xref: /openbmc/linux/mm/vmalloc.c (revision c819e2cf)
1 /*
2  *  linux/mm/vmalloc.c
3  *
4  *  Copyright (C) 1993  Linus Torvalds
5  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
6  *  SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
7  *  Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
8  *  Numa awareness, Christoph Lameter, SGI, June 2005
9  */
10 
11 #include <linux/vmalloc.h>
12 #include <linux/mm.h>
13 #include <linux/module.h>
14 #include <linux/highmem.h>
15 #include <linux/sched.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/interrupt.h>
19 #include <linux/proc_fs.h>
20 #include <linux/seq_file.h>
21 #include <linux/debugobjects.h>
22 #include <linux/kallsyms.h>
23 #include <linux/list.h>
24 #include <linux/rbtree.h>
25 #include <linux/radix-tree.h>
26 #include <linux/rcupdate.h>
27 #include <linux/pfn.h>
28 #include <linux/kmemleak.h>
29 #include <linux/atomic.h>
30 #include <linux/compiler.h>
31 #include <linux/llist.h>
32 
33 #include <asm/uaccess.h>
34 #include <asm/tlbflush.h>
35 #include <asm/shmparam.h>
36 
37 struct vfree_deferred {
38 	struct llist_head list;
39 	struct work_struct wq;
40 };
41 static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
42 
43 static void __vunmap(const void *, int);
44 
45 static void free_work(struct work_struct *w)
46 {
47 	struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
48 	struct llist_node *llnode = llist_del_all(&p->list);
49 	while (llnode) {
50 		void *p = llnode;
51 		llnode = llist_next(llnode);
52 		__vunmap(p, 1);
53 	}
54 }
55 
56 /*** Page table manipulation functions ***/
57 
58 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
59 {
60 	pte_t *pte;
61 
62 	pte = pte_offset_kernel(pmd, addr);
63 	do {
64 		pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
65 		WARN_ON(!pte_none(ptent) && !pte_present(ptent));
66 	} while (pte++, addr += PAGE_SIZE, addr != end);
67 }
68 
69 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
70 {
71 	pmd_t *pmd;
72 	unsigned long next;
73 
74 	pmd = pmd_offset(pud, addr);
75 	do {
76 		next = pmd_addr_end(addr, end);
77 		if (pmd_none_or_clear_bad(pmd))
78 			continue;
79 		vunmap_pte_range(pmd, addr, next);
80 	} while (pmd++, addr = next, addr != end);
81 }
82 
83 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
84 {
85 	pud_t *pud;
86 	unsigned long next;
87 
88 	pud = pud_offset(pgd, addr);
89 	do {
90 		next = pud_addr_end(addr, end);
91 		if (pud_none_or_clear_bad(pud))
92 			continue;
93 		vunmap_pmd_range(pud, addr, next);
94 	} while (pud++, addr = next, addr != end);
95 }
96 
97 static void vunmap_page_range(unsigned long addr, unsigned long end)
98 {
99 	pgd_t *pgd;
100 	unsigned long next;
101 
102 	BUG_ON(addr >= end);
103 	pgd = pgd_offset_k(addr);
104 	do {
105 		next = pgd_addr_end(addr, end);
106 		if (pgd_none_or_clear_bad(pgd))
107 			continue;
108 		vunmap_pud_range(pgd, addr, next);
109 	} while (pgd++, addr = next, addr != end);
110 }
111 
112 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
113 		unsigned long end, pgprot_t prot, struct page **pages, int *nr)
114 {
115 	pte_t *pte;
116 
117 	/*
118 	 * nr is a running index into the array which helps higher level
119 	 * callers keep track of where we're up to.
120 	 */
121 
122 	pte = pte_alloc_kernel(pmd, addr);
123 	if (!pte)
124 		return -ENOMEM;
125 	do {
126 		struct page *page = pages[*nr];
127 
128 		if (WARN_ON(!pte_none(*pte)))
129 			return -EBUSY;
130 		if (WARN_ON(!page))
131 			return -ENOMEM;
132 		set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
133 		(*nr)++;
134 	} while (pte++, addr += PAGE_SIZE, addr != end);
135 	return 0;
136 }
137 
138 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
139 		unsigned long end, pgprot_t prot, struct page **pages, int *nr)
140 {
141 	pmd_t *pmd;
142 	unsigned long next;
143 
144 	pmd = pmd_alloc(&init_mm, pud, addr);
145 	if (!pmd)
146 		return -ENOMEM;
147 	do {
148 		next = pmd_addr_end(addr, end);
149 		if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
150 			return -ENOMEM;
151 	} while (pmd++, addr = next, addr != end);
152 	return 0;
153 }
154 
155 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
156 		unsigned long end, pgprot_t prot, struct page **pages, int *nr)
157 {
158 	pud_t *pud;
159 	unsigned long next;
160 
161 	pud = pud_alloc(&init_mm, pgd, addr);
162 	if (!pud)
163 		return -ENOMEM;
164 	do {
165 		next = pud_addr_end(addr, end);
166 		if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
167 			return -ENOMEM;
168 	} while (pud++, addr = next, addr != end);
169 	return 0;
170 }
171 
172 /*
173  * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
174  * will have pfns corresponding to the "pages" array.
175  *
176  * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
177  */
178 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
179 				   pgprot_t prot, struct page **pages)
180 {
181 	pgd_t *pgd;
182 	unsigned long next;
183 	unsigned long addr = start;
184 	int err = 0;
185 	int nr = 0;
186 
187 	BUG_ON(addr >= end);
188 	pgd = pgd_offset_k(addr);
189 	do {
190 		next = pgd_addr_end(addr, end);
191 		err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
192 		if (err)
193 			return err;
194 	} while (pgd++, addr = next, addr != end);
195 
196 	return nr;
197 }
198 
199 static int vmap_page_range(unsigned long start, unsigned long end,
200 			   pgprot_t prot, struct page **pages)
201 {
202 	int ret;
203 
204 	ret = vmap_page_range_noflush(start, end, prot, pages);
205 	flush_cache_vmap(start, end);
206 	return ret;
207 }
208 
209 int is_vmalloc_or_module_addr(const void *x)
210 {
211 	/*
212 	 * ARM, x86-64 and sparc64 put modules in a special place,
213 	 * and fall back on vmalloc() if that fails. Others
214 	 * just put it in the vmalloc space.
215 	 */
216 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
217 	unsigned long addr = (unsigned long)x;
218 	if (addr >= MODULES_VADDR && addr < MODULES_END)
219 		return 1;
220 #endif
221 	return is_vmalloc_addr(x);
222 }
223 
224 /*
225  * Walk a vmap address to the struct page it maps.
226  */
227 struct page *vmalloc_to_page(const void *vmalloc_addr)
228 {
229 	unsigned long addr = (unsigned long) vmalloc_addr;
230 	struct page *page = NULL;
231 	pgd_t *pgd = pgd_offset_k(addr);
232 
233 	/*
234 	 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
235 	 * architectures that do not vmalloc module space
236 	 */
237 	VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
238 
239 	if (!pgd_none(*pgd)) {
240 		pud_t *pud = pud_offset(pgd, addr);
241 		if (!pud_none(*pud)) {
242 			pmd_t *pmd = pmd_offset(pud, addr);
243 			if (!pmd_none(*pmd)) {
244 				pte_t *ptep, pte;
245 
246 				ptep = pte_offset_map(pmd, addr);
247 				pte = *ptep;
248 				if (pte_present(pte))
249 					page = pte_page(pte);
250 				pte_unmap(ptep);
251 			}
252 		}
253 	}
254 	return page;
255 }
256 EXPORT_SYMBOL(vmalloc_to_page);
257 
258 /*
259  * Map a vmalloc()-space virtual address to the physical page frame number.
260  */
261 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
262 {
263 	return page_to_pfn(vmalloc_to_page(vmalloc_addr));
264 }
265 EXPORT_SYMBOL(vmalloc_to_pfn);
266 
267 
268 /*** Global kva allocator ***/
269 
270 #define VM_LAZY_FREE	0x01
271 #define VM_LAZY_FREEING	0x02
272 #define VM_VM_AREA	0x04
273 
274 static DEFINE_SPINLOCK(vmap_area_lock);
275 /* Export for kexec only */
276 LIST_HEAD(vmap_area_list);
277 static struct rb_root vmap_area_root = RB_ROOT;
278 
279 /* The vmap cache globals are protected by vmap_area_lock */
280 static struct rb_node *free_vmap_cache;
281 static unsigned long cached_hole_size;
282 static unsigned long cached_vstart;
283 static unsigned long cached_align;
284 
285 static unsigned long vmap_area_pcpu_hole;
286 
287 static struct vmap_area *__find_vmap_area(unsigned long addr)
288 {
289 	struct rb_node *n = vmap_area_root.rb_node;
290 
291 	while (n) {
292 		struct vmap_area *va;
293 
294 		va = rb_entry(n, struct vmap_area, rb_node);
295 		if (addr < va->va_start)
296 			n = n->rb_left;
297 		else if (addr >= va->va_end)
298 			n = n->rb_right;
299 		else
300 			return va;
301 	}
302 
303 	return NULL;
304 }
305 
306 static void __insert_vmap_area(struct vmap_area *va)
307 {
308 	struct rb_node **p = &vmap_area_root.rb_node;
309 	struct rb_node *parent = NULL;
310 	struct rb_node *tmp;
311 
312 	while (*p) {
313 		struct vmap_area *tmp_va;
314 
315 		parent = *p;
316 		tmp_va = rb_entry(parent, struct vmap_area, rb_node);
317 		if (va->va_start < tmp_va->va_end)
318 			p = &(*p)->rb_left;
319 		else if (va->va_end > tmp_va->va_start)
320 			p = &(*p)->rb_right;
321 		else
322 			BUG();
323 	}
324 
325 	rb_link_node(&va->rb_node, parent, p);
326 	rb_insert_color(&va->rb_node, &vmap_area_root);
327 
328 	/* address-sort this list */
329 	tmp = rb_prev(&va->rb_node);
330 	if (tmp) {
331 		struct vmap_area *prev;
332 		prev = rb_entry(tmp, struct vmap_area, rb_node);
333 		list_add_rcu(&va->list, &prev->list);
334 	} else
335 		list_add_rcu(&va->list, &vmap_area_list);
336 }
337 
338 static void purge_vmap_area_lazy(void);
339 
340 /*
341  * Allocate a region of KVA of the specified size and alignment, within the
342  * vstart and vend.
343  */
344 static struct vmap_area *alloc_vmap_area(unsigned long size,
345 				unsigned long align,
346 				unsigned long vstart, unsigned long vend,
347 				int node, gfp_t gfp_mask)
348 {
349 	struct vmap_area *va;
350 	struct rb_node *n;
351 	unsigned long addr;
352 	int purged = 0;
353 	struct vmap_area *first;
354 
355 	BUG_ON(!size);
356 	BUG_ON(size & ~PAGE_MASK);
357 	BUG_ON(!is_power_of_2(align));
358 
359 	va = kmalloc_node(sizeof(struct vmap_area),
360 			gfp_mask & GFP_RECLAIM_MASK, node);
361 	if (unlikely(!va))
362 		return ERR_PTR(-ENOMEM);
363 
364 	/*
365 	 * Only scan the relevant parts containing pointers to other objects
366 	 * to avoid false negatives.
