xref: /openbmc/linux/fs/proc/task_mmu.c (revision 161f4089)
1 #include <linux/mm.h>
2 #include <linux/hugetlb.h>
3 #include <linux/huge_mm.h>
4 #include <linux/mount.h>
5 #include <linux/seq_file.h>
6 #include <linux/highmem.h>
7 #include <linux/ptrace.h>
8 #include <linux/slab.h>
9 #include <linux/pagemap.h>
10 #include <linux/mempolicy.h>
11 #include <linux/rmap.h>
12 #include <linux/swap.h>
13 #include <linux/swapops.h>
14 #include <linux/mmu_notifier.h>
15 
16 #include <asm/elf.h>
17 #include <asm/uaccess.h>
18 #include <asm/tlbflush.h>
19 #include "internal.h"
20 
21 void task_mem(struct seq_file *m, struct mm_struct *mm)
22 {
23 	unsigned long data, text, lib, swap;
24 	unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
25 
26 	/*
27 	 * Note: to minimize their overhead, mm maintains hiwater_vm and
28 	 * hiwater_rss only when about to *lower* total_vm or rss.  Any
29 	 * collector of these hiwater stats must therefore get total_vm
30 	 * and rss too, which will usually be the higher.  Barriers? not
31 	 * worth the effort, such snapshots can always be inconsistent.
32 	 */
33 	hiwater_vm = total_vm = mm->total_vm;
34 	if (hiwater_vm < mm->hiwater_vm)
35 		hiwater_vm = mm->hiwater_vm;
36 	hiwater_rss = total_rss = get_mm_rss(mm);
37 	if (hiwater_rss < mm->hiwater_rss)
38 		hiwater_rss = mm->hiwater_rss;
39 
40 	data = mm->total_vm - mm->shared_vm - mm->stack_vm;
41 	text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
42 	lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
43 	swap = get_mm_counter(mm, MM_SWAPENTS);
44 	seq_printf(m,
45 		"VmPeak:\t%8lu kB\n"
46 		"VmSize:\t%8lu kB\n"
47 		"VmLck:\t%8lu kB\n"
48 		"VmPin:\t%8lu kB\n"
49 		"VmHWM:\t%8lu kB\n"
50 		"VmRSS:\t%8lu kB\n"
51 		"VmData:\t%8lu kB\n"
52 		"VmStk:\t%8lu kB\n"
53 		"VmExe:\t%8lu kB\n"
54 		"VmLib:\t%8lu kB\n"
55 		"VmPTE:\t%8lu kB\n"
56 		"VmSwap:\t%8lu kB\n",
57 		hiwater_vm << (PAGE_SHIFT-10),
58 		total_vm << (PAGE_SHIFT-10),
59 		mm->locked_vm << (PAGE_SHIFT-10),
60 		mm->pinned_vm << (PAGE_SHIFT-10),
61 		hiwater_rss << (PAGE_SHIFT-10),
62 		total_rss << (PAGE_SHIFT-10),
63 		data << (PAGE_SHIFT-10),
64 		mm->stack_vm << (PAGE_SHIFT-10), text, lib,
65 		(PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
66 		swap << (PAGE_SHIFT-10));
67 }
68 
69 unsigned long task_vsize(struct mm_struct *mm)
70 {
71 	return PAGE_SIZE * mm->total_vm;
72 }
73 
74 unsigned long task_statm(struct mm_struct *mm,
75 			 unsigned long *shared, unsigned long *text,
76 			 unsigned long *data, unsigned long *resident)
77 {
78 	*shared = get_mm_counter(mm, MM_FILEPAGES);
79 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
80 								>> PAGE_SHIFT;
81 	*data = mm->total_vm - mm->shared_vm;
82 	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
83 	return mm->total_vm;
84 }
85 
86 static void pad_len_spaces(struct seq_file *m, int len)
87 {
88 	len = 25 + sizeof(void*) * 6 - len;
89 	if (len < 1)
90 		len = 1;
91 	seq_printf(m, "%*c", len, ' ');
92 }
93 
94 #ifdef CONFIG_NUMA
95 /*
96  * These functions are for numa_maps but called in generic **maps seq_file
97  * ->start(), ->stop() ops.
98  *
99  * numa_maps scans all vmas under mmap_sem and checks their mempolicy.
100  * Each mempolicy object is controlled by reference counting. The problem here
101  * is how to avoid accessing dead mempolicy object.
102  *
103  * Because we're holding mmap_sem while reading seq_file, it's safe to access
104  * each vma's mempolicy, no vma objects will never drop refs to mempolicy.
105  *
106  * A task's mempolicy (task->mempolicy) has different behavior. task->mempolicy
107  * is set and replaced under mmap_sem but unrefed and cleared under task_lock().
108  * So, without task_lock(), we cannot trust get_vma_policy() because we cannot
109  * gurantee the task never exits under us. But taking task_lock() around
110  * get_vma_plicy() causes lock order problem.
111  *
112  * To access task->mempolicy without lock, we hold a reference count of an
113  * object pointed by task->mempolicy and remember it. This will guarantee
114  * that task->mempolicy points to an alive object or NULL in numa_maps accesses.
115  */
116 static void hold_task_mempolicy(struct proc_maps_private *priv)
117 {
118 	struct task_struct *task = priv->task;
119 
120 	task_lock(task);
121 	priv->task_mempolicy = task->mempolicy;
122 	mpol_get(priv->task_mempolicy);
123 	task_unlock(task);
124 }
125 static void release_task_mempolicy(struct proc_maps_private *priv)
126 {
127 	mpol_put(priv->task_mempolicy);
128 }
129 #else
130 static void hold_task_mempolicy(struct proc_maps_private *priv)
131 {
132 }
133 static void release_task_mempolicy(struct proc_maps_private *priv)
134 {
135 }
136 #endif
137 
138 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
139 {
140 	if (vma && vma != priv->tail_vma) {
141 		struct mm_struct *mm = vma->vm_mm;
142 		release_task_mempolicy(priv);
143 		up_read(&mm->mmap_sem);
144 		mmput(mm);
145 	}
146 }
147 
148 static void *m_start(struct seq_file *m, loff_t *pos)
149 {
150 	struct proc_maps_private *priv = m->private;
151 	unsigned long last_addr = m->version;
152 	struct mm_struct *mm;
153 	struct vm_area_struct *vma, *tail_vma = NULL;
154 	loff_t l = *pos;
155 
156 	/* Clear the per syscall fields in priv */
157 	priv->task = NULL;
158 	priv->tail_vma = NULL;
159 
160 	/*
161 	 * We remember last_addr rather than next_addr to hit with
162 	 * mmap_cache most of the time. We have zero last_addr at
163 	 * the beginning and also after lseek. We will have -1 last_addr
164 	 * after the end of the vmas.
