~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/vmalloc.c

Version: ~ [ linux-5.11 ] ~ [ linux-5.10.17 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.99 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.176 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.221 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.257 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.257 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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         if (!(flags & VM_NO_GUARD))
1328                 size += PAGE_SIZE;
1329 
1330         va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1331         if (IS_ERR(va)) {
1332                 kfree(area);
1333                 return NULL;
1334         }
1335 
1336         setup_vmalloc_vm(area, va, flags, caller);
1337 
1338         return area;
1339 }
1340 
1341 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1342                                 unsigned long start, unsigned long end)
1343 {
1344         return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1345                                   GFP_KERNEL, __builtin_return_address(0));
1346 }
1347 EXPORT_SYMBOL_GPL(__get_vm_area);
1348 
1349 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1350                                        unsigned long start, unsigned long end,
1351                                        const void *caller)
1352 {
1353         return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1354                                   GFP_KERNEL, caller);
1355 }
1356 
1357 /**
1358  *      get_vm_area  -  reserve a contiguous kernel virtual area
1359  *      @size:          size of the area
1360  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
1361  *
1362  *      Search an area of @size in the kernel virtual mapping area,
1363  *      and reserved it for out purposes.  Returns the area descriptor
1364  *      on success or %NULL on failure.
1365  */
1366 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1367 {
1368         return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1369                                   NUMA_NO_NODE, GFP_KERNEL,
1370                                   __builtin_return_address(0));
1371 }
1372 
1373 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1374                                 const void *caller)
1375 {
1376         return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1377                                   NUMA_NO_NODE, GFP_KERNEL, caller);
1378 }
1379 
1380 /**
1381  *      find_vm_area  -  find a continuous kernel virtual area
1382  *      @addr:          base address
1383  *
1384  *      Search for the kernel VM area starting at @addr, and return it.
1385  *      It is up to the caller to do all required locking to keep the returned
1386  *      pointer valid.
1387  */
1388 struct vm_struct *find_vm_area(const void *addr)
1389 {
1390         struct vmap_area *va;
1391 
1392         va = find_vmap_area((unsigned long)addr);
1393         if (va && va->flags & VM_VM_AREA)
1394                 return va->vm;
1395 
1396         return NULL;
1397 }
1398 
1399 /**
1400  *      remove_vm_area  -  find and remove a continuous kernel virtual area
1401  *      @addr:          base address
1402  *
1403  *      Search for the kernel VM area starting at @addr, and remove it.
1404  *      This function returns the found VM area, but using it is NOT safe
1405  *      on SMP machines, except for its size or flags.
1406  */
1407 struct vm_struct *remove_vm_area(const void *addr)
1408 {
1409         struct vmap_area *va;
1410 
1411         va = find_vmap_area((unsigned long)addr);
1412         if (va && va->flags & VM_VM_AREA) {
1413                 struct vm_struct *vm = va->vm;
1414 
1415                 spin_lock(&vmap_area_lock);
1416                 va->vm = NULL;
1417                 va->flags &= ~VM_VM_AREA;
1418                 spin_unlock(&vmap_area_lock);
1419 
1420                 vmap_debug_free_range(va->va_start, va->va_end);
1421                 kasan_free_shadow(vm);
1422                 free_unmap_vmap_area(va);
1423                 vm->size -= PAGE_SIZE;
1424 
1425                 return vm;
1426         }
1427         return NULL;
1428 }
1429 
1430 static void __vunmap(const void *addr, int deallocate_pages)
1431 {
1432         struct vm_struct *area;
1433 
1434         if (!addr)
1435                 return;
1436 
1437         if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1438                         addr))
1439                 return;
1440 
1441         area = remove_vm_area(addr);
1442         if (unlikely(!area)) {
1443                 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1444                                 addr);
1445                 return;
1446         }
1447 
1448         debug_check_no_locks_freed(addr, area->size);
1449         debug_check_no_obj_freed(addr, area->size);
1450 
1451         if (deallocate_pages) {
1452                 int i;
1453 
1454                 for (i = 0; i < area->nr_pages; i++) {
1455                         struct page *page = area->pages[i];
1456 
1457                         BUG_ON(!page);
1458                         __free_page(page);
1459                 }
1460 
1461                 if (area->flags & VM_VPAGES)
1462                         vfree(area->pages);
1463                 else
1464                         kfree(area->pages);
1465         }
1466 
1467         kfree(area);
1468         return;
1469 }
1470  
1471 /**
1472  *      vfree  -  release memory allocated by vmalloc()
1473  *      @addr:          memory base address
1474  *
1475  *      Free the virtually continuous memory area starting at @addr, as
1476  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1477  *      NULL, no operation is performed.
1478  *
1479  *      Must not be called in NMI context (strictly speaking, only if we don't
1480  *      have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1481  *      conventions for vfree() arch-depenedent would be a really bad idea)
1482  *
1483  *      NOTE: assumes that the object at *addr has a size >= sizeof(llist_node)
1484  */
1485 void vfree(const void *addr)
1486 {
1487         BUG_ON(in_nmi());
1488 
1489         kmemleak_free(addr);
1490 
1491         if (!addr)
1492                 return;
1493         if (unlikely(in_interrupt())) {
1494                 struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
1495                 if (llist_add((struct llist_node *)addr, &p->list))
1496                         schedule_work(&p->wq);
1497         } else
1498                 __vunmap(addr, 1);
1499 }
1500 EXPORT_SYMBOL(vfree);
1501 
1502 /**
1503  *      vunmap  -  release virtual mapping obtained by vmap()
1504  *      @addr:          memory base address
1505  *
1506  *      Free the virtually contiguous memory area starting at @addr,
1507  *      which was created from the page array passed to vmap().