367 	 */
368 	kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
369 
370 retry:
371 	spin_lock(&vmap_area_lock);
372 	/*
373 	 * Invalidate cache if we have more permissive parameters.
374 	 * cached_hole_size notes the largest hole noticed _below_
375 	 * the vmap_area cached in free_vmap_cache: if size fits
376 	 * into that hole, we want to scan from vstart to reuse
377 	 * the hole instead of allocating above free_vmap_cache.
378 	 * Note that __free_vmap_area may update free_vmap_cache
379 	 * without updating cached_hole_size or cached_align.
380 	 */
381 	if (!free_vmap_cache ||
382 			size < cached_hole_size ||
383 			vstart < cached_vstart ||
384 			align < cached_align) {
385 nocache:
386 		cached_hole_size = 0;
387 		free_vmap_cache = NULL;
388 	}
389 	/* record if we encounter less permissive parameters */
390 	cached_vstart = vstart;
391 	cached_align = align;
392 
393 	/* find starting point for our search */
394 	if (free_vmap_cache) {
395 		first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
396 		addr = ALIGN(first->va_end, align);
397 		if (addr < vstart)
398 			goto nocache;
399 		if (addr + size < addr)
400 			goto overflow;
401 
402 	} else {
403 		addr = ALIGN(vstart, align);
404 		if (addr + size < addr)
405 			goto overflow;
406 
407 		n = vmap_area_root.rb_node;
408 		first = NULL;
409 
410 		while (n) {
411 			struct vmap_area *tmp;
412 			tmp = rb_entry(n, struct vmap_area, rb_node);
413 			if (tmp->va_end >= addr) {
414 				first = tmp;
415 				if (tmp->va_start <= addr)
416 					break;
417 				n = n->rb_left;
418 			} else
419 				n = n->rb_right;
420 		}
421 
422 		if (!first)
423 			goto found;
424 	}
425 
426 	/* from the starting point, walk areas until a suitable hole is found */
427 	while (addr + size > first->va_start && addr + size <= vend) {
428 		if (addr + cached_hole_size < first->va_start)
429 			cached_hole_size = first->va_start - addr;
430 		addr = ALIGN(first->va_end, align);
431 		if (addr + size < addr)
432 			goto overflow;
433 
434 		if (list_is_last(&first->list, &vmap_area_list))
435 			goto found;
436 
437 		first = list_entry(first->list.next,
438 				struct vmap_area, list);
439 	}
440 
441 found:
442 	if (addr + size > vend)
443 		goto overflow;
444 
445 	va->va_start = addr;
446 	va->va_end = addr + size;
447 	va->flags = 0;
448 	__insert_vmap_area(va);
449 	free_vmap_cache = &va->rb_node;
450 	spin_unlock(&vmap_area_lock);
451 
452 	BUG_ON(va->va_start & (align-1));
453 	BUG_ON(va->va_start < vstart);
454 	BUG_ON(va->va_end > vend);
455 
456 	return va;
457 
458 overflow:
459 	spin_unlock(&vmap_area_lock);
460 	if (!purged) {
461 		purge_vmap_area_lazy();
462 		purged = 1;
463 		goto retry;
464 	}
465 	if (printk_ratelimit())
466 		pr_warn("vmap allocation for size %lu failed: "
467 			"use vmalloc=<size> to increase size.\n", size);
468 	kfree(va);
469 	return ERR_PTR(-EBUSY);
470 }
471 
472 static void __free_vmap_area(struct vmap_area *va)
473 {
474 	BUG_ON(RB_EMPTY_NODE(&va->rb_node));
475 
476 	if (free_vmap_cache) {
477 		if (va->va_end < cached_vstart) {
478 			free_vmap_cache = NULL;
479 		} else {
480 			struct vmap_area *cache;
481 			cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
482 			if (va->va_start <= cache->va_start) {
483 				free_vmap_cache = rb_prev(&va->rb_node);
484 				/*
485 				 * We don't try to update cached_hole_size or
486 				 * cached_align, but it won't go very wrong.
487 				 */
488 			}
489 		}
490 	}
491 	rb_erase(&va->rb_node, &vmap_area_root);
492 	RB_CLEAR_NODE(&va->rb_node);
493 	list_del_rcu(&va->list);
494 
495 	/*
496 	 * Track the highest possible candidate for pcpu area
497 	 * allocation.  Areas outside of vmalloc area can be returned
498 	 * here too, consider only end addresses which fall inside
499 	 * vmalloc area proper.
500 	 */
501 	if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
502 		vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
503 
504 	kfree_rcu(va, rcu_head);
505 }
506 
507 /*
508  * Free a region of KVA allocated by alloc_vmap_area
509  */
510 static void free_vmap_area(struct vmap_area *va)
511 {
512 	spin_lock(&vmap_area_lock);
513 	__free_vmap_area(va);
514 	spin_unlock(&vmap_area_lock);
515 }
516 
517 /*
518  * Clear the pagetable entries of a given vmap_area
519  */
520 static void unmap_vmap_area(struct vmap_area *va)
521 {
522 	vunmap_page_range(va->va_start, va->va_end);
523 }
524 
525 static void vmap_debug_free_range(unsigned long start, unsigned long end)
526 {
527 	/*
528 	 * Unmap page tables and force a TLB flush immediately if
529 	 * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
530 	 * bugs similarly to those in linear kernel virtual address
531 	 * space after a page has been freed.
532 	 *
533 	 * All the lazy freeing logic is still retained, in order to
534 	 * minimise intrusiveness of this debugging feature.
535 	 *
536 	 * This is going to be *slow* (linear kernel virtual address
537 	 * debugging doesn't do a broadcast TLB flush so it is a lot
538 	 * faster).
539 	 */
540 #ifdef CONFIG_DEBUG_PAGEALLOC
541 	vunmap_page_range(start, end);
542 	flush_tlb_kernel_range(start, end);
543 #endif
544 }
545 
546 /*
547  * lazy_max_pages is the maximum amount of virtual address space we gather up
548  * before attempting to purge with a TLB flush.
549  *
550  * There is a tradeoff here: a larger number will cover more kernel page tables
551  * and take slightly longer to purge, but it will linearly reduce the number of
552  * global TLB flushes that must be performed. It would seem natural to scale
553  * this number up linearly with the number of CPUs (because vmapping activity
554  * could also scale linearly with the number of CPUs), however it is likely
555  * that in practice, workloads might be constrained in other ways that mean
556  * vmap activity will not scale linearly with CPUs. Also, I want to be
557  * conservative and not introduce a big latency on huge systems, so go with
558  * a less aggressive log scale. It will still be an improvement over the old
559  * code, and it will be simple to change the scale factor if we find that it
560  * becomes a problem on bigger systems.
561  */
562 static unsigned long lazy_max_pages(void)
563 {
564 	unsigned int log;
565 
566 	log = fls(num_online_cpus());
567 
568 	return log * (32UL * 1024 * 1024 / PAGE_SIZE);
569 }
570 
571 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
572 
573 /* for per-CPU blocks */
574 static void purge_fragmented_blocks_allcpus(void);
575 
576 /*
577  * called before a call to iounmap() if the caller wants vm_area_struct's
578  * immediately freed.
579  */
580 void set_iounmap_nonlazy(void)
581 {
582 	atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
583 }
584 
585 /*
586  * Purges all lazily-freed vmap areas.
587  *
588  * If sync is 0 then don't purge if there is already a purge in progress.
589  * If force_flush is 1, then flush kernel TLBs between *start and *end even
590  * if we found no lazy vmap areas to unmap (callers can use this to optimise
591  * their own TLB flushing).
592  * Returns with *start = min(*start, lowest purged address)
593  *              *end = max(*end, highest purged address)
594  */
595 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
596 					int sync, int force_flush)
597 {
598 	static DEFINE_SPINLOCK(purge_lock);
599 	LIST_HEAD(valist);
600 	struct vmap_area *va;
601 	struct vmap_area *n_va;
602 	int nr = 0;
603 
604 	/*
605 	 * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
606 	 * should not expect such behaviour. This just simplifies locking for
607 	 * the case that isn't actually used at the moment anyway.
608 	 */
609 	if (!sync && !force_flush) {
610 		if (!spin_trylock(&purge_lock))
611 			return;
612 	} else
613 		spin_lock(&purge_lock);
614 
615 	if (sync)
616 		purge_fragmented_blocks_allcpus();
617 
618 	rcu_read_lock();
619 	list_for_each_entry_rcu(va, &vmap_area_list, list) {
620 		if (va->flags & VM_LAZY_FREE) {
621 			if (va->va_start < *start)
622 				*start = va->va_start;
623 			if (va->va_end > *end)
624 				*end = va->va_end;
625 			nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
626 			list_add_tail(&va->purge_list, &valist);
627 			va->flags |= VM_LAZY_FREEING;
628 			va->flags &= ~VM_LAZY_FREE;
629 		}
630 	}
631 	rcu_read_unlock();
632 
633 	if (nr)
634 		atomic_sub(nr, &vmap_lazy_nr);
635 
636 	if (nr || force_flush)
637 		flush_tlb_kernel_range(*start, *end);
638 
639 	if (nr) {
640 		spin_lock(&vmap_area_lock);
641 		list_for_each_entry_safe(va, n_va, &valist, purge_list)
642 			__free_vmap_area(va);
643 		spin_unlock(&vmap_area_lock);
644 	}
645 	spin_unlock(&purge_lock);
646 }
647 
648 /*
649  * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
650  * is already purging.
651  */
652 static void try_purge_vmap_area_lazy(void)
653 {
654 	unsigned long start = ULONG_MAX, end = 0;
655 
656 	__purge_vmap_area_lazy(&start, &end, 0, 0);
657 }
658 
659 /*
660  * Kick off a purge of the outstanding lazy areas.
661  */
662 static void purge_vmap_area_lazy(void)
663 {
664 	unsigned long start = ULONG_MAX, end = 0;
665 
666 	__purge_vmap_area_lazy(&start, &end, 1, 0);
667 }
668 
669 /*
670  * Free a vmap area, caller ensuring that the area has been unmapped
671  * and flush_cache_vunmap had been called for the correct range
672  * previously.
673  */
674 static void free_vmap_area_noflush(struct vmap_area *va)
675 {
676 	va->flags |= VM_LAZY_FREE;
677 	atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
678 	if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
679 		try_purge_vmap_area_lazy();
680 }
681 
682 /*
683  * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
684  * called for the correct range previously.
