to_pagemap(addr, pfn, pm);
694 if (err)
695 return err;
696 }
697
698 cond_resched();
699
700 return err;
701}
702
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700703#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800704static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
705{
706 u64 pme = 0;
707 if (pte_present(pte))
708 pme = PM_PFRAME(pte_pfn(pte) + offset)
709 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
710 return pme;
711}
712
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700713/* This function walks within one hugetlb entry in the single call */
714static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
715 unsigned long addr, unsigned long end,
716 struct mm_walk *walk)
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800717{
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800718 struct pagemapread *pm = walk->private;
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800719 int err = 0;
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700720 u64 pfn;
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800721
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800722 for (; addr != end; addr += PAGE_SIZE) {
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700723 int offset = (addr & ~hmask) >> PAGE_SHIFT;
724 pfn = huge_pte_to_pagemap_entry(*pte, offset);
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800725 err = add_to_pagemap(addr, pfn, pm);
726 if (err)
727 return err;
728 }
729
730 cond_resched();
731
732 return err;
733}
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700734#endif /* HUGETLB_PAGE */
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800735
Matt Mackall85863e42008-02-04 22:29:04 -0800736/*
737 * /proc/pid/pagemap - an array mapping virtual pages to pfns
738 *
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500739 * For each page in the address space, this file contains one 64-bit entry
740 * consisting of the following:
741 *
742 * Bits 0-55 page frame number (PFN) if present
743 * Bits 0-4 swap type if swapped
744 * Bits 5-55 swap offset if swapped
745 * Bits 55-60 page shift (page size = 1<<page shift)
746 * Bit 61 reserved for future use
747 * Bit 62 page swapped
748 * Bit 63 page present
749 *
750 * If the page is not present but in swap, then the PFN contains an
751 * encoding of the swap file number and the page's offset into the
752 * swap. Unmapped pages return a null PFN. This allows determining
Matt Mackall85863e42008-02-04 22:29:04 -0800753 * precisely which pages are mapped (or in swap) and comparing mapped
754 * pages between processes.
755 *
756 * Efficient users of this interface will use /proc/pid/maps to
757 * determine which areas of memory are actually mapped and llseek to
758 * skip over unmapped regions.
759 */
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900760#define PAGEMAP_WALK_SIZE (PMD_SIZE)
Naoya Horiguchiea251c12010-11-24 12:57:13 -0800761#define PAGEMAP_WALK_MASK (PMD_MASK)
Matt Mackall85863e42008-02-04 22:29:04 -0800762static ssize_t pagemap_read(struct file *file, char __user *buf,
763 size_t count, loff_t *ppos)
764{
765 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
Matt Mackall85863e42008-02-04 22:29:04 -0800766 struct mm_struct *mm;
767 struct pagemapread pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800768 int ret = -ESRCH;
Alexey Dobriyanee1e6ab2008-07-21 14:21:36 -0700769 struct mm_walk pagemap_walk = {};
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700770 unsigned long src;
771 unsigned long svpfn;
772 unsigned long start_vaddr;
773 unsigned long end_vaddr;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900774 int copied = 0;
Matt Mackall85863e42008-02-04 22:29:04 -0800775
776 if (!task)
777 goto out;
778
Matt Mackall85863e42008-02-04 22:29:04 -0800779 ret = -EINVAL;
780 /* file position must be aligned */
Thomas Tuttleaae86792008-06-05 22:46:31 -0700781 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700782 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800783
784 ret = 0;
Vitaly Mayatskikh08161782009-04-30 15:08:18 -0700785 if (!count)
786 goto out_task;
787
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900788 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
789 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700790 ret = -ENOMEM;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900791 if (!pm.buffer)
KOSAKI Motohiro98bc93e2011-05-26 16:25:53 -0700792 goto out_task;
793
794 mm = mm_for_maps(task);
795 ret = PTR_ERR(mm);
796 if (!mm || IS_ERR(mm))
797 goto out_free;
Matt Mackall85863e42008-02-04 22:29:04 -0800798
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700799 pagemap_walk.pmd_entry = pagemap_pte_range;
800 pagemap_walk.pte_hole = pagemap_pte_hole;
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700801#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800802 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700803#endif
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700804 pagemap_walk.mm = mm;
805 pagemap_walk.private = &pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800806
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700807 src = *ppos;
808 svpfn = src / PM_ENTRY_BYTES;
809 start_vaddr = svpfn << PAGE_SHIFT;
810 end_vaddr = TASK_SIZE_OF(task);
Matt Mackall85863e42008-02-04 22:29:04 -0800811
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700812 /* watch out for wraparound */
813 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
814 start_vaddr = end_vaddr;
815
816 /*
817 * The odds are that this will stop walking way
818 * before end_vaddr, because the length of the
819 * user buffer is tracked in "pm", and the walk
820 * will stop when we hit the end of the buffer.