165 	 */
166 
167 	if (last_addr == -1UL)
168 		return NULL;
169 
170 	priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
171 	if (!priv->task)
172 		return ERR_PTR(-ESRCH);
173 
174 	mm = mm_access(priv->task, PTRACE_MODE_READ);
175 	if (!mm || IS_ERR(mm))
176 		return mm;
177 	down_read(&mm->mmap_sem);
178 
179 	tail_vma = get_gate_vma(priv->task->mm);
180 	priv->tail_vma = tail_vma;
181 	hold_task_mempolicy(priv);
182 	/* Start with last addr hint */
183 	vma = find_vma(mm, last_addr);
184 	if (last_addr && vma) {
185 		vma = vma->vm_next;
186 		goto out;
187 	}
188 
189 	/*
190 	 * Check the vma index is within the range and do
191 	 * sequential scan until m_index.
192 	 */
193 	vma = NULL;
194 	if ((unsigned long)l < mm->map_count) {
195 		vma = mm->mmap;
196 		while (l-- && vma)
197 			vma = vma->vm_next;
198 		goto out;
199 	}
200 
201 	if (l != mm->map_count)
202 		tail_vma = NULL; /* After gate vma */
203 
204 out:
205 	if (vma)
206 		return vma;
207 
208 	release_task_mempolicy(priv);
209 	/* End of vmas has been reached */
210 	m->version = (tail_vma != NULL)? 0: -1UL;
211 	up_read(&mm->mmap_sem);
212 	mmput(mm);
213 	return tail_vma;
214 }
215 
216 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
217 {
218 	struct proc_maps_private *priv = m->private;
219 	struct vm_area_struct *vma = v;
220 	struct vm_area_struct *tail_vma = priv->tail_vma;
221 
222 	(*pos)++;
223 	if (vma && (vma != tail_vma) && vma->vm_next)
224 		return vma->vm_next;
225 	vma_stop(priv, vma);
226 	return (vma != tail_vma)? tail_vma: NULL;
227 }
228 
229 static void m_stop(struct seq_file *m, void *v)
230 {
231 	struct proc_maps_private *priv = m->private;
232 	struct vm_area_struct *vma = v;
233 
234 	if (!IS_ERR(vma))
235 		vma_stop(priv, vma);
236 	if (priv->task)
237 		put_task_struct(priv->task);
238 }
239 
240 static int do_maps_open(struct inode *inode, struct file *file,
241 			const struct seq_operations *ops)
242 {
243 	struct proc_maps_private *priv;
244 	int ret = -ENOMEM;
245 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
246 	if (priv) {
247 		priv->pid = proc_pid(inode);
248 		ret = seq_open(file, ops);
249 		if (!ret) {
250 			struct seq_file *m = file->private_data;
251 			m->private = priv;
252 		} else {
253 			kfree(priv);
254 		}
255 	}
256 	return ret;
257 }
258 
259 static void
260 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
261 {
262 	struct mm_struct *mm = vma->vm_mm;
263 	struct file *file = vma->vm_file;
264 	struct proc_maps_private *priv = m->private;
265 	struct task_struct *task = priv->task;
266 	vm_flags_t flags = vma->vm_flags;
267 	unsigned long ino = 0;
268 	unsigned long long pgoff = 0;
269 	unsigned long start, end;
270 	dev_t dev = 0;
271 	int len;
272 	const char *name = NULL;
273 
274 	if (file) {
275 		struct inode *inode = file_inode(vma->vm_file);
276 		dev = inode->i_sb->s_dev;
277 		ino = inode->i_ino;
278 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
279 	}
280 
281 	/* We don't show the stack guard page in /proc/maps */
282 	start = vma->vm_start;
283 	if (stack_guard_page_start(vma, start))
284 		start += PAGE_SIZE;
285 	end = vma->vm_end;
286 	if (stack_guard_page_end(vma, end))
287 		end -= PAGE_SIZE;
288 
289 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
290 			start,
291 			end,
292 			flags & VM_READ ? 'r' : '-',
293 			flags & VM_WRITE ? 'w' : '-',
294 			flags & VM_EXEC ? 'x' : '-',
295 			flags & VM_MAYSHARE ? 's' : 'p',
296 			pgoff,
297 			MAJOR(dev), MINOR(dev), ino, &len);
298 
299 	/*
300 	 * Print the dentry name for named mappings, and a
301 	 * special [heap] marker for the heap:
302 	 */
303 	if (file) {
304 		pad_len_spaces(m, len);
305 		seq_path(m, &file->f_path, "\n");
306 		goto done;
307 	}
308 
309 	name = arch_vma_name(vma);
310 	if (!name) {
311 		pid_t tid;
312 
313 		if (!mm) {
314 			name = "[vdso]";
315 			goto done;
316 		}
317 
318 		if (vma->vm_start <= mm->brk &&
319 		    vma->vm_end >= mm->start_brk) {
320 			name = "[heap]";
321 			goto done;
322 		}
323 
324 		tid = vm_is_stack(task, vma, is_pid);
325 
326 		if (tid != 0) {
327 			/*
328 			 * Thread stack in /proc/PID/task/TID/maps or
329 			 * the main process stack.
330 			 */
331 			if (!is_pid || (vma->vm_start <= mm->start_stack &&
332 			    vma->vm_end >= mm->start_stack)) {
333 				name = "[stack]";
334 			} else {
335 				/* Thread stack in /proc/PID/maps */
336 				pad_len_spaces(m, len);
337 				seq_printf(m, "[stack:%d]", tid);
338 			}
339 		}
340 	}
341 
342 done:
343 	if (name) {
344 		pad_len_spaces(m, len);
345 		seq_puts(m, name);
346 	}
347 	seq_putc(m, '\n');
348 }
349 
350 static int show_map(struct seq_file *m, void *v, int is_pid)
351 {
352 	struct vm_area_struct *vma = v;
353 	struct proc_maps_private *priv = m->private;
354 	struct task_struct *task = priv->task;
355 
356 	show_map_vma(m, vma, is_pid);
357 
358 	if (m->count < m->size)  /* vma is copied successfully */
359 		m->version = (vma != get_gate_vma(task->mm))
360 			? vma->vm_start : 0;
361 	return 0;
362 }
363 
364 static int show_pid_map(struct seq_file *m, void *v)
365 {
366 	return show_map(m, v, 1);
367 }
368 
369 static int show_tid_map(struct seq_file *m, void *v)
370 {
371 	return show_map(m, v, 0);
372 }
373 
374 static const struct seq_operations proc_pid_maps_op = {
375 	.start	= m_start,
376 	.next	= m_next,
377 	.stop	= m_stop,
378 	.show	= show_pid_map
379 };
380 
381 static const struct seq_operations proc_tid_maps_op = {
382 	.start	= m_start,
383 	.next	= m_next,
384 	.stop	= m_stop,
385 	.show	= show_tid_map
386 };
387 
388 static int pid_maps_open(struct inode *inode, struct file *file)
389 {
390 	return do_maps_open(inode, file, &proc_pid_maps_op);
391 }
392 
393 static int tid_maps_open(struct inode *inode, struct file *file)
394 {
395 	return do_maps_open(inode, file, &proc_tid_maps_op);
396 }
397 
398 const struct file_operations proc_pid_maps_operations = {
399 	.open		= pid_maps_open,
400 	.read		= seq_read,
401 	.llseek		= seq_lseek,
402 	.release	= seq_release_private,
403 };
404 
405 const struct file_operations proc_tid_maps_operations = {
406 	.open		= tid_maps_open,
407 	.read		= seq_read,
408 	.llseek		= seq_lseek,
409 	.release	= seq_release_private,
410 };
411 
412 /*
413  * Proportional Set Size(PSS): my share of RSS.