1508  *
1509  *      Must not be called in interrupt context.
1510  */
1511 void vunmap(const void *addr)
1512 {
1513         BUG_ON(in_interrupt());
1514         might_sleep();
1515         if (addr)
1516                 __vunmap(addr, 0);
1517 }
1518 EXPORT_SYMBOL(vunmap);
1519 
1520 /**
1521  *      vmap  -  map an array of pages into virtually contiguous space
1522  *      @pages:         array of page pointers
1523  *      @count:         number of pages to map
1524  *      @flags:         vm_area->flags
1525  *      @prot:          page protection for the mapping
1526  *
1527  *      Maps @count pages from @pages into contiguous kernel virtual
1528  *      space.
1529  */
1530 void *vmap(struct page **pages, unsigned int count,
1531                 unsigned long flags, pgprot_t prot)
1532 {
1533         struct vm_struct *area;
1534 
1535         might_sleep();
1536 
1537         if (count > totalram_pages)
1538                 return NULL;
1539 
1540         area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1541                                         __builtin_return_address(0));
1542         if (!area)
1543                 return NULL;
1544 
1545         if (map_vm_area(area, prot, pages)) {
1546                 vunmap(area->addr);
1547                 return NULL;
1548         }
1549 
1550         return area->addr;
1551 }
1552 EXPORT_SYMBOL(vmap);
1553 
1554 static void *__vmalloc_node(unsigned long size, unsigned long align,
1555                             gfp_t gfp_mask, pgprot_t prot,
1556                             int node, const void *caller);
1557 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1558                                  pgprot_t prot, int node)
1559 {
1560         const int order = 0;
1561         struct page **pages;
1562         unsigned int nr_pages, array_size, i;
1563         const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1564         const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1565 
1566         nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1567         array_size = (nr_pages * sizeof(struct page *));
1568 
1569         area->nr_pages = nr_pages;
1570         /* Please note that the recursion is strictly bounded. */
1571         if (array_size > PAGE_SIZE) {
1572                 pages = __vmalloc_node(array_size, 1, nested_gfp|__GFP_HIGHMEM,
1573                                 PAGE_KERNEL, node, area->caller);
1574                 area->flags |= VM_VPAGES;
1575         } else {
1576                 pages = kmalloc_node(array_size, nested_gfp, node);
1577         }
1578         area->pages = pages;
1579         if (!area->pages) {
1580                 remove_vm_area(area->addr);
1581                 kfree(area);
1582                 return NULL;
1583         }
1584 
1585         for (i = 0; i < area->nr_pages; i++) {
1586                 struct page *page;
1587 
1588                 if (node == NUMA_NO_NODE)
1589                         page = alloc_page(alloc_mask);
1590                 else
1591                         page = alloc_pages_node(node, alloc_mask, order);
1592 
1593                 if (unlikely(!page)) {
1594                         /* Successfully allocated i pages, free them in __vunmap() */
1595                         area->nr_pages = i;
1596                         goto fail;
1597                 }
1598                 area->pages[i] = page;
1599                 if (gfp_mask & __GFP_WAIT)
1600                         cond_resched();
1601         }
1602 
1603         if (map_vm_area(area, prot, pages))
1604                 goto fail;
1605         return area->addr;
1606 
1607 fail:
1608         warn_alloc_failed(gfp_mask, order,
1609                           "vmalloc: allocation failure, allocated %ld of %ld bytes\n",
1610                           (area->nr_pages*PAGE_SIZE), area->size);
1611         vfree(area->addr);
1612         return NULL;
1613 }
1614 
1615 /**
1616  *      __vmalloc_node_range  -  allocate virtually contiguous memory
1617  *      @size:          allocation size
1618  *      @align:         desired alignment
1619  *      @start:         vm area range start
1620  *      @end:           vm area range end
1621  *      @gfp_mask:      flags for the page level allocator
1622  *      @prot:          protection mask for the allocated pages
1623  *      @vm_flags:      additional vm area flags (e.g. %VM_NO_GUARD)
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, unsigned long vm_flags, int node,
1634                         const void *caller)
1635 {
1636         struct vm_struct *area;
1637         void *addr;
1638         unsigned long real_size = size;
1639 
1640         size = PAGE_ALIGN(size);
1641         if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1642                 goto fail;
1643 
1644         area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1645                                 vm_flags, start, end, node, gfp_mask, caller);
1646         if (!area)
1647                 goto fail;
1648 
1649         addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1650         if (!addr)
1651                 return NULL;
1652 
1653         /*
1654          * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1655          * flag. It means that vm_struct is not fully initialized.
1656          * Now, it is fully initialized, so remove this flag here.
1657          */
1658         clear_vm_uninitialized_flag(area);
1659 
1660         /*
1661          * A ref_count = 2 is needed because vm_struct allocated in
1662          * __get_vm_area_node() contains a reference to the virtual address of
1663          * the vmalloc'ed block.