685  */
686 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
687 {
688 	unmap_vmap_area(va);
689 	free_vmap_area_noflush(va);
690 }
691 
692 /*
693  * Free and unmap a vmap area
694  */
695 static void free_unmap_vmap_area(struct vmap_area *va)
696 {
697 	flush_cache_vunmap(va->va_start, va->va_end);
698 	free_unmap_vmap_area_noflush(va);
699 }
700 
701 static struct vmap_area *find_vmap_area(unsigned long addr)
702 {
703 	struct vmap_area *va;
704 
705 	spin_lock(&vmap_area_lock);
706 	va = __find_vmap_area(addr);
707 	spin_unlock(&vmap_area_lock);
708 
709 	return va;
710 }
711 
712 static void free_unmap_vmap_area_addr(unsigned long addr)
713 {
714 	struct vmap_area *va;
715 
716 	va = find_vmap_area(addr);
717 	BUG_ON(!va);
718 	free_unmap_vmap_area(va);
719 }
720 
721 
722 /*** Per cpu kva allocator ***/
723 
724 /*
725  * vmap space is limited especially on 32 bit architectures. Ensure there is
726  * room for at least 16 percpu vmap blocks per CPU.
727  */
728 /*
729  * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
730  * to #define VMALLOC_SPACE		(VMALLOC_END-VMALLOC_START). Guess
731  * instead (we just need a rough idea)
732  */
733 #if BITS_PER_LONG == 32
734 #define VMALLOC_SPACE		(128UL*1024*1024)
735 #else
736 #define VMALLOC_SPACE		(128UL*1024*1024*1024)
737 #endif
738 
739 #define VMALLOC_PAGES		(VMALLOC_SPACE / PAGE_SIZE)
740 #define VMAP_MAX_ALLOC		BITS_PER_LONG	/* 256K with 4K pages */
741 #define VMAP_BBMAP_BITS_MAX	1024	/* 4MB with 4K pages */
742 #define VMAP_BBMAP_BITS_MIN	(VMAP_MAX_ALLOC*2)
743 #define VMAP_MIN(x, y)		((x) < (y) ? (x) : (y)) /* can't use min() */
744 #define VMAP_MAX(x, y)		((x) > (y) ? (x) : (y)) /* can't use max() */
745 #define VMAP_BBMAP_BITS		\
746 		VMAP_MIN(VMAP_BBMAP_BITS_MAX,	\
747 		VMAP_MAX(VMAP_BBMAP_BITS_MIN,	\
748 			VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
749 
750 #define VMAP_BLOCK_SIZE		(VMAP_BBMAP_BITS * PAGE_SIZE)
751 
752 static bool vmap_initialized __read_mostly = false;
753 
754 struct vmap_block_queue {
755 	spinlock_t lock;
756 	struct list_head free;
757 };
758 
759 struct vmap_block {
760 	spinlock_t lock;
761 	struct vmap_area *va;
762 	unsigned long free, dirty;
763 	DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
764 	struct list_head free_list;
765 	struct rcu_head rcu_head;
766 	struct list_head purge;
767 };
768 
769 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
770 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
771 
772 /*
773  * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
774  * in the free path. Could get rid of this if we change the API to return a
775  * "cookie" from alloc, to be passed to free. But no big deal yet.
776  */
777 static DEFINE_SPINLOCK(vmap_block_tree_lock);
778 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
779 
780 /*
781  * We should probably have a fallback mechanism to allocate virtual memory
782  * out of partially filled vmap blocks. However vmap block sizing should be
783  * fairly reasonable according to the vmalloc size, so it shouldn't be a
784  * big problem.
785  */
786 
787 static unsigned long addr_to_vb_idx(unsigned long addr)
788 {
789 	addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
790 	addr /= VMAP_BLOCK_SIZE;
791 	return addr;
792 }
793 
794 static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
795 {
796 	struct vmap_block_queue *vbq;
797 	struct vmap_block *vb;
798 	struct vmap_area *va;
799 	unsigned long vb_idx;
800 	int node, err;
801 
802 	node = numa_node_id();
803 
804 	vb = kmalloc_node(sizeof(struct vmap_block),
805 			gfp_mask & GFP_RECLAIM_MASK, node);
806 	if (unlikely(!vb))
807 		return ERR_PTR(-ENOMEM);
808 
809 	va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
810 					VMALLOC_START, VMALLOC_END,
811 					node, gfp_mask);
812 	if (IS_ERR(va)) {
813 		kfree(vb);
814 		return ERR_CAST(va);
815 	}
816 
817 	err = radix_tree_preload(gfp_mask);
818 	if (unlikely(err)) {
819 		kfree(vb);
820 		free_vmap_area(va);
821 		return ERR_PTR(err);
822 	}
823 
824 	spin_lock_init(&vb->lock);
825 	vb->va = va;
826 	vb->free = VMAP_BBMAP_BITS;
827 	vb->dirty = 0;
828 	bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
829 	INIT_LIST_HEAD(&vb->free_list);
830 
831 	vb_idx = addr_to_vb_idx(va->va_start);
832 	spin_lock(&vmap_block_tree_lock);
833 	err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
834 	spin_unlock(&vmap_block_tree_lock);
835 	BUG_ON(err);
836 	radix_tree_preload_end();
837 
838 	vbq = &get_cpu_var(vmap_block_queue);
839 	spin_lock(&vbq->lock);
840 	list_add_rcu(&vb->free_list, &vbq->free);
841 	spin_unlock(&vbq->lock);
842 	put_cpu_var(vmap_block_queue);
843 
844 	return vb;
845 }
846 
847 static void free_vmap_block(struct vmap_block *vb)
848 {
849 	struct vmap_block *tmp;
850 	unsigned long vb_idx;
851 
852 	vb_idx = addr_to_vb_idx(vb->va->va_start);
853 	spin_lock(&vmap_block_tree_lock);
854 	tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
855 	spin_unlock(&vmap_block_tree_lock);
856 	BUG_ON(tmp != vb);
857 
858 	free_vmap_area_noflush(vb->va);
859 	kfree_rcu(vb, rcu_head);
860 }
861 
862 static void purge_fragmented_blocks(int cpu)
863 {
864 	LIST_HEAD(purge);
865 	struct vmap_block *vb;
866 	struct vmap_block *n_vb;
867 	struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
868 
869 	rcu_read_lock();
870 	list_for_each_entry_rcu(vb, &vbq->free, free_list) {
871 
872 		if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
873 			continue;
874 
875 		spin_lock(&vb->lock);
876 		if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
877 			vb->free = 0; /* prevent further allocs after releasing lock */
878 			vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
879 			bitmap_fill(vb->dirty_map, VMAP_BBMAP_BITS);
880 			spin_lock(&vbq->lock);
881 			list_del_rcu(&vb->free_list);
882 			spin_unlock(&vbq->lock);
883 			spin_unlock(&vb->lock);
884 			list_add_tail(&vb->purge, &purge);
885 		} else
886 			spin_unlock(&vb->lock);
887 	}
888 	rcu_read_unlock();
889 
890 	list_for_each_entry_safe(vb, n_vb, &purge, purge) {
891 		list_del(&vb->purge);
892 		free_vmap_block(vb);
893 	}
894 }
895 
896 static void purge_fragmented_blocks_allcpus(void)
897 {
898 	int cpu;
899 
900 	for_each_possible_cpu(cpu)
901 		purge_fragmented_blocks(cpu);
902 }
903 
904 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
905 {
906 	struct vmap_block_queue *vbq;
907 	struct vmap_block *vb;
908 	unsigned long addr = 0;
909 	unsigned int order;
910 
911 	BUG_ON(size & ~PAGE_MASK);
912 	BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
913 	if (WARN_ON(size == 0)) {
914 		/*
915 		 * Allocating 0 bytes isn't what caller wants since
916 		 * get_order(0) returns funny result. Just warn and terminate
917 		 * early.
918 		 */
919 		return NULL;
920 	}
921 	order = get_order(size);
922 
923 again:
924 	rcu_read_lock();
925 	vbq = &get_cpu_var(vmap_block_queue);
926 	list_for_each_entry_rcu(vb, &vbq->free, free_list) {
927 		int i;
928 
929 		spin_lock(&vb->lock);
930 		if (vb->free < 1UL << order)
931 			goto next;
932 
933 		i = VMAP_BBMAP_BITS - vb->free;
934 		addr = vb->va->va_start + (i << PAGE_SHIFT);
935 		BUG_ON(addr_to_vb_idx(addr) !=
936 				addr_to_vb_idx(vb->va->va_start));
937 		vb->free -= 1UL << order;
938 		if (vb->free == 0) {
939 			spin_lock(&vbq->lock);
940 			list_del_rcu(&vb->free_list);
941 			spin_unlock(&vbq->lock);
942 		}
943 		spin_unlock(&vb->lock);
944 		break;
945 next:
946 		spin_unlock(&vb->lock);
947 	}
948 
949 	put_cpu_var(vmap_block_queue);
950 	rcu_read_unlock();
951 
952 	if (!addr) {
953 		vb = new_vmap_block(gfp_mask);
954 		if (IS_ERR(vb))
955 			return vb;
956 		goto again;
957 	}
958 
959 	return (void *)addr;
960 }
961 
962 static void vb_free(const void *addr, unsigned long size)
963 {
964 	unsigned long offset;
965 	unsigned long vb_idx;
966 	unsigned int order;
967 	struct vmap_block *vb;
968 
969 	BUG_ON(size & ~PAGE_MASK);
970 	BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
971 
972 	flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
973 
974 	order = get_order(size);
975 
976 	offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
977 
978 	vb_idx = addr_to_vb_idx((unsigned long)addr);
979 	rcu_read_lock();
980 	vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
981 	rcu_read_unlock();
982 	BUG_ON(!vb);
983 
984 	vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
985 
986 	spin_lock(&vb->lock);
987 	BUG_ON(bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order));
988 
989 	vb->dirty += 1UL << order;
990 	if (vb->dirty == VMAP_BBMAP_BITS) {
991 		BUG_ON(vb->free);
992 		spin_unlock(&vb->lock);
993 		free_vmap_block(vb);
994 	} else
995 		spin_unlock(&vb->lock);
996 }
997 
998 /**
999  * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1000  *
1001  * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1002  * to amortize TLB flushing overheads. What this means is that any page you
1003  * have now, may, in a former life, have been mapped into kernel virtual
1004  * address by the vmap layer and so there might be some CPUs with TLB entries
1005  * still referencing that page (additional to the regular 1:1 kernel mapping).
1006  *
1007  * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1008  * be sure that none of the pages we have control over will have any aliases
1009  * from the vmap layer.
