xref: /openbmc/linux/fs/proc/task_mmu.c (revision 12eb4683)
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) *
66 		 atomic_long_read(&mm->nr_ptes)) >> 10,
67 		swap << (PAGE_SHIFT-10));
68 }
69 
70 unsigned long task_vsize(struct mm_struct *mm)
71 {
72 	return PAGE_SIZE * mm->total_vm;
73 }
74 
75 unsigned long task_statm(struct mm_struct *mm,
76 			 unsigned long *shared, unsigned long *text,
77 			 unsigned long *data, unsigned long *resident)
78 {
79 	*shared = get_mm_counter(mm, MM_FILEPAGES);
80 	*text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
81 								>> PAGE_SHIFT;
82 	*data = mm->total_vm - mm->shared_vm;
83 	*resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
84 	return mm->total_vm;
85 }
86 
87 #ifdef CONFIG_NUMA
88 /*
89  * These functions are for numa_maps but called in generic **maps seq_file
90  * ->start(), ->stop() ops.
91  *
92  * numa_maps scans all vmas under mmap_sem and checks their mempolicy.
93  * Each mempolicy object is controlled by reference counting. The problem here
94  * is how to avoid accessing dead mempolicy object.
95  *
96  * Because we're holding mmap_sem while reading seq_file, it's safe to access
97  * each vma's mempolicy, no vma objects will never drop refs to mempolicy.
98  *
99  * A task's mempolicy (task->mempolicy) has different behavior. task->mempolicy
100  * is set and replaced under mmap_sem but unrefed and cleared under task_lock().
101  * So, without task_lock(), we cannot trust get_vma_policy() because we cannot
102  * gurantee the task never exits under us. But taking task_lock() around
103  * get_vma_plicy() causes lock order problem.
104  *
105  * To access task->mempolicy without lock, we hold a reference count of an
106  * object pointed by task->mempolicy and remember it. This will guarantee
107  * that task->mempolicy points to an alive object or NULL in numa_maps accesses.
108  */
109 static void hold_task_mempolicy(struct proc_maps_private *priv)
110 {
111 	struct task_struct *task = priv->task;
112 
113 	task_lock(task);
114 	priv->task_mempolicy = task->mempolicy;
115 	mpol_get(priv->task_mempolicy);
116 	task_unlock(task);
117 }
118 static void release_task_mempolicy(struct proc_maps_private *priv)
119 {
120 	mpol_put(priv->task_mempolicy);
121 }
122 #else
123 static void hold_task_mempolicy(struct proc_maps_private *priv)
124 {
125 }
126 static void release_task_mempolicy(struct proc_maps_private *priv)
127 {
128 }
129 #endif
130 
131 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
132 {
133 	if (vma && vma != priv->tail_vma) {
134 		struct mm_struct *mm = vma->vm_mm;
135 		release_task_mempolicy(priv);
136 		up_read(&mm->mmap_sem);
137 		mmput(mm);
138 	}
139 }
140 
141 static void *m_start(struct seq_file *m, loff_t *pos)
142 {
143 	struct proc_maps_private *priv = m->private;
144 	unsigned long last_addr = m->version;
145 	struct mm_struct *mm;
146 	struct vm_area_struct *vma, *tail_vma = NULL;
147 	loff_t l = *pos;
148 
149 	/* Clear the per syscall fields in priv */
150 	priv->task = NULL;
151 	priv->tail_vma = NULL;
152 
153 	/*
154 	 * We remember last_addr rather than next_addr to hit with
155 	 * mmap_cache most of the time. We have zero last_addr at
156 	 * the beginning and also after lseek. We will have -1 last_addr
157 	 * after the end of the vmas.
158 	 */
159 
160 	if (last_addr == -1UL)
161 		return NULL;
162 
163 	priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
164 	if (!priv->task)
165 		return ERR_PTR(-ESRCH);
166 
167 	mm = mm_access(priv->task, PTRACE_MODE_READ);
168 	if (!mm || IS_ERR(mm))
169 		return mm;
170 	down_read(&mm->mmap_sem);
171 
172 	tail_vma = get_gate_vma(priv->task->mm);
173 	priv->tail_vma = tail_vma;
174 	hold_task_mempolicy(priv);
175 	/* Start with last addr hint */
176 	vma = find_vma(mm, last_addr);
177 	if (last_addr && vma) {
178 		vma = vma->vm_next;
179 		goto out;
180 	}
181 
182 	/*
183 	 * Check the vma index is within the range and do
184 	 * sequential scan until m_index.
185 	 */
186 	vma = NULL;
187 	if ((unsigned long)l < mm->map_count) {
188 		vma = mm->mmap;
189 		while (l-- && vma)
190 			vma = vma->vm_next;
191 		goto out;
192 	}
193 
194 	if (l != mm->map_count)
195 		tail_vma = NULL; /* After gate vma */
196 
197 out:
198 	if (vma)
199 		return vma;
200 
201 	release_task_mempolicy(priv);
202 	/* End of vmas has been reached */
203 	m->version = (tail_vma != NULL)? 0: -1UL;
204 	up_read(&mm->mmap_sem);
205 	mmput(mm);
206 	return tail_vma;
207 }
208 
209 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
210 {
211 	struct proc_maps_private *priv = m->private;
212 	struct vm_area_struct *vma = v;
213 	struct vm_area_struct *tail_vma = priv->tail_vma;
214 
215 	(*pos)++;
216 	if (vma && (vma != tail_vma) && vma->vm_next)
217 		return vma->vm_next;
218 	vma_stop(priv, vma);
219 	return (vma != tail_vma)? tail_vma: NULL;
220 }
221 
222 static void m_stop(struct seq_file *m, void *v)
223 {
224 	struct proc_maps_private *priv = m->private;
225 	struct vm_area_struct *vma = v;
226 
227 	if (!IS_ERR(vma))
228 		vma_stop(priv, vma);
229 	if (priv->task)
230 		put_task_struct(priv->task);
231 }
232 