1664          */
1665         kmemleak_alloc(addr, real_size, 2, gfp_mask);
1666 
1667         return addr;
1668 
1669 fail:
1670         warn_alloc_failed(gfp_mask, 0,
1671                           "vmalloc: allocation failure: %lu bytes\n",
1672                           real_size);
1673         return NULL;
1674 }
1675 
1676 /**
1677  *      __vmalloc_node  -  allocate virtually contiguous memory
1678  *      @size:          allocation size
1679  *      @align:         desired alignment
1680  *      @gfp_mask:      flags for the page level allocator
1681  *      @prot:          protection mask for the allocated pages
1682  *      @node:          node to use for allocation or NUMA_NO_NODE
1683  *      @caller:        caller's return address
1684  *
1685  *      Allocate enough pages to cover @size from the page level
1686  *      allocator with @gfp_mask flags.  Map them into contiguous
1687  *      kernel virtual space, using a pagetable protection of @prot.
1688  */
1689 static void *__vmalloc_node(unsigned long size, unsigned long align,
1690                             gfp_t gfp_mask, pgprot_t prot,
1691                             int node, const void *caller)
1692 {
1693         return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1694                                 gfp_mask, prot, 0, node, caller);
1695 }
1696 
1697 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1698 {
1699         return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1700                                 __builtin_return_address(0));
1701 }
1702 EXPORT_SYMBOL(__vmalloc);
1703 
1704 static inline void *__vmalloc_node_flags(unsigned long size,
1705                                         int node, gfp_t flags)
1706 {
1707         return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1708                                         node, __builtin_return_address(0));
1709 }
1710 
1711 /**
1712  *      vmalloc  -  allocate virtually contiguous memory
1713  *      @size:          allocation size
1714  *      Allocate enough pages to cover @size from the page level
1715  *      allocator and map them into contiguous kernel virtual space.
1716  *
1717  *      For tight control over page level allocator and protection flags
1718  *      use __vmalloc() instead.
1719  */
1720 void *vmalloc(unsigned long size)
1721 {
1722         return __vmalloc_node_flags(size, NUMA_NO_NODE,
1723                                     GFP_KERNEL | __GFP_HIGHMEM);
1724 }
1725 EXPORT_SYMBOL(vmalloc);
1726 
1727 /**
1728  *      vzalloc - allocate virtually contiguous memory with zero fill
1729  *      @size:  allocation size
1730  *      Allocate enough pages to cover @size from the page level
1731  *      allocator and map them into contiguous kernel virtual space.
1732  *      The memory allocated is set to zero.
1733  *
1734  *      For tight control over page level allocator and protection flags
1735  *      use __vmalloc() instead.
1736  */
1737 void *vzalloc(unsigned long size)
1738 {
1739         return __vmalloc_node_flags(size, NUMA_NO_NODE,
1740                                 GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1741 }
1742 EXPORT_SYMBOL(vzalloc);
1743 
1744 /**
1745  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1746  * @size: allocation size
1747  *
1748  * The resulting memory area is zeroed so it can be mapped to userspace
1749  * without leaking data.
1750  */
1751 void *vmalloc_user(unsigned long size)
1752 {
1753         struct vm_struct *area;
1754         void *ret;
1755 
1756         ret = __vmalloc_node(size, SHMLBA,
1757                              GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1758                              PAGE_KERNEL, NUMA_NO_NODE,
1759                              __builtin_return_address(0));
1760         if (ret) {
1761                 area = find_vm_area(ret);
1762                 area->flags |= VM_USERMAP;
1763         }
1764         return ret;
1765 }
1766 EXPORT_SYMBOL(vmalloc_user);
1767 
1768 /**
1769  *      vmalloc_node  -  allocate memory on a specific node
1770  *      @size:          allocation size
1771  *      @node:          numa node
1772  *
1773  *      Allocate enough pages to cover @size from the page level
1774  *      allocator and map them into contiguous kernel virtual space.
1775  *
1776  *      For tight control over page level allocator and protection flags
1777  *      use __vmalloc() instead.
1778  */
1779 void *vmalloc_node(unsigned long size, int node)
1780 {
1781         return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1782                                         node, __builtin_return_address(0));
1783 }
1784 EXPORT_SYMBOL(vmalloc_node);
1785 
1786 /**
1787  * vzalloc_node - allocate memory on a specific node with zero fill
1788  * @size:       allocation size
1789  * @node:       numa node
1790  *
1791  * Allocate enough pages to cover @size from the page level
1792  * allocator and map them into contiguous kernel virtual space.
1793  * The memory allocated is set to zero.
1794  *
1795  * For tight control over page level allocator and protection flags
1796  * use __vmalloc_node() instead.
1797  */
1798 void *vzalloc_node(unsigned long size, int node)
1799 {
1800         return __vmalloc_node_flags(size, node,
1801                          GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO);
1802 }
1803 EXPORT_SYMBOL(vzalloc_node);
1804 
1805 #ifndef PAGE_KERNEL_EXEC
1806 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1807 #endif
1808 
1809 /**
1810  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
1811  *      @size:          allocation size
1812  *
1813  *      Kernel-internal function to allocate enough pages to cover @size
1814  *      the page level allocator and map them into contiguous and
1815  *      executable kernel virtual space.
1816  *
1817  *      For tight control over page level allocator and protection flags
1818  *      use __vmalloc() instead.
1819  */
1820 
1821 void *vmalloc_exec(unsigned long size)
1822 {
1823         return __vmalloc_node(size, 1, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1824                               NUMA_NO_NODE, __builtin_return_address(0));
1825 }
1826 
1827 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1828 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1829 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1830 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1831 #else
1832 #define GFP_VMALLOC32 GFP_KERNEL
1833 #endif
1834 
1835 /**
1836  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
1837  *      @size:          allocation size
1838  *
1839  *      Allocate enough 32bit PA addressable pages to cover @size from the
1840  *      page level allocator and map them into contiguous kernel virtual space.
