mm: re-architect the VM_UNPAGED logic
[linux-2.6.git] / fs / proc / task_mmu.c
1 #include <linux/mm.h>
2 #include <linux/hugetlb.h>
3 #include <linux/mount.h>
4 #include <linux/seq_file.h>
5 #include <linux/highmem.h>
6 #include <linux/pagemap.h>
7 #include <linux/mempolicy.h>
8
9 #include <asm/elf.h>
10 #include <asm/uaccess.h>
11 #include <asm/tlbflush.h>
12 #include "internal.h"
13
14 char *task_mem(struct mm_struct *mm, char *buffer)
15 {
16         unsigned long data, text, lib;
17         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
18
19         /*
20          * Note: to minimize their overhead, mm maintains hiwater_vm and
21          * hiwater_rss only when about to *lower* total_vm or rss.  Any
22          * collector of these hiwater stats must therefore get total_vm
23          * and rss too, which will usually be the higher.  Barriers? not
24          * worth the effort, such snapshots can always be inconsistent.
25          */
26         hiwater_vm = total_vm = mm->total_vm;
27         if (hiwater_vm < mm->hiwater_vm)
28                 hiwater_vm = mm->hiwater_vm;
29         hiwater_rss = total_rss = get_mm_rss(mm);
30         if (hiwater_rss < mm->hiwater_rss)
31                 hiwater_rss = mm->hiwater_rss;
32
33         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
34         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
35         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
36         buffer += sprintf(buffer,
37                 "VmPeak:\t%8lu kB\n"
38                 "VmSize:\t%8lu kB\n"
39                 "VmLck:\t%8lu kB\n"
40                 "VmHWM:\t%8lu kB\n"
41                 "VmRSS:\t%8lu kB\n"
42                 "VmData:\t%8lu kB\n"
43                 "VmStk:\t%8lu kB\n"
44                 "VmExe:\t%8lu kB\n"
45                 "VmLib:\t%8lu kB\n"
46                 "VmPTE:\t%8lu kB\n",
47                 hiwater_vm << (PAGE_SHIFT-10),
48                 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
49                 mm->locked_vm << (PAGE_SHIFT-10),
50                 hiwater_rss << (PAGE_SHIFT-10),
51                 total_rss << (PAGE_SHIFT-10),
52                 data << (PAGE_SHIFT-10),
53                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
54                 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
55         return buffer;
56 }
57
58 unsigned long task_vsize(struct mm_struct *mm)
59 {
60         return PAGE_SIZE * mm->total_vm;
61 }
62
63 int task_statm(struct mm_struct *mm, int *shared, int *text,
64                int *data, int *resident)
65 {
66         *shared = get_mm_counter(mm, file_rss);
67         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
68                                                                 >> PAGE_SHIFT;
69         *data = mm->total_vm - mm->shared_vm;
70         *resident = *shared + get_mm_counter(mm, anon_rss);
71         return mm->total_vm;
72 }
73
74 int proc_exe_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
75 {
76         struct vm_area_struct * vma;
77         int result = -ENOENT;
78         struct task_struct *task = proc_task(inode);
79         struct mm_struct * mm = get_task_mm(task);
80
81         if (!mm)
82                 goto out;
83         down_read(&mm->mmap_sem);
84
85         vma = mm->mmap;
86         while (vma) {
87                 if ((vma->vm_flags & VM_EXECUTABLE) && vma->vm_file)
88                         break;
89                 vma = vma->vm_next;
90         }
91
92         if (vma) {
93                 *mnt = mntget(vma->vm_file->f_vfsmnt);
94                 *dentry = dget(vma->vm_file->f_dentry);
95                 result = 0;
96         }
97
98         up_read(&mm->mmap_sem);
99         mmput(mm);
100 out:
101         return result;
102 }
103
104 static void pad_len_spaces(struct seq_file *m, int len)
105 {
106         len = 25 + sizeof(void*) * 6 - len;
107         if (len < 1)
108                 len = 1;
109         seq_printf(m, "%*c", len, ' ');
110 }
111
112 struct mem_size_stats
113 {
114         unsigned long resident;
115         unsigned long shared_clean;
116         unsigned long shared_dirty;
117         unsigned long private_clean;
118         unsigned long private_dirty;
119 };
120
121 static int show_map_internal(struct seq_file *m, void *v, struct mem_size_stats *mss)
122 {
123         struct task_struct *task = m->private;
124         struct vm_area_struct *vma = v;
125         struct mm_struct *mm = vma->vm_mm;
126         struct file *file = vma->vm_file;
127         int flags = vma->vm_flags;
128         unsigned long ino = 0;
129         dev_t dev = 0;