233 static int do_maps_open(struct inode *inode, struct file *file,
234 			const struct seq_operations *ops)
235 {
236 	struct proc_maps_private *priv;
237 	int ret = -ENOMEM;
238 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
239 	if (priv) {
240 		priv->pid = proc_pid(inode);
241 		ret = seq_open(file, ops);
242 		if (!ret) {
243 			struct seq_file *m = file->private_data;
244 			m->private = priv;
245 		} else {
246 			kfree(priv);
247 		}
248 	}
249 	return ret;
250 }
251 
252 static void
253 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
254 {
255 	struct mm_struct *mm = vma->vm_mm;
256 	struct file *file = vma->vm_file;
257 	struct proc_maps_private *priv = m->private;
258 	struct task_struct *task = priv->task;
259 	vm_flags_t flags = vma->vm_flags;
260 	unsigned long ino = 0;
261 	unsigned long long pgoff = 0;
262 	unsigned long start, end;
263 	dev_t dev = 0;
264 	const char *name = NULL;
265 
266 	if (file) {
267 		struct inode *inode = file_inode(vma->vm_file);
268 		dev = inode->i_sb->s_dev;
269 		ino = inode->i_ino;
270 		pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
271 	}
272 
273 	/* We don't show the stack guard page in /proc/maps */
274 	start = vma->vm_start;
275 	if (stack_guard_page_start(vma, start))
276 		start += PAGE_SIZE;
277 	end = vma->vm_end;
278 	if (stack_guard_page_end(vma, end))
279 		end -= PAGE_SIZE;
280 
281 	seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
282 	seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ",
283 			start,
284 			end,
285 			flags & VM_READ ? 'r' : '-',
286 			flags & VM_WRITE ? 'w' : '-',
287 			flags & VM_EXEC ? 'x' : '-',
288 			flags & VM_MAYSHARE ? 's' : 'p',
289 			pgoff,
290 			MAJOR(dev), MINOR(dev), ino);
291 
292 	/*
293 	 * Print the dentry name for named mappings, and a
294 	 * special [heap] marker for the heap:
295 	 */
296 	if (file) {
297 		seq_pad(m, ' ');
298 		seq_path(m, &file->f_path, "\n");
299 		goto done;
300 	}
301 
302 	name = arch_vma_name(vma);
303 	if (!name) {
304 		pid_t tid;
305 
306 		if (!mm) {
307 			name = "[vdso]";
308 			goto done;
309 		}
310 
311 		if (vma->vm_start <= mm->brk &&
312 		    vma->vm_end >= mm->start_brk) {
313 			name = "[heap]";
314 			goto done;
315 		}
316 
317 		tid = vm_is_stack(task, vma, is_pid);
318 
319 		if (tid != 0) {
320 			/*
321 			 * Thread stack in /proc/PID/task/TID/maps or
322 			 * the main process stack.
323 			 */
324 			if (!is_pid || (vma->vm_start <= mm->start_stack &&
325 			    vma->vm_end >= mm->start_stack)) {
326 				name = "[stack]";
327 			} else {
328 				/* Thread stack in /proc/PID/maps */
329 				seq_pad(m, ' ');
330 				seq_printf(m, "[stack:%d]", tid);
331 			}
332 		}
333 	}
334 
335 done:
336 	if (name) {
337 		seq_pad(m, ' ');
338 		seq_puts(m, name);
339 	}
340 	seq_putc(m, '\n');
341 }
342 
343 static int show_map(struct seq_file *m, void *v, int is_pid)
344 {
345 	struct vm_area_struct *vma = v;
346 	struct proc_maps_private *priv = m->private;
347 	struct task_struct *task = priv->task;
348 
349 	show_map_vma(m, vma, is_pid);
350 
351 	if (m->count < m->size)  /* vma is copied successfully */
352 		m->version = (vma != get_gate_vma(task->mm))
353 			? vma->vm_start : 0;
354 	return 0;
355 }
356 
357 static int show_pid_map(struct seq_file *m, void *v)
358 {
359 	return show_map(m, v, 1);
360 }
361 
362 static int show_tid_map(struct seq_file *m, void *v)
363 {
364 	return show_map(m, v, 0);
365 }
366 
367 static const struct seq_operations proc_pid_maps_op = {
368 	.start	= m_start,
369 	.next	= m_next,
370 	.stop	= m_stop,
371 	.show	= show_pid_map
372 };
373 
374 static const struct seq_operations proc_tid_maps_op = {
375 	.start	= m_start,
376 	.next	= m_next,
377 	.stop	= m_stop,
378 	.show	= show_tid_map
379 };
380 
381 static int pid_maps_open(struct inode *inode, struct file *file)
382 {
383 	return do_maps_open(inode, file, &proc_pid_maps_op);
384 }
385 
386 static int tid_maps_open(struct inode *inode, struct file *file)
387 {
388 	return do_maps_open(inode, file, &proc_tid_maps_op);
389 }
390 
391 const struct file_operations proc_pid_maps_operations = {
392 	.open		= pid_maps_open,
393 	.read		= seq_read,
394 	.llseek		= seq_lseek,
395 	.release	= seq_release_private,
396 };
397 
398 const struct file_operations proc_tid_maps_operations = {
399 	.open		= tid_maps_open,
400 	.read		= seq_read,
401 	.llseek		= seq_lseek,
402 	.release	= seq_release_private,
403 };
404 
405 /*
406  * Proportional Set Size(PSS): my share of RSS.
407  *
408  * PSS of a process is the count of pages it has in memory, where each
409  * page is divided by the number of processes sharing it.  So if a
410  * process has 1000 pages all to itself, and 1000 shared with one other
411  * process, its PSS will be 1500.
412  *
413  * To keep (accumulated) division errors low, we adopt a 64bit
414  * fixed-point pss counter to minimize division errors. So (pss >>
415  * PSS_SHIFT) would be the real byte count.
416  *
417  * A shift of 12 before division means (assuming 4K page size):
418  * 	- 1M 3-user-pages add up to 8KB errors;
419  * 	- supports mapcount up to 2^24, or 16M;
420  * 	- supports PSS up to 2^52 bytes, or 4PB.
