]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | #include <linux/mm.h> |
615d6e87 | 2 | #include <linux/vmacache.h> |
1da177e4 | 3 | #include <linux/hugetlb.h> |
22e057c5 | 4 | #include <linux/huge_mm.h> |
1da177e4 LT |
5 | #include <linux/mount.h> |
6 | #include <linux/seq_file.h> | |
e070ad49 | 7 | #include <linux/highmem.h> |
5096add8 | 8 | #include <linux/ptrace.h> |
5a0e3ad6 | 9 | #include <linux/slab.h> |
6e21c8f1 CL |
10 | #include <linux/pagemap.h> |
11 | #include <linux/mempolicy.h> | |
22e057c5 | 12 | #include <linux/rmap.h> |
85863e47 MM |
13 | #include <linux/swap.h> |
14 | #include <linux/swapops.h> | |
0f8975ec | 15 | #include <linux/mmu_notifier.h> |
33c3fc71 | 16 | #include <linux/page_idle.h> |
6a15a370 | 17 | #include <linux/shmem_fs.h> |
e070ad49 | 18 | |
1da177e4 LT |
19 | #include <asm/elf.h> |
20 | #include <asm/uaccess.h> | |
e070ad49 | 21 | #include <asm/tlbflush.h> |
1da177e4 LT |
22 | #include "internal.h" |
23 | ||
df5f8314 | 24 | void task_mem(struct seq_file *m, struct mm_struct *mm) |
1da177e4 | 25 | { |
dc6c9a35 | 26 | unsigned long data, text, lib, swap, ptes, pmds; |
365e9c87 HD |
27 | unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; |
28 | ||
29 | /* | |
30 | * Note: to minimize their overhead, mm maintains hiwater_vm and | |
31 | * hiwater_rss only when about to *lower* total_vm or rss. Any | |
32 | * collector of these hiwater stats must therefore get total_vm | |
33 | * and rss too, which will usually be the higher. Barriers? not | |
34 | * worth the effort, such snapshots can always be inconsistent. | |
35 | */ | |
36 | hiwater_vm = total_vm = mm->total_vm; | |
37 | if (hiwater_vm < mm->hiwater_vm) | |
38 | hiwater_vm = mm->hiwater_vm; | |
39 | hiwater_rss = total_rss = get_mm_rss(mm); | |
40 | if (hiwater_rss < mm->hiwater_rss) | |
41 | hiwater_rss = mm->hiwater_rss; | |
1da177e4 LT |
42 | |
43 | data = mm->total_vm - mm->shared_vm - mm->stack_vm; | |
44 | text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10; | |
45 | lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text; | |
b084d435 | 46 | swap = get_mm_counter(mm, MM_SWAPENTS); |
dc6c9a35 KS |
47 | ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes); |
48 | pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm); | |
df5f8314 | 49 | seq_printf(m, |
365e9c87 | 50 | "VmPeak:\t%8lu kB\n" |
1da177e4 LT |
51 | "VmSize:\t%8lu kB\n" |
52 | "VmLck:\t%8lu kB\n" | |
bc3e53f6 | 53 | "VmPin:\t%8lu kB\n" |
365e9c87 | 54 | "VmHWM:\t%8lu kB\n" |
1da177e4 LT |
55 | "VmRSS:\t%8lu kB\n" |
56 | "VmData:\t%8lu kB\n" | |
57 | "VmStk:\t%8lu kB\n" | |
58 | "VmExe:\t%8lu kB\n" | |
59 | "VmLib:\t%8lu kB\n" | |
b084d435 | 60 | "VmPTE:\t%8lu kB\n" |
dc6c9a35 | 61 | "VmPMD:\t%8lu kB\n" |
b084d435 | 62 | "VmSwap:\t%8lu kB\n", |
365e9c87 | 63 | hiwater_vm << (PAGE_SHIFT-10), |
314e51b9 | 64 | total_vm << (PAGE_SHIFT-10), |
1da177e4 | 65 | mm->locked_vm << (PAGE_SHIFT-10), |
bc3e53f6 | 66 | mm->pinned_vm << (PAGE_SHIFT-10), |
365e9c87 HD |
67 | hiwater_rss << (PAGE_SHIFT-10), |
68 | total_rss << (PAGE_SHIFT-10), | |
1da177e4 LT |
69 | data << (PAGE_SHIFT-10), |
70 | mm->stack_vm << (PAGE_SHIFT-10), text, lib, | |
dc6c9a35 KS |
71 | ptes >> 10, |
72 | pmds >> 10, | |
b084d435 | 73 | swap << (PAGE_SHIFT-10)); |
5d317b2b | 74 | hugetlb_report_usage(m, mm); |
1da177e4 LT |
75 | } |
76 | ||
77 | unsigned long task_vsize(struct mm_struct *mm) | |
78 | { | |
79 | return PAGE_SIZE * mm->total_vm; | |
80 | } | |
81 | ||
a2ade7b6 AD |
82 | unsigned long task_statm(struct mm_struct *mm, |
83 | unsigned long *shared, unsigned long *text, | |
84 | unsigned long *data, unsigned long *resident) | |
1da177e4 | 85 | { |
d559db08 | 86 | *shared = get_mm_counter(mm, MM_FILEPAGES); |
1da177e4 LT |
87 | *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) |
88 | >> PAGE_SHIFT; | |
89 | *data = mm->total_vm - mm->shared_vm; | |
d559db08 | 90 | *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); |
1da177e4 LT |
91 | return mm->total_vm; |
92 | } | |
93 | ||
9e781440 KH |
94 | #ifdef CONFIG_NUMA |
95 | /* | |
498f2371 | 96 | * Save get_task_policy() for show_numa_map(). |
9e781440 KH |
97 | */ |
98 | static void hold_task_mempolicy(struct proc_maps_private *priv) | |
99 | { | |
100 | struct task_struct *task = priv->task; | |
101 | ||
102 | task_lock(task); | |
498f2371 | 103 | priv->task_mempolicy = get_task_policy(task); |
9e781440 KH |
104 | mpol_get(priv->task_mempolicy); |
105 | task_unlock(task); | |
106 | } | |
107 | static void release_task_mempolicy(struct proc_maps_private *priv) | |
108 | { | |
109 | mpol_put(priv->task_mempolicy); | |
110 | } | |
111 | #else | |
112 | static void hold_task_mempolicy(struct proc_maps_private *priv) | |
113 | { | |
114 | } | |
115 | static void release_task_mempolicy(struct proc_maps_private *priv) | |
116 | { | |
117 | } | |
118 | #endif | |
119 | ||
59b4bf12 | 120 | static void vma_stop(struct proc_maps_private *priv) |
a6198797 | 121 | { |
59b4bf12 ON |
122 | struct mm_struct *mm = priv->mm; |
123 | ||
124 | release_task_mempolicy(priv); | |
125 | up_read(&mm->mmap_sem); | |
126 | mmput(mm); | |
a6198797 | 127 | } |
ec4dd3eb | 128 | |
ad2a00e4 ON |
129 | static struct vm_area_struct * |
130 | m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma) | |
131 | { | |
132 | if (vma == priv->tail_vma) | |
133 | return NULL; | |
134 | return vma->vm_next ?: priv->tail_vma; | |
135 | } | |
136 | ||
b8c20a9b ON |
137 | static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma) |
138 | { | |
139 | if (m->count < m->size) /* vma is copied successfully */ | |
140 | m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL; | |
141 | } | |
142 | ||
0c255321 | 143 | static void *m_start(struct seq_file *m, loff_t *ppos) |
e070ad49 | 144 | { |
a6198797 | 145 | struct proc_maps_private *priv = m->private; |
b8c20a9b | 146 | unsigned long last_addr = m->version; |
a6198797 | 147 | struct mm_struct *mm; |
0c255321 ON |
148 | struct vm_area_struct *vma; |
149 | unsigned int pos = *ppos; | |
a6198797 | 150 | |
b8c20a9b ON |
151 | /* See m_cache_vma(). Zero at the start or after lseek. */ |
152 | if (last_addr == -1UL) | |
153 | return NULL; | |
154 | ||
2c03376d | 155 | priv->task = get_proc_task(priv->inode); |
a6198797 | 156 | if (!priv->task) |
ec6fd8a4 | 157 | return ERR_PTR(-ESRCH); |
a6198797 | 158 | |
29a40ace ON |
159 | mm = priv->mm; |
160 | if (!mm || !atomic_inc_not_zero(&mm->mm_users)) | |
161 | return NULL; | |
a6198797 | 162 | |
0c255321 | 163 | down_read(&mm->mmap_sem); |
9e781440 | 164 | hold_task_mempolicy(priv); |
0c255321 | 165 | priv->tail_vma = get_gate_vma(mm); |
a6198797 | 166 | |
b8c20a9b ON |
167 | if (last_addr) { |
168 | vma = find_vma(mm, last_addr); | |
169 | if (vma && (vma = m_next_vma(priv, vma))) | |
170 | return vma; | |
171 | } | |
172 | ||
173 | m->version = 0; | |
0c255321 | 174 | if (pos < mm->map_count) { |
557c2d8a ON |
175 | for (vma = mm->mmap; pos; pos--) { |
176 | m->version = vma->vm_start; | |
a6198797 | 177 | vma = vma->vm_next; |
557c2d8a | 178 | } |
a6198797 | 179 | return vma; |
0c255321 | 180 | } |
a6198797 | 181 | |
557c2d8a | 182 | /* we do not bother to update m->version in this case */ |
0c255321 ON |
183 | if (pos == mm->map_count && priv->tail_vma) |
184 | return priv->tail_vma; | |
59b4bf12 ON |
185 | |
186 | vma_stop(priv); | |
187 | return NULL; | |
a6198797 MM |
188 | } |
189 | ||
190 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
191 | { | |
192 | struct proc_maps_private *priv = m->private; | |
ad2a00e4 | 193 | struct vm_area_struct *next; |
a6198797 MM |
194 | |
195 | (*pos)++; | |
ad2a00e4 | 196 | next = m_next_vma(priv, v); |
59b4bf12 ON |
197 | if (!next) |
198 | vma_stop(priv); | |
199 | return next; | |
a6198797 MM |
200 | } |
201 | ||
202 | static void m_stop(struct seq_file *m, void *v) | |
203 | { | |
204 | struct proc_maps_private *priv = m->private; | |