1010  */
1011 void vm_unmap_aliases(void)
1012 {
1013 	unsigned long start = ULONG_MAX, end = 0;
1014 	int cpu;
1015 	int flush = 0;
1016 
1017 	if (unlikely(!vmap_initialized))
1018 		return;
1019 
1020 	for_each_possible_cpu(cpu) {
1021 		struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1022 		struct vmap_block *vb;
1023 
1024 		rcu_read_lock();
1025 		list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1026 			int i, j;
1027 
1028 			spin_lock(&vb->lock);
1029 			i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
1030 			if (i < VMAP_BBMAP_BITS) {
1031 				unsigned long s, e;
1032 
1033 				j = find_last_bit(vb->dirty_map,
1034 							VMAP_BBMAP_BITS);
1035 				j = j + 1; /* need exclusive index */
1036 
1037 				s = vb->va->va_start + (i << PAGE_SHIFT);
1038 				e = vb->va->va_start + (j << PAGE_SHIFT);
1039 				flush = 1;
1040 
1041 				if (s < start)
1042 					start = s;
1043 				if (e > end)
1044 					end = e;
1045 			}
1046 			spin_unlock(&vb->lock);
1047 		}
1048 		rcu_read_unlock();
1049 	}
1050 
1051 	__purge_vmap_area_lazy(&start, &end, 1, flush);
1052 }
1053 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1054 
1055 /**
1056  * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1057  * @mem: the pointer returned by vm_map_ram
1058  * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1059  */
1060 void vm_unmap_ram(const void *mem, unsigned int count)
1061 {
1062 	unsigned long size = count << PAGE_SHIFT;
1063 	unsigned long addr = (unsigned long)mem;
1064 
1065 	BUG_ON(!addr);
1066 	BUG_ON(addr < VMALLOC_START);
1067 	BUG_ON(addr > VMALLOC_END);
1068 	BUG_ON(addr & (PAGE_SIZE-1));
1069 
1070 	debug_check_no_locks_freed(mem, size);
1071 	vmap_debug_free_range(addr, addr+size);
1072 
1073 	if (likely(count <= VMAP_MAX_ALLOC))
1074 		vb_free(mem, size);
1075 	else
1076 		free_unmap_vmap_area_addr(addr);
1077 }
1078 EXPORT_SYMBOL(vm_unmap_ram);
1079 
1080 /**
1081  * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1082  * @pages: an array of pointers to the pages to be mapped
1083  * @count: number of pages
1084  * @node: prefer to allocate data structures on this node
1085  * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1086  *
1087  * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1088  * faster than vmap so it's good.  But if you mix long-life and short-life
1089  * objects with vm_map_ram(), it could consume lots of address space through
1090  * fragmentation (especially on a 32bit machine).  You could see failures in
1091  * the end.  Please use this function for short-lived objects.
1092  *
1093  * Returns: a pointer to the address that has been mapped, or %NULL on failure
1094  */
1095 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1096 {
1097 	unsigned long size = count << PAGE_SHIFT;
1098 	unsigned long addr;
1099 	void *mem;
1100 
1101 	if (likely(count <= VMAP_MAX_ALLOC)) {
1102 		mem = vb_alloc(size, GFP_KERNEL);
1103 		if (IS_ERR(mem))
1104 			return NULL;
1105 		addr = (unsigned long)mem;
1106 	} else {
1107 		struct vmap_area *va;
1108 		va = alloc_vmap_area(size, PAGE_SIZE,
1109 				VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1110 		if (IS_ERR(va))
1111 			return NULL;
1112 
1113 		addr = va->va_start;
1114 		mem = (void *)addr;
1115 	}
1116 	if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1117 		vm_unmap_ram(mem, count);
1118 		return NULL;
1119 	}
1120 	return mem;
1121 }
1122 EXPORT_SYMBOL(vm_map_ram);
1123 
1124 static struct vm_struct *vmlist __initdata;
1125 /**
1126  * vm_area_add_early - add vmap area early during boot
1127  * @vm: vm_struct to add
1128  *
1129  * This function is used to add fixed kernel vm area to vmlist before
1130  * vmalloc_init() is called.  @vm->addr, @vm->size, and @vm->flags
1131  * should contain proper values and the other fields should be zero.
1132  *
1133  * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1134  */
1135 void __init vm_area_add_early(struct vm_struct *vm)
1136 {
1137 	struct vm_struct *tmp, **p;
1138 
1139 	BUG_ON(vmap_initialized);
1140 	for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1141 		if (tmp->addr >= vm->addr) {
1142 			BUG_ON(tmp->addr < vm->addr + vm->size);
1143 			break;
1144 		} else
1145 			BUG_ON(tmp->addr + tmp->size > vm->addr);
1146 	}
1147 	vm->next = *p;
1148 	*p = vm;
1149 }
1150 
1151 /**
1152  * vm_area_register_early - register vmap area early during boot
1153  * @vm: vm_struct to register
1154  * @align: requested alignment
1155  *
1156  * This function is used to register kernel vm area before
1157  * vmalloc_init() is called.  @vm->size and @vm->flags should contain
1158  * proper values on entry and other fields should be zero.  On return,
1159  * vm->addr contains the allocated address.
1160  *
1161  * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1162  */
1163 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1164 {
1165 	static size_t vm_init_off __initdata;
1166 	unsigned long addr;
1167 
1168 	addr = ALIGN(VMALLOC_START + vm_init_off, align);
1169 	vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1170 
1171 	vm->addr = (void *)addr;
1172 
1173 	vm_area_add_early(vm);
1174 }
1175 
1176 void __init vmalloc_init(void)
1177 {
1178 	struct vmap_area *va;
1179 	struct vm_struct *tmp;
1180 	int i;
1181 
1182 	for_each_possible_cpu(i) {
1183 		struct vmap_block_queue *vbq;
1184 		struct vfree_deferred *p;
1185 
1186 		vbq = &per_cpu(vmap_block_queue, i);
1187 		spin_lock_init(&vbq->lock);
1188 		INIT_LIST_HEAD(&vbq->free);
1189 		p = &per_cpu(vfree_deferred, i);
1190 		init_llist_head(&p->list);
1191 		INIT_WORK(&p->wq, free_work);
1192 	}
1193 
1194 	/* Import existing vmlist entries. */
1195 	for (tmp = vmlist; tmp; tmp = tmp->next) {
1196 		va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1197 		va->flags = VM_VM_AREA;
1198 		va->va_start = (unsigned long)tmp->addr;
1199 		va->va_end = va->va_start + tmp->size;
1200 		va->vm = tmp;
1201 		__insert_vmap_area(va);
1202 	}
1203 
1204 	vmap_area_pcpu_hole = VMALLOC_END;
1205 
1206 	vmap_initialized = true;
1207 }
1208 
1209 /**
1210  * map_kernel_range_noflush - map kernel VM area with the specified pages
1211  * @addr: start of the VM area to map
1212  * @size: size of the VM area to map
1213  * @prot: page protection flags to use
1214  * @pages: pages to map
1215  *
1216  * Map PFN_UP(@size) pages at @addr.  The VM area @addr and @size
1217  * specify should have been allocated using get_vm_area() and its
1218  * friends.
1219  *
1220  * NOTE:
1221  * This function does NOT do any cache flushing.  The caller is
1222  * responsible for calling flush_cache_vmap() on to-be-mapped areas
1223  * before calling this function.
1224  *
1225  * RETURNS:
1226  * The number of pages mapped on success, -errno on failure.
1227  */
1228 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1229 			     pgprot_t prot, struct page **pages)
1230 {
1231 	return vmap_page_range_noflush(addr, addr + size, prot, pages);
1232 }
1233 
1234 /**
1235  * unmap_kernel_range_noflush - unmap kernel VM area
1236  * @addr: start of the VM area to unmap
1237  * @size: size of the VM area to unmap
1238  *
1239  * Unmap PFN_UP(@size) pages at @addr.  The VM area @addr and @size
1240  * specify should have been allocated using get_vm_area() and its
1241  * friends.
1242  *
1243  * NOTE:
1244  * This function does NOT do any cache flushing.  The caller is
1245  * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1246  * before calling this function and flush_tlb_kernel_range() after.
1247  */
1248 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1249 {
1250 	vunmap_page_range(addr, addr + size);
1251 }
1252 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1253 
1254 /**
1255  * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1256  * @addr: start of the VM area to unmap
1257  * @size: size of the VM area to unmap
1258  *
1259  * Similar to unmap_kernel_range_noflush() but flushes vcache before
1260  * the unmapping and tlb after.
1261  */
1262 void unmap_kernel_range(unsigned long addr, unsigned long size)
1263 {
1264 	unsigned long end = addr + size;
1265 
1266 	flush_cache_vunmap(addr, end);
1267 	vunmap_page_range(addr, end);
1268 	flush_tlb_kernel_range(addr, end);
1269 }
1270 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1271 
1272 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1273 {
1274 	unsigned long addr = (unsigned long)area->addr;
1275 	unsigned long end = addr + get_vm_area_size(area);
1276 	int err;
1277 
1278 	err = vmap_page_range(addr, end, prot, pages);
1279 
1280 	return err > 0 ? 0 : err;
1281 }
1282 EXPORT_SYMBOL_GPL(map_vm_area);
1283 
1284 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1285 			      unsigned long flags, const void *caller)
1286 {
1287 	spin_lock(&vmap_area_lock);
1288 	vm->flags = flags;
1289 	vm->addr = (void *)va->va_start;
1290 	vm->size = va->va_end - va->va_start;
1291 	vm->caller = caller;
1292 	va->vm = vm;
1293 	va->flags |= VM_VM_AREA;
1294 	spin_unlock(&vmap_area_lock);
1295 }
1296 
1297 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1298 {
1299 	/*
1300 	 * Before removing VM_UNINITIALIZED,
1301 	 * we should make sure that vm has proper values.
1302 	 * Pair with smp_rmb() in show_numa_info().
1303 	 */
1304 	smp_wmb();
1305 	vm->flags &= ~VM_UNINITIALIZED;
1306 }
1307 
1308 static struct vm_struct *__get_vm_area_node(unsigned long size,
1309 		unsigned long align, unsigned long flags, unsigned long start,
1310 		unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1311 {
1312 	struct vmap_area *va;
1313 	struct vm_struct *area;
1314 
1315 	BUG_ON(in_interrupt());
1316 	if (flags & VM_IOREMAP)
1317 		align = 1ul << clamp(fls(size), PAGE_SHIFT, IOREMAP_MAX_ORDER);
1318 
1319 	size = PAGE_ALIGN(size);
1320 	if (unlikely(!size))
1321 		return NULL;
1322 
1323 	area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1324 	if (unlikely(!area))
1325 		return NULL;
1326 
1327 	/*
1328 	 * We always allocate a guard page.
1329 	 */
1330 	size += PAGE_SIZE;
1331 
1332 	va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1333 	if (IS_ERR(va)) {
1334 		kfree(area);
1335 		return NULL;
1336 	}
1337 
1338 	setup_vmalloc_vm(area, va, flags, caller);
1339 
1340 	return area;
1341 }
1342 
1343 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1344 				unsigned long start, unsigned long end)
1345 {
1346 	return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1347 				  GFP_KERNEL, __builtin_return_address(0));
1348 }
1349 EXPORT_SYMBOL_GPL(__get_vm_area);
1350 
1351 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1352 				       unsigned long start, unsigned long end,
1353 				       const void *caller)
1354 {
1355 	return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1356 				  GFP_KERNEL, caller);
1357 }
1358 
1359 /**
1360  *	get_vm_area  -  reserve a contiguous kernel virtual area
1361  *	@size:		size of the area
1362  *	@flags:		%VM_IOREMAP for I/O mappings or VM_ALLOC
1363  *
1364  *	Search an area of @size in the kernel virtual mapping area,
1365  *	and reserved it for out purposes.  Returns the area descriptor
1366  *	on success or %NULL on failure.