421  */
422 #define PSS_SHIFT 12
423 
424 #ifdef CONFIG_PROC_PAGE_MONITOR
425 struct mem_size_stats {
426 	struct vm_area_struct *vma;
427 	unsigned long resident;
428 	unsigned long shared_clean;
429 	unsigned long shared_dirty;
430 	unsigned long private_clean;
431 	unsigned long private_dirty;
432 	unsigned long referenced;
433 	unsigned long anonymous;
434 	unsigned long anonymous_thp;
435 	unsigned long swap;
436 	unsigned long nonlinear;
437 	u64 pss;
438 };
439 
440 
441 static void smaps_pte_entry(pte_t ptent, unsigned long addr,
442 		unsigned long ptent_size, struct mm_walk *walk)
443 {
444 	struct mem_size_stats *mss = walk->private;
445 	struct vm_area_struct *vma = mss->vma;
446 	pgoff_t pgoff = linear_page_index(vma, addr);
447 	struct page *page = NULL;
448 	int mapcount;
449 
450 	if (pte_present(ptent)) {
451 		page = vm_normal_page(vma, addr, ptent);
452 	} else if (is_swap_pte(ptent)) {
453 		swp_entry_t swpent = pte_to_swp_entry(ptent);
454 
455 		if (!non_swap_entry(swpent))
456 			mss->swap += ptent_size;
457 		else if (is_migration_entry(swpent))
458 			page = migration_entry_to_page(swpent);
459 	} else if (pte_file(ptent)) {
460 		if (pte_to_pgoff(ptent) != pgoff)
461 			mss->nonlinear += ptent_size;
462 	}
463 
464 	if (!page)
465 		return;
466 
467 	if (PageAnon(page))
468 		mss->anonymous += ptent_size;
469 
470 	if (page->index != pgoff)
471 		mss->nonlinear += ptent_size;
472 
473 	mss->resident += ptent_size;
474 	/* Accumulate the size in pages that have been accessed. */
475 	if (pte_young(ptent) || PageReferenced(page))
476 		mss->referenced += ptent_size;
477 	mapcount = page_mapcount(page);
478 	if (mapcount >= 2) {
479 		if (pte_dirty(ptent) || PageDirty(page))
480 			mss->shared_dirty += ptent_size;
481 		else
482 			mss->shared_clean += ptent_size;
483 		mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
484 	} else {
485 		if (pte_dirty(ptent) || PageDirty(page))
486 			mss->private_dirty += ptent_size;
487 		else
488 			mss->private_clean += ptent_size;
489 		mss->pss += (ptent_size << PSS_SHIFT);
490 	}
491 }
492 
493 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
494 			   struct mm_walk *walk)
495 {
496 	struct mem_size_stats *mss = walk->private;
497 	struct vm_area_struct *vma = mss->vma;
498 	pte_t *pte;
499 	spinlock_t *ptl;
500 
501 	if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
502 		smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk);
503 		spin_unlock(ptl);
504 		mss->anonymous_thp += HPAGE_PMD_SIZE;
505 		return 0;
506 	}
507 
508 	if (pmd_trans_unstable(pmd))
509 		return 0;
510 	/*
511 	 * The mmap_sem held all the way back in m_start() is what
512 	 * keeps khugepaged out of here and from collapsing things
513 	 * in here.
514 	 */
515 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
516 	for (; addr != end; pte++, addr += PAGE_SIZE)
517 		smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
518 	pte_unmap_unlock(pte - 1, ptl);
519 	cond_resched();
520 	return 0;
521 }
522 
523 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
524 {
525 	/*
526 	 * Don't forget to update Documentation/ on changes.
527 	 */
528 	static const char mnemonics[BITS_PER_LONG][2] = {
529 		/*
530 		 * In case if we meet a flag we don't know about.
531 		 */
532 		[0 ... (BITS_PER_LONG-1)] = "??",
533 
534 		[ilog2(VM_READ)]	= "rd",
535 		[ilog2(VM_WRITE)]	= "wr",
536 		[ilog2(VM_EXEC)]	= "ex",
537 		[ilog2(VM_SHARED)]	= "sh",
538 		[ilog2(VM_MAYREAD)]	= "mr",
539 		[ilog2(VM_MAYWRITE)]	= "mw",
540 		[ilog2(VM_MAYEXEC)]	= "me",
541 		[ilog2(VM_MAYSHARE)]	= "ms",
542 		[ilog2(VM_GROWSDOWN)]	= "gd",
543 		[ilog2(VM_PFNMAP)]	= "pf",
544 		[ilog2(VM_DENYWRITE)]	= "dw",
545 		[ilog2(VM_LOCKED)]	= "lo",
546 		[ilog2(VM_IO)]		= "io",
547 		[ilog2(VM_SEQ_READ)]	= "sr",
548 		[ilog2(VM_RAND_READ)]	= "rr",
549 		[ilog2(VM_DONTCOPY)]	= "dc",
550 		[ilog2(VM_DONTEXPAND)]	= "de",
551 		[ilog2(VM_ACCOUNT)]	= "ac",
552 		[ilog2(VM_NORESERVE)]	= "nr",
553 		[ilog2(VM_HUGETLB)]	= "ht",
554 		[ilog2(VM_NONLINEAR)]	= "nl",
555 		[ilog2(VM_ARCH_1)]	= "ar",
556 		[ilog2(VM_DONTDUMP)]	= "dd",
557 #ifdef CONFIG_MEM_SOFT_DIRTY
558 		[ilog2(VM_SOFTDIRTY)]	= "sd",
559 #endif
560 		[ilog2(VM_MIXEDMAP)]	= "mm",
561 		[ilog2(VM_HUGEPAGE)]	= "hg",
562 		[ilog2(VM_NOHUGEPAGE)]	= "nh",
563 		[ilog2(VM_MERGEABLE)]	= "mg",
564 	};
565 	size_t i;
566 
567 	seq_puts(m, "VmFlags: ");
568 	for (i = 0; i < BITS_PER_LONG; i++) {
569 		if (vma->vm_flags & (1UL << i)) {
570 			seq_printf(m, "%c%c ",
571 				   mnemonics[i][0], mnemonics[i][1]);
572 		}
573 	}
574 	seq_putc(m, '\n');
575 }
576 
577 static int show_smap(struct seq_file *m, void *v, int is_pid)
578 {
579 	struct proc_maps_private *priv = m->private;
580 	struct task_struct *task = priv->task;
581 	struct vm_area_struct *vma = v;
582 	struct mem_size_stats mss;
583 	struct mm_walk smaps_walk = {
584 		.pmd_entry = smaps_pte_range,
585 		.mm = vma->vm_mm,
586 		.private = &mss,
587 	};
588 
589 	memset(&mss, 0, sizeof mss);
590 	mss.vma = vma;
591 	/* mmap_sem is held in m_start */
592 	if (vma->vm_mm && !is_vm_hugetlb_page(vma))
593 		walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
594 
595 	show_map_vma(m, vma, is_pid);
596 
597 	seq_printf(m,
598 		   "Size:           %8lu kB\n"
599 		   "Rss:            %8lu kB\n"
600 		   "Pss:            %8lu kB\n"
601 		   "Shared_Clean:   %8lu kB\n"
602 		   "Shared_Dirty:   %8lu kB\n"
603 		   "Private_Clean:  %8lu kB\n"
604 		   "Private_Dirty:  %8lu kB\n"
605 		   "Referenced:     %8lu kB\n"
606 		   "Anonymous:      %8lu kB\n"
607 		   "AnonHugePages:  %8lu kB\n"
608 		   "Swap:           %8lu kB\n"
609 		   "KernelPageSize: %8lu kB\n"
610 		   "MMUPageSize:    %8lu kB\n"
611 		   "Locked:         %8lu kB\n",
612 		   (vma->vm_end - vma->vm_start) >> 10,
613 		   mss.resident >> 10,
614 		   (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
615 		   mss.shared_clean  >> 10,
616 		   mss.shared_dirty  >> 10,
617 		   mss.private_clean >> 10,
618 		   mss.private_dirty >> 10,
619 		   mss.referenced >> 10,
620 		   mss.anonymous >> 10,
621 		   mss.anonymous_thp >> 10,
622 		   mss.swap >> 10,
623 		   vma_kernel_pagesize(vma) >> 10,
624 		   vma_mmu_pagesize(vma) >> 10,
625 		   (vma->vm_flags & VM_LOCKED) ?