a6198797 | 205 | |
59b4bf12 ON |
206 | if (!IS_ERR_OR_NULL(v)) |
207 | vma_stop(priv); | |
0d5f5f45 | 208 | if (priv->task) { |
a6198797 | 209 | put_task_struct(priv->task); |
0d5f5f45 ON |
210 | priv->task = NULL; |
211 | } | |
a6198797 MM |
212 | } |
213 | ||
4db7d0ee ON |
214 | static int proc_maps_open(struct inode *inode, struct file *file, |
215 | const struct seq_operations *ops, int psize) | |
216 | { | |
217 | struct proc_maps_private *priv = __seq_open_private(file, ops, psize); | |
218 | ||
219 | if (!priv) | |
220 | return -ENOMEM; | |
221 | ||
2c03376d | 222 | priv->inode = inode; |
29a40ace ON |
223 | priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); |
224 | if (IS_ERR(priv->mm)) { | |
225 | int err = PTR_ERR(priv->mm); | |
226 | ||
227 | seq_release_private(inode, file); | |
228 | return err; | |
229 | } | |
230 | ||
4db7d0ee ON |
231 | return 0; |
232 | } | |
233 | ||
29a40ace ON |
234 | static int proc_map_release(struct inode *inode, struct file *file) |
235 | { | |
236 | struct seq_file *seq = file->private_data; | |
237 | struct proc_maps_private *priv = seq->private; | |
238 | ||
239 | if (priv->mm) | |
240 | mmdrop(priv->mm); | |
241 | ||
242 | return seq_release_private(inode, file); | |
243 | } | |
244 | ||
a6198797 | 245 | static int do_maps_open(struct inode *inode, struct file *file, |
03a44825 | 246 | const struct seq_operations *ops) |
a6198797 | 247 | { |
4db7d0ee ON |
248 | return proc_maps_open(inode, file, ops, |
249 | sizeof(struct proc_maps_private)); | |
a6198797 | 250 | } |
e070ad49 | 251 | |
58cb6548 ON |
252 | static pid_t pid_of_stack(struct proc_maps_private *priv, |
253 | struct vm_area_struct *vma, bool is_pid) | |
254 | { | |
255 | struct inode *inode = priv->inode; | |
256 | struct task_struct *task; | |
257 | pid_t ret = 0; | |
258 | ||
259 | rcu_read_lock(); | |
260 | task = pid_task(proc_pid(inode), PIDTYPE_PID); | |
261 | if (task) { | |
262 | task = task_of_stack(task, vma, is_pid); | |
263 | if (task) | |
264 | ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info); | |
265 | } | |
266 | rcu_read_unlock(); | |
267 | ||
268 | return ret; | |
269 | } | |
270 | ||
b7643757 SP |
271 | static void |
272 | show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid) | |
1da177e4 | 273 | { |
e070ad49 ML |
274 | struct mm_struct *mm = vma->vm_mm; |
275 | struct file *file = vma->vm_file; | |
b7643757 | 276 | struct proc_maps_private *priv = m->private; |
ca16d140 | 277 | vm_flags_t flags = vma->vm_flags; |
1da177e4 | 278 | unsigned long ino = 0; |
6260a4b0 | 279 | unsigned long long pgoff = 0; |
a09a79f6 | 280 | unsigned long start, end; |
1da177e4 | 281 | dev_t dev = 0; |
b7643757 | 282 | const char *name = NULL; |
1da177e4 LT |
283 | |
284 | if (file) { | |
496ad9aa | 285 | struct inode *inode = file_inode(vma->vm_file); |
1da177e4 LT |
286 | dev = inode->i_sb->s_dev; |
287 | ino = inode->i_ino; | |
6260a4b0 | 288 | pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; |
1da177e4 LT |
289 | } |
290 | ||
d7824370 LT |
291 | /* We don't show the stack guard page in /proc/maps */ |
292 | start = vma->vm_start; | |
a09a79f6 MP |
293 | if (stack_guard_page_start(vma, start)) |
294 | start += PAGE_SIZE; | |
295 | end = vma->vm_end; | |
296 | if (stack_guard_page_end(vma, end)) | |
297 | end -= PAGE_SIZE; | |
d7824370 | 298 | |
652586df TH |
299 | seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); |
300 | seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ", | |
d7824370 | 301 | start, |
a09a79f6 | 302 | end, |
1da177e4 LT |
303 | flags & VM_READ ? 'r' : '-', |
304 | flags & VM_WRITE ? 'w' : '-', | |
305 | flags & VM_EXEC ? 'x' : '-', | |
306 | flags & VM_MAYSHARE ? 's' : 'p', | |
6260a4b0 | 307 | pgoff, |
652586df | 308 | MAJOR(dev), MINOR(dev), ino); |
1da177e4 LT |
309 | |
310 | /* | |
311 | * Print the dentry name for named mappings, and a | |
312 | * special [heap] marker for the heap: | |
313 | */ | |
e070ad49 | 314 | if (file) { |
652586df | 315 | seq_pad(m, ' '); |
2726d566 | 316 | seq_file_path(m, file, "\n"); |
b7643757 SP |
317 | goto done; |
318 | } | |
319 | ||
78d683e8 AL |
320 | if (vma->vm_ops && vma->vm_ops->name) { |
321 | name = vma->vm_ops->name(vma); | |
322 | if (name) | |
323 | goto done; | |
324 | } | |
325 | ||
b7643757 SP |
326 | name = arch_vma_name(vma); |
327 | if (!name) { | |
328 | pid_t tid; | |
329 | ||
330 | if (!mm) { | |
331 | name = "[vdso]"; | |
332 | goto done; | |
333 | } | |
334 | ||
335 | if (vma->vm_start <= mm->brk && | |
336 | vma->vm_end >= mm->start_brk) { | |
337 | name = "[heap]"; | |
338 | goto done; | |
339 | } | |
340 | ||
58cb6548 | 341 | tid = pid_of_stack(priv, vma, is_pid); |
b7643757 SP |
342 | if (tid != 0) { |
343 | /* | |
344 | * Thread stack in /proc/PID/task/TID/maps or | |
345 | * the main process stack. | |
346 | */ | |
347 | if (!is_pid || (vma->vm_start <= mm->start_stack && | |
348 | vma->vm_end >= mm->start_stack)) { | |
349 | name = "[stack]"; | |
e6e5494c | 350 | } else { |
b7643757 | 351 | /* Thread stack in /proc/PID/maps */ |
652586df | 352 | seq_pad(m, ' '); |
b7643757 | 353 | seq_printf(m, "[stack:%d]", tid); |
1da177e4 | 354 | } |
e6e5494c | 355 | } |
b7643757 SP |
356 | } |
357 | ||
358 | done: | |
359 | if (name) { | |
652586df | 360 | seq_pad(m, ' '); |
b7643757 | 361 | seq_puts(m, name); |
1da177e4 LT |
362 | } |
363 | seq_putc(m, '\n'); | |
7c88db0c JK |
364 | } |
365 | ||
b7643757 | 366 | static int show_map(struct seq_file *m, void *v, int is_pid) |
7c88db0c | 367 | { |
ebb6cdde | 368 | show_map_vma(m, v, is_pid); |
b8c20a9b | 369 | m_cache_vma(m, v); |
1da177e4 LT |
370 | return 0; |
371 | } | |
372 | ||
b7643757 SP |
373 | static int show_pid_map(struct seq_file *m, void *v) |
374 | { | |
375 | return show_map(m, v, 1); | |
376 | } | |
377 | ||
378 | static int show_tid_map(struct seq_file *m, void *v) | |
379 | { | |
380 | return show_map(m, v, 0); | |
381 | } | |
382 | ||
03a44825 | 383 | static const struct seq_operations proc_pid_maps_op = { |
a6198797 MM |
384 | .start = m_start, |
385 | .next = m_next, | |
386 | .stop = m_stop, | |
b7643757 SP |
387 | .show = show_pid_map |
388 | }; | |
389 | ||
390 | static const struct seq_operations proc_tid_maps_op = { | |
391 | .start = m_start, | |
392 | .next = m_next, | |
393 | .stop = m_stop, | |
394 | .show = show_tid_map | |
a6198797 MM |
395 | }; |
396 | ||
b7643757 | 397 | static int pid_maps_open(struct inode *inode, struct file *file) |
a6198797 MM |
398 | { |
399 | return do_maps_open(inode, file, &proc_pid_maps_op); | |
400 | } | |
401 | ||
b7643757 SP |
402 | static int tid_maps_open(struct inode *inode, struct file *file) |
403 | { | |
404 | return do_maps_open(inode, file, &proc_tid_maps_op); | |
405 | } | |
406 | ||
407 | const struct file_operations proc_pid_maps_operations = { | |
408 | .open = pid_maps_open, | |
409 | .read = seq_read, | |
410 | .llseek = seq_lseek, | |
29a40ace | 411 | .release = proc_map_release, |
b7643757 SP |
412 | }; |
413 | ||
414 | const struct file_operations proc_tid_maps_operations = { | |
415 | .open = tid_maps_open, | |
a6198797 MM |
416 | .read = seq_read, |
417 | .llseek = seq_lseek, | |
29a40ace | 418 | .release = proc_map_release, |
a6198797 MM |
419 | }; |
420 | ||
421 | /* | |
422 | * Proportional Set Size(PSS): my share of RSS. | |
423 | * | |
424 | * PSS of a process is the count of pages it has in memory, where each | |
425 | * page is divided by the number of processes sharing it. So if a | |
426 | * process has 1000 pages all to itself, and 1000 shared with one other | |
427 | * process, its PSS will be 1500. | |
428 | * | |
429 | * To keep (accumulated) division errors low, we adopt a 64bit | |
430 | * fixed-point pss counter to minimize division errors. So (pss >> | |
431 | * PSS_SHIFT) would be the real byte count. | |
432 | * | |
433 | * A shift of 12 before division means (assuming 4K page size): | |