1841  */
1842 void *vmalloc_32(unsigned long size)
1843 {
1844         return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1845                               NUMA_NO_NODE, __builtin_return_address(0));
1846 }
1847 EXPORT_SYMBOL(vmalloc_32);
1848 
1849 /**
1850  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1851  *      @size:          allocation size
1852  *
1853  * The resulting memory area is 32bit addressable and zeroed so it can be
1854  * mapped to userspace without leaking data.
1855  */
1856 void *vmalloc_32_user(unsigned long size)
1857 {
1858         struct vm_struct *area;
1859         void *ret;
1860 
1861         ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1862                              NUMA_NO_NODE, __builtin_return_address(0));
1863         if (ret) {
1864                 area = find_vm_area(ret);
1865                 area->flags |= VM_USERMAP;
1866         }
1867         return ret;
1868 }
1869 EXPORT_SYMBOL(vmalloc_32_user);
1870 
1871 /*
1872  * small helper routine , copy contents to buf from addr.
1873  * If the page is not present, fill zero.
1874  */
1875 
1876 static int aligned_vread(char *buf, char *addr, unsigned long count)
1877 {
1878         struct page *p;
1879         int copied = 0;
1880 
1881         while (count) {
1882                 unsigned long offset, length;
1883 
1884                 offset = (unsigned long)addr & ~PAGE_MASK;
1885                 length = PAGE_SIZE - offset;
1886                 if (length > count)
1887                         length = count;
1888                 p = vmalloc_to_page(addr);
1889                 /*
1890                  * To do safe access to this _mapped_ area, we need
1891                  * lock. But adding lock here means that we need to add
1892                  * overhead of vmalloc()/vfree() calles for this _debug_
1893                  * interface, rarely used. Instead of that, we'll use
1894                  * kmap() and get small overhead in this access function.
1895                  */
1896                 if (p) {
1897                         /*
1898                          * we can expect USER0 is not used (see vread/vwrite's
1899                          * function description)
1900                          */
1901                         void *map = kmap_atomic(p);
1902                         memcpy(buf, map + offset, length);
1903                         kunmap_atomic(map);
1904                 } else
1905                         memset(buf, 0, length);
1906 
1907                 addr += length;
1908                 buf += length;
1909                 copied += length;
1910                 count -= length;
1911         }
1912         return copied;
1913 }
1914 
1915 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
1916 {
1917         struct page *p;
1918         int copied = 0;
1919 
1920         while (count) {
1921                 unsigned long offset, length;
1922 
1923                 offset = (unsigned long)addr & ~PAGE_MASK;
1924                 length = PAGE_SIZE - offset;
1925                 if (length > count)
1926                         length = count;
1927                 p = vmalloc_to_page(addr);
1928                 /*
1929                  * To do safe access to this _mapped_ area, we need
1930                  * lock. But adding lock here means that we need to add
1931                  * overhead of vmalloc()/vfree() calles for this _debug_
1932                  * interface, rarely used. Instead of that, we'll use
1933                  * kmap() and get small overhead in this access function.
1934                  */
1935                 if (p) {
1936                         /*
1937                          * we can expect USER0 is not used (see vread/vwrite's
1938                          * function description)
1939                          */
1940                         void *map = kmap_atomic(p);
1941                         memcpy(map + offset, buf, length);
1942                         kunmap_atomic(map);
1943                 }
1944                 addr += length;
1945                 buf += length;
1946                 copied += length;
1947                 count -= length;
1948         }
1949         return copied;
1950 }
1951 
1952 /**
1953  *      vread() -  read vmalloc area in a safe way.
1954  *      @buf:           buffer for reading data
1955  *      @addr:          vm address.
1956  *      @count:         number of bytes to be read.
1957  *
1958  *      Returns # of bytes which addr and buf should be increased.
1959  *      (same number to @count). Returns 0 if [addr...addr+count) doesn't
1960  *      includes any intersect with alive vmalloc area.
1961  *
1962  *      This function checks that addr is a valid vmalloc'ed area, and
1963  *      copy data from that area to a given buffer. If the given memory range
1964  *      of [addr...addr+count) includes some valid address, data is copied to
1965  *      proper area of @buf. If there are memory holes, they'll be zero-filled.
1966  *      IOREMAP area is treated as memory hole and no copy is done.
1967  *
1968  *      If [addr...addr+count) doesn't includes any intersects with alive
1969  *      vm_struct area, returns 0. @buf should be kernel's buffer.
1970  *
1971  *      Note: In usual ops, vread() is never necessary because the caller
1972  *      should know vmalloc() area is valid and can use memcpy().
1973  *      This is for routines which have to access vmalloc area without
1974  *      any informaion, as /dev/kmem.