626 			(unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
627 
628 	if (vma->vm_flags & VM_NONLINEAR)
629 		seq_printf(m, "Nonlinear:      %8lu kB\n",
630 				mss.nonlinear >> 10);
631 
632 	show_smap_vma_flags(m, vma);
633 
634 	if (m->count < m->size)  /* vma is copied successfully */
635 		m->version = (vma != get_gate_vma(task->mm))
636 			? vma->vm_start : 0;
637 	return 0;
638 }
639 
640 static int show_pid_smap(struct seq_file *m, void *v)
641 {
642 	return show_smap(m, v, 1);
643 }
644 
645 static int show_tid_smap(struct seq_file *m, void *v)
646 {
647 	return show_smap(m, v, 0);
648 }
649 
650 static const struct seq_operations proc_pid_smaps_op = {
651 	.start	= m_start,
652 	.next	= m_next,
653 	.stop	= m_stop,
654 	.show	= show_pid_smap
655 };
656 
657 static const struct seq_operations proc_tid_smaps_op = {
658 	.start	= m_start,
659 	.next	= m_next,
660 	.stop	= m_stop,
661 	.show	= show_tid_smap
662 };
663 
664 static int pid_smaps_open(struct inode *inode, struct file *file)
665 {
666 	return do_maps_open(inode, file, &proc_pid_smaps_op);
667 }
668 
669 static int tid_smaps_open(struct inode *inode, struct file *file)
670 {
671 	return do_maps_open(inode, file, &proc_tid_smaps_op);
672 }
673 
674 const struct file_operations proc_pid_smaps_operations = {
675 	.open		= pid_smaps_open,
676 	.read		= seq_read,
677 	.llseek		= seq_lseek,
678 	.release	= seq_release_private,
679 };
680 
681 const struct file_operations proc_tid_smaps_operations = {
682 	.open		= tid_smaps_open,
683 	.read		= seq_read,
684 	.llseek		= seq_lseek,
685 	.release	= seq_release_private,
686 };
687 
688 /*
689  * We do not want to have constant page-shift bits sitting in
690  * pagemap entries and are about to reuse them some time soon.
691  *
692  * Here's the "migration strategy":
693  * 1. when the system boots these bits remain what they are,
694  *    but a warning about future change is printed in log;
695  * 2. once anyone clears soft-dirty bits via clear_refs file,
696  *    these flag is set to denote, that user is aware of the
697  *    new API and those page-shift bits change their meaning.
698  *    The respective warning is printed in dmesg;
699  * 3. In a couple of releases we will remove all the mentions
700  *    of page-shift in pagemap entries.
701  */
702 
703 static bool soft_dirty_cleared __read_mostly;
704 
705 enum clear_refs_types {
706 	CLEAR_REFS_ALL = 1,
707 	CLEAR_REFS_ANON,
708 	CLEAR_REFS_MAPPED,
709 	CLEAR_REFS_SOFT_DIRTY,
710 	CLEAR_REFS_LAST,
711 };
712 
713 struct clear_refs_private {
714 	struct vm_area_struct *vma;
715 	enum clear_refs_types type;
716 };
717 
718 static inline void clear_soft_dirty(struct vm_area_struct *vma,
719 		unsigned long addr, pte_t *pte)
720 {
721 #ifdef CONFIG_MEM_SOFT_DIRTY
722 	/*
723 	 * The soft-dirty tracker uses #PF-s to catch writes
724 	 * to pages, so write-protect the pte as well. See the
725 	 * Documentation/vm/soft-dirty.txt for full description
726 	 * of how soft-dirty works.
727 	 */
728 	pte_t ptent = *pte;
729 
730 	if (pte_present(ptent)) {
731 		ptent = pte_wrprotect(ptent);
732 		ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY);
733 	} else if (is_swap_pte(ptent)) {
734 		ptent = pte_swp_clear_soft_dirty(ptent);
735 	} else if (pte_file(ptent)) {
736 		ptent = pte_file_clear_soft_dirty(ptent);
737 	}
738 
739 	if (vma->vm_flags & VM_SOFTDIRTY)
740 		vma->vm_flags &= ~VM_SOFTDIRTY;
741 
742 	set_pte_at(vma->vm_mm, addr, pte, ptent);
743 #endif
744 }
745 
746 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
747 				unsigned long end, struct mm_walk *walk)
748 {
749 	struct clear_refs_private *cp = walk->private;
750 	struct vm_area_struct *vma = cp->vma;
751 	pte_t *pte, ptent;
752 	spinlock_t *ptl;
753 	struct page *page;
754 
755 	split_huge_page_pmd(vma, addr, pmd);
756 	if (pmd_trans_unstable(pmd))
757 		return 0;
758 
759 	pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
760 	for (; addr != end; pte++, addr += PAGE_SIZE) {
761 		ptent = *pte;
762 
763 		if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
764 			clear_soft_dirty(vma, addr, pte);
765 			continue;
766 		}
767 
768 		if (!pte_present(ptent))
769 			continue;
770 
771 		page = vm_normal_page(vma, addr, ptent);
772 		if (!page)
773 			continue;
774 
775 		/* Clear accessed and referenced bits. */
776 		ptep_test_and_clear_young(vma, addr, pte);
777 		ClearPageReferenced(page);
778 	}
779 	pte_unmap_unlock(pte - 1, ptl);
780 	cond_resched();
781 	return 0;
782 }
783 
784 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
785 				size_t count, loff_t *ppos)
786 {
787 	struct task_struct *task;
788 	char buffer[PROC_NUMBUF];
789 	struct mm_struct *mm;
790 	struct vm_area_struct *vma;
791 	enum clear_refs_types type;
792 	int itype;
793 	int rv;
794 
795 	memset(buffer, 0, sizeof(buffer));
796 	if (count > sizeof(buffer) - 1)
797 		count = sizeof(buffer) - 1;
798 	if (copy_from_user(buffer, buf, count))
799 		return -EFAULT;
800 	rv = kstrtoint(strstrip(buffer), 10, &itype);
801 	if (rv < 0)
802 		return rv;
803 	type = (enum clear_refs_types)itype;
804 	if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
805 		return -EINVAL;
806 
807 	if (type == CLEAR_REFS_SOFT_DIRTY) {
808 		soft_dirty_cleared = true;
809 		pr_warn_once("The pagemap bits 55-60 has changed their meaning! "
810 				"See the linux/Documentation/vm/pagemap.txt for details.\n");
811 	}
812 
813 	task = get_proc_task(file_inode(file));
814 	if (!task)
815 		return -ESRCH;
816 	mm = get_task_mm(task);
817 	if (mm) {
818 		struct clear_refs_private cp = {
819 			.type = type,
820 		};
821 		struct mm_walk clear_refs_walk = {
822 			.pmd_entry = clear_refs_pte_range,
823 			.mm = mm,
824 			.private = &cp,
825 		};
826 		down_read(&mm->mmap_sem);
827 		if (type == CLEAR_REFS_SOFT_DIRTY)
828 			mmu_notifier_invalidate_range_start(mm, 0, -1);
829 		for (vma = mm->mmap; vma; vma = vma->vm_next) {
830 			cp.vma = vma;
831 			if (is_vm_hugetlb_page(vma))
832 				continue;
833 			/*
834 			 * Writing 1 to /proc/pid/clear_refs affects all pages.