434 | * - 1M 3-user-pages add up to 8KB errors; | |
435 | * - supports mapcount up to 2^24, or 16M; | |
436 | * - supports PSS up to 2^52 bytes, or 4PB. | |
437 | */ | |
438 | #define PSS_SHIFT 12 | |
439 | ||
1e883281 | 440 | #ifdef CONFIG_PROC_PAGE_MONITOR |
214e471f | 441 | struct mem_size_stats { |
a6198797 MM |
442 | unsigned long resident; |
443 | unsigned long shared_clean; | |
444 | unsigned long shared_dirty; | |
445 | unsigned long private_clean; | |
446 | unsigned long private_dirty; | |
447 | unsigned long referenced; | |
b40d4f84 | 448 | unsigned long anonymous; |
4031a219 | 449 | unsigned long anonymous_thp; |
214e471f | 450 | unsigned long swap; |
25ee01a2 NH |
451 | unsigned long shared_hugetlb; |
452 | unsigned long private_hugetlb; | |
a6198797 | 453 | u64 pss; |
8334b962 | 454 | u64 swap_pss; |
c261e7d9 | 455 | bool check_shmem_swap; |
a6198797 MM |
456 | }; |
457 | ||
c164e038 KS |
458 | static void smaps_account(struct mem_size_stats *mss, struct page *page, |
459 | unsigned long size, bool young, bool dirty) | |
460 | { | |
461 | int mapcount; | |
462 | ||
463 | if (PageAnon(page)) | |
464 | mss->anonymous += size; | |
465 | ||
466 | mss->resident += size; | |
467 | /* Accumulate the size in pages that have been accessed. */ | |
33c3fc71 | 468 | if (young || page_is_young(page) || PageReferenced(page)) |
c164e038 KS |
469 | mss->referenced += size; |
470 | mapcount = page_mapcount(page); | |
471 | if (mapcount >= 2) { | |
472 | u64 pss_delta; | |
473 | ||
474 | if (dirty || PageDirty(page)) | |
475 | mss->shared_dirty += size; | |
476 | else | |
477 | mss->shared_clean += size; | |
478 | pss_delta = (u64)size << PSS_SHIFT; | |
479 | do_div(pss_delta, mapcount); | |
480 | mss->pss += pss_delta; | |
481 | } else { | |
482 | if (dirty || PageDirty(page)) | |
483 | mss->private_dirty += size; | |
484 | else | |
485 | mss->private_clean += size; | |
486 | mss->pss += (u64)size << PSS_SHIFT; | |
487 | } | |
488 | } | |
ae11c4d9 | 489 | |
c261e7d9 | 490 | #ifdef CONFIG_SHMEM |
c261e7d9 VB |
491 | static int smaps_pte_hole(unsigned long addr, unsigned long end, |
492 | struct mm_walk *walk) | |
493 | { | |
494 | struct mem_size_stats *mss = walk->private; | |
495 | ||
48131e03 VB |
496 | mss->swap += shmem_partial_swap_usage( |
497 | walk->vma->vm_file->f_mapping, addr, end); | |
c261e7d9 VB |
498 | |
499 | return 0; | |
500 | } | |
c261e7d9 VB |
501 | #endif |
502 | ||
c164e038 KS |
503 | static void smaps_pte_entry(pte_t *pte, unsigned long addr, |
504 | struct mm_walk *walk) | |
ae11c4d9 DH |
505 | { |
506 | struct mem_size_stats *mss = walk->private; | |
14eb6fdd | 507 | struct vm_area_struct *vma = walk->vma; |
b1d4d9e0 | 508 | struct page *page = NULL; |
ae11c4d9 | 509 | |
c164e038 KS |
510 | if (pte_present(*pte)) { |
511 | page = vm_normal_page(vma, addr, *pte); | |
512 | } else if (is_swap_pte(*pte)) { | |
513 | swp_entry_t swpent = pte_to_swp_entry(*pte); | |
ae11c4d9 | 514 | |
8334b962 MK |
515 | if (!non_swap_entry(swpent)) { |
516 | int mapcount; | |
517 | ||
c164e038 | 518 | mss->swap += PAGE_SIZE; |
8334b962 MK |
519 | mapcount = swp_swapcount(swpent); |
520 | if (mapcount >= 2) { | |
521 | u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; | |
522 | ||
523 | do_div(pss_delta, mapcount); | |
524 | mss->swap_pss += pss_delta; | |
525 | } else { | |
526 | mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; | |
527 | } | |
528 | } else if (is_migration_entry(swpent)) | |
b1d4d9e0 | 529 | page = migration_entry_to_page(swpent); |
c261e7d9 VB |
530 | } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap |
531 | && pte_none(*pte))) { | |
48131e03 VB |
532 | page = find_get_entry(vma->vm_file->f_mapping, |
533 | linear_page_index(vma, addr)); | |
534 | if (!page) | |
535 | return; | |
536 | ||
537 | if (radix_tree_exceptional_entry(page)) | |
538 | mss->swap += PAGE_SIZE; | |
539 | else | |
540 | page_cache_release(page); | |
541 | ||
542 | return; | |
b1d4d9e0 | 543 | } |
ae11c4d9 | 544 | |
ae11c4d9 DH |
545 | if (!page) |
546 | return; | |
c164e038 | 547 | smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte)); |
ae11c4d9 DH |
548 | } |
549 | ||
c164e038 KS |
550 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
551 | static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, | |
552 | struct mm_walk *walk) | |
553 | { | |
554 | struct mem_size_stats *mss = walk->private; | |
14eb6fdd | 555 | struct vm_area_struct *vma = walk->vma; |
c164e038 KS |
556 | struct page *page; |
557 | ||
558 | /* FOLL_DUMP will return -EFAULT on huge zero page */ | |
559 | page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); | |
560 | if (IS_ERR_OR_NULL(page)) | |
561 | return; | |
562 | mss->anonymous_thp += HPAGE_PMD_SIZE; | |
563 | smaps_account(mss, page, HPAGE_PMD_SIZE, | |
564 | pmd_young(*pmd), pmd_dirty(*pmd)); | |
565 | } | |
566 | #else | |
567 | static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, | |
568 | struct mm_walk *walk) | |
569 | { | |
570 | } | |
571 | #endif | |
572 | ||
b3ae5acb | 573 | static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
2165009b | 574 | struct mm_walk *walk) |
e070ad49 | 575 | { |
14eb6fdd | 576 | struct vm_area_struct *vma = walk->vma; |
ae11c4d9 | 577 | pte_t *pte; |
705e87c0 | 578 | spinlock_t *ptl; |
e070ad49 | 579 | |
bf929152 | 580 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
c164e038 | 581 | smaps_pmd_entry(pmd, addr, walk); |
bf929152 | 582 | spin_unlock(ptl); |
025c5b24 | 583 | return 0; |
22e057c5 | 584 | } |
1a5a9906 AA |
585 | |
586 | if (pmd_trans_unstable(pmd)) | |
587 | return 0; | |
22e057c5 DH |
588 | /* |
589 | * The mmap_sem held all the way back in m_start() is what | |
590 | * keeps khugepaged out of here and from collapsing things | |
591 | * in here. | |
592 | */ | |
705e87c0 | 593 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
ae11c4d9 | 594 | for (; addr != end; pte++, addr += PAGE_SIZE) |
c164e038 | 595 | smaps_pte_entry(pte, addr, walk); |
705e87c0 HD |
596 | pte_unmap_unlock(pte - 1, ptl); |
597 | cond_resched(); | |
b3ae5acb | 598 | return 0; |
e070ad49 ML |
599 | } |
600 | ||
834f82e2 CG |
601 | static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) |
602 | { | |
603 | /* | |
604 | * Don't forget to update Documentation/ on changes. | |
605 | */ | |
606 | static const char mnemonics[BITS_PER_LONG][2] = { | |
607 | /* | |
608 | * In case if we meet a flag we don't know about. | |
609 | */ | |
610 | [0 ... (BITS_PER_LONG-1)] = "??", | |
611 | ||
612 | [ilog2(VM_READ)] = "rd", | |
613 | [ilog2(VM_WRITE)] = "wr", | |
614 | [ilog2(VM_EXEC)] = "ex", | |
615 | [ilog2(VM_SHARED)] = "sh", | |
616 | [ilog2(VM_MAYREAD)] = "mr", | |
617 | [ilog2(VM_MAYWRITE)] = "mw", | |
618 | [ilog2(VM_MAYEXEC)] = "me", | |
619 | [ilog2(VM_MAYSHARE)] = "ms", | |
620 | [ilog2(VM_GROWSDOWN)] = "gd", | |
621 | [ilog2(VM_PFNMAP)] = "pf", | |
622 | [ilog2(VM_DENYWRITE)] = "dw", | |
4aae7e43 QR |
623 | #ifdef CONFIG_X86_INTEL_MPX |
624 | [ilog2(VM_MPX)] = "mp", | |
625 | #endif | |
834f82e2 CG |
626 | [ilog2(VM_LOCKED)] = "lo", |
627 | [ilog2(VM_IO)] = "io", | |
628 | [ilog2(VM_SEQ_READ)] = "sr", | |
629 | [ilog2(VM_RAND_READ)] = "rr", | |
630 | [ilog2(VM_DONTCOPY)] = "dc", | |
631 | [ilog2(VM_DONTEXPAND)] = "de", | |
632 | [ilog2(VM_ACCOUNT)] = "ac", | |
633 | [ilog2(VM_NORESERVE)] = "nr", | |
634 | [ilog2(VM_HUGETLB)] = "ht", | |
834f82e2 CG |
635 | [ilog2(VM_ARCH_1)] = "ar", |
636 | [ilog2(VM_DONTDUMP)] = "dd", | |
ec8e41ae NH |
637 | #ifdef CONFIG_MEM_SOFT_DIRTY |
638 | [ilog2(VM_SOFTDIRTY)] = "sd", | |
639 | #endif | |
834f82e2 CG |
640 | [ilog2(VM_MIXEDMAP)] = "mm", |
641 | [ilog2(VM_HUGEPAGE)] = "hg", | |
642 | [ilog2(VM_NOHUGEPAGE)] = "nh", | |
643 | [ilog2(VM_MERGEABLE)] = "mg", | |
16ba6f81 AA |
644 | [ilog2(VM_UFFD_MISSING)]= "um", |
645 | [ilog2(VM_UFFD_WP)] = "uw", | |
834f82e2 CG |
646 | }; |