821 */
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900822 ret = 0;
823 while (count && (start_vaddr < end_vaddr)) {
824 int len;
825 unsigned long end;
Matt Mackall85863e42008-02-04 22:29:04 -0800826
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900827 pm.pos = 0;
Naoya Horiguchiea251c12010-11-24 12:57:13 -0800828 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900829 /* overflow ? */
830 if (end < start_vaddr || end > end_vaddr)
831 end = end_vaddr;
832 down_read(&mm->mmap_sem);
833 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
834 up_read(&mm->mmap_sem);
835 start_vaddr = end;
836
837 len = min(count, PM_ENTRY_BYTES * pm.pos);
Dan Carpenter309361e02010-04-06 13:45:39 +0300838 if (copy_to_user(buf, pm.buffer, len)) {
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900839 ret = -EFAULT;
KOSAKI Motohiro98bc93e2011-05-26 16:25:53 -0700840 goto out_mm;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900841 }
842 copied += len;
843 buf += len;
844 count -= len;
Matt Mackall85863e42008-02-04 22:29:04 -0800845 }
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900846 *ppos += copied;
847 if (!ret || ret == PM_END_OF_BUFFER)
848 ret = copied;
849
Marcelo Tosattifb393802008-03-13 12:32:35 -0700850out_mm:
851 mmput(mm);
KOSAKI Motohiro98bc93e2011-05-26 16:25:53 -0700852out_free:
853 kfree(pm.buffer);
Matt Mackall85863e42008-02-04 22:29:04 -0800854out_task:
855 put_task_struct(task);
856out:
857 return ret;
858}
859
860const struct file_operations proc_pagemap_operations = {
861 .llseek = mem_lseek, /* borrow this */
862 .read = pagemap_read,
863};
Matt Mackall1e883282008-02-04 22:29:07 -0800864#endif /* CONFIG_PROC_PAGE_MONITOR */
Matt Mackall85863e42008-02-04 22:29:04 -0800865
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700866#ifdef CONFIG_NUMA
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700867
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700868struct numa_maps {
869 struct vm_area_struct *vma;
870 unsigned long pages;
871 unsigned long anon;
872 unsigned long active;
873 unsigned long writeback;
874 unsigned long mapcount_max;
875 unsigned long dirty;
876 unsigned long swapcache;
877 unsigned long node[MAX_NUMNODES];
878};
879
Stephen Wilson5b52fc82011-05-24 17:12:49 -0700880struct numa_maps_private {
881 struct proc_maps_private proc_maps;
882 struct numa_maps md;
883};
884
Dave Hanseneb4866d2011-09-20 15:19:38 -0700885static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
886 unsigned long nr_pages)
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700887{
888 int count = page_mapcount(page);
889
Dave Hanseneb4866d2011-09-20 15:19:38 -0700890 md->pages += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700891 if (pte_dirty || PageDirty(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -0700892 md->dirty += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700893
894 if (PageSwapCache(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -0700895 md->swapcache += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700896
897 if (PageActive(page) || PageUnevictable(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -0700898 md->active += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700899
900 if (PageWriteback(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -0700901 md->writeback += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700902
903 if (PageAnon(page))
Dave Hanseneb4866d2011-09-20 15:19:38 -0700904 md->anon += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700905
906 if (count > md->mapcount_max)
907 md->mapcount_max = count;
908
Dave Hanseneb4866d2011-09-20 15:19:38 -0700909 md->node[page_to_nid(page)] += nr_pages;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700910}
911
Dave Hansen3200a8a2011-09-20 15:19:39 -0700912static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
913 unsigned long addr)
914{
915 struct page *page;
916 int nid;
917
918 if (!pte_present(pte))
919 return NULL;
920
921 page = vm_normal_page(vma, addr, pte);
922 if (!page)
923 return NULL;
924
925 if (PageReserved(page))
926 return NULL;
927
928 nid = page_to_nid(page);
929 if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
930 return NULL;
931
932 return page;
933}
934
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700935static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
936 unsigned long end, struct mm_walk *walk)
937{
938 struct numa_maps *md;
939 spinlock_t *ptl;
940 pte_t *orig_pte;
941 pte_t *pte;
942
943 md = walk->private;
Dave Hansen32ef4382011-09-20 15:19:41 -0700944 spin_lock(&walk->mm->page_table_lock);
945 if (pmd_trans_huge(*pmd)) {
946 if (pmd_trans_splitting(*pmd)) {
947 spin_unlock(&walk->mm->page_table_lock);
948 wait_split_huge_page(md->vma->anon_vma, pmd);
949 } else {
950 pte_t huge_pte = *(pte_t *)pmd;
951 struct page *page;
952
953 page = can_gather_numa_stats(huge_pte, md->vma, addr);
954 if (page)
955 gather_stats(page, md, pte_dirty(huge_pte),
956 HPAGE_PMD_SIZE/PAGE_SIZE);
957 spin_unlock(&walk->mm->page_table_lock);
958 return 0;
959 }
960 } else {
961 spin_unlock(&walk->mm->page_table_lock);
962 }
963
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700964 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
965 do {
Dave Hansen3200a8a2011-09-20 15:19:39 -0700966 struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700967 if (!page)
968 continue;
Dave Hanseneb4866d2011-09-20 15:19:38 -0700969 gather_stats(page, md, pte_dirty(*pte), 1);
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700970
971 } while (pte++, addr += PAGE_SIZE, addr != end);
972 pte_unmap_unlock(orig_pte, ptl);
973 return 0;
974}
975#ifdef CONFIG_HUGETLB_PAGE
976static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
977 unsigned long addr, unsigned long end, struct mm_walk *walk)
978{
979 struct numa_maps *md;
980 struct page *page;
981
982 if (pte_none(*pte))
983 return 0;
984
985 page = pte_page(*pte);
986 if (!page)
987 return 0;
988
989 md = walk->private;
Dave Hanseneb4866d2011-09-20 15:19:38 -0700990 gather_stats(page, md, pte_dirty(*pte), 1);
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700991 return 0;
992}
993
994#else
995static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
996 unsigned long addr, unsigned long end, struct mm_walk *walk)
997{
998 return 0;
999}
1000#endif
1001
1002/*
1003 * Display pages allocated per node and memory policy via /proc.