835 			 *
836 			 * Writing 2 to /proc/pid/clear_refs only affects
837 			 * Anonymous pages.
838 			 *
839 			 * Writing 3 to /proc/pid/clear_refs only affects file
840 			 * mapped pages.
841 			 */
842 			if (type == CLEAR_REFS_ANON && vma->vm_file)
843 				continue;
844 			if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
845 				continue;
846 			walk_page_range(vma->vm_start, vma->vm_end,
847 					&clear_refs_walk);
848 		}
849 		if (type == CLEAR_REFS_SOFT_DIRTY)
850 			mmu_notifier_invalidate_range_end(mm, 0, -1);
851 		flush_tlb_mm(mm);
852 		up_read(&mm->mmap_sem);
853 		mmput(mm);
854 	}
855 	put_task_struct(task);
856 
857 	return count;
858 }
859 
860 const struct file_operations proc_clear_refs_operations = {
861 	.write		= clear_refs_write,
862 	.llseek		= noop_llseek,
863 };
864 
865 typedef struct {
866 	u64 pme;
867 } pagemap_entry_t;
868 
869 struct pagemapread {
870 	int pos, len;		/* units: PM_ENTRY_BYTES, not bytes */
871 	pagemap_entry_t *buffer;
872 	bool v2;
873 };
874 
875 #define PAGEMAP_WALK_SIZE	(PMD_SIZE)
876 #define PAGEMAP_WALK_MASK	(PMD_MASK)
877 
878 #define PM_ENTRY_BYTES      sizeof(pagemap_entry_t)
879 #define PM_STATUS_BITS      3
880 #define PM_STATUS_OFFSET    (64 - PM_STATUS_BITS)
881 #define PM_STATUS_MASK      (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
882 #define PM_STATUS(nr)       (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
883 #define PM_PSHIFT_BITS      6
884 #define PM_PSHIFT_OFFSET    (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
885 #define PM_PSHIFT_MASK      (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
886 #define __PM_PSHIFT(x)      (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
887 #define PM_PFRAME_MASK      ((1LL << PM_PSHIFT_OFFSET) - 1)
888 #define PM_PFRAME(x)        ((x) & PM_PFRAME_MASK)
889 /* in "new" pagemap pshift bits are occupied with more status bits */
890 #define PM_STATUS2(v2, x)   (__PM_PSHIFT(v2 ? x : PAGE_SHIFT))
891 
892 #define __PM_SOFT_DIRTY      (1LL)
893 #define PM_PRESENT          PM_STATUS(4LL)
894 #define PM_SWAP             PM_STATUS(2LL)
895 #define PM_FILE             PM_STATUS(1LL)
896 #define PM_NOT_PRESENT(v2)  PM_STATUS2(v2, 0)
897 #define PM_END_OF_BUFFER    1
898 
899 static inline pagemap_entry_t make_pme(u64 val)
900 {
901 	return (pagemap_entry_t) { .pme = val };
902 }
903 
904 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
905 			  struct pagemapread *pm)
906 {
907 	pm->buffer[pm->pos++] = *pme;
908 	if (pm->pos >= pm->len)
909 		return PM_END_OF_BUFFER;
910 	return 0;
911 }
912 
913 static int pagemap_pte_hole(unsigned long start, unsigned long end,
914 				struct mm_walk *walk)
915 {
916 	struct pagemapread *pm = walk->private;
917 	unsigned long addr;
918 	int err = 0;
919 	pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
920 
921 	for (addr = start; addr < end; addr += PAGE_SIZE) {
922 		err = add_to_pagemap(addr, &pme, pm);
923 		if (err)
924 			break;
925 	}
926 	return err;
927 }
928 
929 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
930 		struct vm_area_struct *vma, unsigned long addr, pte_t pte)
931 {
932 	u64 frame, flags;
933 	struct page *page = NULL;
934 	int flags2 = 0;
935 
936 	if (pte_present(pte)) {
937 		frame = pte_pfn(pte);
938 		flags = PM_PRESENT;
939 		page = vm_normal_page(vma, addr, pte);
940 		if (pte_soft_dirty(pte))
941 			flags2 |= __PM_SOFT_DIRTY;
942 	} else if (is_swap_pte(pte)) {
943 		swp_entry_t entry;
944 		if (pte_swp_soft_dirty(pte))
945 			flags2 |= __PM_SOFT_DIRTY;
946 		entry = pte_to_swp_entry(pte);
947 		frame = swp_type(entry) |
948 			(swp_offset(entry) << MAX_SWAPFILES_SHIFT);
949 		flags = PM_SWAP;
950 		if (is_migration_entry(entry))
951 			page = migration_entry_to_page(entry);
952 	} else {
953 		if (vma->vm_flags & VM_SOFTDIRTY)
954 			flags2 |= __PM_SOFT_DIRTY;
955 		*pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
956 		return;
957 	}
958 
959 	if (page && !PageAnon(page))
960 		flags |= PM_FILE;
961 	if ((vma->vm_flags & VM_SOFTDIRTY))
962 		flags2 |= __PM_SOFT_DIRTY;
963 
964 	*pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags);
965 }
966 
967 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
968 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
969 		pmd_t pmd, int offset, int pmd_flags2)
970 {
971 	/*
972 	 * Currently pmd for thp is always present because thp can not be
973 	 * swapped-out, migrated, or HWPOISONed (split in such cases instead.)