647 | size_t i; | |
648 | ||
649 | seq_puts(m, "VmFlags: "); | |
650 | for (i = 0; i < BITS_PER_LONG; i++) { | |
651 | if (vma->vm_flags & (1UL << i)) { | |
652 | seq_printf(m, "%c%c ", | |
653 | mnemonics[i][0], mnemonics[i][1]); | |
654 | } | |
655 | } | |
656 | seq_putc(m, '\n'); | |
657 | } | |
658 | ||
25ee01a2 NH |
659 | #ifdef CONFIG_HUGETLB_PAGE |
660 | static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, | |
661 | unsigned long addr, unsigned long end, | |
662 | struct mm_walk *walk) | |
663 | { | |
664 | struct mem_size_stats *mss = walk->private; | |
665 | struct vm_area_struct *vma = walk->vma; | |
666 | struct page *page = NULL; | |
667 | ||
668 | if (pte_present(*pte)) { | |
669 | page = vm_normal_page(vma, addr, *pte); | |
670 | } else if (is_swap_pte(*pte)) { | |
671 | swp_entry_t swpent = pte_to_swp_entry(*pte); | |
672 | ||
673 | if (is_migration_entry(swpent)) | |
674 | page = migration_entry_to_page(swpent); | |
675 | } | |
676 | if (page) { | |
677 | int mapcount = page_mapcount(page); | |
678 | ||
679 | if (mapcount >= 2) | |
680 | mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); | |
681 | else | |
682 | mss->private_hugetlb += huge_page_size(hstate_vma(vma)); | |
683 | } | |
684 | return 0; | |
685 | } | |
686 | #endif /* HUGETLB_PAGE */ | |
687 | ||
b7643757 | 688 | static int show_smap(struct seq_file *m, void *v, int is_pid) |
e070ad49 ML |
689 | { |
690 | struct vm_area_struct *vma = v; | |
e070ad49 | 691 | struct mem_size_stats mss; |
2165009b DH |
692 | struct mm_walk smaps_walk = { |
693 | .pmd_entry = smaps_pte_range, | |
25ee01a2 NH |
694 | #ifdef CONFIG_HUGETLB_PAGE |
695 | .hugetlb_entry = smaps_hugetlb_range, | |
696 | #endif | |
2165009b DH |
697 | .mm = vma->vm_mm, |
698 | .private = &mss, | |
699 | }; | |
e070ad49 ML |
700 | |
701 | memset(&mss, 0, sizeof mss); | |
c261e7d9 VB |
702 | |
703 | #ifdef CONFIG_SHMEM | |
704 | if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { | |
6a15a370 VB |
705 | /* |
706 | * For shared or readonly shmem mappings we know that all | |
707 | * swapped out pages belong to the shmem object, and we can | |
708 | * obtain the swap value much more efficiently. For private | |
709 | * writable mappings, we might have COW pages that are | |
710 | * not affected by the parent swapped out pages of the shmem | |
711 | * object, so we have to distinguish them during the page walk. | |
712 | * Unless we know that the shmem object (or the part mapped by | |
713 | * our VMA) has no swapped out pages at all. | |
714 | */ | |
715 | unsigned long shmem_swapped = shmem_swap_usage(vma); | |
716 | ||
717 | if (!shmem_swapped || (vma->vm_flags & VM_SHARED) || | |
718 | !(vma->vm_flags & VM_WRITE)) { | |
719 | mss.swap = shmem_swapped; | |
720 | } else { | |
721 | mss.check_shmem_swap = true; | |
722 | smaps_walk.pte_hole = smaps_pte_hole; | |
723 | } | |
c261e7d9 VB |
724 | } |
725 | #endif | |
726 | ||
d82ef020 | 727 | /* mmap_sem is held in m_start */ |
14eb6fdd | 728 | walk_page_vma(vma, &smaps_walk); |
4752c369 | 729 | |
b7643757 | 730 | show_map_vma(m, vma, is_pid); |
4752c369 MM |
731 | |
732 | seq_printf(m, | |
733 | "Size: %8lu kB\n" | |
734 | "Rss: %8lu kB\n" | |
735 | "Pss: %8lu kB\n" | |
736 | "Shared_Clean: %8lu kB\n" | |
737 | "Shared_Dirty: %8lu kB\n" | |
738 | "Private_Clean: %8lu kB\n" | |
739 | "Private_Dirty: %8lu kB\n" | |
214e471f | 740 | "Referenced: %8lu kB\n" |
b40d4f84 | 741 | "Anonymous: %8lu kB\n" |
4031a219 | 742 | "AnonHugePages: %8lu kB\n" |
25ee01a2 NH |
743 | "Shared_Hugetlb: %8lu kB\n" |
744 | "Private_Hugetlb: %7lu kB\n" | |
08fba699 | 745 | "Swap: %8lu kB\n" |
8334b962 | 746 | "SwapPss: %8lu kB\n" |
3340289d | 747 | "KernelPageSize: %8lu kB\n" |
2d90508f NK |
748 | "MMUPageSize: %8lu kB\n" |
749 | "Locked: %8lu kB\n", | |
4752c369 MM |
750 | (vma->vm_end - vma->vm_start) >> 10, |
751 | mss.resident >> 10, | |
752 | (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), | |
753 | mss.shared_clean >> 10, | |
754 | mss.shared_dirty >> 10, | |
755 | mss.private_clean >> 10, | |
756 | mss.private_dirty >> 10, | |
214e471f | 757 | mss.referenced >> 10, |
b40d4f84 | 758 | mss.anonymous >> 10, |
4031a219 | 759 | mss.anonymous_thp >> 10, |
25ee01a2 NH |
760 | mss.shared_hugetlb >> 10, |
761 | mss.private_hugetlb >> 10, | |
08fba699 | 762 | mss.swap >> 10, |
8334b962 | 763 | (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)), |
3340289d | 764 | vma_kernel_pagesize(vma) >> 10, |
2d90508f NK |
765 | vma_mmu_pagesize(vma) >> 10, |
766 | (vma->vm_flags & VM_LOCKED) ? | |
767 | (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0); | |
4752c369 | 768 | |
834f82e2 | 769 | show_smap_vma_flags(m, vma); |
b8c20a9b | 770 | m_cache_vma(m, vma); |
7c88db0c | 771 | return 0; |
e070ad49 ML |
772 | } |
773 | ||
b7643757 SP |
774 | static int show_pid_smap(struct seq_file *m, void *v) |
775 | { | |
776 | return show_smap(m, v, 1); | |
777 | } | |
778 | ||
779 | static int show_tid_smap(struct seq_file *m, void *v) | |
780 | { | |
781 | return show_smap(m, v, 0); | |
782 | } | |
783 | ||
03a44825 | 784 | static const struct seq_operations proc_pid_smaps_op = { |
a6198797 MM |
785 | .start = m_start, |
786 | .next = m_next, | |
787 | .stop = m_stop, | |
b7643757 SP |
788 | .show = show_pid_smap |
789 | }; | |
790 | ||
791 | static const struct seq_operations proc_tid_smaps_op = { | |
792 | .start = m_start, | |
793 | .next = m_next, | |
794 | .stop = m_stop, | |
795 | .show = show_tid_smap | |
a6198797 MM |
796 | }; |
797 | ||
b7643757 | 798 | static int pid_smaps_open(struct inode *inode, struct file *file) |
a6198797 MM |
799 | { |
800 | return do_maps_open(inode, file, &proc_pid_smaps_op); | |
801 | } | |
802 | ||
b7643757 SP |
803 | static int tid_smaps_open(struct inode *inode, struct file *file) |
804 | { | |
805 | return do_maps_open(inode, file, &proc_tid_smaps_op); | |
806 | } | |
807 | ||
808 | const struct file_operations proc_pid_smaps_operations = { | |
809 | .open = pid_smaps_open, | |
810 | .read = seq_read, | |
811 | .llseek = seq_lseek, | |
29a40ace | 812 | .release = proc_map_release, |
b7643757 SP |
813 | }; |
814 | ||
815 | const struct file_operations proc_tid_smaps_operations = { | |
816 | .open = tid_smaps_open, | |
a6198797 MM |
817 | .read = seq_read, |
818 | .llseek = seq_lseek, | |
29a40ace | 819 | .release = proc_map_release, |
a6198797 MM |
820 | }; |
821 | ||
040fa020 PE |
822 | enum clear_refs_types { |
823 | CLEAR_REFS_ALL = 1, | |
824 | CLEAR_REFS_ANON, | |
825 | CLEAR_REFS_MAPPED, | |
0f8975ec | 826 | CLEAR_REFS_SOFT_DIRTY, |
695f0559 | 827 | CLEAR_REFS_MM_HIWATER_RSS, |
040fa020 PE |
828 | CLEAR_REFS_LAST, |
829 | }; | |
830 | ||
af9de7eb | 831 | struct clear_refs_private { |
0f8975ec | 832 | enum clear_refs_types type; |
af9de7eb PE |
833 | }; |
834 | ||
7d5b3bfa | 835 | #ifdef CONFIG_MEM_SOFT_DIRTY |
0f8975ec PE |
836 | static inline void clear_soft_dirty(struct vm_area_struct *vma, |
837 | unsigned long addr, pte_t *pte) | |
838 | { | |
0f8975ec PE |
839 | /* |
840 | * The soft-dirty tracker uses #PF-s to catch writes | |
841 | * to pages, so write-protect the pte as well. See the | |
842 | * Documentation/vm/soft-dirty.txt for full description | |
843 | * of how soft-dirty works. | |
844 | */ | |
845 | pte_t ptent = *pte; | |
179ef71c CG |
846 | |
847 | if (pte_present(ptent)) { | |
326c2597 | 848 | ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte); |
179ef71c | 849 | ptent = pte_wrprotect(ptent); |
a7b76174 | 850 | ptent = pte_clear_soft_dirty(ptent); |
326c2597 | 851 | ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent); |
179ef71c CG |
852 | } else if (is_swap_pte(ptent)) { |
853 | ptent = pte_swp_clear_soft_dirty(ptent); | |
326c2597 | 854 | set_pte_at(vma->vm_mm, addr, pte, ptent); |
179ef71c | 855 | } |
0f8975ec | 856 | } |
5d3875a0 LD |
857 | #else |
858 | static inline void clear_soft_dirty(struct vm_area_struct *vma, | |