1004 */
1005static int show_numa_map(struct seq_file *m, void *v)
1006{
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001007 struct numa_maps_private *numa_priv = m->private;
1008 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001009 struct vm_area_struct *vma = v;
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001010 struct numa_maps *md = &numa_priv->md;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001011 struct file *file = vma->vm_file;
1012 struct mm_struct *mm = vma->vm_mm;
1013 struct mm_walk walk = {};
1014 struct mempolicy *pol;
1015 int n;
1016 char buffer[50];
1017
1018 if (!mm)
1019 return 0;
1020
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001021 /* Ensure we start with an empty set of numa_maps statistics. */
1022 memset(md, 0, sizeof(*md));
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001023
1024 md->vma = vma;
1025
1026 walk.hugetlb_entry = gather_hugetbl_stats;
1027 walk.pmd_entry = gather_pte_stats;
1028 walk.private = md;
1029 walk.mm = mm;
1030
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001031 pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001032 mpol_to_str(buffer, sizeof(buffer), pol, 0);
1033 mpol_cond_put(pol);
1034
1035 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1036
1037 if (file) {
1038 seq_printf(m, " file=");
1039 seq_path(m, &file->f_path, "\n\t= ");
1040 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1041 seq_printf(m, " heap");
1042 } else if (vma->vm_start <= mm->start_stack &&
1043 vma->vm_end >= mm->start_stack) {
1044 seq_printf(m, " stack");
1045 }
1046
Andrew Mortonfc360bd2011-10-31 17:06:32 -07001047 if (is_vm_hugetlb_page(vma))
1048 seq_printf(m, " huge");
1049
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001050 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1051
1052 if (!md->pages)
1053 goto out;
1054
1055 if (md->anon)
1056 seq_printf(m, " anon=%lu", md->anon);
1057
1058 if (md->dirty)
1059 seq_printf(m, " dirty=%lu", md->dirty);
1060
1061 if (md->pages != md->anon && md->pages != md->dirty)
1062 seq_printf(m, " mapped=%lu", md->pages);
1063
1064 if (md->mapcount_max > 1)
1065 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1066
1067 if (md->swapcache)
1068 seq_printf(m, " swapcache=%lu", md->swapcache);
1069
1070 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1071 seq_printf(m, " active=%lu", md->active);
1072
1073 if (md->writeback)
1074 seq_printf(m, " writeback=%lu", md->writeback);
1075
1076 for_each_node_state(n, N_HIGH_MEMORY)
1077 if (md->node[n])
1078 seq_printf(m, " N%d=%lu", n, md->node[n]);
1079out:
1080 seq_putc(m, '\n');
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001081
1082 if (m->count < m->size)
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001083 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001084 return 0;
1085}
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001086
Jan Engelhardt03a44822008-02-08 04:21:19 -08001087static const struct seq_operations proc_pid_numa_maps_op = {
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08001088 .start = m_start,
1089 .next = m_next,
1090 .stop = m_stop,
Alexey Dobriyan3bbfe052008-10-10 03:27:16 +04001091 .show = show_numa_map,
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001092};
Eric W. Biederman662795d2006-06-26 00:25:48 -07001093
1094static int numa_maps_open(struct inode *inode, struct file *file)
1095{
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001096 struct numa_maps_private *priv;
1097 int ret = -ENOMEM;
1098 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1099 if (priv) {
1100 priv->proc_maps.pid = proc_pid(inode);
1101 ret = seq_open(file, &proc_pid_numa_maps_op);
1102 if (!ret) {
1103 struct seq_file *m = file->private_data;
1104 m->private = priv;
1105 } else {
1106 kfree(priv);
1107 }
1108 }
1109 return ret;
Eric W. Biederman662795d2006-06-26 00:25:48 -07001110}
1111
Arjan van de Ven00977a52007-02-12 00:55:34 -08001112const struct file_operations proc_numa_maps_operations = {
Eric W. Biederman662795d2006-06-26 00:25:48 -07001113 .open = numa_maps_open,
1114 .read = seq_read,
1115 .llseek = seq_lseek,
Eric W. Biederman99f89552006-06-26 00:25:55 -07001116 .release = seq_release_private,
Eric W. Biederman662795d2006-06-26 00:25:48 -07001117};
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001118#endif /* CONFIG_NUMA */