974 	 * This if-check is just to prepare for future implementation.
975 	 */
976 	if (pmd_present(pmd))
977 		*pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset)
978 				| PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT);
979 	else
980 		*pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2));
981 }
982 #else
983 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
984 		pmd_t pmd, int offset, int pmd_flags2)
985 {
986 }
987 #endif
988 
989 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
990 			     struct mm_walk *walk)
991 {
992 	struct vm_area_struct *vma;
993 	struct pagemapread *pm = walk->private;
994 	spinlock_t *ptl;
995 	pte_t *pte;
996 	int err = 0;
997 	pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2));
998 
999 	/* find the first VMA at or above 'addr' */
1000 	vma = find_vma(walk->mm, addr);
1001 	if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) {
1002 		int pmd_flags2;
1003 
1004 		if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd))
1005 			pmd_flags2 = __PM_SOFT_DIRTY;
1006 		else
1007 			pmd_flags2 = 0;
1008 
1009 		for (; addr != end; addr += PAGE_SIZE) {
1010 			unsigned long offset;
1011 
1012 			offset = (addr & ~PAGEMAP_WALK_MASK) >>
1013 					PAGE_SHIFT;
1014 			thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2);
1015 			err = add_to_pagemap(addr, &pme, pm);
1016 			if (err)
1017 				break;
1018 		}
1019 		spin_unlock(ptl);
1020 		return err;
1021 	}
1022 
1023 	if (pmd_trans_unstable(pmd))
1024 		return 0;
1025 	for (; addr != end; addr += PAGE_SIZE) {
1026 		int flags2;
1027 
1028 		/* check to see if we've left 'vma' behind
1029 		 * and need a new, higher one */
1030 		if (vma && (addr >= vma->vm_end)) {
1031 			vma = find_vma(walk->mm, addr);
1032 			if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1033 				flags2 = __PM_SOFT_DIRTY;
1034 			else
1035 				flags2 = 0;
1036 			pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2));
1037 		}
1038 
1039 		/* check that 'vma' actually covers this address,
1040 		 * and that it isn't a huge page vma */
1041 		if (vma && (vma->vm_start <= addr) &&
1042 		    !is_vm_hugetlb_page(vma)) {
1043 			pte = pte_offset_map(pmd, addr);
1044 			pte_to_pagemap_entry(&pme, pm, vma, addr, *pte);
1045 			/* unmap before userspace copy */
1046 			pte_unmap(pte);
1047 		}
1048 		err = add_to_pagemap(addr, &pme, pm);
1049 		if (err)
1050 			return err;
1051 	}
1052 
1053 	cond_resched();
1054 
1055 	return err;
1056 }
1057 
1058 #ifdef CONFIG_HUGETLB_PAGE
1059 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm,
1060 					pte_t pte, int offset, int flags2)
1061 {
1062 	if (pte_present(pte))
1063 		*pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset)	|
1064 				PM_STATUS2(pm->v2, flags2)		|
1065 				PM_PRESENT);
1066 	else
1067 		*pme = make_pme(PM_NOT_PRESENT(pm->v2)			|
1068 				PM_STATUS2(pm->v2, flags2));
1069 }
1070 
1071 /* This function walks within one hugetlb entry in the single call */
1072 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
1073 				 unsigned long addr, unsigned long end,
1074 				 struct mm_walk *walk)
1075 {
1076 	struct pagemapread *pm = walk->private;
1077 	struct vm_area_struct *vma;
1078 	int err = 0;
1079 	int flags2;
1080 	pagemap_entry_t pme;
1081 
1082 	vma = find_vma(walk->mm, addr);
1083 	WARN_ON_ONCE(!vma);
1084 
1085 	if (vma && (vma->vm_flags & VM_SOFTDIRTY))
1086 		flags2 = __PM_SOFT_DIRTY;
1087 	else
1088 		flags2 = 0;
1089 
1090 	for (; addr != end; addr += PAGE_SIZE) {
1091 		int offset = (addr & ~hmask) >> PAGE_SHIFT;
1092 		huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2);
1093 		err = add_to_pagemap(addr, &pme, pm);
1094 		if (err)
1095 			return err;
1096 	}
1097 
1098 	cond_resched();
1099 
1100 	return err;
1101 }
1102 #endif /* HUGETLB_PAGE */
1103 
1104 /*
1105  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1106  *
1107  * For each page in the address space, this file contains one 64-bit entry
1108  * consisting of the following:
1109  *
1110  * Bits 0-54  page frame number (PFN) if present
1111  * Bits 0-4   swap type if swapped
1112  * Bits 5-54  swap offset if swapped
1113  * Bits 55-60 page shift (page size = 1<<page shift)
1114  * Bit  61    page is file-page or shared-anon
1115  * Bit  62    page swapped
1116  * Bit  63    page present
1117  *
1118  * If the page is not present but in swap, then the PFN contains an
1119  * encoding of the swap file number and the page's offset into the
1120  * swap. Unmapped pages return a null PFN. This allows determining
1121  * precisely which pages are mapped (or in swap) and comparing mapped
1122  * pages between processes.
1123  *
1124  * Efficient users of this interface will use /proc/pid/maps to
1125  * determine which areas of memory are actually mapped and llseek to
1126  * skip over unmapped regions.