1975  *
1976  */
1977 
1978 long vread(char *buf, char *addr, unsigned long count)
1979 {
1980         struct vmap_area *va;
1981         struct vm_struct *vm;
1982         char *vaddr, *buf_start = buf;
1983         unsigned long buflen = count;
1984         unsigned long n;
1985 
1986         /* Don't allow overflow */
1987         if ((unsigned long) addr + count < count)
1988                 count = -(unsigned long) addr;
1989 
1990         spin_lock(&vmap_area_lock);
1991         list_for_each_entry(va, &vmap_area_list, list) {
1992                 if (!count)
1993                         break;
1994 
1995                 if (!(va->flags & VM_VM_AREA))
1996                         continue;
1997 
1998                 vm = va->vm;
1999                 vaddr = (char *) vm->addr;
2000                 if (addr >= vaddr + get_vm_area_size(vm))
2001                         continue;
2002                 while (addr < vaddr) {
2003                         if (count == 0)
2004                                 goto finished;
2005                         *buf = '\0';
2006                         buf++;
2007                         addr++;
2008                         count--;
2009                 }
2010                 n = vaddr + get_vm_area_size(vm) - addr;
2011                 if (n > count)
2012                         n = count;
2013                 if (!(vm->flags & VM_IOREMAP))
2014                         aligned_vread(buf, addr, n);
2015                 else /* IOREMAP area is treated as memory hole */
2016                         memset(buf, 0, n);
2017                 buf += n;
2018                 addr += n;
2019                 count -= n;
2020         }
2021 finished:
2022         spin_unlock(&vmap_area_lock);
2023 
2024         if (buf == buf_start)
2025                 return 0;
2026         /* zero-fill memory holes */
2027         if (buf != buf_start + buflen)
2028                 memset(buf, 0, buflen - (buf - buf_start));
2029 
2030         return buflen;
2031 }
2032 
2033 /**
2034  *      vwrite() -  write vmalloc area in a safe way.
2035  *      @buf:           buffer for source data
2036  *      @addr:          vm address.
2037  *      @count:         number of bytes to be read.
2038  *
2039  *      Returns # of bytes which addr and buf should be incresed.
2040  *      (same number to @count).
2041  *      If [addr...addr+count) doesn't includes any intersect with valid
2042  *      vmalloc area, returns 0.
2043  *
2044  *      This function checks that addr is a valid vmalloc'ed area, and
2045  *      copy data from a buffer to the given addr. If specified range of
2046  *      [addr...addr+count) includes some valid address, data is copied from
2047  *      proper area of @buf. If there are memory holes, no copy to hole.
2048  *      IOREMAP area is treated as memory hole and no copy is done.
2049  *
2050  *      If [addr...addr+count) doesn't includes any intersects with alive
2051  *      vm_struct area, returns 0. @buf should be kernel's buffer.
2052  *
2053  *      Note: In usual ops, vwrite() is never necessary because the caller
2054  *      should know vmalloc() area is valid and can use memcpy().
2055  *      This is for routines which have to access vmalloc area without
2056  *      any informaion, as /dev/kmem.
2057  */
2058 
2059 long vwrite(char *buf, char *addr, unsigned long count)
2060 {
2061         struct vmap_area *va;
2062         struct vm_struct *vm;
2063         char *vaddr;
2064         unsigned long n, buflen;
2065         int copied = 0;
2066 
2067         /* Don't allow overflow */
2068         if ((unsigned long) addr + count < count)
2069                 count = -(unsigned long) addr;
2070         buflen = count;
2071 
2072         spin_lock(&vmap_area_lock);
2073         list_for_each_entry(va, &vmap_area_list, list) {
2074                 if (!count)
2075                         break;
2076 
2077                 if (!(va->flags & VM_VM_AREA))
2078                         continue;
2079 
2080                 vm = va->vm;
2081                 vaddr = (char *) vm->addr;
2082                 if (addr >= vaddr + get_vm_area_size(vm))
2083                         continue;
2084                 while (addr < vaddr) {
2085                         if (count == 0)
2086                                 goto finished;
2087                         buf++;
2088                         addr++;
2089                         count--;
2090                 }
2091                 n = vaddr + get_vm_area_size(vm) - addr;
2092                 if (n > count)
2093                         n = count;
2094                 if (!(vm->flags & VM_IOREMAP)) {
2095                         aligned_vwrite(buf, addr, n);
2096                         copied++;
2097                 }
2098                 buf += n;
2099                 addr += n;
2100                 count -= n;
2101         }
2102 finished:
2103         spin_unlock(&vmap_area_lock);
2104         if (!copied)
2105                 return 0;
2106         return buflen;
2107 }
2108 
2109 /**
2110  *      remap_vmalloc_range_partial  -  map vmalloc pages to userspace
2111  *      @vma:           vma to cover
2112  *      @uaddr:         target user address to start at
2113  *      @kaddr:         virtual address of vmalloc kernel memory
2114  *      @size:          size of map area
2115  *
2116  *      Returns:        0 for success, -Exxx on failure
2117  *
2118  *      This function checks that @kaddr is a valid vmalloc'ed area,
2119  *      and that it is big enough to cover the range starting at
2120  *      @uaddr in @vma. Will return failure if that criteria isn't
2121  *      met.
2122  *
2123  *      Similar to remap_pfn_range() (see mm/memory.c)
2124  */
2125 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2126                                 void *kaddr, unsigned long size)
2127 {
2128         struct vm_struct *area;
2129 
2130         size = PAGE_ALIGN(size);
2131 
2132         if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2133                 return -EINVAL;
2134 
2135         area = find_vm_area(kaddr);
2136         if (!area)
2137                 return -EINVAL;
2138 
2139         if (!(area->flags & VM_USERMAP))
2140                 return -EINVAL;
2141 
2142         if (kaddr + size > area->addr + area->size)
2143                 return -EINVAL;
2144 
2145         do {
2146                 struct page *page = vmalloc_to_page(kaddr);
2147                 int ret;
2148 
2149                 ret = vm_insert_page(vma, uaddr, page);
2150                 if (ret)
2151                         return ret;
2152 
2153                 uaddr += PAGE_SIZE;
2154                 kaddr += PAGE_SIZE;
2155                 size -= PAGE_SIZE;
2156         } while (size > 0);
2157 
2158         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2159 
2160         return 0;
2161 }
2162 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2163 
2164 /**
2165  *      remap_vmalloc_range  -  map vmalloc pages to userspace
2166  *      @vma:           vma to cover (map full range of vma)
2167  *      @addr:          vmalloc memory
2168  *      @pgoff:         number of pages into addr before first page to map
2169  *
2170  *      Returns:        0 for success, -Exxx on failure
2171  *
2172  *      This function checks that addr is a valid vmalloc'ed area, and
2173  *      that it is big enough to cover the vma. Will return failure if
2174  *      that criteria isn't met.
