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