414  *
415  * PSS of a process is the count of pages it has in memory, where each
416  * page is divided by the number of processes sharing it.  So if a
417  * process has 1000 pages all to itself, and 1000 shared with one other
418  * process, its PSS will be 1500.
419  *
420  * To keep (accumulated) division errors low, we adopt a 64bit
421  * fixed-point pss counter to minimize division errors. So (pss >>
422  * PSS_SHIFT) would be the real byte count.
423  *
424  * A shift of 12 before division means (assuming 4K page size):
425  * 	- 1M 3-user-pages add up to 8KB errors;
426  * 	- supports mapcount up to 2^24, or 16M;
427  * 	- supports PSS up to 2^52 bytes, or 4PB.
428  */
429 #define PSS_SHIFT 12
430 
431 #ifdef CONFIG_PROC_PAGE_MONITOR
432 struct mem_size_stats {
433 	struct vm_area_struct *vma;
434 	unsigned long resident;
435 	unsigned long shared_clean;
436 	unsigned long shared_dirty;
437 	unsigned long private_clean;
438 	unsigned long private_dirty;
439 	unsigned long referenced;
440 	unsigned long anonymous;
441 	unsigned long anonymous_thp;
442 	unsigned long swap;
443 	unsigned long nonlinear;
444 	u64 pss;
445 };
446 
447 
448 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
449 		unsigned long ptent_size, struct mm_walk *walk)
450 {
451 	struct mem_size_stats *mss = walk->private;
452 	struct vm_area_struct *vma = mss->vma;
453 	pgoff_t pgoff = linear_page_index(vma, addr);
454 	struct page *page = NULL;
455 	int mapcount;
456 
457 	if (pte_present(ptent)) {
458 		page = vm_normal_page(vma, addr, ptent);
459 	} else if (is_swap_pte(ptent)) {
460 		swp_entry_t swpent = pte_to_swp_entry(ptent);
461 
462 		if (!non_swap_entry(swpent))
463 			mss->swap += ptent_size;
464 		else if (is_migration_entry(swpent))
465 			page = migration_entry_to_page(swpent);
466 	} else if (pte_file(ptent)) {
467 		if (pte_to_pgoff(ptent) != pgoff)
468 			mss->nonlinear += ptent_size;
469 	}
470 
471 	if (!page)
472 		return;
473 
474 	if (PageAnon(page))
475 		mss->anonymous += ptent_size;
476 
477 	if (page->index != pgoff)
478 		mss->nonlinear += ptent_size;
479 
480 	mss->resident += ptent_size;
481 	/* Accumulate the size in pages that have been accessed. */
482 	if (pte_young(ptent) || PageReferenced(page))
483 		mss->referenced += ptent_size;
484 	mapcount = page_mapcount(page);
485 	if (mapcount >= 2) {
486 		if (pte_dirty(ptent) || PageDirty(page))
487 			mss->shared_dirty += ptent_size;
488 		else
489 			mss->shared_clean += ptent_size;
490 		mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
491 	} else {
492 		if (pte_dirty(ptent) || PageDirty(page))
493 			mss->private_dirty += ptent_size;
494 		else
495 			mss->private_clean += ptent_size;
496 		mss->pss += (ptent_size << PSS_SHIFT);
497 	}
498 }
499 
500 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
501 			   struct mm_walk *walk)
502 {
503 	struct mem_size_stats *mss = walk->private;
504 	struct vm_area_struct *vma = mss->vma;
505 	pte_t *pte;
506 	spinlock_t *ptl;
507 
508 	if (pmd_trans_huge_lock(pmd, vma) == 1) {
509 		smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
510 		spin_unlock(&walk->mm->page_table_lock);
511 		mss->anonymous_thp += HPAGE_PMD_SIZE;
512 		return 0;
513 	}
514 
515 	if (pmd_trans_unstable(pmd))
516 		return 0;
517 	/*
518 	 * The mmap_sem held all the way back in m_start() is what
519 	 * keeps khugepaged out of here and from collapsing things
520 	 * in here.
521 	 */
522 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
523 	for (; addr != end; pte++, addr += PAGE_SIZE)
524 		smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
525 	pte_unmap_unlock(pte - 1, ptl);
526 	cond_resched();
527 	return 0;
528 }
529 
530 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
531 {
532 	/*
533 	 * Don't forget to update Documentation/ on changes.
534 	 */
535 	static const char mnemonics[BITS_PER_LONG][2] = {
536 		/*
537 		 * In case if we meet a flag we don't know about.