2175  *
2176  *      Similar to remap_pfn_range() (see mm/memory.c)
2177  */
2178 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2179                                                 unsigned long pgoff)
2180 {
2181         return remap_vmalloc_range_partial(vma, vma->vm_start,
2182                                            addr + (pgoff << PAGE_SHIFT),
2183                                            vma->vm_end - vma->vm_start);
2184 }
2185 EXPORT_SYMBOL(remap_vmalloc_range);
2186 
2187 /*
2188  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2189  * have one.
2190  */
2191 void __weak vmalloc_sync_all(void)
2192 {
2193 }
2194 
2195 
2196 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2197 {
2198         pte_t ***p = data;
2199 
2200         if (p) {
2201                 *(*p) = pte;
2202                 (*p)++;
2203         }
2204         return 0;
2205 }
2206 
2207 /**
2208  *      alloc_vm_area - allocate a range of kernel address space
2209  *      @size:          size of the area
2210  *      @ptes:          returns the PTEs for the address space
2211  *
2212  *      Returns:        NULL on failure, vm_struct on success
2213  *
2214  *      This function reserves a range of kernel address space, and
2215  *      allocates pagetables to map that range.  No actual mappings
2216  *      are created.
2217  *
2218  *      If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2219  *      allocated for the VM area are returned.
2220  */
2221 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2222 {
2223         struct vm_struct *area;
2224 
2225         area = get_vm_area_caller(size, VM_IOREMAP,
2226                                 __builtin_return_address(0));
2227         if (area == NULL)
2228                 return NULL;
2229 
2230         /*
2231          * This ensures that page tables are constructed for this region
2232          * of kernel virtual address space and mapped into init_mm.
2233          */
2234         if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2235                                 size, f, ptes ? &ptes : NULL)) {
2236                 free_vm_area(area);
2237                 return NULL;
2238         }
2239 
2240         return area;
2241 }
2242 EXPORT_SYMBOL_GPL(alloc_vm_area);
2243 
2244 void free_vm_area(struct vm_struct *area)
2245 {
2246         struct vm_struct *ret;
2247         ret = remove_vm_area(area->addr);
2248         BUG_ON(ret != area);
2249         kfree(area);
2250 }
2251 EXPORT_SYMBOL_GPL(free_vm_area);
2252 
2253 #ifdef CONFIG_SMP
2254 static struct vmap_area *node_to_va(struct rb_node *n)
2255 {
2256         return n ? rb_entry(n, struct vmap_area, rb_node) : NULL;
2257 }
2258 
2259 /**
2260  * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2261  * @end: target address
2262  * @pnext: out arg for the next vmap_area
2263  * @pprev: out arg for the previous vmap_area
2264  *
2265  * Returns: %true if either or both of next and prev are found,
2266  *          %false if no vmap_area exists
2267  *
2268  * Find vmap_areas end addresses of which enclose @end.  ie. if not
2269  * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2270  */
2271 static bool pvm_find_next_prev(unsigned long end,
2272                                struct vmap_area **pnext,
2273                                struct vmap_area **pprev)
2274 {
2275         struct rb_node *n = vmap_area_root.rb_node;
2276         struct vmap_area *va = NULL;
2277 
2278         while (n) {
2279                 va = rb_entry(n, struct vmap_area, rb_node);
2280                 if (end < va->va_end)
2281                         n = n->rb_left;
2282                 else if (end > va->va_end)
2283                         n = n->rb_right;
2284                 else
2285                         break;
2286         }
2287 
2288         if (!va)
2289                 return false;
2290 
2291         if (va->va_end > end) {
2292                 *pnext = va;
2293                 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2294         } else {
2295                 *pprev = va;
2296                 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2297         }
2298         return true;
2299 }
2300 
2301 /**
2302  * pvm_determine_end - find the highest aligned address between two vmap_areas
2303  * @pnext: in/out arg for the next vmap_area
2304  * @pprev: in/out arg for the previous vmap_area
2305  * @align: alignment
2306  *
2307  * Returns: determined end address
2308  *
2309  * Find the highest aligned address between *@pnext and *@pprev below
2310  * VMALLOC_END.  *@pnext and *@pprev are adjusted so that the aligned
2311  * down address is between the end addresses of the two vmap_areas.
2312  *
2313  * Please note that the address returned by this function may fall
2314  * inside *@pnext vmap_area.  The caller is responsible for checking
2315  * that.
2316  */
2317 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2318                                        struct vmap_area **pprev,
2319                                        unsigned long align)
2320 {
2321         const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2322         unsigned long addr;
2323 
2324         if (*pnext)
2325                 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2326         else
2327                 addr = vmalloc_end;
2328 
2329         while (*pprev && (*pprev)->va_end > addr) {
2330                 *pnext = *pprev;
2331                 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2332         }
2333 
2334         return addr;
2335 }
2336 
2337 /**
2338  * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2339  * @offsets: array containing offset of each area
2340  * @sizes: array containing size of each area
2341  * @nr_vms: the number of areas to allocate
2342  * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2343  *
2344  * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2345  *          vm_structs on success, %NULL on failure
2346  *
2347  * Percpu allocator wants to use congruent vm areas so that it can
2348  * maintain the offsets among percpu areas.  This function allocates
2349  * congruent vmalloc areas for it with GFP_KERNEL.  These areas tend to
2350  * be scattered pretty far, distance between two areas easily going up
2351  * to gigabytes.  To avoid interacting with regular vmallocs, these
2352  * areas are allocated from top.