859 | unsigned long addr, pte_t *pte) | |
860 | { | |
861 | } | |
862 | #endif | |
0f8975ec | 863 | |
5d3875a0 | 864 | #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) |
7d5b3bfa KS |
865 | static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, |
866 | unsigned long addr, pmd_t *pmdp) | |
867 | { | |
326c2597 | 868 | pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp); |
7d5b3bfa KS |
869 | |
870 | pmd = pmd_wrprotect(pmd); | |
a7b76174 | 871 | pmd = pmd_clear_soft_dirty(pmd); |
7d5b3bfa KS |
872 | |
873 | if (vma->vm_flags & VM_SOFTDIRTY) | |
874 | vma->vm_flags &= ~VM_SOFTDIRTY; | |
875 | ||
876 | set_pmd_at(vma->vm_mm, addr, pmdp, pmd); | |
877 | } | |
7d5b3bfa | 878 | #else |
7d5b3bfa KS |
879 | static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, |
880 | unsigned long addr, pmd_t *pmdp) | |
881 | { | |
882 | } | |
883 | #endif | |
884 | ||
a6198797 | 885 | static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, |
2165009b | 886 | unsigned long end, struct mm_walk *walk) |
a6198797 | 887 | { |
af9de7eb | 888 | struct clear_refs_private *cp = walk->private; |
5c64f52a | 889 | struct vm_area_struct *vma = walk->vma; |
a6198797 MM |
890 | pte_t *pte, ptent; |
891 | spinlock_t *ptl; | |
892 | struct page *page; | |
893 | ||
7d5b3bfa KS |
894 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
895 | if (cp->type == CLEAR_REFS_SOFT_DIRTY) { | |
896 | clear_soft_dirty_pmd(vma, addr, pmd); | |
897 | goto out; | |
898 | } | |
899 | ||
900 | page = pmd_page(*pmd); | |
901 | ||
902 | /* Clear accessed and referenced bits. */ | |
903 | pmdp_test_and_clear_young(vma, addr, pmd); | |
33c3fc71 | 904 | test_and_clear_page_young(page); |
7d5b3bfa KS |
905 | ClearPageReferenced(page); |
906 | out: | |
907 | spin_unlock(ptl); | |
908 | return 0; | |
909 | } | |
910 | ||
1a5a9906 AA |
911 | if (pmd_trans_unstable(pmd)) |
912 | return 0; | |
03319327 | 913 | |
a6198797 MM |
914 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
915 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
916 | ptent = *pte; | |
a6198797 | 917 | |
0f8975ec PE |
918 | if (cp->type == CLEAR_REFS_SOFT_DIRTY) { |
919 | clear_soft_dirty(vma, addr, pte); | |
920 | continue; | |
921 | } | |
922 | ||
179ef71c CG |
923 | if (!pte_present(ptent)) |
924 | continue; | |
925 | ||
a6198797 MM |
926 | page = vm_normal_page(vma, addr, ptent); |
927 | if (!page) | |
928 | continue; | |
929 | ||
930 | /* Clear accessed and referenced bits. */ | |
931 | ptep_test_and_clear_young(vma, addr, pte); | |
33c3fc71 | 932 | test_and_clear_page_young(page); |
a6198797 MM |
933 | ClearPageReferenced(page); |
934 | } | |
935 | pte_unmap_unlock(pte - 1, ptl); | |
936 | cond_resched(); | |
937 | return 0; | |
938 | } | |
939 | ||
5c64f52a NH |
940 | static int clear_refs_test_walk(unsigned long start, unsigned long end, |
941 | struct mm_walk *walk) | |
942 | { | |
943 | struct clear_refs_private *cp = walk->private; | |
944 | struct vm_area_struct *vma = walk->vma; | |
945 | ||
48684a65 NH |
946 | if (vma->vm_flags & VM_PFNMAP) |
947 | return 1; | |
948 | ||
5c64f52a NH |
949 | /* |
950 | * Writing 1 to /proc/pid/clear_refs affects all pages. | |
951 | * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. | |
952 | * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. | |
953 | * Writing 4 to /proc/pid/clear_refs affects all pages. | |
954 | */ | |
955 | if (cp->type == CLEAR_REFS_ANON && vma->vm_file) | |
956 | return 1; | |
957 | if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) | |
958 | return 1; | |
959 | return 0; | |
960 | } | |
961 | ||
f248dcb3 MM |
962 | static ssize_t clear_refs_write(struct file *file, const char __user *buf, |
963 | size_t count, loff_t *ppos) | |
b813e931 | 964 | { |
f248dcb3 | 965 | struct task_struct *task; |
fb92a4b0 | 966 | char buffer[PROC_NUMBUF]; |
f248dcb3 | 967 | struct mm_struct *mm; |
b813e931 | 968 | struct vm_area_struct *vma; |
040fa020 PE |
969 | enum clear_refs_types type; |
970 | int itype; | |
0a8cb8e3 | 971 | int rv; |
b813e931 | 972 | |
f248dcb3 MM |
973 | memset(buffer, 0, sizeof(buffer)); |
974 | if (count > sizeof(buffer) - 1) | |
975 | count = sizeof(buffer) - 1; | |
976 | if (copy_from_user(buffer, buf, count)) | |
977 | return -EFAULT; | |
040fa020 | 978 | rv = kstrtoint(strstrip(buffer), 10, &itype); |
0a8cb8e3 AD |
979 | if (rv < 0) |
980 | return rv; | |
040fa020 PE |
981 | type = (enum clear_refs_types)itype; |
982 | if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) | |
f248dcb3 | 983 | return -EINVAL; |
541c237c | 984 | |
496ad9aa | 985 | task = get_proc_task(file_inode(file)); |
f248dcb3 MM |
986 | if (!task) |
987 | return -ESRCH; | |
988 | mm = get_task_mm(task); | |
989 | if (mm) { | |
af9de7eb | 990 | struct clear_refs_private cp = { |
0f8975ec | 991 | .type = type, |
af9de7eb | 992 | }; |
20cbc972 AM |
993 | struct mm_walk clear_refs_walk = { |
994 | .pmd_entry = clear_refs_pte_range, | |
5c64f52a | 995 | .test_walk = clear_refs_test_walk, |
20cbc972 | 996 | .mm = mm, |
af9de7eb | 997 | .private = &cp, |
20cbc972 | 998 | }; |
695f0559 PC |
999 | |
1000 | if (type == CLEAR_REFS_MM_HIWATER_RSS) { | |
1001 | /* | |
1002 | * Writing 5 to /proc/pid/clear_refs resets the peak | |
1003 | * resident set size to this mm's current rss value. | |
1004 | */ | |
1005 | down_write(&mm->mmap_sem); | |
1006 | reset_mm_hiwater_rss(mm); | |
1007 | up_write(&mm->mmap_sem); | |
1008 | goto out_mm; | |
1009 | } | |
1010 | ||
f248dcb3 | 1011 | down_read(&mm->mmap_sem); |
64e45507 PF |
1012 | if (type == CLEAR_REFS_SOFT_DIRTY) { |
1013 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1014 | if (!(vma->vm_flags & VM_SOFTDIRTY)) | |
1015 | continue; | |
1016 | up_read(&mm->mmap_sem); | |
1017 | down_write(&mm->mmap_sem); | |
1018 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1019 | vma->vm_flags &= ~VM_SOFTDIRTY; | |
1020 | vma_set_page_prot(vma); | |
1021 | } | |
1022 | downgrade_write(&mm->mmap_sem); | |
1023 | break; | |
1024 | } | |
0f8975ec | 1025 | mmu_notifier_invalidate_range_start(mm, 0, -1); |
64e45507 | 1026 | } |
5c64f52a | 1027 | walk_page_range(0, ~0UL, &clear_refs_walk); |
0f8975ec PE |
1028 | if (type == CLEAR_REFS_SOFT_DIRTY) |
1029 | mmu_notifier_invalidate_range_end(mm, 0, -1); | |
f248dcb3 MM |
1030 | flush_tlb_mm(mm); |
1031 | up_read(&mm->mmap_sem); | |
695f0559 | 1032 | out_mm: |
f248dcb3 MM |
1033 | mmput(mm); |
1034 | } | |
1035 | put_task_struct(task); | |
fb92a4b0 VL |
1036 | |
1037 | return count; | |
b813e931 DR |
1038 | } |
1039 | ||
f248dcb3 MM |
1040 | const struct file_operations proc_clear_refs_operations = { |
1041 | .write = clear_refs_write, | |
6038f373 | 1042 | .llseek = noop_llseek, |
f248dcb3 MM |
1043 | }; |
1044 | ||
092b50ba NH |
1045 | typedef struct { |
1046 | u64 pme; | |
1047 | } pagemap_entry_t; | |
1048 | ||
85863e47 | 1049 | struct pagemapread { |
8c829622 | 1050 | int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ |
092b50ba | 1051 | pagemap_entry_t *buffer; |
1c90308e | 1052 | bool show_pfn; |
85863e47 MM |
1053 | }; |
1054 | ||
5aaabe83 NH |
1055 | #define PAGEMAP_WALK_SIZE (PMD_SIZE) |
1056 | #define PAGEMAP_WALK_MASK (PMD_MASK) | |
1057 | ||
deb94544 KK |
1058 | #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) |
1059 | #define PM_PFRAME_BITS 55 | |
1060 | #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) | |
1061 | #define PM_SOFT_DIRTY BIT_ULL(55) | |
77bb499b | 1062 | #define PM_MMAP_EXCLUSIVE BIT_ULL(56) |
deb94544 KK |
1063 | #define PM_FILE BIT_ULL(61) |
1064 | #define PM_SWAP BIT_ULL(62) | |
1065 | #define PM_PRESENT BIT_ULL(63) | |
1066 | ||
85863e47 MM |
1067 | #define PM_END_OF_BUFFER 1 |
1068 | ||
deb94544 | 1069 | static inline pagemap_entry_t make_pme(u64 frame, u64 flags) |
092b50ba | 1070 | { |
deb94544 | 1071 | return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; |
092b50ba NH |