538 		 */
539 		[0 ... (BITS_PER_LONG-1)] = "??",
540 
541 		[ilog2(VM_READ)]	= "rd",
542 		[ilog2(VM_WRITE)]	= "wr",
543 		[ilog2(VM_EXEC)]	= "ex",
544 		[ilog2(VM_SHARED)]	= "sh",
545 		[ilog2(VM_MAYREAD)]	= "mr",
546 		[ilog2(VM_MAYWRITE)]	= "mw",
547 		[ilog2(VM_MAYEXEC)]	= "me",
548 		[ilog2(VM_MAYSHARE)]	= "ms",
549 		[ilog2(VM_GROWSDOWN)]	= "gd",
550 		[ilog2(VM_PFNMAP)]	= "pf",
551 		[ilog2(VM_DENYWRITE)]	= "dw",
552 		[ilog2(VM_LOCKED)]	= "lo",
553 		[ilog2(VM_IO)]		= "io",
554 		[ilog2(VM_SEQ_READ)]	= "sr",
555 		[ilog2(VM_RAND_READ)]	= "rr",
556 		[ilog2(VM_DONTCOPY)]	= "dc",
557 		[ilog2(VM_DONTEXPAND)]	= "de",
558 		[ilog2(VM_ACCOUNT)]	= "ac",
559 		[ilog2(VM_NORESERVE)]	= "nr",
560 		[ilog2(VM_HUGETLB)]	= "ht",
561 		[ilog2(VM_NONLINEAR)]	= "nl",
562 		[ilog2(VM_ARCH_1)]	= "ar",
563 		[ilog2(VM_DONTDUMP)]	= "dd",
564 		[ilog2(VM_MIXEDMAP)]	= "mm",
565 		[ilog2(VM_HUGEPAGE)]	= "hg",
566 		[ilog2(VM_NOHUGEPAGE)]	= "nh",
567 		[ilog2(VM_MERGEABLE)]	= "mg",
568 	};
569 	size_t i;
570 
571 	seq_puts(m, "VmFlags: ");
572 	for (i = 0; i < BITS_PER_LONG; i++) {
573 		if (vma->vm_flags & (1UL << i)) {
574 			seq_printf(m, "%c%c ",
575 				   mnemonics[i][0], mnemonics[i][1]);
576 		}
577 	}
578 	seq_putc(m, '\n');
579 }
580 
581 static int show_smap(struct seq_file *m, void *v, int is_pid)
582 {
583 	struct proc_maps_private *priv = m->private;
584 	struct task_struct *task = priv->task;
585 	struct vm_area_struct *vma = v;
586 	struct mem_size_stats mss;
587 	struct mm_walk smaps_walk = {
588 		.pmd_entry = smaps_pte_range,
589 		.mm = vma->vm_mm,
590 		.private = &mss,
591 	};
592 
593 	memset(&mss, 0, sizeof mss);
594 	mss.vma = vma;
595 	/* mmap_sem is held in m_start */
596 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
597 		walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
598 
599 	show_map_vma(m, vma, is_pid);
600 
601 	seq_printf(m,
602 		   "Size:           %8lu kB\n"
603 		   "Rss:            %8lu kB\n"
604 		   "Pss:            %8lu kB\n"
605 		   "Shared_Clean:   %8lu kB\n"
606 		   "Shared_Dirty:   %8lu kB\n"
607 		   "Private_Clean:  %8lu kB\n"
608 		   "Private_Dirty:  %8lu kB\n"
609 		   "Referenced:     %8lu kB\n"
610 		   "Anonymous:      %8lu kB\n"
611 		   "AnonHugePages:  %8lu kB\n"
612 		   "Swap:           %8lu kB\n"
613 		   "KernelPageSize: %8lu kB\n"
614 		   "MMUPageSize:    %8lu kB\n"
615 		   "Locked:         %8lu kB\n",
616 		   (vma->vm_end - vma->vm_start) >> 10,
617 		   mss.resident >> 10,
618 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
619 		   mss.shared_clean  >> 10,
620 		   mss.shared_dirty  >> 10,
621 		   mss.private_clean >> 10,
622 		   mss.private_dirty >> 10,
623 		   mss.referenced >> 10,
624 		   mss.anonymous >> 10,
625 		   mss.anonymous_thp >> 10,
626 		   mss.swap >> 10,
627 		   vma_kernel_pagesize(vma) >> 10,
628 		   vma_mmu_pagesize(vma) >> 10,
629 		   (vma->vm_flags & VM_LOCKED) ?
630 			(unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
631 
632 	if (vma->vm_flags & VM_NONLINEAR)
633 		seq_printf(m, "Nonlinear:      %8lu kB\n",
634 				mss.nonlinear >> 10);
635 
636 	show_smap_vma_flags(m, vma);
637 
638 	if (m->count < m->size)  /* vma is copied successfully */
639 		m->version = (vma != get_gate_vma(task->mm))
640 			? vma->vm_start : 0;
641 	return 0;
642 }
643 
644 static int show_pid_smap(struct seq_file *m, void *v)
645 {
646 	return show_smap(m, v, 1);
647 }
648 
649 static int show_tid_smap(struct seq_file *m, void *v)
650 {
651 	return show_smap(m, v, 0);
652 }
653 
654 static const struct seq_operations proc_pid_smaps_op = {
655 	.start	= m_start,
656 	.next	= m_next,
657 	.stop	= m_stop,
658 	.show	= show_pid_smap
659 };
660 
661 static const struct seq_operations proc_tid_smaps_op = {
662 	.start	= m_start,
663 	.next	= m_next,
664 	.stop	= m_stop,
665 	.show	= show_tid_smap
666 };
667 
668 static int pid_smaps_open(struct inode *inode, struct file *file)
669 {
670 	return do_maps_open(inode, file, &proc_pid_smaps_op);
671 }
672 
673 static int tid_smaps_open(struct inode *inode, struct file *file)
674 {
675 	return do_maps_open(inode, file, &proc_tid_smaps_op);
676 }
677 
678 const struct file_operations proc_pid_smaps_operations = {
679 	.open		= pid_smaps_open,
680 	.read		= seq_read,
681 	.llseek		= seq_lseek,
682 	.release	= seq_release_private,
683 };
684 
685 const struct file_operations proc_tid_smaps_operations = {
686 	.open		= tid_smaps_open,
687 	.read		= seq_read,
688 	.llseek		= seq_lseek,
689 	.release	= seq_release_private,
690 };
691 
692 /*
693  * We do not want to have constant page-shift bits sitting in
694  * pagemap entries and are about to reuse them some time soon.
695  *
696  * Here's the "migration strategy":
697  * 1. when the system boots these bits remain what they are,
698  *    but a warning about future change is printed in log;
699  * 2. once anyone clears soft-dirty bits via clear_refs file,
700  *    these flag is set to denote, that user is aware of the
701  *    new API and those page-shift bits change their meaning.
702  *    The respective warning is printed in dmesg;
703  * 3. In a couple of releases we will remove all the mentions
704  *    of page-shift in pagemap entries.
705  */
706 
707 static bool soft_dirty_cleared __read_mostly;
708 
709 enum clear_refs_types {
710 	CLEAR_REFS_ALL = 1,
711 	CLEAR_REFS_ANON,
712 	CLEAR_REFS_MAPPED,
713 	CLEAR_REFS_SOFT_DIRTY,
714 	CLEAR_REFS_LAST,
715 };
716 
717 struct clear_refs_private {
718 	struct vm_area_struct *vma;
719 	enum clear_refs_types type;
720 };
721 
722 static inline void clear_soft_dirty(struct vm_area_struct *vma,
723 		unsigned long addr, pte_t *pte)
724 {
725 #ifdef CONFIG_MEM_SOFT_DIRTY
726 	/*
727 	 * The soft-dirty tracker uses #PF-s to catch writes
728 	 * to pages, so write-protect the pte as well. See the
729 	 * Documentation/vm/soft-dirty.txt for full description
730 	 * of how soft-dirty works.