1367  */
1368 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1369 {
1370 	return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1371 				  NUMA_NO_NODE, GFP_KERNEL,
1372 				  __builtin_return_address(0));
1373 }
1374 
1375 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1376 				const void *caller)
1377 {
1378 	return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1379 				  NUMA_NO_NODE, GFP_KERNEL, caller);
1380 }
1381 
1382 /**
1383  *	find_vm_area  -  find a continuous kernel virtual area
1384  *	@addr:		base address
1385  *
1386  *	Search for the kernel VM area starting at @addr, and return it.
1387  *	It is up to the caller to do all required locking to keep the returned
1388  *	pointer valid.
1389  */
1390 struct vm_struct *find_vm_area(const void *addr)
1391 {
1392 	struct vmap_area *va;
1393 
1394 	va = find_vmap_area((unsigned long)addr);
1395 	if (va && va->flags & VM_VM_AREA)
1396 		return va->vm;
1397 
1398 	return NULL;
1399 }
1400 
1401 /**
1402  *	remove_vm_area  -  find and remove a continuous kernel virtual area
1403  *	@addr:		base address
1404  *
1405  *	Search for the kernel VM area starting at @addr, and remove it.
1406  *	This function returns the found VM area, but using it is NOT safe
1407  *	on SMP machines, except for its size or flags.
1408  */
1409 struct vm_struct *remove_vm_area(const void *addr)
1410 {
1411 	struct vmap_area *va;
1412 
1413 	va = find_vmap_area((unsigned long)addr);
1414 	if (va && va->flags & VM_VM_AREA) {
1415 		struct vm_struct *vm = va->vm;
1416 
1417 		spin_lock(&vmap_area_lock);
1418 		va->vm = NULL;
1419 		va->flags &= ~VM_VM_AREA;
1420 		spin_unlock(&vmap_area_lock);
1421 
1422 		vmap_debug_free_range(va->va_start, va->va_end);
1423 		free_unmap_vmap_area(va);
1424 		vm->size -= PAGE_SIZE;
1425 
1426 		return vm;
1427 	}
1428 	return NULL;
1429 }
1430 
1431 static void __vunmap(const void *addr, int deallocate_pages)
1432 {
1433 	struct vm_struct *area;
1434 
1435 	if (!addr)
1436 		return;
1437 
1438 	if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1439 			addr))
1440 		return;
1441 
1442 	area = remove_vm_area(addr);
1443 	if (unlikely(!area)) {
1444 		WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1445 				addr);
1446 		return;
1447 	}
1448 
1449 	debug_check_no_locks_freed(addr, area->size);
1450 	debug_check_no_obj_freed(addr, area->size);
1451 
1452 	if (deallocate_pages) {
1453 		int i;
1454 
1455 		for (i = 0; i < area->nr_pages; i++) {
1456 			struct page *page = area->pages[i];
1457 
1458 			BUG_ON(!page);
1459 			__free_page(page);
1460 		}
1461 
1462 		if (area->flags & VM_VPAGES)
1463 			vfree(area->pages);
1464 		else
1465 			kfree(area->pages);
1466 	}
1467 
1468 	kfree(area);
1469 	return;
1470 }
1471 
1472 /**
1473  *	vfree  -  release memory allocated by vmalloc()
1474  *	@addr:		memory base address
1475  *
1476  *	Free the virtually continuous memory area starting at @addr, as
1477  *	obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1478  *	NULL, no operation is performed.
1479  *
1480  *	Must not be called in NMI context (strictly speaking, only if we don't
1481  *	have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1482  *	conventions for vfree() arch-depenedent would be a really bad idea)
1483  *
1484  *	NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1485  */
1486 void vfree(const void *addr)
1487 {
1488 	BUG_ON(in_nmi());
1489 
1490 	kmemleak_free(addr);
1491 
1492 	if (!addr)
1493 		return;
1494 	if (unlikely(in_interrupt())) {
1495 		struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
1496 		if (llist_add((struct llist_node *)addr, &p->list))
1497 			schedule_work(&p->wq);
1498 	} else
1499 		__vunmap(addr, 1);
1500 }
1501 EXPORT_SYMBOL(vfree);
1502 
1503 /**
1504  *	vunmap  -  release virtual mapping obtained by vmap()
1505  *	@addr:		memory base address
1506  *
1507  *	Free the virtually contiguous memory area starting at @addr,
1508  *	which was created from the page array passed to vmap().
1509  *
1510  *	Must not be called in interrupt context.
1511  */
1512 void vunmap(const void *addr)
1513 {
1514 	BUG_ON(in_interrupt());
1515 	might_sleep();
1516 	if (addr)
1517 		__vunmap(addr, 0);
1518 }
1519 EXPORT_SYMBOL(vunmap);
1520 
1521 /**
1522  *	vmap  -  map an array of pages into virtually contiguous space
1523  *	@pages:		array of page pointers
1524  *	@count:		number of pages to map
1525  *	@flags:		vm_area->flags
1526  *	@prot:		page protection for the mapping
1527  *
1528  *	Maps @count pages from @pages into contiguous kernel virtual
1529  *	space.
1530  */
1531 void *vmap(struct page **pages, unsigned int count,
1532 		unsigned long flags, pgprot_t prot)
1533 {
1534 	struct vm_struct *area;
1535 
1536 	might_sleep();
1537 
1538 	if (count > totalram_pages)
1539 		return NULL;
1540 
1541 	area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1542 					__builtin_return_address(0));
1543 	if (!area)
1544 		return NULL;
1545 
1546 	if (map_vm_area(area, prot, pages)) {
1547 		vunmap(area->addr);
1548 		return NULL;
1549 	}
1550 
1551 	return area->addr;
1552 }
1553 EXPORT_SYMBOL(vmap);
1554 
1555 static void *__vmalloc_node(unsigned long size, unsigned long align,
1556 			    gfp_t gfp_mask, pgprot_t prot,
1557 			    int node, const void *caller);
1558 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1559 				 pgprot_t prot, int node)
1560 {
1561 	const int order = 0;
1562 	struct page **pages;
1563 	unsigned int nr_pages, array_size, i;
1564 	const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1565 	const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1566 
1567 	nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1568 	array_size = (nr_pages * sizeof(struct page *));
1569 
1570 	area->nr_pages = nr_pages;
1571 	/* Please note that the recursion is strictly bounded. */
1572 	if (array_size > PAGE_SIZE) {
1573 		pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1574 				PAGE_KERNEL, node, area->caller);
1575 		area->flags |= VM_VPAGES;
1576 	} else {
1577 		pages = kmalloc_node(array_size, nested_gfp, node);
1578 	}
1579 	area->pages = pages;
1580 	if (!area->pages) {
1581 		remove_vm_area(area->addr);
1582 		kfree(area);
1583 		return NULL;
1584 	}
1585 
1586 	for (i = 0; i < area->nr_pages; i++) {
1587 		struct page *page;
1588 
1589 		if (node == NUMA_NO_NODE)
1590 			page = alloc_page(alloc_mask);
1591 		else
1592 			page = alloc_pages_node(node, alloc_mask, order);
1593 
1594 		if (unlikely(!page)) {
1595 			/* Successfully allocated i pages, free them in __vunmap() */
1596 			area->nr_pages = i;
1597 			goto fail;
1598 		}
1599 		area->pages[i] = page;
1600 		if (gfp_mask & __GFP_WAIT)
1601 			cond_resched();
1602 	}
1603 
1604 	if (map_vm_area(area, prot, pages))
1605 		goto fail;
1606 	return area->addr;
1607 
1608 fail:
1609 	warn_alloc_failed(gfp_mask, order,
1610 			  "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1611 			  (area->nr_pages*PAGE_SIZE), area->size);
1612 	vfree(area->addr);
1613 	return NULL;
1614 }
1615 
1616 /**
1617  *	__vmalloc_node_range  -  allocate virtually contiguous memory
1618  *	@size:		allocation size
1619  *	@align:		desired alignment
1620  *	@start:		vm area range start
1621  *	@end:		vm area range end
1622  *	@gfp_mask:	flags for the page level allocator
1623  *	@prot:		protection mask for the allocated pages
1624  *	@node:		node to use for allocation or NUMA_NO_NODE
1625  *	@caller:	caller's return address
1626  *
1627  *	Allocate enough pages to cover @size from the page level
1628  *	allocator with @gfp_mask flags.  Map them into contiguous
1629  *	kernel virtual space, using a pagetable protection of @prot.
1630  */
1631 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1632 			unsigned long start, unsigned long end, gfp_t gfp_mask,
1633 			pgprot_t prot, int node, const void *caller)
1634 {
1635 	struct vm_struct *area;
1636 	void *addr;
1637 	unsigned long real_size = size;
1638 
1639 	size = PAGE_ALIGN(size);
1640 	if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1641 		goto fail;
1642 
1643 	area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED,
1644 				  start, end, node, gfp_mask, caller);
1645 	if (!area)
1646 		goto fail;
1647 
1648 	addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1649 	if (!addr)
1650 		return NULL;
1651 
1652 	/*
1653 	 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1654 	 * flag. It means that vm_struct is not fully initialized.
1655 	 * Now, it is fully initialized, so remove this flag here.
1656 	 */
1657 	clear_vm_uninitialized_flag(area);
1658 
1659 	/*
1660 	 * A ref_count = 2 is needed because vm_struct allocated in
1661 	 * __get_vm_area_node() contains a reference to the virtual address of
1662 	 * the vmalloc'ed block.
1663 	 */
1664 	kmemleak_alloc(addr, real_size, 2, gfp_mask);
1665 
1666 	return addr;
1667 
1668 fail:
1669 	warn_alloc_failed(gfp_mask, 0,
1670 			  "vmalloc: allocation failure: %lu bytes\n",
1671 			  real_size);
1672 	return NULL;
1673 }
1674 
1675 /**
1676  *	__vmalloc_node  -  allocate virtually contiguous memory
1677  *	@size:		allocation size
1678  *	@align:		desired alignment
1679  *	@gfp_mask:	flags for the page level allocator
1680  *	@prot:		protection mask for the allocated pages
1681  *	@node:		node to use for allocation or NUMA_NO_NODE
1682  *	@caller:	caller's return address
1683  *
1684  *	Allocate enough pages to cover @size from the page level
1685  *	allocator with @gfp_mask flags.  Map them into contiguous
1686  *	kernel virtual space, using a pagetable protection of @prot.
1687  */
1688 static void *__vmalloc_node(unsigned long size, unsigned long align,
1689 			    gfp_t gfp_mask, pgprot_t prot,
1690 			    int node, const void *caller)
1691 {
1692 	return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1693 				gfp_mask, prot, node, caller);
1694 }
1695 
1696 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1697 {
1698 	return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1699 				__builtin_return_address(0));
1700 }
1701 EXPORT_SYMBOL(__vmalloc);
1702 
1703 static inline void *__vmalloc_node_flags(unsigned long size,
1704 					int node, gfp_t flags)
1705 {
1706 	return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1707 					node, __builtin_return_address(0));
1708 }
1709 
1710 /**
1711  *	vmalloc  -  allocate virtually contiguous memory
1712  *	@size:		allocation size
1713  *	Allocate enough pages to cover @size from the page level
1714  *	allocator and map them into contiguous kernel virtual space.