1127  */
1128 static ssize_t pagemap_read(struct file *file, char __user *buf,
1129 			    size_t count, loff_t *ppos)
1130 {
1131 	struct task_struct *task = get_proc_task(file_inode(file));
1132 	struct mm_struct *mm;
1133 	struct pagemapread pm;
1134 	int ret = -ESRCH;
1135 	struct mm_walk pagemap_walk = {};
1136 	unsigned long src;
1137 	unsigned long svpfn;
1138 	unsigned long start_vaddr;
1139 	unsigned long end_vaddr;
1140 	int copied = 0;
1141 
1142 	if (!task)
1143 		goto out;
1144 
1145 	ret = -EINVAL;
1146 	/* file position must be aligned */
1147 	if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1148 		goto out_task;
1149 
1150 	ret = 0;
1151 	if (!count)
1152 		goto out_task;
1153 
1154 	pm.v2 = soft_dirty_cleared;
1155 	pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1156 	pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY);
1157 	ret = -ENOMEM;
1158 	if (!pm.buffer)
1159 		goto out_task;
1160 
1161 	mm = mm_access(task, PTRACE_MODE_READ);
1162 	ret = PTR_ERR(mm);
1163 	if (!mm || IS_ERR(mm))
1164 		goto out_free;
1165 
1166 	pagemap_walk.pmd_entry = pagemap_pte_range;
1167 	pagemap_walk.pte_hole = pagemap_pte_hole;
1168 #ifdef CONFIG_HUGETLB_PAGE
1169 	pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1170 #endif
1171 	pagemap_walk.mm = mm;
1172 	pagemap_walk.private = &pm;
1173 
1174 	src = *ppos;
1175 	svpfn = src / PM_ENTRY_BYTES;
1176 	start_vaddr = svpfn << PAGE_SHIFT;
1177 	end_vaddr = TASK_SIZE_OF(task);
1178 
1179 	/* watch out for wraparound */
1180 	if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
1181 		start_vaddr = end_vaddr;
1182 
1183 	/*
1184 	 * The odds are that this will stop walking way
1185 	 * before end_vaddr, because the length of the
1186 	 * user buffer is tracked in "pm", and the walk
1187 	 * will stop when we hit the end of the buffer.
1188 	 */
1189 	ret = 0;
1190 	while (count && (start_vaddr < end_vaddr)) {
1191 		int len;
1192 		unsigned long end;
1193 
1194 		pm.pos = 0;
1195 		end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1196 		/* overflow ? */
1197 		if (end < start_vaddr || end > end_vaddr)
1198 			end = end_vaddr;
1199 		down_read(&mm->mmap_sem);
1200 		ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1201 		up_read(&mm->mmap_sem);
1202 		start_vaddr = end;
1203 
1204 		len = min(count, PM_ENTRY_BYTES * pm.pos);
1205 		if (copy_to_user(buf, pm.buffer, len)) {
1206 			ret = -EFAULT;
1207 			goto out_mm;
1208 		}
1209 		copied += len;
1210 		buf += len;
1211 		count -= len;
1212 	}
1213 	*ppos += copied;
1214 	if (!ret || ret == PM_END_OF_BUFFER)
1215 		ret = copied;
1216 
1217 out_mm:
1218 	mmput(mm);
1219 out_free:
1220 	kfree(pm.buffer);
1221 out_task:
1222 	put_task_struct(task);
1223 out:
1224 	return ret;
1225 }
1226 
1227 static int pagemap_open(struct inode *inode, struct file *file)
1228 {
1229 	pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about "
1230 			"to stop being page-shift some time soon. See the "
1231 			"linux/Documentation/vm/pagemap.txt for details.\n");
1232 	return 0;
1233 }
1234 
1235 const struct file_operations proc_pagemap_operations = {
1236 	.llseek		= mem_lseek, /* borrow this */
1237 	.read		= pagemap_read,
1238 	.open		= pagemap_open,
1239 };
1240 #endif /* CONFIG_PROC_PAGE_MONITOR */
1241 
1242 #ifdef CONFIG_NUMA
1243 
1244 struct numa_maps {
1245 	struct vm_area_struct *vma;
1246 	unsigned long pages;
1247 	unsigned long anon;
1248 	unsigned long active;
1249 	unsigned long writeback;
1250 	unsigned long mapcount_max;
1251 	unsigned long dirty;
1252 	unsigned long swapcache;
1253 	unsigned long node[MAX_NUMNODES];
1254 };
1255 
1256 struct numa_maps_private {
1257 	struct proc_maps_private proc_maps;
1258 	struct numa_maps md;
1259 };
1260 
1261 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1262 			unsigned long nr_pages)
1263 {
1264 	int count = page_mapcount(page);
1265 
1266 	md->pages += nr_pages;
1267 	if (pte_dirty || PageDirty(page))
1268 		md->dirty += nr_pages;
1269 
1270 	if (PageSwapCache(page))
1271 		md->swapcache += nr_pages;
1272 
1273 	if (PageActive(page) || PageUnevictable(page))
1274 		md->active += nr_pages;
1275 
1276 	if (PageWriteback(page))
1277 		md->writeback += nr_pages;
1278 
1279 	if (PageAnon(page))
1280 		md->anon += nr_pages;
1281 
1282 	if (count > md->mapcount_max)
1283 		md->mapcount_max = count;
1284 
1285 	md->node[page_to_nid(page)] += nr_pages;
1286 }
1287 
1288 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1289 		unsigned long addr)
1290 {
1291 	struct page *page;
1292 	int nid;
1293 
1294 	if (!pte_present(pte))
1295 		return NULL;
1296 
1297 	page = vm_normal_page(vma, addr, pte);
1298 	if (!page)
1299 		return NULL;
1300 
1301 	if (PageReserved(page))
1302 		return NULL;
1303 
1304 	nid = page_to_nid(page);
1305 	if (!node_isset(nid, node_states[N_MEMORY]))
1306 		return NULL;
1307 
1308 	return page;
1309 }
1310 
1311 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1312 		unsigned long end, struct mm_walk *walk)
1313 {
1314 	struct numa_maps *md;
1315 	spinlock_t *ptl;
1316 	pte_t *orig_pte;
1317 	pte_t *pte;
1318 
1319 	md = walk->private;
1320 
1321 	if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) {
1322 		pte_t huge_pte = *(pte_t *)pmd;
1323 		struct page *page;
1324 
1325 		page = can_gather_numa_stats(huge_pte, md->vma, addr);
1326 		if (page)
1327 			gather_stats(page, md, pte_dirty(huge_pte),
1328 				     HPAGE_PMD_SIZE/PAGE_SIZE);
1329 		spin_unlock(ptl);
1330 		return 0;
1331 	}
1332 
1333 	if (pmd_trans_unstable(pmd))
1334 		return 0;
1335 	orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1336 	do {
1337 		struct page *page = can_gather_numa_stats(*pte, md->vma, addr);
1338 		if (!page)
1339 			continue;
1340 		gather_stats(page, md, pte_dirty(*pte), 1);
1341 
1342 	} while (pte++, addr += PAGE_SIZE, addr != end);
1343 	pte_unmap_unlock(orig_pte, ptl);
1344 	return 0;
1345 }
1346 #ifdef CONFIG_HUGETLB_PAGE
1347 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1348 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1349 {
1350 	struct numa_maps *md;
1351 	struct page *page;
1352 
1353 	if (pte_none(*pte))
1354 		return 0;
1355 
1356 	page = pte_page(*pte);
1357 	if (!page)
1358 		return 0;
1359 
1360 	md = walk->private;
1361 	gather_stats(page, md, pte_dirty(*pte), 1);
1362 	return 0;
1363 }
1364 
1365 #else
1366 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
1367 		unsigned long addr, unsigned long end, struct mm_walk *walk)
1368 {
1369 	return 0;
1370 }
1371 #endif
1372 
1373 /*
1374  * Display pages allocated per node and memory policy via /proc.