1072 | } |
1073 | ||
1074 | static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, | |
85863e47 MM |
1075 | struct pagemapread *pm) |
1076 | { | |
092b50ba | 1077 | pm->buffer[pm->pos++] = *pme; |
d82ef020 | 1078 | if (pm->pos >= pm->len) |
aae8679b | 1079 | return PM_END_OF_BUFFER; |
85863e47 MM |
1080 | return 0; |
1081 | } | |
1082 | ||
1083 | static int pagemap_pte_hole(unsigned long start, unsigned long end, | |
2165009b | 1084 | struct mm_walk *walk) |
85863e47 | 1085 | { |
2165009b | 1086 | struct pagemapread *pm = walk->private; |
68b5a652 | 1087 | unsigned long addr = start; |
85863e47 | 1088 | int err = 0; |
092b50ba | 1089 | |
68b5a652 PF |
1090 | while (addr < end) { |
1091 | struct vm_area_struct *vma = find_vma(walk->mm, addr); | |
deb94544 | 1092 | pagemap_entry_t pme = make_pme(0, 0); |
87e6d49a PF |
1093 | /* End of address space hole, which we mark as non-present. */ |
1094 | unsigned long hole_end; | |
68b5a652 | 1095 | |
87e6d49a PF |
1096 | if (vma) |
1097 | hole_end = min(end, vma->vm_start); | |
1098 | else | |
1099 | hole_end = end; | |
1100 | ||
1101 | for (; addr < hole_end; addr += PAGE_SIZE) { | |
1102 | err = add_to_pagemap(addr, &pme, pm); | |
1103 | if (err) | |
1104 | goto out; | |
68b5a652 PF |
1105 | } |
1106 | ||
87e6d49a PF |
1107 | if (!vma) |
1108 | break; | |
1109 | ||
1110 | /* Addresses in the VMA. */ | |
1111 | if (vma->vm_flags & VM_SOFTDIRTY) | |
deb94544 | 1112 | pme = make_pme(0, PM_SOFT_DIRTY); |
87e6d49a | 1113 | for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { |
68b5a652 PF |
1114 | err = add_to_pagemap(addr, &pme, pm); |
1115 | if (err) | |
1116 | goto out; | |
1117 | } | |
85863e47 | 1118 | } |
68b5a652 | 1119 | out: |
85863e47 MM |
1120 | return err; |
1121 | } | |
1122 | ||
deb94544 | 1123 | static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, |
052fb0d6 | 1124 | struct vm_area_struct *vma, unsigned long addr, pte_t pte) |
85863e47 | 1125 | { |
deb94544 | 1126 | u64 frame = 0, flags = 0; |
052fb0d6 | 1127 | struct page *page = NULL; |
85863e47 | 1128 | |
052fb0d6 | 1129 | if (pte_present(pte)) { |
1c90308e KK |
1130 | if (pm->show_pfn) |
1131 | frame = pte_pfn(pte); | |
deb94544 | 1132 | flags |= PM_PRESENT; |
052fb0d6 | 1133 | page = vm_normal_page(vma, addr, pte); |
e9cdd6e7 | 1134 | if (pte_soft_dirty(pte)) |
deb94544 | 1135 | flags |= PM_SOFT_DIRTY; |
052fb0d6 | 1136 | } else if (is_swap_pte(pte)) { |
179ef71c CG |
1137 | swp_entry_t entry; |
1138 | if (pte_swp_soft_dirty(pte)) | |
deb94544 | 1139 | flags |= PM_SOFT_DIRTY; |
179ef71c | 1140 | entry = pte_to_swp_entry(pte); |
052fb0d6 KK |
1141 | frame = swp_type(entry) | |
1142 | (swp_offset(entry) << MAX_SWAPFILES_SHIFT); | |
deb94544 | 1143 | flags |= PM_SWAP; |
052fb0d6 KK |
1144 | if (is_migration_entry(entry)) |
1145 | page = migration_entry_to_page(entry); | |
052fb0d6 KK |
1146 | } |
1147 | ||
1148 | if (page && !PageAnon(page)) | |
1149 | flags |= PM_FILE; | |
77bb499b KK |
1150 | if (page && page_mapcount(page) == 1) |
1151 | flags |= PM_MMAP_EXCLUSIVE; | |
deb94544 KK |
1152 | if (vma->vm_flags & VM_SOFTDIRTY) |
1153 | flags |= PM_SOFT_DIRTY; | |
052fb0d6 | 1154 | |
deb94544 | 1155 | return make_pme(frame, flags); |
bcf8039e DH |
1156 | } |
1157 | ||
356515e7 | 1158 | static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, |
2165009b | 1159 | struct mm_walk *walk) |
85863e47 | 1160 | { |
f995ece2 | 1161 | struct vm_area_struct *vma = walk->vma; |
2165009b | 1162 | struct pagemapread *pm = walk->private; |
bf929152 | 1163 | spinlock_t *ptl; |
05fbf357 | 1164 | pte_t *pte, *orig_pte; |
85863e47 MM |
1165 | int err = 0; |
1166 | ||
356515e7 KK |
1167 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
1168 | if (pmd_trans_huge_lock(pmdp, vma, &ptl) == 1) { | |
1169 | u64 flags = 0, frame = 0; | |
1170 | pmd_t pmd = *pmdp; | |
0f8975ec | 1171 | |
356515e7 | 1172 | if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd)) |
deb94544 | 1173 | flags |= PM_SOFT_DIRTY; |
d9104d1c | 1174 | |
356515e7 KK |
1175 | /* |
1176 | * Currently pmd for thp is always present because thp | |
1177 | * can not be swapped-out, migrated, or HWPOISONed | |
1178 | * (split in such cases instead.) | |
1179 | * This if-check is just to prepare for future implementation. | |
1180 | */ | |
1181 | if (pmd_present(pmd)) { | |
77bb499b KK |
1182 | struct page *page = pmd_page(pmd); |
1183 | ||
1184 | if (page_mapcount(page) == 1) | |
1185 | flags |= PM_MMAP_EXCLUSIVE; | |
1186 | ||
356515e7 | 1187 | flags |= PM_PRESENT; |
1c90308e KK |
1188 | if (pm->show_pfn) |
1189 | frame = pmd_pfn(pmd) + | |
1190 | ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
356515e7 KK |
1191 | } |
1192 | ||
025c5b24 | 1193 | for (; addr != end; addr += PAGE_SIZE) { |
356515e7 | 1194 | pagemap_entry_t pme = make_pme(frame, flags); |
025c5b24 | 1195 | |
092b50ba | 1196 | err = add_to_pagemap(addr, &pme, pm); |
025c5b24 NH |
1197 | if (err) |
1198 | break; | |
1c90308e | 1199 | if (pm->show_pfn && (flags & PM_PRESENT)) |
356515e7 | 1200 | frame++; |
5aaabe83 | 1201 | } |
bf929152 | 1202 | spin_unlock(ptl); |
025c5b24 | 1203 | return err; |
5aaabe83 NH |
1204 | } |
1205 | ||
356515e7 | 1206 | if (pmd_trans_unstable(pmdp)) |
45f83cef | 1207 | return 0; |
356515e7 | 1208 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
81d0fa62 | 1209 | |
f995ece2 NH |
1210 | /* |
1211 | * We can assume that @vma always points to a valid one and @end never | |
1212 | * goes beyond vma->vm_end. | |
1213 | */ | |
356515e7 | 1214 | orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); |
f995ece2 NH |
1215 | for (; addr < end; pte++, addr += PAGE_SIZE) { |
1216 | pagemap_entry_t pme; | |
05fbf357 | 1217 | |
deb94544 | 1218 | pme = pte_to_pagemap_entry(pm, vma, addr, *pte); |
f995ece2 | 1219 | err = add_to_pagemap(addr, &pme, pm); |
05fbf357 | 1220 | if (err) |
81d0fa62 | 1221 | break; |
85863e47 | 1222 | } |
f995ece2 | 1223 | pte_unmap_unlock(orig_pte, ptl); |
85863e47 MM |
1224 | |
1225 | cond_resched(); | |
1226 | ||
1227 | return err; | |
1228 | } | |
1229 | ||
1a5cb814 | 1230 | #ifdef CONFIG_HUGETLB_PAGE |
116354d1 | 1231 | /* This function walks within one hugetlb entry in the single call */ |
356515e7 | 1232 | static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, |
116354d1 NH |
1233 | unsigned long addr, unsigned long end, |
1234 | struct mm_walk *walk) | |
5dc37642 | 1235 | { |
5dc37642 | 1236 | struct pagemapread *pm = walk->private; |
f995ece2 | 1237 | struct vm_area_struct *vma = walk->vma; |
356515e7 | 1238 | u64 flags = 0, frame = 0; |
5dc37642 | 1239 | int err = 0; |
356515e7 | 1240 | pte_t pte; |
5dc37642 | 1241 | |
f995ece2 | 1242 | if (vma->vm_flags & VM_SOFTDIRTY) |
deb94544 | 1243 | flags |= PM_SOFT_DIRTY; |
d9104d1c | 1244 | |
356515e7 KK |
1245 | pte = huge_ptep_get(ptep); |
1246 | if (pte_present(pte)) { | |
1247 | struct page *page = pte_page(pte); | |
1248 | ||
1249 | if (!PageAnon(page)) | |
1250 | flags |= PM_FILE; | |
1251 | ||
77bb499b KK |
1252 | if (page_mapcount(page) == 1) |
1253 | flags |= PM_MMAP_EXCLUSIVE; | |
1254 | ||
356515e7 | 1255 | flags |= PM_PRESENT; |
1c90308e KK |
1256 | if (pm->show_pfn) |
1257 | frame = pte_pfn(pte) + | |
1258 | ((addr & ~hmask) >> PAGE_SHIFT); | |
356515e7 KK |
1259 | } |
1260 | ||
5dc37642 | 1261 | for (; addr != end; addr += PAGE_SIZE) { |
356515e7 KK |
1262 | pagemap_entry_t pme = make_pme(frame, flags); |
1263 | ||
092b50ba | 1264 | err = add_to_pagemap(addr, &pme, pm); |
5dc37642 NH |
1265 | if (err) |
1266 | return err; | |
1c90308e | 1267 | if (pm->show_pfn && (flags & PM_PRESENT)) |
356515e7 | 1268 | frame++; |
5dc37642 NH |
1269 | } |
1270 | ||
1271 | cond_resched(); | |
1272 | ||
1273 | return err; | |
1274 | } | |
1a5cb814 | 1275 | #endif /* HUGETLB_PAGE */ |
5dc37642 | 1276 | |
85863e47 MM |
1277 | /* |
1278 | * /proc/pid/pagemap - an array mapping virtual pages to pfns | |