1715  *
1716  *	For tight control over page level allocator and protection flags
1717  *	use __vmalloc() instead.
1718  */
1719 void *vmalloc(unsigned long size)
1720 {
1721 	return __vmalloc_node_flags(size, NUMA_NO_NODE,
1722 				    GFP_KERNEL | __GFP_HIGHMEM);
1723 }
1724 EXPORT_SYMBOL(vmalloc);
1725 
1726 /**
1727  *	vzalloc - allocate virtually contiguous memory with zero fill
1728  *	@size:	allocation size
1729  *	Allocate enough pages to cover @size from the page level
1730  *	allocator and map them into contiguous kernel virtual space.
1731  *	The memory allocated is set to zero.
1732  *
1733  *	For tight control over page level allocator and protection flags
1734  *	use __vmalloc() instead.
1735  */
1736 void *vzalloc(unsigned long size)
1737 {
1738 	return __vmalloc_node_flags(size, NUMA_NO_NODE,
1739 				GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1740 }
1741 EXPORT_SYMBOL(vzalloc);
1742 
1743 /**
1744  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1745  * @size: allocation size
1746  *
1747  * The resulting memory area is zeroed so it can be mapped to userspace
1748  * without leaking data.
1749  */
1750 void *vmalloc_user(unsigned long size)
1751 {
1752 	struct vm_struct *area;
1753 	void *ret;
1754 
1755 	ret = __vmalloc_node(size, SHMLBA,
1756 			     GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1757 			     PAGE_KERNEL, NUMA_NO_NODE,
1758 			     __builtin_return_address(0));
1759 	if (ret) {
1760 		area = find_vm_area(ret);
1761 		area->flags |= VM_USERMAP;
1762 	}
1763 	return ret;
1764 }
1765 EXPORT_SYMBOL(vmalloc_user);
1766 
1767 /**
1768  *	vmalloc_node  -  allocate memory on a specific node
1769  *	@size:		allocation size
1770  *	@node:		numa node
1771  *
1772  *	Allocate enough pages to cover @size from the page level
1773  *	allocator and map them into contiguous kernel virtual space.
1774  *
1775  *	For tight control over page level allocator and protection flags
1776  *	use __vmalloc() instead.
1777  */
1778 void *vmalloc_node(unsigned long size, int node)
1779 {
1780 	return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1781 					node, __builtin_return_address(0));
1782 }
1783 EXPORT_SYMBOL(vmalloc_node);
1784 
1785 /**
1786  * vzalloc_node - allocate memory on a specific node with zero fill
1787  * @size:	allocation size
1788  * @node:	numa node
1789  *
1790  * Allocate enough pages to cover @size from the page level
1791  * allocator and map them into contiguous kernel virtual space.
1792  * The memory allocated is set to zero.
1793  *
1794  * For tight control over page level allocator and protection flags
1795  * use __vmalloc_node() instead.
1796  */
1797 void *vzalloc_node(unsigned long size, int node)
1798 {
1799 	return __vmalloc_node_flags(size, node,
1800 			 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1801 }
1802 EXPORT_SYMBOL(vzalloc_node);
1803 
1804 #ifndef PAGE_KERNEL_EXEC
1805 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1806 #endif
1807 
1808 /**
1809  *	vmalloc_exec  -  allocate virtually contiguous, executable memory
1810  *	@size:		allocation size
1811  *
1812  *	Kernel-internal function to allocate enough pages to cover @size
1813  *	the page level allocator and map them into contiguous and
1814  *	executable kernel virtual space.
1815  *
1816  *	For tight control over page level allocator and protection flags
1817  *	use __vmalloc() instead.
1818  */
1819 
1820 void *vmalloc_exec(unsigned long size)
1821 {
1822 	return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1823 			      NUMA_NO_NODE, __builtin_return_address(0));
1824 }
1825 
1826 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1827 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1828 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1829 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1830 #else
1831 #define GFP_VMALLOC32 GFP_KERNEL
1832 #endif
1833 
1834 /**
1835  *	vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
1836  *	@size:		allocation size
1837  *
1838  *	Allocate enough 32bit PA addressable pages to cover @size from the
1839  *	page level allocator and map them into contiguous kernel virtual space.
1840  */
1841 void *vmalloc_32(unsigned long size)
1842 {
1843 	return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1844 			      NUMA_NO_NODE, __builtin_return_address(0));
1845 }
1846 EXPORT_SYMBOL(vmalloc_32);
1847 
1848 /**
1849  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1850  *	@size:		allocation size
1851  *
1852  * The resulting memory area is 32bit addressable and zeroed so it can be
1853  * mapped to userspace without leaking data.
1854  */
1855 void *vmalloc_32_user(unsigned long size)
1856 {
1857 	struct vm_struct *area;
1858 	void *ret;
1859 
1860 	ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1861 			     NUMA_NO_NODE, __builtin_return_address(0));
1862 	if (ret) {
1863 		area = find_vm_area(ret);
1864 		area->flags |= VM_USERMAP;
1865 	}
1866 	return ret;
1867 }
1868 EXPORT_SYMBOL(vmalloc_32_user);
1869 
1870 /*
1871  * small helper routine , copy contents to buf from addr.
1872  * If the page is not present, fill zero.
1873  */
1874 
1875 static int aligned_vread(char *buf, char *addr, unsigned long count)
1876 {
1877 	struct page *p;
1878 	int copied = 0;
1879 
1880 	while (count) {
1881 		unsigned long offset, length;
1882 
1883 		offset = (unsigned long)addr & ~PAGE_MASK;
1884 		length = PAGE_SIZE - offset;
1885 		if (length > count)
1886 			length = count;
1887 		p = vmalloc_to_page(addr);
1888 		/*
1889 		 * To do safe access to this _mapped_ area, we need
1890 		 * lock. But adding lock here means that we need to add
1891 		 * overhead of vmalloc()/vfree() calles for this _debug_
1892 		 * interface, rarely used. Instead of that, we'll use
1893 		 * kmap() and get small overhead in this access function.
1894 		 */
1895 		if (p) {
1896 			/*
1897 			 * we can expect USER0 is not used (see vread/vwrite's
1898 			 * function description)
1899 			 */
1900 			void *map = kmap_atomic(p);
1901 			memcpy(buf, map + offset, length);
1902 			kunmap_atomic(map);
1903 		} else
1904 			memset(buf, 0, length);
1905 
1906 		addr += length;
1907 		buf += length;
1908 		copied += length;
1909 		count -= length;
1910 	}
1911 	return copied;
1912 }
1913 
1914 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1915 {
1916 	struct page *p;
1917 	int copied = 0;
1918 
1919 	while (count) {
1920 		unsigned long offset, length;
1921 
1922 		offset = (unsigned long)addr & ~PAGE_MASK;
1923 		length = PAGE_SIZE - offset;
1924 		if (length > count)
1925 			length = count;
1926 		p = vmalloc_to_page(addr);
1927 		/*
1928 		 * To do safe access to this _mapped_ area, we need
1929 		 * lock. But adding lock here means that we need to add
1930 		 * overhead of vmalloc()/vfree() calles for this _debug_
1931 		 * interface, rarely used. Instead of that, we'll use
1932 		 * kmap() and get small overhead in this access function.
1933 		 */
1934 		if (p) {
1935 			/*
1936 			 * we can expect USER0 is not used (see vread/vwrite's
1937 			 * function description)
1938 			 */
1939 			void *map = kmap_atomic(p);
1940 			memcpy(map + offset, buf, length);
1941 			kunmap_atomic(map);
1942 		}
1943 		addr += length;
1944 		buf += length;
1945 		copied += length;
1946 		count -= length;
1947 	}
1948 	return copied;
1949 }
1950 
1951 /**
1952  *	vread() -  read vmalloc area in a safe way.
1953  *	@buf:		buffer for reading data
1954  *	@addr:		vm address.
1955  *	@count:		number of bytes to be read.
1956  *
1957  *	Returns # of bytes which addr and buf should be increased.
1958  *	(same number to @count). Returns 0 if [addr...addr+count) doesn't
1959  *	includes any intersect with alive vmalloc area.
1960  *
1961  *	This function checks that addr is a valid vmalloc'ed area, and
1962  *	copy data from that area to a given buffer. If the given memory range
1963  *	of [addr...addr+count) includes some valid address, data is copied to
1964  *	proper area of @buf. If there are memory holes, they'll be zero-filled.
1965  *	IOREMAP area is treated as memory hole and no copy is done.
1966  *
1967  *	If [addr...addr+count) doesn't includes any intersects with alive
1968  *	vm_struct area, returns 0. @buf should be kernel's buffer.
1969  *
1970  *	Note: In usual ops, vread() is never necessary because the caller
1971  *	should know vmalloc() area is valid and can use memcpy().
1972  *	This is for routines which have to access vmalloc area without
1973  *	any informaion, as /dev/kmem.
1974  *
1975  */
1976 
1977 long vread(char *buf, char *addr, unsigned long count)
1978 {
1979 	struct vmap_area *va;
1980 	struct vm_struct *vm;
1981 	char *vaddr, *buf_start = buf;
1982 	unsigned long buflen = count;
1983 	unsigned long n;
1984 
1985 	/* Don't allow overflow */
1986 	if ((unsigned long) addr + count < count)
1987 		count = -(unsigned long) addr;
1988 
1989 	spin_lock(&vmap_area_lock);
1990 	list_for_each_entry(va, &vmap_area_list, list) {
1991 		if (!count)
1992 			break;
1993 
1994 		if (!(va->flags & VM_VM_AREA))
1995 			continue;
1996 
1997 		vm = va->vm;
1998 		vaddr = (char *) vm->addr;
1999 		if (addr >= vaddr + get_vm_area_size(vm))
2000 			continue;
2001 		while (addr < vaddr) {
2002 			if (count == 0)
2003 				goto finished;
2004 			*buf = '\0';
2005 			buf++;
2006 			addr++;
2007 			count--;
2008 		}
2009 		n = vaddr + get_vm_area_size(vm) - addr;
2010 		if (n > count)
2011 			n = count;
2012 		if (!(vm->flags & VM_IOREMAP))
2013 			aligned_vread(buf, addr, n);
2014 		else /* IOREMAP area is treated as memory hole */
2015 			memset(buf, 0, n);
2016 		buf += n;
2017 		addr += n;
2018 		count -= n;
2019 	}
2020 finished:
2021 	spin_unlock(&vmap_area_lock);
2022 
2023 	if (buf == buf_start)
2024 		return 0;
2025 	/* zero-fill memory holes */
2026 	if (buf != buf_start + buflen)
2027 		memset(buf, 0, buflen - (buf - buf_start));
2028 
2029 	return buflen;
2030 }
2031 
2032 /**
2033  *	vwrite() -  write vmalloc area in a safe way.