130         int len;
131
132         if (file) {
133                 struct inode *inode = vma->vm_file->f_dentry->d_inode;
134                 dev = inode->i_sb->s_dev;
135                 ino = inode->i_ino;
136         }
137
138         seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
139                         vma->vm_start,
140                         vma->vm_end,
141                         flags & VM_READ ? 'r' : '-',
142                         flags & VM_WRITE ? 'w' : '-',
143                         flags & VM_EXEC ? 'x' : '-',
144                         flags & VM_MAYSHARE ? 's' : 'p',
145                         vma->vm_pgoff << PAGE_SHIFT,
146                         MAJOR(dev), MINOR(dev), ino, &len);
147
148         /*
149          * Print the dentry name for named mappings, and a
150          * special [heap] marker for the heap:
151          */
152         if (file) {
153                 pad_len_spaces(m, len);
154                 seq_path(m, file->f_vfsmnt, file->f_dentry, "\n");
155         } else {
156                 if (mm) {
157                         if (vma->vm_start <= mm->start_brk &&
158                                                 vma->vm_end >= mm->brk) {
159                                 pad_len_spaces(m, len);
160                                 seq_puts(m, "[heap]");
161                         } else {
162                                 if (vma->vm_start <= mm->start_stack &&
163                                         vma->vm_end >= mm->start_stack) {
164
165                                         pad_len_spaces(m, len);
166                                         seq_puts(m, "[stack]");
167                                 }
168                         }
169                 } else {
170                         pad_len_spaces(m, len);
171                         seq_puts(m, "[vdso]");
172                 }
173         }
174         seq_putc(m, '\n');
175
176         if (mss)
177                 seq_printf(m,
178                            "Size:          %8lu kB\n"
179                            "Rss:           %8lu kB\n"
180                            "Shared_Clean:  %8lu kB\n"
181                            "Shared_Dirty:  %8lu kB\n"
182                            "Private_Clean: %8lu kB\n"
183                            "Private_Dirty: %8lu kB\n",
184                            (vma->vm_end - vma->vm_start) >> 10,
185                            mss->resident >> 10,
186                            mss->shared_clean  >> 10,
187                            mss->shared_dirty  >> 10,
188                            mss->private_clean >> 10,
189                            mss->private_dirty >> 10);
190
191         if (m->count < m->size)  /* vma is copied successfully */
192                 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
193         return 0;
194 }
195
196 static int show_map(struct seq_file *m, void *v)
197 {
198         return show_map_internal(m, v, NULL);
199 }
200
201 static void smaps_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
202                                 unsigned long addr, unsigned long end,
203                                 struct mem_size_stats *mss)
204 {
205         pte_t *pte, ptent;
206         spinlock_t *ptl;
207         unsigned long pfn;
208         struct page *page;
209
210         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
211         do {
212                 ptent = *pte;
213                 if (!pte_present(ptent))
214                         continue;
215
216                 mss->resident += PAGE_SIZE;
217                 pfn = pte_pfn(ptent);
218                 if (!pfn_valid(pfn))
219                         continue;
220
221                 page = pfn_to_page(pfn);
222                 if (page_count(page) >= 2) {
223                         if (pte_dirty(ptent))
224                                 mss->shared_dirty += PAGE_SIZE;
225                         else
226                                 mss->shared_clean += PAGE_SIZE;
227                 } else {
228                         if (pte_dirty(ptent))
229                                 mss->private_dirty += PAGE_SIZE;
230                         else
231                                 mss->private_clean += PAGE_SIZE;
232                 }
233         } while (pte++, addr += PAGE_SIZE, addr != end);
234         pte_unmap_unlock(pte - 1, ptl);
235         cond_resched();
236 }
237
238 static inline void smaps_pmd_range(struct vm_area_struct *vma, pud_t *pud,
239                                 unsigned long addr, unsigned long end,
240                                 struct mem_size_stats *mss)