1279 | * | |
f16278c6 HR |
1280 | * For each page in the address space, this file contains one 64-bit entry |
1281 | * consisting of the following: | |
1282 | * | |
052fb0d6 | 1283 | * Bits 0-54 page frame number (PFN) if present |
f16278c6 | 1284 | * Bits 0-4 swap type if swapped |
052fb0d6 | 1285 | * Bits 5-54 swap offset if swapped |
deb94544 | 1286 | * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt) |
77bb499b KK |
1287 | * Bit 56 page exclusively mapped |
1288 | * Bits 57-60 zero | |
052fb0d6 | 1289 | * Bit 61 page is file-page or shared-anon |
f16278c6 HR |
1290 | * Bit 62 page swapped |
1291 | * Bit 63 page present | |
1292 | * | |
1293 | * If the page is not present but in swap, then the PFN contains an | |
1294 | * encoding of the swap file number and the page's offset into the | |
1295 | * swap. Unmapped pages return a null PFN. This allows determining | |
85863e47 MM |
1296 | * precisely which pages are mapped (or in swap) and comparing mapped |
1297 | * pages between processes. | |
1298 | * | |
1299 | * Efficient users of this interface will use /proc/pid/maps to | |
1300 | * determine which areas of memory are actually mapped and llseek to | |
1301 | * skip over unmapped regions. | |
1302 | */ | |
1303 | static ssize_t pagemap_read(struct file *file, char __user *buf, | |
1304 | size_t count, loff_t *ppos) | |
1305 | { | |
a06db751 | 1306 | struct mm_struct *mm = file->private_data; |
85863e47 | 1307 | struct pagemapread pm; |
ee1e6ab6 | 1308 | struct mm_walk pagemap_walk = {}; |
5d7e0d2b AM |
1309 | unsigned long src; |
1310 | unsigned long svpfn; | |
1311 | unsigned long start_vaddr; | |
1312 | unsigned long end_vaddr; | |
a06db751 | 1313 | int ret = 0, copied = 0; |
85863e47 | 1314 | |
a06db751 | 1315 | if (!mm || !atomic_inc_not_zero(&mm->mm_users)) |
85863e47 MM |
1316 | goto out; |
1317 | ||
85863e47 MM |
1318 | ret = -EINVAL; |
1319 | /* file position must be aligned */ | |
aae8679b | 1320 | if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) |
a06db751 | 1321 | goto out_mm; |
85863e47 MM |
1322 | |
1323 | ret = 0; | |
08161786 | 1324 | if (!count) |
a06db751 | 1325 | goto out_mm; |
08161786 | 1326 | |
1c90308e KK |
1327 | /* do not disclose physical addresses: attack vector */ |
1328 | pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); | |
1329 | ||
8c829622 | 1330 | pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); |
1331 | pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY); | |
5d7e0d2b | 1332 | ret = -ENOMEM; |
d82ef020 | 1333 | if (!pm.buffer) |
a06db751 | 1334 | goto out_mm; |
85863e47 | 1335 | |
356515e7 | 1336 | pagemap_walk.pmd_entry = pagemap_pmd_range; |
5d7e0d2b | 1337 | pagemap_walk.pte_hole = pagemap_pte_hole; |
1a5cb814 | 1338 | #ifdef CONFIG_HUGETLB_PAGE |
5dc37642 | 1339 | pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; |
1a5cb814 | 1340 | #endif |
5d7e0d2b AM |
1341 | pagemap_walk.mm = mm; |
1342 | pagemap_walk.private = ± | |
1343 | ||
1344 | src = *ppos; | |
1345 | svpfn = src / PM_ENTRY_BYTES; | |
1346 | start_vaddr = svpfn << PAGE_SHIFT; | |
a06db751 | 1347 | end_vaddr = mm->task_size; |
5d7e0d2b AM |
1348 | |
1349 | /* watch out for wraparound */ | |
a06db751 | 1350 | if (svpfn > mm->task_size >> PAGE_SHIFT) |
5d7e0d2b AM |
1351 | start_vaddr = end_vaddr; |
1352 | ||
1353 | /* | |
1354 | * The odds are that this will stop walking way | |
1355 | * before end_vaddr, because the length of the | |
1356 | * user buffer is tracked in "pm", and the walk | |
1357 | * will stop when we hit the end of the buffer. | |
1358 | */ | |
d82ef020 KH |
1359 | ret = 0; |
1360 | while (count && (start_vaddr < end_vaddr)) { | |
1361 | int len; | |
1362 | unsigned long end; | |
1363 | ||
1364 | pm.pos = 0; | |
ea251c1d | 1365 | end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; |
d82ef020 KH |
1366 | /* overflow ? */ |
1367 | if (end < start_vaddr || end > end_vaddr) | |
1368 | end = end_vaddr; | |
1369 | down_read(&mm->mmap_sem); | |
1370 | ret = walk_page_range(start_vaddr, end, &pagemap_walk); | |
1371 | up_read(&mm->mmap_sem); | |
1372 | start_vaddr = end; | |
1373 | ||
1374 | len = min(count, PM_ENTRY_BYTES * pm.pos); | |
309361e0 | 1375 | if (copy_to_user(buf, pm.buffer, len)) { |
d82ef020 | 1376 | ret = -EFAULT; |
a06db751 | 1377 | goto out_free; |
d82ef020 KH |
1378 | } |
1379 | copied += len; | |
1380 | buf += len; | |
1381 | count -= len; | |
85863e47 | 1382 | } |
d82ef020 KH |
1383 | *ppos += copied; |
1384 | if (!ret || ret == PM_END_OF_BUFFER) | |
1385 | ret = copied; | |
1386 | ||
98bc93e5 KM |
1387 | out_free: |
1388 | kfree(pm.buffer); | |
a06db751 KK |
1389 | out_mm: |
1390 | mmput(mm); | |
85863e47 MM |
1391 | out: |
1392 | return ret; | |
1393 | } | |
1394 | ||
541c237c PE |
1395 | static int pagemap_open(struct inode *inode, struct file *file) |
1396 | { | |
a06db751 KK |
1397 | struct mm_struct *mm; |
1398 | ||
a06db751 KK |
1399 | mm = proc_mem_open(inode, PTRACE_MODE_READ); |
1400 | if (IS_ERR(mm)) | |
1401 | return PTR_ERR(mm); | |
1402 | file->private_data = mm; | |
1403 | return 0; | |
1404 | } | |
1405 | ||
1406 | static int pagemap_release(struct inode *inode, struct file *file) | |
1407 | { | |
1408 | struct mm_struct *mm = file->private_data; | |
1409 | ||
1410 | if (mm) | |
1411 | mmdrop(mm); | |
541c237c PE |
1412 | return 0; |
1413 | } | |
1414 | ||
85863e47 MM |
1415 | const struct file_operations proc_pagemap_operations = { |
1416 | .llseek = mem_lseek, /* borrow this */ | |
1417 | .read = pagemap_read, | |
541c237c | 1418 | .open = pagemap_open, |
a06db751 | 1419 | .release = pagemap_release, |
85863e47 | 1420 | }; |
1e883281 | 1421 | #endif /* CONFIG_PROC_PAGE_MONITOR */ |
85863e47 | 1422 | |
6e21c8f1 | 1423 | #ifdef CONFIG_NUMA |
6e21c8f1 | 1424 | |
f69ff943 | 1425 | struct numa_maps { |
f69ff943 SW |
1426 | unsigned long pages; |
1427 | unsigned long anon; | |
1428 | unsigned long active; | |
1429 | unsigned long writeback; | |
1430 | unsigned long mapcount_max; | |
1431 | unsigned long dirty; | |
1432 | unsigned long swapcache; | |
1433 | unsigned long node[MAX_NUMNODES]; | |
1434 | }; | |
1435 | ||
5b52fc89 SW |
1436 | struct numa_maps_private { |
1437 | struct proc_maps_private proc_maps; | |
1438 | struct numa_maps md; | |
1439 | }; | |
1440 | ||
eb4866d0 DH |
1441 | static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, |
1442 | unsigned long nr_pages) | |
f69ff943 SW |
1443 | { |
1444 | int count = page_mapcount(page); | |
1445 | ||
eb4866d0 | 1446 | md->pages += nr_pages; |
f69ff943 | 1447 | if (pte_dirty || PageDirty(page)) |
eb4866d0 | 1448 | md->dirty += nr_pages; |
f69ff943 SW |
1449 | |
1450 | if (PageSwapCache(page)) | |
eb4866d0 | 1451 | md->swapcache += nr_pages; |
f69ff943 SW |
1452 | |
1453 | if (PageActive(page) || PageUnevictable(page)) | |
eb4866d0 | 1454 | md->active += nr_pages; |
f69ff943 SW |
1455 | |
1456 | if (PageWriteback(page)) | |
eb4866d0 | 1457 | md->writeback += nr_pages; |
f69ff943 SW |
1458 | |
1459 | if (PageAnon(page)) | |
eb4866d0 | 1460 | md->anon += nr_pages; |
f69ff943 SW |
1461 | |
1462 | if (count > md->mapcount_max) | |
1463 | md->mapcount_max = count; | |
1464 | ||
eb4866d0 | 1465 | md->node[page_to_nid(page)] += nr_pages; |
f69ff943 SW |
1466 | } |
1467 | ||
3200a8aa DH |
1468 | static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, |
1469 | unsigned long addr) | |
1470 | { | |
1471 | struct page *page; | |
1472 | int nid; | |
1473 | ||
1474 | if (!pte_present(pte)) | |
1475 | return NULL; | |
1476 | ||
1477 | page = vm_normal_page(vma, addr, pte); | |
1478 | if (!page) | |
1479 | return NULL; | |
1480 | ||
1481 | if (PageReserved(page)) | |
1482 | return NULL; | |
1483 | ||
1484 | nid = page_to_nid(page); | |
4ff1b2c2 | 1485 | if (!node_isset(nid, node_states[N_MEMORY])) |
3200a8aa DH |
1486 | return NULL; |
1487 | ||
1488 | return page; | |
1489 | } | |
1490 | ||
f69ff943 SW |