2034  *	@buf:		buffer for source data
2035  *	@addr:		vm address.
2036  *	@count:		number of bytes to be read.
2037  *
2038  *	Returns # of bytes which addr and buf should be incresed.
2039  *	(same number to @count).
2040  *	If [addr...addr+count) doesn't includes any intersect with valid
2041  *	vmalloc area, returns 0.
2042  *
2043  *	This function checks that addr is a valid vmalloc'ed area, and
2044  *	copy data from a buffer to the given addr. If specified range of
2045  *	[addr...addr+count) includes some valid address, data is copied from
2046  *	proper area of @buf. If there are memory holes, no copy to hole.
2047  *	IOREMAP area is treated as memory hole and no copy is done.
2048  *
2049  *	If [addr...addr+count) doesn't includes any intersects with alive
2050  *	vm_struct area, returns 0. @buf should be kernel's buffer.
2051  *
2052  *	Note: In usual ops, vwrite() is never necessary because the caller
2053  *	should know vmalloc() area is valid and can use memcpy().
2054  *	This is for routines which have to access vmalloc area without
2055  *	any informaion, as /dev/kmem.
2056  */
2057 
2058 long vwrite(char *buf, char *addr, unsigned long count)
2059 {
2060 	struct vmap_area *va;
2061 	struct vm_struct *vm;
2062 	char *vaddr;
2063 	unsigned long n, buflen;
2064 	int copied = 0;
2065 
2066 	/* Don't allow overflow */
2067 	if ((unsigned long) addr + count < count)
2068 		count = -(unsigned long) addr;
2069 	buflen = count;
2070 
2071 	spin_lock(&vmap_area_lock);
2072 	list_for_each_entry(va, &vmap_area_list, list) {
2073 		if (!count)
2074 			break;
2075 
2076 		if (!(va->flags & VM_VM_AREA))
2077 			continue;
2078 
2079 		vm = va->vm;
2080 		vaddr = (char *) vm->addr;
2081 		if (addr >= vaddr + get_vm_area_size(vm))
2082 			continue;
2083 		while (addr < vaddr) {
2084 			if (count == 0)
2085 				goto finished;
2086 			buf++;
2087 			addr++;
2088 			count--;
2089 		}
2090 		n = vaddr + get_vm_area_size(vm) - addr;
2091 		if (n > count)
2092 			n = count;
2093 		if (!(vm->flags & VM_IOREMAP)) {
2094 			aligned_vwrite(buf, addr, n);
2095 			copied++;
2096 		}
2097 		buf += n;
2098 		addr += n;
2099 		count -= n;
2100 	}
2101 finished:
2102 	spin_unlock(&vmap_area_lock);
2103 	if (!copied)
2104 		return 0;
2105 	return buflen;
2106 }
2107 
2108 /**
2109  *	remap_vmalloc_range_partial  -  map vmalloc pages to userspace
2110  *	@vma:		vma to cover
2111  *	@uaddr:		target user address to start at
2112  *	@kaddr:		virtual address of vmalloc kernel memory
2113  *	@size:		size of map area
2114  *
2115  *	Returns:	0 for success, -Exxx on failure
2116  *
2117  *	This function checks that @kaddr is a valid vmalloc'ed area,
2118  *	and that it is big enough to cover the range starting at
2119  *	@uaddr in @vma. Will return failure if that criteria isn't
2120  *	met.
2121  *
2122  *	Similar to remap_pfn_range() (see mm/memory.c)
2123  */
2124 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2125 				void *kaddr, unsigned long size)
2126 {
2127 	struct vm_struct *area;
2128 
2129 	size = PAGE_ALIGN(size);
2130 
2131 	if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2132 		return -EINVAL;
2133 
2134 	area = find_vm_area(kaddr);
2135 	if (!area)
2136 		return -EINVAL;
2137 
2138 	if (!(area->flags & VM_USERMAP))
2139 		return -EINVAL;
2140 
2141 	if (kaddr + size > area->addr + area->size)
2142 		return -EINVAL;
2143 
2144 	do {
2145 		struct page *page = vmalloc_to_page(kaddr);
2146 		int ret;
2147 
2148 		ret = vm_insert_page(vma, uaddr, page);
2149 		if (ret)
2150 			return ret;
2151 
2152 		uaddr += PAGE_SIZE;
2153 		kaddr += PAGE_SIZE;
2154 		size -= PAGE_SIZE;
2155 	} while (size > 0);
2156 
2157 	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2158 
2159 	return 0;
2160 }
2161 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2162 
2163 /**
2164  *	remap_vmalloc_range  -  map vmalloc pages to userspace
2165  *	@vma:		vma to cover (map full range of vma)
2166  *	@addr:		vmalloc memory
2167  *	@pgoff:		number of pages into addr before first page to map
2168  *
2169  *	Returns:	0 for success, -Exxx on failure
2170  *
2171  *	This function checks that addr is a valid vmalloc'ed area, and
2172  *	that it is big enough to cover the vma. Will return failure if
2173  *	that criteria isn't met.
2174  *
2175  *	Similar to remap_pfn_range() (see mm/memory.c)
2176  */
2177 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2178 						unsigned long pgoff)
2179 {
2180 	return remap_vmalloc_range_partial(vma, vma->vm_start,
2181 					   addr + (pgoff << PAGE_SHIFT),
2182 					   vma->vm_end - vma->vm_start);
2183 }
2184 EXPORT_SYMBOL(remap_vmalloc_range);
2185 
2186 /*
2187  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2188  * have one.
2189  */
2190 void __weak vmalloc_sync_all(void)
2191 {
2192 }
2193 
2194 
2195 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2196 {
2197 	pte_t ***p = data;
2198 
2199 	if (p) {
2200 		*(*p) = pte;
2201 		(*p)++;
2202 	}
2203 	return 0;
2204 }
2205 
2206 /**
2207  *	alloc_vm_area - allocate a range of kernel address space
2208  *	@size:		size of the area
2209  *	@ptes:		returns the PTEs for the address space
2210  *
2211  *	Returns:	NULL on failure, vm_struct on success
2212  *
2213  *	This function reserves a range of kernel address space, and
2214  *	allocates pagetables to map that range.  No actual mappings
2215  *	are created.
2216  *
2217  *	If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2218  *	allocated for the VM area are returned.
2219  */
2220 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2221 {
2222 	struct vm_struct *area;
2223 
2224 	area = get_vm_area_caller(size, VM_IOREMAP,
2225 				__builtin_return_address(0));
2226 	if (area == NULL)
2227 		return NULL;
2228 
2229 	/*
2230 	 * This ensures that page tables are constructed for this region
2231 	 * of kernel virtual address space and mapped into init_mm.
2232 	 */
2233 	if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2234 				size, f, ptes ? &ptes : NULL)) {
2235 		free_vm_area(area);
2236 		return NULL;
2237 	}
2238 
2239 	return area;
2240 }
2241 EXPORT_SYMBOL_GPL(alloc_vm_area);
2242 
2243 void free_vm_area(struct vm_struct *area)
2244 {
2245 	struct vm_struct *ret;
2246 	ret = remove_vm_area(area->addr);
2247 	BUG_ON(ret != area);
2248 	kfree(area);
2249 }
2250 EXPORT_SYMBOL_GPL(free_vm_area);
2251 
2252 #ifdef CONFIG_SMP
2253 static struct vmap_area *node_to_va(struct rb_node *n)
2254 {
2255 	return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2256 }
2257 
2258 /**
2259  * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2260  * @end: target address
2261  * @pnext: out arg for the next vmap_area
2262  * @pprev: out arg for the previous vmap_area
2263  *
2264  * Returns: %true if either or both of next and prev are found,
2265  *	    %false if no vmap_area exists
2266  *
2267  * Find vmap_areas end addresses of which enclose @end.  ie. if not
2268  * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2269  */
2270 static bool pvm_find_next_prev(unsigned long end,
2271 			       struct vmap_area **pnext,
2272 			       struct vmap_area **pprev)
2273 {
2274 	struct rb_node *n = vmap_area_root.rb_node;
2275 	struct vmap_area *va = NULL;
2276 
2277 	while (n) {
2278 		va = rb_entry(n, struct vmap_area, rb_node);
2279 		if (end < va->va_end)
2280 			n = n->rb_left;
2281 		else if (end > va->va_end)
2282 			n = n->rb_right;
2283 		else
2284 			break;
2285 	}
2286 
2287 	if (!va)
2288 		return false;
2289 
2290 	if (va->va_end > end) {
2291 		*pnext = va;
2292 		*pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2293 	} else {
2294 		*pprev = va;
2295 		*pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2296 	}
2297 	return true;
2298 }
2299 
2300 /**
2301  * pvm_determine_end - find the highest aligned address between two vmap_areas
2302  * @pnext: in/out arg for the next vmap_area
2303  * @pprev: in/out arg for the previous vmap_area
2304  * @align: alignment
2305  *
2306  * Returns: determined end address
2307  *
2308  * Find the highest aligned address between *@pnext and *@pprev below
2309  * VMALLOC_END.  *@pnext and *@pprev are adjusted so that the aligned
2310  * down address is between the end addresses of the two vmap_areas.
2311  *
2312  * Please note that the address returned by this function may fall
2313  * inside *@pnext vmap_area.  The caller is responsible for checking
2314  * that.
2315  */
2316 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2317 				       struct vmap_area **pprev,
2318 				       unsigned long align)
2319 {
2320 	const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2321 	unsigned long addr;
2322 
2323 	if (*pnext)
2324 		addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2325 	else
2326 		addr = vmalloc_end;
2327 
2328 	while (*pprev && (*pprev)->va_end > addr) {
2329 		*pnext = *pprev;
2330 		*pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2331 	}
2332 
2333 	return addr;
2334 }
2335 
2336 /**
2337  * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2338  * @offsets: array containing offset of each area
2339  * @sizes: array containing size of each area
2340  * @nr_vms: the number of areas to allocate
2341  * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2342  *
2343  * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2344  *	    vm_structs on success, %NULL on failure
2345  *
2346  * Percpu allocator wants to use congruent vm areas so that it can
2347  * maintain the offsets among percpu areas.  This function allocates
2348  * congruent vmalloc areas for it with GFP_KERNEL.  These areas tend to
2349  * be scattered pretty far, distance between two areas easily going up
2350  * to gigabytes.  To avoid interacting with regular vmallocs, these
2351  * areas are allocated from top.
2352  *
2353  * Despite its complicated look, this allocator is rather simple.  It
2354  * does everything top-down and scans areas from the end looking for
2355  * matching slot.  While scanning, if any of the areas overlaps with
2356  * existing vmap_area, the base address is pulled down to fit the
2357  * area.  Scanning is repeated till all the areas fit and then all
2358  * necessary data structres are inserted and the result is returned.