731 	 */
732 	pte_t ptent = *pte;
733 
734 	if (pte_present(ptent)) {
735 		ptent = pte_wrprotect(ptent);
736 		ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY);
737 	} else if (is_swap_pte(ptent)) {
738 		ptent = pte_swp_clear_soft_dirty(ptent);
739 	} else if (pte_file(ptent)) {
740 		ptent = pte_file_clear_soft_dirty(ptent);
741 	}
742 
743 	if (vma->vm_flags & VM_SOFTDIRTY)
744 		vma->vm_flags &= ~VM_SOFTDIRTY;
745 
746 	set_pte_at(vma->vm_mm, addr, pte, ptent);
747 #endif
748 }
749 
750 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
751 				unsigned long end, struct mm_walk *walk)
752 {
753 	struct clear_refs_private *cp = walk->private;
754 	struct vm_area_struct *vma = cp->vma;
755 	pte_t *pte, ptent;
756 	spinlock_t *ptl;
757 	struct page *page;
758 
759 	split_huge_page_pmd(vma, addr, pmd);
760 	if (pmd_trans_unstable(pmd))
761 		return 0;
762 
763 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
764 	for (; addr != end; pte++, addr += PAGE_SIZE) {
765 		ptent = *pte;
766 
767 		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
768 			clear_soft_dirty(vma, addr, pte);
769 			continue;
770 		}
771 
772 		if (!pte_present(ptent))
773 			continue;
774 
775 		page = vm_normal_page(vma, addr, ptent);
776 		if (!page)
777 			continue;
778 
779 		/* Clear accessed and referenced bits. */
780 		ptep_test_and_clear_young(vma, addr, pte);
781 		ClearPageReferenced(page);
782 	}
783 	pte_unmap_unlock(pte - 1, ptl);
784 	cond_resched();
785 	return 0;
786 }
787 
788 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
789 				size_t count, loff_t *ppos)
790 {
791 	struct task_struct *task;
792 	char buffer[PROC_NUMBUF];
793 	struct mm_struct *mm;
794 	struct vm_area_struct *vma;
795 	enum clear_refs_types type;
796 	int itype;
797 	int rv;
798 
799 	memset(buffer, 0, sizeof(buffer));
800 	if (count > sizeof(buffer) - 1)
801 		count = sizeof(buffer) - 1;
802 	if (copy_from_user(buffer, buf, count))
803 		return -EFAULT;
804 	rv = kstrtoint(strstrip(buffer), 10, &itype);
805 	if (rv < 0)
806 		return rv;
807 	type = (enum clear_refs_types)itype;
808 	if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
809 		return -EINVAL;
810 
811 	if (type == CLEAR_REFS_SOFT_DIRTY) {
812 		soft_dirty_cleared = true;
813 		pr_warn_once("The pagemap bits 55-60 has changed their meaning! "
814 				"See the linux/Documentation/vm/pagemap.txt for details.\n");
815 	}
816 
817 	task = get_proc_task(file_inode(file));
818 	if (!task)
819 		return -ESRCH;
820 	mm = get_task_mm(task);
821 	if (mm) {
822 		struct clear_refs_private cp = {
823 			.type = type,
824 		};
825 		struct mm_walk clear_refs_walk = {
826 			.pmd_entry = clear_refs_pte_range,
827 			.mm = mm,
828 			.private = &cp,
829 		};
830 		down_read(&mm->mmap_sem);
831 		if (type == CLEAR_REFS_SOFT_DIRTY)
832 			mmu_notifier_invalidate_range_start(mm, 0, -1);
833 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
834 			cp.vma = vma;
835 			if (is_vm_hugetlb_page(vma))
836 				continue;
837 			/*
838 			 * Writing 1 to /proc/pid/clear_refs affects all pages.
839 			 *
840 			 * Writing 2 to /proc/pid/clear_refs only affects
841 			 * Anonymous pages.
842 			 *
843 			 * Writing 3 to /proc/pid/clear_refs only affects file
844 			 * mapped pages.
845 			 */
846 			if (type == CLEAR_REFS_ANON && vma->vm_file)
847 				continue;
848 			if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
849 				continue;
850 			walk_page_range(vma->vm_start, vma->vm_end,
851 					&clear_refs_walk);
852 		}
853 		if (type == CLEAR_REFS_SOFT_DIRTY)
854 			mmu_notifier_invalidate_range_end(mm, 0, -1);
855 		flush_tlb_mm(mm);
856 		up_read(&mm->mmap_sem);
857 		mmput(mm);
858 	}
859 	put_task_struct(task);
860 
861 	return count;
862 }
863 
864 const struct file_operations proc_clear_refs_operations = {
865 	.write		= clear_refs_write,
866 	.llseek		= noop_llseek,
867 };
868 
869 typedef struct {
870 	u64 pme;
871 } pagemap_entry_t;
872 
873 struct pagemapread {
874 	int pos, len;		/* units: PM_ENTRY_BYTES, not bytes */
875 	pagemap_entry_t *buffer;
876 	bool v2;
877 };
878 
879 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
880 #define PAGEMAP_WALK_MASK	(PMD_MASK)
881 
882 #define PM_ENTRY_BYTES      sizeof(pagemap_entry_t)
883 #define PM_STATUS_BITS      3
884 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
885 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
886 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
887 #define PM_PSHIFT_BITS      6
888 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
889 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
890 #define __PM_PSHIFT(x)      (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
891 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
892 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
893 /* in "new" pagemap pshift bits are occupied with more status bits */
894 #define PM_STATUS2(v2, x)   (__PM_PSHIFT(v2 ? x : PAGE_SHIFT))
895 
896 #define __PM_SOFT_DIRTY      (1LL)
897 #define PM_PRESENT          PM_STATUS(4LL)
898 #define PM_SWAP             PM_STATUS(2LL)
899 #define PM_FILE             PM_STATUS(1LL)
900 #define PM_NOT_PRESENT(v2)  PM_STATUS2(v2, 0)
901 #define PM_END_OF_BUFFER    1
902 
903 static inline pagemap_entry_t make_pme(u64 val)
904 {
905 	return (pagemap_entry_t) { .pme = val };
906 }
907 
908 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
909 			  struct pagemapread *pm)
910 {
911 	pm->buffer[pm->pos++] = *pme;
912 	if (pm->pos >= pm->len)
913 		return PM_END_OF_BUFFER;
914 	return 0;
915 }
916 
917 static int pagemap_pte_hole(unsigned long start, unsigned long end,
918 				struct mm_walk *walk)
919 {
920 	struct pagemapread *pm = walk->private;
921 	unsigned long addr;
922 	int err = 0;
923 	pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
924 
925 	for (addr = start; addr < end; addr += PAGE_SIZE) {
926 		err = add_to_pagemap(addr, &pme, pm);
927 		if (err)
928 			break;
929 	}
930 	return err;
931 }
932 
933 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
934 		struct vm_area_struct *vma, unsigned long addr, pte_t pte)
935 {
936 	u64 frame, flags;
937 	struct page *page = NULL;
938 	int flags2 = 0;
939 
940 	if (pte_present(pte)) {
941 		frame = pte_pfn(pte);
942 		flags = PM_PRESENT;
943 		page = vm_normal_page(vma, addr, pte);
944 		if (pte_soft_dirty(pte))
945 			flags2 |= __PM_SOFT_DIRTY;
946 	} else if (is_swap_pte(pte)) {
947 		swp_entry_t entry;
948 		if (pte_swp_soft_dirty(pte))
949 			flags2 |= __PM_SOFT_DIRTY;
950 		entry = pte_to_swp_entry(pte);
951 		frame = swp_type(entry) |
952 			(swp_offset(entry) << MAX_SWAPFILES_SHIFT);
953 		flags = PM_SWAP;
954 		if (is_migration_entry(entry))
955 			page = migration_entry_to_page(entry);
956 	} else {
957 		if (vma->vm_flags & VM_SOFTDIRTY)
958 			flags2 |= __PM_SOFT_DIRTY;
959 		*pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
960 		return;
961 	}
962 
963 	if (page && !PageAnon(page))
964 		flags |= PM_FILE;
965 	if ((vma->vm_flags & VM_SOFTDIRTY))
966 		flags2 |= __PM_SOFT_DIRTY;
967 
968 	*pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags);
969 }
970 
971 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
972 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
973 		pmd_t pmd, int offset, int pmd_flags2)
974 {
975 	/*
976 	 * Currently pmd for thp is always present because thp can not be
977 	 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
978 	 * This if-check is just to prepare for future implementation.