1491 | static int gather_pte_stats(pmd_t *pmd, unsigned long addr, |
1492 | unsigned long end, struct mm_walk *walk) | |
1493 | { | |
d85f4d6d NH |
1494 | struct numa_maps *md = walk->private; |
1495 | struct vm_area_struct *vma = walk->vma; | |
f69ff943 SW |
1496 | spinlock_t *ptl; |
1497 | pte_t *orig_pte; | |
1498 | pte_t *pte; | |
1499 | ||
d85f4d6d | 1500 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
025c5b24 NH |
1501 | pte_t huge_pte = *(pte_t *)pmd; |
1502 | struct page *page; | |
1503 | ||
d85f4d6d | 1504 | page = can_gather_numa_stats(huge_pte, vma, addr); |
025c5b24 NH |
1505 | if (page) |
1506 | gather_stats(page, md, pte_dirty(huge_pte), | |
1507 | HPAGE_PMD_SIZE/PAGE_SIZE); | |
bf929152 | 1508 | spin_unlock(ptl); |
025c5b24 | 1509 | return 0; |
32ef4384 DH |
1510 | } |
1511 | ||
1a5a9906 AA |
1512 | if (pmd_trans_unstable(pmd)) |
1513 | return 0; | |
f69ff943 SW |
1514 | orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); |
1515 | do { | |
d85f4d6d | 1516 | struct page *page = can_gather_numa_stats(*pte, vma, addr); |
f69ff943 SW |
1517 | if (!page) |
1518 | continue; | |
eb4866d0 | 1519 | gather_stats(page, md, pte_dirty(*pte), 1); |
f69ff943 SW |
1520 | |
1521 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
1522 | pte_unmap_unlock(orig_pte, ptl); | |
1523 | return 0; | |
1524 | } | |
1525 | #ifdef CONFIG_HUGETLB_PAGE | |
632fd60f | 1526 | static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, |
f69ff943 SW |
1527 | unsigned long addr, unsigned long end, struct mm_walk *walk) |
1528 | { | |
1529 | struct numa_maps *md; | |
1530 | struct page *page; | |
1531 | ||
d4c54919 | 1532 | if (!pte_present(*pte)) |
f69ff943 SW |
1533 | return 0; |
1534 | ||
1535 | page = pte_page(*pte); | |
1536 | if (!page) | |
1537 | return 0; | |
1538 | ||
1539 | md = walk->private; | |
eb4866d0 | 1540 | gather_stats(page, md, pte_dirty(*pte), 1); |
f69ff943 SW |
1541 | return 0; |
1542 | } | |
1543 | ||
1544 | #else | |
632fd60f | 1545 | static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, |
f69ff943 SW |
1546 | unsigned long addr, unsigned long end, struct mm_walk *walk) |
1547 | { | |
1548 | return 0; | |
1549 | } | |
1550 | #endif | |
1551 | ||
1552 | /* | |
1553 | * Display pages allocated per node and memory policy via /proc. | |
1554 | */ | |
b7643757 | 1555 | static int show_numa_map(struct seq_file *m, void *v, int is_pid) |
f69ff943 | 1556 | { |
5b52fc89 SW |
1557 | struct numa_maps_private *numa_priv = m->private; |
1558 | struct proc_maps_private *proc_priv = &numa_priv->proc_maps; | |
f69ff943 | 1559 | struct vm_area_struct *vma = v; |
5b52fc89 | 1560 | struct numa_maps *md = &numa_priv->md; |
f69ff943 SW |
1561 | struct file *file = vma->vm_file; |
1562 | struct mm_struct *mm = vma->vm_mm; | |
d85f4d6d NH |
1563 | struct mm_walk walk = { |
1564 | .hugetlb_entry = gather_hugetlb_stats, | |
1565 | .pmd_entry = gather_pte_stats, | |
1566 | .private = md, | |
1567 | .mm = mm, | |
1568 | }; | |
f69ff943 | 1569 | struct mempolicy *pol; |
948927ee DR |
1570 | char buffer[64]; |
1571 | int nid; | |
f69ff943 SW |
1572 | |
1573 | if (!mm) | |
1574 | return 0; | |
1575 | ||
5b52fc89 SW |
1576 | /* Ensure we start with an empty set of numa_maps statistics. */ |
1577 | memset(md, 0, sizeof(*md)); | |
f69ff943 | 1578 | |
498f2371 ON |
1579 | pol = __get_vma_policy(vma, vma->vm_start); |
1580 | if (pol) { | |
1581 | mpol_to_str(buffer, sizeof(buffer), pol); | |
1582 | mpol_cond_put(pol); | |
1583 | } else { | |
1584 | mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); | |
1585 | } | |
f69ff943 SW |
1586 | |
1587 | seq_printf(m, "%08lx %s", vma->vm_start, buffer); | |
1588 | ||
1589 | if (file) { | |
17c2b4ee | 1590 | seq_puts(m, " file="); |
2726d566 | 1591 | seq_file_path(m, file, "\n\t= "); |
f69ff943 | 1592 | } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { |
17c2b4ee | 1593 | seq_puts(m, " heap"); |
b7643757 | 1594 | } else { |
58cb6548 | 1595 | pid_t tid = pid_of_stack(proc_priv, vma, is_pid); |
b7643757 SP |
1596 | if (tid != 0) { |
1597 | /* | |
1598 | * Thread stack in /proc/PID/task/TID/maps or | |
1599 | * the main process stack. | |
1600 | */ | |
1601 | if (!is_pid || (vma->vm_start <= mm->start_stack && | |
1602 | vma->vm_end >= mm->start_stack)) | |
17c2b4ee | 1603 | seq_puts(m, " stack"); |
b7643757 SP |
1604 | else |
1605 | seq_printf(m, " stack:%d", tid); | |
1606 | } | |
f69ff943 SW |
1607 | } |
1608 | ||
fc360bd9 | 1609 | if (is_vm_hugetlb_page(vma)) |
17c2b4ee | 1610 | seq_puts(m, " huge"); |
fc360bd9 | 1611 | |
d85f4d6d NH |
1612 | /* mmap_sem is held by m_start */ |
1613 | walk_page_vma(vma, &walk); | |
f69ff943 SW |
1614 | |
1615 | if (!md->pages) | |
1616 | goto out; | |
1617 | ||
1618 | if (md->anon) | |
1619 | seq_printf(m, " anon=%lu", md->anon); | |
1620 | ||
1621 | if (md->dirty) | |
1622 | seq_printf(m, " dirty=%lu", md->dirty); | |
1623 | ||
1624 | if (md->pages != md->anon && md->pages != md->dirty) | |
1625 | seq_printf(m, " mapped=%lu", md->pages); | |
1626 | ||
1627 | if (md->mapcount_max > 1) | |
1628 | seq_printf(m, " mapmax=%lu", md->mapcount_max); | |
1629 | ||
1630 | if (md->swapcache) | |
1631 | seq_printf(m, " swapcache=%lu", md->swapcache); | |
1632 | ||
1633 | if (md->active < md->pages && !is_vm_hugetlb_page(vma)) | |
1634 | seq_printf(m, " active=%lu", md->active); | |
1635 | ||
1636 | if (md->writeback) | |
1637 | seq_printf(m, " writeback=%lu", md->writeback); | |
1638 | ||
948927ee DR |
1639 | for_each_node_state(nid, N_MEMORY) |
1640 | if (md->node[nid]) | |
1641 | seq_printf(m, " N%d=%lu", nid, md->node[nid]); | |
198d1597 RA |
1642 | |
1643 | seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); | |
f69ff943 SW |
1644 | out: |
1645 | seq_putc(m, '\n'); | |
b8c20a9b | 1646 | m_cache_vma(m, vma); |
f69ff943 SW |
1647 | return 0; |
1648 | } | |
5b52fc89 | 1649 | |
b7643757 SP |
1650 | static int show_pid_numa_map(struct seq_file *m, void *v) |
1651 | { | |
1652 | return show_numa_map(m, v, 1); | |
1653 | } | |
1654 | ||
1655 | static int show_tid_numa_map(struct seq_file *m, void *v) | |
1656 | { | |
1657 | return show_numa_map(m, v, 0); | |
1658 | } | |
1659 | ||
03a44825 | 1660 | static const struct seq_operations proc_pid_numa_maps_op = { |
b7643757 SP |
1661 | .start = m_start, |
1662 | .next = m_next, | |
1663 | .stop = m_stop, | |
1664 | .show = show_pid_numa_map, | |
6e21c8f1 | 1665 | }; |
662795de | 1666 | |
b7643757 SP |
1667 | static const struct seq_operations proc_tid_numa_maps_op = { |
1668 | .start = m_start, | |
1669 | .next = m_next, | |
1670 | .stop = m_stop, | |
1671 | .show = show_tid_numa_map, | |
1672 | }; | |
1673 | ||
1674 | static int numa_maps_open(struct inode *inode, struct file *file, | |
1675 | const struct seq_operations *ops) | |
662795de | 1676 | { |
4db7d0ee ON |
1677 | return proc_maps_open(inode, file, ops, |
1678 | sizeof(struct numa_maps_private)); | |
662795de EB |
1679 | } |
1680 | ||
b7643757 SP |
1681 | static int pid_numa_maps_open(struct inode *inode, struct file *file) |
1682 | { | |
1683 | return numa_maps_open(inode, file, &proc_pid_numa_maps_op); | |
1684 | } | |
1685 | ||
1686 | static int tid_numa_maps_open(struct inode *inode, struct file *file) | |
1687 | { | |
1688 | return numa_maps_open(inode, file, &proc_tid_numa_maps_op); | |
1689 | } | |
1690 | ||
1691 | const struct file_operations proc_pid_numa_maps_operations = { | |
1692 | .open = pid_numa_maps_open, | |
1693 | .read = seq_read, | |
1694 | .llseek = seq_lseek, | |
29a40ace | 1695 | .release = proc_map_release, |
b7643757 SP |
1696 | }; |
1697 | ||
1698 | const struct file_operations proc_tid_numa_maps_operations = { | |
1699 | .open = tid_numa_maps_open, | |
662795de EB |
1700 | .read = seq_read, |
1701 | .llseek = seq_lseek, | |
29a40ace | 1702 | .release = proc_map_release, |
662795de | 1703 | }; |
f69ff943 | 1704 | #endif /* CONFIG_NUMA */ |