241 {
242         pmd_t *pmd;
243         unsigned long next;
244
245         pmd = pmd_offset(pud, addr);
246         do {
247                 next = pmd_addr_end(addr, end);
248                 if (pmd_none_or_clear_bad(pmd))
249                         continue;
250                 smaps_pte_range(vma, pmd, addr, next, mss);
251         } while (pmd++, addr = next, addr != end);
252 }
253
254 static inline void smaps_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
255                                 unsigned long addr, unsigned long end,
256                                 struct mem_size_stats *mss)
257 {
258         pud_t *pud;
259         unsigned long next;
260
261         pud = pud_offset(pgd, addr);
262         do {
263                 next = pud_addr_end(addr, end);
264                 if (pud_none_or_clear_bad(pud))
265                         continue;
266                 smaps_pmd_range(vma, pud, addr, next, mss);
267         } while (pud++, addr = next, addr != end);
268 }
269
270 static inline void smaps_pgd_range(struct vm_area_struct *vma,
271                                 unsigned long addr, unsigned long end,
272                                 struct mem_size_stats *mss)
273 {
274         pgd_t *pgd;
275         unsigned long next;
276
277         pgd = pgd_offset(vma->vm_mm, addr);
278         do {
279                 next = pgd_addr_end(addr, end);
280                 if (pgd_none_or_clear_bad(pgd))
281                         continue;
282                 smaps_pud_range(vma, pgd, addr, next, mss);
283         } while (pgd++, addr = next, addr != end);
284 }
285
286 static int show_smap(struct seq_file *m, void *v)
287 {
288         struct vm_area_struct *vma = v;
289         struct mem_size_stats mss;
290
291         memset(&mss, 0, sizeof mss);
292         if (vma->vm_mm)
293                 smaps_pgd_range(vma, vma->vm_start, vma->vm_end, &mss);
294         return show_map_internal(m, v, &mss);
295 }
296
297 static void *m_start(struct seq_file *m, loff_t *pos)
298 {
299         struct task_struct *task = m->private;
300         unsigned long last_addr = m->version;
301         struct mm_struct *mm;
302         struct vm_area_struct *vma, *tail_vma;
303         loff_t l = *pos;
304
305         /*
306          * We remember last_addr rather than next_addr to hit with
307          * mmap_cache most of the time. We have zero last_addr at
308          * the beginning and also after lseek. We will have -1 last_addr
309          * after the end of the vmas.
310          */
311
312         if (last_addr == -1UL)
313                 return NULL;
314
315         mm = get_task_mm(task);
316         if (!mm)
317                 return NULL;
318
319         tail_vma = get_gate_vma(task);
320         down_read(&mm->mmap_sem);
321
322         /* Start with last addr hint */
323         if (last_addr && (vma = find_vma(mm, last_addr))) {
324                 vma = vma->vm_next;
325                 goto out;
326         }
327
328         /*
329          * Check the vma index is within the range and do
330          * sequential scan until m_index.
331          */
332         vma = NULL;
333         if ((unsigned long)l < mm->map_count) {
334                 vma = mm->mmap;
335                 while (l-- && vma)
336                         vma = vma->vm_next;
337                 goto out;
338         }
339
340         if (l != mm->map_count)
341                 tail_vma = NULL; /* After gate vma */
342
343 out:
344         if (vma)
345                 return vma;
346
347         /* End of vmas has been reached */
348         m->version = (tail_vma != NULL)? 0: -1UL;
349         up_read(&mm->mmap_sem);
350         mmput(mm);
351         return tail_vma;
352 }
353
354 static void m_stop(struct seq_file *m, void *v)
355 {
356         struct task_struct *task = m->private;
357         struct vm_area_struct *vma = v;
358         if (vma && vma != get_gate_vma(task)) {
359                 struct mm_struct *mm = vma->vm_mm;
360                 up_read(&mm->mmap_sem);
361                 mmput(mm);
362         }
363 }
364
365 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
366 {
367         struct task_struct *task = m->private;
368         struct vm_area_struct *vma = v;
369         struct vm_area_struct *tail_vma = get_gate_vma(task);
370
371         (*pos)++;
372         if (vma && (vma != tail_vma) && vma->vm_next)
373                 return vma->vm_next;
374         m_stop(m, v);
375         return (vma != tail_vma)? tail_vma: NULL;
376 }
377
378 struct seq_operations proc_pid_maps_op = {
379         .start  = m_start,
380         .next   = m_next,