2353  *
2354  * Despite its complicated look, this allocator is rather simple.  It
2355  * does everything top-down and scans areas from the end looking for
2356  * matching slot.  While scanning, if any of the areas overlaps with
2357  * existing vmap_area, the base address is pulled down to fit the
2358  * area.  Scanning is repeated till all the areas fit and then all
2359  * necessary data structres are inserted and the result is returned.
2360  */
2361 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2362                                      const size_t *sizes, int nr_vms,
2363                                      size_t align)
2364 {
2365         const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2366         const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2367         struct vmap_area **vas, *prev, *next;
2368         struct vm_struct **vms;
2369         int area, area2, last_area, term_area;
2370         unsigned long base, start, end, last_end;
2371         bool purged = false;
2372 
2373         /* verify parameters and allocate data structures */
2374         BUG_ON(align & ~PAGE_MASK || !is_power_of_2(align));
2375         for (last_area = 0, area = 0; area < nr_vms; area++) {
2376                 start = offsets[area];
2377                 end = start + sizes[area];
2378 
2379                 /* is everything aligned properly? */
2380                 BUG_ON(!IS_ALIGNED(offsets[area], align));
2381                 BUG_ON(!IS_ALIGNED(sizes[area], align));
2382 
2383                 /* detect the area with the highest address */
2384                 if (start > offsets[last_area])
2385                         last_area = area;
2386 
2387                 for (area2 = 0; area2 < nr_vms; area2++) {
2388                         unsigned long start2 = offsets[area2];
2389                         unsigned long end2 = start2 + sizes[area2];
2390 
2391                         if (area2 == area)
2392                                 continue;
2393 
2394                         BUG_ON(start2 >= start && start2 < end);
2395                         BUG_ON(end2 <= end && end2 > start);
2396                 }
2397         }
2398         last_end = offsets[last_area] + sizes[last_area];
2399 
2400         if (vmalloc_end - vmalloc_start < last_end) {
2401                 WARN_ON(true);
2402                 return NULL;
2403         }
2404 
2405         vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2406         vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2407         if (!vas || !vms)
2408                 goto err_free2;
2409 
2410         for (area = 0; area < nr_vms; area++) {
2411                 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2412                 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2413                 if (!vas[area] || !vms[area])
2414                         goto err_free;
2415         }
2416 retry:
2417         spin_lock(&vmap_area_lock);
2418 
2419         /* start scanning - we scan from the top, begin with the last area */
2420         area = term_area = last_area;
2421         start = offsets[area];
2422         end = start + sizes[area];
2423 
2424         if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2425                 base = vmalloc_end - last_end;
2426                 goto found;
2427         }
2428         base = pvm_determine_end(&next, &prev, align) - end;
2429 
2430         while (true) {
2431                 BUG_ON(next && next->va_end <= base + end);
2432                 BUG_ON(prev && prev->va_end > base + end);
2433 
2434                 /*
2435                  * base might have underflowed, add last_end before
2436                  * comparing.
2437                  */
2438                 if (base + last_end < vmalloc_start + last_end) {
2439                         spin_unlock(&vmap_area_lock);
2440                         if (!purged) {
2441                                 purge_vmap_area_lazy();
2442                                 purged = true;
2443                                 goto retry;
2444                         }
2445                         goto err_free;
2446                 }
2447 
2448                 /*
2449                  * If next overlaps, move base downwards so that it's
2450                  * right below next and then recheck.
2451                  */
2452                 if (next && next->va_start < base + end) {
2453                         base = pvm_determine_end(&next, &prev, align) - end;
2454                         term_area = area;
2455                         continue;
2456                 }
2457 
2458                 /*
2459                  * If prev overlaps, shift down next and prev and move
2460                  * base so that it's right below new next and then
2461                  * recheck.
2462                  */
2463                 if (prev && prev->va_end > base + start)  {
2464                         next = prev;
2465                         prev = node_to_va(rb_prev(&next->rb_node));
2466                         base = pvm_determine_end(&next, &prev, align) - end;
2467                         term_area = area;
2468                         continue;
2469                 }
2470 
2471                 /*
2472                  * This area fits, move on to the previous one.  If
2473                  * the previous one is the terminal one, we're done.