1375  */
1376 static int show_numa_map(struct seq_file *m, void *v, int is_pid)
1377 {
1378 	struct numa_maps_private *numa_priv = m->private;
1379 	struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1380 	struct vm_area_struct *vma = v;
1381 	struct numa_maps *md = &numa_priv->md;
1382 	struct file *file = vma->vm_file;
1383 	struct task_struct *task = proc_priv->task;
1384 	struct mm_struct *mm = vma->vm_mm;
1385 	struct mm_walk walk = {};
1386 	struct mempolicy *pol;
1387 	char buffer[64];
1388 	int nid;
1389 
1390 	if (!mm)
1391 		return 0;
1392 
1393 	/* Ensure we start with an empty set of numa_maps statistics. */
1394 	memset(md, 0, sizeof(*md));
1395 
1396 	md->vma = vma;
1397 
1398 	walk.hugetlb_entry = gather_hugetbl_stats;
1399 	walk.pmd_entry = gather_pte_stats;
1400 	walk.private = md;
1401 	walk.mm = mm;
1402 
1403 	pol = get_vma_policy(task, vma, vma->vm_start);
1404 	mpol_to_str(buffer, sizeof(buffer), pol);
1405 	mpol_cond_put(pol);
1406 
1407 	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1408 
1409 	if (file) {
1410 		seq_printf(m, " file=");
1411 		seq_path(m, &file->f_path, "\n\t= ");
1412 	} else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1413 		seq_printf(m, " heap");
1414 	} else {
1415 		pid_t tid = vm_is_stack(task, vma, is_pid);
1416 		if (tid != 0) {
1417 			/*
1418 			 * Thread stack in /proc/PID/task/TID/maps or
1419 			 * the main process stack.
1420 			 */
1421 			if (!is_pid || (vma->vm_start <= mm->start_stack &&
1422 			    vma->vm_end >= mm->start_stack))
1423 				seq_printf(m, " stack");
1424 			else
1425 				seq_printf(m, " stack:%d", tid);
1426 		}
1427 	}
1428 
1429 	if (is_vm_hugetlb_page(vma))
1430 		seq_printf(m, " huge");
1431 
1432 	walk_page_range(vma->vm_start, vma->vm_end, &walk);
1433 
1434 	if (!md->pages)
1435 		goto out;
1436 
1437 	if (md->anon)
1438 		seq_printf(m, " anon=%lu", md->anon);
1439 
1440 	if (md->dirty)
1441 		seq_printf(m, " dirty=%lu", md->dirty);
1442 
1443 	if (md->pages != md->anon && md->pages != md->dirty)
1444 		seq_printf(m, " mapped=%lu", md->pages);
1445 
1446 	if (md->mapcount_max > 1)
1447 		seq_printf(m, " mapmax=%lu", md->mapcount_max);
1448 
1449 	if (md->swapcache)
1450 		seq_printf(m, " swapcache=%lu", md->swapcache);
1451 
1452 	if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1453 		seq_printf(m, " active=%lu", md->active);
1454 
1455 	if (md->writeback)
1456 		seq_printf(m, " writeback=%lu", md->writeback);
1457 
1458 	for_each_node_state(nid, N_MEMORY)
1459 		if (md->node[nid])
1460 			seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1461 out:
1462 	seq_putc(m, '\n');
1463 
1464 	if (m->count < m->size)
1465 		m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
1466 	return 0;
1467 }
1468 
1469 static int show_pid_numa_map(struct seq_file *m, void *v)
1470 {
1471 	return show_numa_map(m, v, 1);
1472 }
1473 
1474 static int show_tid_numa_map(struct seq_file *m, void *v)
1475 {
1476 	return show_numa_map(m, v, 0);
1477 }
1478 
1479 static const struct seq_operations proc_pid_numa_maps_op = {
1480 	.start  = m_start,
1481 	.next   = m_next,
1482 	.stop   = m_stop,
1483 	.show   = show_pid_numa_map,
1484 };
1485 
1486 static const struct seq_operations proc_tid_numa_maps_op = {
1487 	.start  = m_start,
1488 	.next   = m_next,
1489 	.stop   = m_stop,
1490 	.show   = show_tid_numa_map,
1491 };
1492 
1493 static int numa_maps_open(struct inode *inode, struct file *file,
1494 			  const struct seq_operations *ops)
1495 {
1496 	struct numa_maps_private *priv;
1497 	int ret = -ENOMEM;
1498 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1499 	if (priv) {
1500 		priv->proc_maps.pid = proc_pid(inode);
1501 		ret = seq_open(file, ops);
1502 		if (!ret) {
1503 			struct seq_file *m = file->private_data;
1504 			m->private = priv;
1505 		} else {
1506 			kfree(priv);
1507 		}
1508 	}
1509 	return ret;
1510 }
1511 
1512 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1513 {
1514 	return numa_maps_open(inode, file, &proc_pid_numa_maps_op);
1515 }
1516 
1517 static int tid_numa_maps_open(struct inode *inode, struct file *file)
1518 {
1519 	return numa_maps_open(inode, file, &proc_tid_numa_maps_op);
1520 }
1521 
1522 const struct file_operations proc_pid_numa_maps_operations = {
1523 	.open		= pid_numa_maps_open,
1524 	.read		= seq_read,
1525 	.llseek		= seq_lseek,
1526 	.release	= seq_release_private,
1527 };
1528 
1529 const struct file_operations proc_tid_numa_maps_operations = {
1530 	.open		= tid_numa_maps_open,
1531 	.read		= seq_read,
1532 	.llseek		= seq_lseek,
1533 	.release	= seq_release_private,
1534 };
1535 #endif /* CONFIG_NUMA */
1536