381         .stop   = m_stop,
382         .show   = show_map
383 };
384
385 struct seq_operations proc_pid_smaps_op = {
386         .start  = m_start,
387         .next   = m_next,
388         .stop   = m_stop,
389         .show   = show_smap
390 };
391
392 #ifdef CONFIG_NUMA
393
394 struct numa_maps {
395         unsigned long pages;
396         unsigned long anon;
397         unsigned long mapped;
398         unsigned long mapcount_max;
399         unsigned long node[MAX_NUMNODES];
400 };
401
402 /*
403  * Calculate numa node maps for a vma
404  */
405 static struct numa_maps *get_numa_maps(struct vm_area_struct *vma)
406 {
407         int i;
408         struct page *page;
409         unsigned long vaddr;
410         struct numa_maps *md = kmalloc(sizeof(struct numa_maps), GFP_KERNEL);
411
412         if (!md)
413                 return NULL;
414         md->pages = 0;
415         md->anon = 0;
416         md->mapped = 0;
417         md->mapcount_max = 0;
418         for_each_node(i)
419                 md->node[i] =0;
420
421         for (vaddr = vma->vm_start; vaddr < vma->vm_end; vaddr += PAGE_SIZE) {
422                 page = follow_page(vma, vaddr, 0);
423                 if (page) {
424                         int count = page_mapcount(page);
425
426                         if (count)
427                                 md->mapped++;
428                         if (count > md->mapcount_max)
429                                 md->mapcount_max = count;
430                         md->pages++;
431                         if (PageAnon(page))
432                                 md->anon++;
433                         md->node[page_to_nid(page)]++;
434                 }
435                 cond_resched();
436         }
437         return md;
438 }
439
440 static int show_numa_map(struct seq_file *m, void *v)
441 {
442         struct task_struct *task = m->private;
443         struct vm_area_struct *vma = v;
444         struct mempolicy *pol;
445         struct numa_maps *md;
446         struct zone **z;
447         int n;
448         int first;
449
450         if (!vma->vm_mm)
451                 return 0;
452
453         md = get_numa_maps(vma);
454         if (!md)
455                 return 0;
456
457         seq_printf(m, "%08lx", vma->vm_start);
458         pol = get_vma_policy(task, vma, vma->vm_start);
459         /* Print policy */
460         switch (pol->policy) {
461         case MPOL_PREFERRED:
462                 seq_printf(m, " prefer=%d", pol->v.preferred_node);
463                 break;
464         case MPOL_BIND:
465                 seq_printf(m, " bind={");
466                 first = 1;
467                 for (z = pol->v.zonelist->zones; *z; z++) {
468
469                         if (!first)
470                                 seq_putc(m, ',');
471                         else
472                                 first = 0;
473                         seq_printf(m, "%d/%s", (*z)->zone_pgdat->node_id,
474                                         (*z)->name);
475                 }
476                 seq_putc(m, '}');
477                 break;
478         case MPOL_INTERLEAVE:
479                 seq_printf(m, " interleave={");
480                 first = 1;
481                 for_each_node(n) {
482                         if (node_isset(n, pol->v.nodes)) {
483                                 if (!first)
484                                         seq_putc(m,',');
485                                 else
486                                         first = 0;
487                                 seq_printf(m, "%d",n);
488                         }
489                 }
490                 seq_putc(m, '}');
491                 break;
492         default:
493                 seq_printf(m," default");
494                 break;
495         }
496         seq_printf(m, " MaxRef=%lu Pages=%lu Mapped=%lu",
497                         md->mapcount_max, md->pages, md->mapped);
498         if (md->anon)
499                 seq_printf(m," Anon=%lu",md->anon);
500
501         for_each_online_node(n) {
502                 if (md->node[n])
503                         seq_printf(m, " N%d=%lu", n, md->node[n]);
504         }
505         seq_putc(m, '\n');
506         kfree(md);
507         if (m->count < m->size)  /* vma is copied successfully */
508                 m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
509         return 0;
510 }
511
512 struct seq_operations proc_pid_numa_maps_op = {
513         .start  = m_start,
514         .next   = m_next,
515         .stop   = m_stop,
516         .show   = show_numa_map
517 };
518 #endif