979 	 */
980 	if (pmd_present(pmd))
981 		*pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
982 				| PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT);
983 	else
984 		*pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2));
985 }
986 #else
987 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
988 		pmd_t pmd, int offset, int pmd_flags2)
989 {
990 }
991 #endif
992 
993 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
994 			     struct mm_walk *walk)
995 {
996 	struct vm_area_struct *vma;
997 	struct pagemapread *pm = walk->private;
998 	pte_t *pte;
999 	int err = 0;
1000 	pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
1001 
1002 	/* find the first VMA at or above 'addr' */
1003 	vma = find_vma(walk->mm, addr);
1004 	if (vma && pmd_trans_huge_lock(pmd, vma) == 1) {
1005 		int pmd_flags2;
1006 
1007 		if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd))
1008 			pmd_flags2 = __PM_SOFT_DIRTY;
1009 		else
1010 			pmd_flags2 = 0;
1011 
1012 		for (; addr != end; addr += PAGE_SIZE) {
1013 			unsigned long offset;
1014 
1015 			offset = (addr & ~PAGEMAP_WALK_MASK) >>
1016 					PAGE_SHIFT;
1017 			thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2);
1018 			err = add_to_pagemap(addr, &pme, pm);
1019 			if (err)
1020 				break;
1021 		}
1022 		spin_unlock(&walk->mm->page_table_lock);
1023 		return err;
1024 	}
1025 
1026 	if (pmd_trans_unstable(pmd))
1027 		return 0;
1028 	for (; addr != end; addr += PAGE_SIZE) {
1029 		int flags2;
1030 
1031 		/* check to see if we've left 'vma' behind
1032 		 * and need a new, higher one */
1033 		if (vma && (addr >= vma->vm_end)) {
1034 			vma = find_vma(walk->mm, addr);
1035 			if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1036 				flags2 = __PM_SOFT_DIRTY;
1037 			else
1038 				flags2 = 0;
1039 			pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
1040 		}
1041 
1042 		/* check that 'vma' actually covers this address,
1043 		 * and that it isn't a huge page vma */
1044 		if (vma && (vma->vm_start <= addr) &&
1045 		    !is_vm_hugetlb_page(vma)) {
1046 			pte = pte_offset_map(pmd, addr);
1047 			pte_to_pagemap_entry(&pme, pm, vma, addr, *pte);
1048 			/* unmap before userspace copy */
1049 			pte_unmap(pte);
1050 		}
1051 		err = add_to_pagemap(addr, &pme, pm);
1052 		if (err)
1053 			return err;
1054 	}
1055 
1056 	cond_resched();
1057 
1058 	return err;
1059 }
1060 
1061 #ifdef CONFIG_HUGETLB_PAGE
1062 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1063 					pte_t pte, int offset, int flags2)
1064 {
1065 	if (pte_present(pte))
1066 		*pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset)	|
1067 				PM_STATUS2(pm->v2, flags2)		|
1068 				PM_PRESENT);
1069 	else
1070 		*pme = make_pme(PM_NOT_PRESENT(pm->v2)			|
1071 				PM_STATUS2(pm->v2, flags2));
1072 }
1073 
1074 /* This function walks within one hugetlb entry in the single call */
1075 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1076 				 unsigned long addr, unsigned long end,
1077 				 struct mm_walk *walk)
1078 {
1079 	struct pagemapread *pm = walk->private;
1080 	struct vm_area_struct *vma;
1081 	int err = 0;
1082 	int flags2;
1083 	pagemap_entry_t pme;
1084 
1085 	vma = find_vma(walk->mm, addr);
1086 	WARN_ON_ONCE(!vma);
1087 
1088 	if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1089 		flags2 = __PM_SOFT_DIRTY;
1090 	else
1091 		flags2 = 0;
1092 
1093 	for (; addr != end; addr += PAGE_SIZE) {
1094 		int offset = (addr & ~hmask) >> PAGE_SHIFT;
1095 		huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2);
1096 		err = add_to_pagemap(addr, &pme, pm);
1097 		if (err)
1098 			return err;
1099 	}
1100 
1101 	cond_resched();
1102 
1103 	return err;
1104 }
1105 #endif /* HUGETLB_PAGE */
1106 
1107 /*
1108  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1109  *
1110  * For each page in the address space, this file contains one 64-bit entry
1111  * consisting of the following:
1112  *
1113  * Bits 0-54  page frame number (PFN) if present
1114  * Bits 0-4   swap type if swapped
1115  * Bits 5-54  swap offset if swapped
1116  * Bits 55-60 page shift (page size = 1<<page shift)
1117  * Bit  61    page is file-page or shared-anon
1118  * Bit  62    page swapped
1119  * Bit  63    page present
1120  *
1121  * If the page is not present but in swap, then the PFN contains an
1122  * encoding of the swap file number and the page's offset into the
1123  * swap. Unmapped pages return a null PFN. This allows determining
1124  * precisely which pages are mapped (or in swap) and comparing mapped
1125  * pages between processes.