2359  */
2360 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2361 				     const size_t *sizes, int nr_vms,
2362 				     size_t align)
2363 {
2364 	const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2365 	const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2366 	struct vmap_area **vas, *prev, *next;
2367 	struct vm_struct **vms;
2368 	int area, area2, last_area, term_area;
2369 	unsigned long base, start, end, last_end;
2370 	bool purged = false;
2371 
2372 	/* verify parameters and allocate data structures */
2373 	BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2374 	for (last_area = 0, area = 0; area < nr_vms; area++) {
2375 		start = offsets[area];
2376 		end = start + sizes[area];
2377 
2378 		/* is everything aligned properly? */
2379 		BUG_ON(!IS_ALIGNED(offsets[area], align));
2380 		BUG_ON(!IS_ALIGNED(sizes[area], align));
2381 
2382 		/* detect the area with the highest address */
2383 		if (start > offsets[last_area])
2384 			last_area = area;
2385 
2386 		for (area2 = 0; area2 < nr_vms; area2++) {
2387 			unsigned long start2 = offsets[area2];
2388 			unsigned long end2 = start2 + sizes[area2];
2389 
2390 			if (area2 == area)
2391 				continue;
2392 
2393 			BUG_ON(start2 >= start && start2 < end);
2394 			BUG_ON(end2 <= end && end2 > start);
2395 		}
2396 	}
2397 	last_end = offsets[last_area] + sizes[last_area];
2398 
2399 	if (vmalloc_end - vmalloc_start < last_end) {
2400 		WARN_ON(true);
2401 		return NULL;
2402 	}
2403 
2404 	vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2405 	vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2406 	if (!vas || !vms)
2407 		goto err_free2;
2408 
2409 	for (area = 0; area < nr_vms; area++) {
2410 		vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2411 		vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2412 		if (!vas[area] || !vms[area])
2413 			goto err_free;
2414 	}
2415 retry:
2416 	spin_lock(&vmap_area_lock);
2417 
2418 	/* start scanning - we scan from the top, begin with the last area */
2419 	area = term_area = last_area;
2420 	start = offsets[area];
2421 	end = start + sizes[area];
2422 
2423 	if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2424 		base = vmalloc_end - last_end;
2425 		goto found;
2426 	}
2427 	base = pvm_determine_end(&next, &prev, align) - end;
2428 
2429 	while (true) {
2430 		BUG_ON(next && next->va_end <= base + end);
2431 		BUG_ON(prev && prev->va_end > base + end);
2432 
2433 		/*
2434 		 * base might have underflowed, add last_end before
2435 		 * comparing.
2436 		 */
2437 		if (base + last_end < vmalloc_start + last_end) {
2438 			spin_unlock(&vmap_area_lock);
2439 			if (!purged) {
2440 				purge_vmap_area_lazy();
2441 				purged = true;
2442 				goto retry;
2443 			}
2444 			goto err_free;
2445 		}
2446 
2447 		/*
2448 		 * If next overlaps, move base downwards so that it's
2449 		 * right below next and then recheck.
2450 		 */
2451 		if (next && next->va_start < base + end) {
2452 			base = pvm_determine_end(&next, &prev, align) - end;
2453 			term_area = area;
2454 			continue;
2455 		}
2456 
2457 		/*
2458 		 * If prev overlaps, shift down next and prev and move
2459 		 * base so that it's right below new next and then
2460 		 * recheck.
2461 		 */
2462 		if (prev && prev->va_end > base + start)  {
2463 			next = prev;
2464 			prev = node_to_va(rb_prev(&next->rb_node));
2465 			base = pvm_determine_end(&next, &prev, align) - end;
2466 			term_area = area;
2467 			continue;
2468 		}
2469 
2470 		/*
2471 		 * This area fits, move on to the previous one.  If
2472 		 * the previous one is the terminal one, we're done.
2473 		 */
2474 		area = (area + nr_vms - 1) % nr_vms;
2475 		if (area == term_area)
2476 			break;
2477 		start = offsets[area];
2478 		end = start + sizes[area];
2479 		pvm_find_next_prev(base + end, &next, &prev);
2480 	}
2481 found:
2482 	/* we've found a fitting base, insert all va's */
2483 	for (area = 0; area < nr_vms; area++) {
2484 		struct vmap_area *va = vas[area];
2485 
2486 		va->va_start = base + offsets[area];
2487 		va->va_end = va->va_start + sizes[area];
2488 		__insert_vmap_area(va);
2489 	}
2490 
2491 	vmap_area_pcpu_hole = base + offsets[last_area];
2492 
2493 	spin_unlock(&vmap_area_lock);
2494 
2495 	/* insert all vm's */
2496 	for (area = 0; area < nr_vms; area++)
2497 		setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2498 				 pcpu_get_vm_areas);
2499 
2500 	kfree(vas);
2501 	return vms;
2502 
2503 err_free:
2504 	for (area = 0; area < nr_vms; area++) {
2505 		kfree(vas[area]);
2506 		kfree(vms[area]);
2507 	}
2508 err_free2:
2509 	kfree(vas);
2510 	kfree(vms);
2511 	return NULL;
2512 }
2513 
2514 /**
2515  * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2516  * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2517  * @nr_vms: the number of allocated areas
2518  *
2519  * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2520  */
2521 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2522 {
2523 	int i;
2524 
2525 	for (i = 0; i < nr_vms; i++)
2526 		free_vm_area(vms[i]);
2527 	kfree(vms);
2528 }
2529 #endif	/* CONFIG_SMP */
2530 
2531 #ifdef CONFIG_PROC_FS
2532 static void *s_start(struct seq_file *m, loff_t *pos)
2533 	__acquires(&vmap_area_lock)
2534 {
2535 	loff_t n = *pos;
2536 	struct vmap_area *va;
2537 
2538 	spin_lock(&vmap_area_lock);
2539 	va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2540 	while (n > 0 && &va->list != &vmap_area_list) {
2541 		n--;
2542 		va = list_entry(va->list.next, typeof(*va), list);
2543 	}
2544 	if (!n && &va->list != &vmap_area_list)
2545 		return va;
2546 
2547 	return NULL;
2548 
2549 }
2550 
2551 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2552 {
2553 	struct vmap_area *va = p, *next;
2554 
2555 	++*pos;
2556 	next = list_entry(va->list.next, typeof(*va), list);
2557 	if (&next->list != &vmap_area_list)
2558 		return next;
2559 
2560 	return NULL;
2561 }
2562 
2563 static void s_stop(struct seq_file *m, void *p)
2564 	__releases(&vmap_area_lock)
2565 {
2566 	spin_unlock(&vmap_area_lock);
2567 }
2568 
2569 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2570 {
2571 	if (IS_ENABLED(CONFIG_NUMA)) {
2572 		unsigned int nr, *counters = m->private;
2573 
2574 		if (!counters)
2575 			return;
2576 
2577 		if (v->flags & VM_UNINITIALIZED)
2578 			return;
2579 		/* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2580 		smp_rmb();
2581 
2582 		memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2583 
2584 		for (nr = 0; nr < v->nr_pages; nr++)
2585 			counters[page_to_nid(v->pages[nr])]++;
2586 
2587 		for_each_node_state(nr, N_HIGH_MEMORY)
2588 			if (counters[nr])
2589 				seq_printf(m, " N%u=%u", nr, counters[nr]);
2590 	}
2591 }
2592 
2593 static int s_show(struct seq_file *m, void *p)
2594 {
2595 	struct vmap_area *va = p;
2596 	struct vm_struct *v;
2597 
2598 	/*
2599 	 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2600 	 * behalf of vmap area is being tear down or vm_map_ram allocation.
2601 	 */
2602 	if (!(va->flags & VM_VM_AREA))
2603 		return 0;
2604 
2605 	v = va->vm;
2606 
2607 	seq_printf(m, "0x%pK-0x%pK %7ld",
2608 		v->addr, v->addr + v->size, v->size);
2609 
2610 	if (v->caller)
2611 		seq_printf(m, " %pS", v->caller);
2612 
2613 	if (v->nr_pages)
2614 		seq_printf(m, " pages=%d", v->nr_pages);
2615 
2616 	if (v->phys_addr)
2617 		seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2618 
2619 	if (v->flags & VM_IOREMAP)
2620 		seq_puts(m, " ioremap");
2621 
2622 	if (v->flags & VM_ALLOC)
2623 		seq_puts(m, " vmalloc");
2624 
2625 	if (v->flags & VM_MAP)
2626 		seq_puts(m, " vmap");
2627 
2628 	if (v->flags & VM_USERMAP)
2629 		seq_puts(m, " user");
2630 
2631 	if (v->flags & VM_VPAGES)
2632 		seq_puts(m, " vpages");
2633 
2634 	show_numa_info(m, v);
2635 	seq_putc(m, '\n');
2636 	return 0;
2637 }
2638 
2639 static const struct seq_operations vmalloc_op = {
2640 	.start = s_start,
2641 	.next = s_next,
2642 	.stop = s_stop,
2643 	.show = s_show,
2644 };
2645 
2646 static int vmalloc_open(struct inode *inode, struct file *file)
2647 {
2648 	if (IS_ENABLED(CONFIG_NUMA))
2649 		return seq_open_private(file, &vmalloc_op,
2650 					nr_node_ids * sizeof(unsigned int));
2651 	else
2652 		return seq_open(file, &vmalloc_op);
2653 }
2654 
2655 static const struct file_operations proc_vmalloc_operations = {
2656 	.open		= vmalloc_open,
2657 	.read		= seq_read,
2658 	.llseek		= seq_lseek,
2659 	.release	= seq_release_private,
2660 };
2661 
2662 static int __init proc_vmalloc_init(void)
2663 {
2664 	proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2665 	return 0;
2666 }
2667 module_init(proc_vmalloc_init);
2668 
2669 void get_vmalloc_info(struct vmalloc_info *vmi)
2670 {
2671 	struct vmap_area *va;
2672 	unsigned long free_area_size;
2673 	unsigned long prev_end;
2674 
2675 	vmi->used = 0;
2676 	vmi->largest_chunk = 0;
2677 
2678 	prev_end = VMALLOC_START;
2679 
2680 	rcu_read_lock();
2681 
2682 	if (list_empty(&vmap_area_list)) {
2683 		vmi->largest_chunk = VMALLOC_TOTAL;
2684 		goto out;
2685 	}
2686 
2687 	list_for_each_entry_rcu(va, &vmap_area_list, list) {
2688 		unsigned long addr = va->va_start;
2689 
2690 		/*
2691 		 * Some archs keep another range for modules in vmalloc space
2692 		 */
2693 		if (addr < VMALLOC_START)
2694 			continue;
2695 		if (addr >= VMALLOC_END)
2696 			break;
2697 
2698 		if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2699 			continue;
2700 
2701 		vmi->used += (va->va_end - va->va_start);
2702 
2703 		free_area_size = addr - prev_end;
2704 		if (vmi->largest_chunk < free_area_size)
2705 			vmi->largest_chunk = free_area_size;
2706 
2707 		prev_end = va->va_end;
2708 	}
2709 
2710 	if (VMALLOC_END - prev_end > vmi->largest_chunk)
2711 		vmi->largest_chunk = VMALLOC_END - prev_end;
2712 
2713 out:
2714 	rcu_read_unlock();
2715 }
2716 #endif
2717 
2718