1126  *
1127  * Efficient users of this interface will use /proc/pid/maps to
1128  * determine which areas of memory are actually mapped and llseek to
1129  * skip over unmapped regions.
1130  */
1131 static ssize_t pagemap_read(struct file *file, char __user *buf,
1132 			    size_t count, loff_t *ppos)
1133 {
1134 	struct task_struct *task = get_proc_task(file_inode(file));
1135 	struct mm_struct *mm;
1136 	struct pagemapread pm;
1137 	int ret = -ESRCH;
1138 	struct mm_walk pagemap_walk = {};
1139 	unsigned long src;
1140 	unsigned long svpfn;
1141 	unsigned long start_vaddr;
1142 	unsigned long end_vaddr;
1143 	int copied = 0;
1144 
1145 	if (!task)
1146 		goto out;
1147 
1148 	ret = -EINVAL;
1149 	/* file position must be aligned */
1150 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1151 		goto out_task;
1152 
1153 	ret = 0;
1154 	if (!count)
1155 		goto out_task;
1156 
1157 	pm.v2 = soft_dirty_cleared;
1158 	pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1159 	pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1160 	ret = -ENOMEM;
1161 	if (!pm.buffer)
1162 		goto out_task;
1163 
1164 	mm = mm_access(task, PTRACE_MODE_READ);
1165 	ret = PTR_ERR(mm);
1166 	if (!mm || IS_ERR(mm))
1167 		goto out_free;
1168 
1169 	pagemap_walk.pmd_entry = pagemap_pte_range;
1170 	pagemap_walk.pte_hole = pagemap_pte_hole;
1171 #ifdef CONFIG_HUGETLB_PAGE
1172 	pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1173 #endif
1174 	pagemap_walk.mm = mm;
1175 	pagemap_walk.private = &pm;
1176 
1177 	src = *ppos;
1178 	svpfn = src / PM_ENTRY_BYTES;
1179 	start_vaddr = svpfn << PAGE_SHIFT;
1180 	end_vaddr = TASK_SIZE_OF(task);
1181 
1182 	/* watch out for wraparound */
1183 	if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1184 		start_vaddr = end_vaddr;
1185 
1186 	/*
1187 	 * The odds are that this will stop walking way
1188 	 * before end_vaddr, because the length of the
1189 	 * user buffer is tracked in "pm", and the walk
1190 	 * will stop when we hit the end of the buffer.
1191 	 */
1192 	ret = 0;
1193 	while (count && (start_vaddr < end_vaddr)) {
1194 		int len;
1195 		unsigned long end;
1196 
1197 		pm.pos = 0;
1198 		end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1199 		/* overflow ? */
1200 		if (end < start_vaddr || end > end_vaddr)
1201 			end = end_vaddr;
1202 		down_read(&mm->mmap_sem);
1203 		ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1204 		up_read(&mm->mmap_sem);
1205 		start_vaddr = end;
1206 
1207 		len = min(count, PM_ENTRY_BYTES * pm.pos);
1208 		if (copy_to_user(buf, pm.buffer, len)) {
1209 			ret = -EFAULT;
1210 			goto out_mm;
1211 		}
1212 		copied += len;
1213 		buf += len;
1214 		count -= len;
1215 	}
1216 	*ppos += copied;
1217 	if (!ret || ret == PM_END_OF_BUFFER)
1218 		ret = copied;
1219 
1220 out_mm:
1221 	mmput(mm);
1222 out_free:
1223 	kfree(pm.buffer);
1224 out_task:
1225 	put_task_struct(task);
1226 out:
1227 	return ret;
1228 }
1229 
1230 static int pagemap_open(struct inode *inode, struct file *file)
1231 {
1232 	pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about "
1233 			"to stop being page-shift some time soon. See the "
1234 			"linux/Documentation/vm/pagemap.txt for details.\n");
1235 	return 0;
1236 }
1237 
1238 const struct file_operations proc_pagemap_operations = {
1239 	.llseek		= mem_lseek, /* borrow this */
1240 	.read		= pagemap_read,
1241 	.open		= pagemap_open,
1242 };
1243 #endif /* CONFIG_PROC_PAGE_MONITOR */
1244 
1245 #ifdef CONFIG_NUMA
1246 
1247 struct numa_maps {
1248 	struct vm_area_struct *vma;
1249 	unsigned long pages;
1250 	unsigned long anon;
1251 	unsigned long active;
1252 	unsigned long writeback;
1253 	unsigned long mapcount_max;
1254 	unsigned long dirty;
1255 	unsigned long swapcache;
1256 	unsigned long node[MAX_NUMNODES];
1257 };
1258 
1259 struct numa_maps_private {
1260 	struct proc_maps_private proc_maps;
1261 	struct numa_maps md;
1262 };
1263 
1264 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1265 			unsigned long nr_pages)
1266 {
1267 	int count = page_mapcount(page);
1268 
1269 	md->pages += nr_pages;
1270 	if (pte_dirty || PageDirty(page))
1271 		md->dirty += nr_pages;
1272 
1273 	if (PageSwapCache(page))
1274 		md->swapcache += nr_pages;
1275 
1276 	if (PageActive(page) || PageUnevictable(page))
1277 		md->active += nr_pages;
1278 
1279 	if (PageWriteback(page))
1280 		md->writeback += nr_pages;
1281 
1282 	if (PageAnon(page))
1283 		md->anon += nr_pages;
1284 
1285 	if (count > md->mapcount_max)
1286 		md->mapcount_max = count;
1287 
1288 	md->node[page_to_nid(page)] += nr_pages;
1289 }
1290 
1291 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1292 		unsigned long addr)
1293 {
1294 	struct page *page;
1295 	int nid;
1296 
1297 	if (!pte_present(pte))
1298 		return NULL;
1299 
1300 	page = vm_normal_page(vma, addr, pte);
1301 	if (!page)
1302 		return NULL;
1303 
1304 	if (PageReserved(page))
1305 		return NULL;
1306 
1307 	nid = page_to_nid(page);
1308 	if (!node_isset(nid, node_states[N_MEMORY]))
1309 		return NULL;
1310 
1311 	return page;
1312 }
1313 
1314 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1315 		unsigned long end, struct mm_walk *walk)
1316 {
1317 	struct numa_maps *md;
1318 	spinlock_t *ptl;
1319 	pte_t *orig_pte;
1320 	pte_t *pte;
1321 
1322 	md = walk->private;
1323 
1324 	if (pmd_trans_huge_lock(pmd, md->vma) == 1) {
1325 		pte_t huge_pte = *(pte_t *)pmd;
1326 		struct page *page;
1327 
1328 		page = can_gather_numa_stats(huge_pte, md->vma, addr);
1329 		if (page)
1330 			gather_stats(page, md, pte_dirty(huge_pte),
1331 				     HPAGE_PMD_SIZE/PAGE_SIZE);
1332 		spin_unlock(&walk->mm->page_table_lock);
1333 		return 0;
1334 	}
1335 
1336 	if (pmd_trans_unstable(pmd))
1337 		return 0;
1338 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1339 	do {
1340 		struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1341 		if (!page)
1342 			continue;
1343 		gather_stats(page, md, pte_dirty(*pte), 1);
1344 
1345 	} while (pte++, addr += PAGE_SIZE, addr != end);
1346 	pte_unmap_unlock(orig_pte, ptl);
1347 	return 0;
1348 }
1349 #ifdef CONFIG_HUGETLB_PAGE
1350 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1351 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1352 {
1353 	struct numa_maps *md;
1354 	struct page *page;
1355 
1356 	if (pte_none(*pte))
1357 		return 0;
1358 
1359 	page = pte_page(*pte);
1360 	if (!page)
1361 		return 0;
1362 
1363 	md = walk->private;
1364 	gather_stats(page, md, pte_dirty(*pte), 1);
1365 	return 0;
1366 }
1367 
1368 #else
1369 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1370 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1371 {
1372 	return 0;
1373 }
1374 #endif
1375 
1376 /*
1377  * Display pages allocated per node and memory policy via /proc.
1378  */
1379 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1380 {
1381 	struct numa_maps_private *numa_priv = m->private;
1382 	struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1383 	struct vm_area_struct *vma = v;
1384 	struct numa_maps *md = &numa_priv->md;
1385 	struct file *file = vma->vm_file;
1386 	struct task_struct *task = proc_priv->task;
1387 	struct mm_struct *mm = vma->vm_mm;
1388 	struct mm_walk walk = {};
1389 	struct mempolicy *pol;
1390 	int n;
1391 	char buffer[50];
1392 
1393 	if (!mm)
1394 		return 0;
1395 
1396 	/* Ensure we start with an empty set of numa_maps statistics. */
1397 	memset(md, 0, sizeof(*md));
1398 
1399 	md->vma = vma;
1400 
1401 	walk.hugetlb_entry = gather_hugetbl_stats;
1402 	walk.pmd_entry = gather_pte_stats;
1403 	walk.private = md;
1404 	walk.mm = mm;
1405 
1406 	pol = get_vma_policy(task, vma, vma->vm_start);
1407 	n = mpol_to_str(buffer, sizeof(buffer), pol);
1408 	mpol_cond_put(pol);
1409 	if (n < 0)
1410 		return n;
1411 
1412 	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1413 
1414 	if (file) {
1415 		seq_printf(m, " file=");
1416 		seq_path(m, &file->f_path, "\n\t= ");
1417 	} else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1418 		seq_printf(m, " heap");
1419 	} else {
1420 		pid_t tid = vm_is_stack(task, vma, is_pid);
1421 		if (tid != 0) {
1422 			/*
1423 			 * Thread stack in /proc/PID/task/TID/maps or
1424 			 * the main process stack.
1425 			 */
1426 			if (!is_pid || (vma->vm_start <= mm->start_stack &&
1427 			    vma->vm_end >= mm->start_stack))
1428 				seq_printf(m, " stack");
1429 			else
1430 				seq_printf(m, " stack:%d", tid);
1431 		}
1432 	}
1433 
1434 	if (is_vm_hugetlb_page(vma))
1435 		seq_printf(m, " huge");
1436 
1437 	walk_page_range(vma->vm_start, vma->vm_end, &walk);
1438 
1439 	if (!md->pages)
1440 		goto out;
1441 
1442 	if (md->anon)
1443 		seq_printf(m, " anon=%lu", md->anon);
1444 
1445 	if (md->dirty)
1446 		seq_printf(m, " dirty=%lu", md->dirty);
1447 
1448 	if (md->pages != md->anon && md->pages != md->dirty)
1449 		seq_printf(m, " mapped=%lu", md->pages);
1450 
1451 	if (md->mapcount_max > 1)
1452 		seq_printf(m, " mapmax=%lu", md->mapcount_max);
1453 
1454 	if (md->swapcache)
1455 		seq_printf(m, " swapcache=%lu", md->swapcache);
1456 
1457 	if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1458 		seq_printf(m, " active=%lu", md->active);
1459 
1460 	if (md->writeback)
1461 		seq_printf(m, " writeback=%lu", md->writeback);
1462 
1463 	for_each_node_state(n, N_MEMORY)
1464 		if (md->node[n])
1465 			seq_printf(m, " N%d=%lu", n, md->node[n]);
1466 out:
1467 	seq_putc(m, '\n');
1468 
1469 	if (m->count < m->size)
1470 		m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1471 	return 0;
1472 }
1473 
1474 static int show_pid_numa_map(struct seq_file *m, void *v)
1475 {
1476 	return show_numa_map(m, v, 1);
1477 }
1478 
1479 static int show_tid_numa_map(struct seq_file *m, void *v)
1480 {
1481 	return show_numa_map(m, v, 0);
1482 }
1483 
1484 static const struct seq_operations proc_pid_numa_maps_op = {
1485 	.start  = m_start,
1486 	.next   = m_next,
1487 	.stop   = m_stop,
1488 	.show   = show_pid_numa_map,
1489 };
1490 
1491 static const struct seq_operations proc_tid_numa_maps_op = {
1492 	.start  = m_start,
1493 	.next   = m_next,
1494 	.stop   = m_stop,
1495 	.show   = show_tid_numa_map,
1496 };
1497 
1498 static int numa_maps_open(struct inode *inode, struct file *file,
1499 			  const struct seq_operations *ops)
1500 {
1501 	struct numa_maps_private *priv;
1502 	int ret = -ENOMEM;
1503 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1504 	if (priv) {
1505 		priv->proc_maps.pid = proc_pid(inode);
1506 		ret = seq_open(file, ops);
1507 		if (!ret) {
1508 			struct seq_file *m = file->private_data;
1509 			m->private = priv;
1510 		} else {
1511 			kfree(priv);
1512 		}
1513 	}
1514 	return ret;
1515 }
1516 
1517 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1518 {
1519 	return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1520 }
1521 
1522 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1523 {
1524 	return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1525 }
1526 
1527 const struct file_operations proc_pid_numa_maps_operations = {
1528 	.open		= pid_numa_maps_open,
1529 	.read		= seq_read,
1530 	.llseek		= seq_lseek,
1531 	.release	= seq_release_private,
1532 };
1533 
1534 const struct file_operations proc_tid_numa_maps_operations = {
1535 	.open		= tid_numa_maps_open,
1536 	.read		= seq_read,
1537 	.llseek		= seq_lseek,
1538 	.release	= seq_release_private,
1539 };
1540 #endif /* CONFIG_NUMA */
1541