2474                  */
2475                 area = (area + nr_vms - 1) % nr_vms;
2476                 if (area == term_area)
2477                         break;
2478                 start = offsets[area];
2479                 end = start + sizes[area];
2480                 pvm_find_next_prev(base + end, &next, &prev);
2481         }
2482 found:
2483         /* we've found a fitting base, insert all va's */
2484         for (area = 0; area < nr_vms; area++) {
2485                 struct vmap_area *va = vas[area];
2486 
2487                 va->va_start = base + offsets[area];
2488                 va->va_end = va->va_start + sizes[area];
2489                 __insert_vmap_area(va);
2490         }
2491 
2492         vmap_area_pcpu_hole = base + offsets[last_area];
2493 
2494         spin_unlock(&vmap_area_lock);
2495 
2496         /* insert all vm's */
2497         for (area = 0; area < nr_vms; area++)
2498                 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2499                                  pcpu_get_vm_areas);
2500 
2501         kfree(vas);
2502         return vms;
2503 
2504 err_free:
2505         for (area = 0; area < nr_vms; area++) {
2506                 kfree(vas[area]);
2507                 kfree(vms[area]);
2508         }
2509 err_free2:
2510         kfree(vas);
2511         kfree(vms);
2512         return NULL;
2513 }
2514 
2515 /**
2516  * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2517  * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2518  * @nr_vms: the number of allocated areas
2519  *
2520  * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2521  */
2522 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2523 {
2524         int i;
2525 
2526         for (i = 0; i < nr_vms; i++)
2527                 free_vm_area(vms[i]);
2528         kfree(vms);
2529 }
2530 #endif  /* CONFIG_SMP */
2531 
2532 #ifdef CONFIG_PROC_FS
2533 static void *s_start(struct seq_file *m, loff_t *pos)
2534         __acquires(&vmap_area_lock)
2535 {
2536         loff_t n = *pos;
2537         struct vmap_area *va;
2538 
2539         spin_lock(&vmap_area_lock);
2540         va = list_entry((&vmap_area_list)->next, typeof(*va), list);
2541         while (n > 0 && &va->list != &vmap_area_list) {
2542                 n--;
2543                 va = list_entry(va->list.next, typeof(*va), list);
2544         }
2545         if (!n && &va->list != &vmap_area_list)
2546                 return va;
2547 
2548         return NULL;
2549 
2550 }
2551 
2552 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2553 {
2554         struct vmap_area *va = p, *next;
2555 
2556         ++*pos;
2557         next = list_entry(va->list.next, typeof(*va), list);
2558         if (&next->list != &vmap_area_list)
2559                 return next;
2560 
2561         return NULL;
2562 }
2563 
2564 static void s_stop(struct seq_file *m, void *p)
2565         __releases(&vmap_area_lock)
2566 {
2567         spin_unlock(&vmap_area_lock);
2568 }
2569 
2570 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2571 {
2572         if (IS_ENABLED(CONFIG_NUMA)) {
2573                 unsigned int nr, *counters = m->private;
2574 
2575                 if (!counters)
2576                         return;
2577 
2578                 if (v->flags & VM_UNINITIALIZED)
2579                         return;
2580                 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2581                 smp_rmb();
2582 
2583                 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2584 
2585                 for (nr = 0; nr < v->nr_pages; nr++)
2586                         counters[page_to_nid(v->pages[nr])]++;
2587 
2588                 for_each_node_state(nr, N_HIGH_MEMORY)
2589                         if (counters[nr])
2590                                 seq_printf(m, " N%u=%u", nr, counters[nr]);
2591         }
2592 }
2593 
2594 static int s_show(struct seq_file *m, void *p)
2595 {
2596         struct vmap_area *va = p;
2597         struct vm_struct *v;
2598 
2599         /*
2600          * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2601          * behalf of vmap area is being tear down or vm_map_ram allocation.
2602          */
2603         if (!(va->flags & VM_VM_AREA))
2604                 return 0;
2605 
2606         v = va->vm;
2607 
2608         seq_printf(m, "0x%pK-0x%pK %7ld",
2609                 v->addr, v->addr + v->size, v->size);
2610 
2611         if (v->caller)
2612                 seq_printf(m, " %pS", v->caller);
2613 
2614         if (v->nr_pages)
2615                 seq_printf(m, " pages=%d", v->nr_pages);
2616 
2617         if (v->phys_addr)
2618                 seq_printf(m, " phys=%llx", (unsigned long long)v->phys_addr);
2619 
2620         if (v->flags & VM_IOREMAP)
2621                 seq_puts(m, " ioremap");
2622 
2623         if (v->flags & VM_ALLOC)
2624                 seq_puts(m, " vmalloc");
2625 
2626         if (v->flags & VM_MAP)
2627                 seq_puts(m, " vmap");
2628 
2629         if (v->flags & VM_USERMAP)
2630                 seq_puts(m, " user");
2631 
2632         if (v->flags & VM_VPAGES)
2633                 seq_puts(m, " vpages");
2634 
2635         show_numa_info(m, v);
2636         seq_putc(m, '\n');
2637         return 0;
2638 }
2639 
2640 static const struct seq_operations vmalloc_op = {
2641         .start = s_start,
2642         .next = s_next,
2643         .stop = s_stop,
2644         .show = s_show,
2645 };
2646 
2647 static int vmalloc_open(struct inode *inode, struct file *file)
2648 {
2649         if (IS_ENABLED(CONFIG_NUMA))
2650                 return seq_open_private(file, &vmalloc_op,
2651                                         nr_node_ids * sizeof(unsigned int));
2652         else
2653                 return seq_open(file, &vmalloc_op);
2654 }
2655 
2656 static const struct file_operations proc_vmalloc_operations = {
2657         .open           = vmalloc_open,
2658         .read           = seq_read,
2659         .llseek         = seq_lseek,
2660         .release        = seq_release_private,
2661 };
2662 
2663 static int __init proc_vmalloc_init(void)
2664 {
2665         proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2666         return 0;
2667 }
2668 module_init(proc_vmalloc_init);
2669 
2670 void get_vmalloc_info(struct vmalloc_info *vmi)
2671 {
2672         struct vmap_area *va;
2673         unsigned long free_area_size;
2674         unsigned long prev_end;
2675 
2676         vmi->used = 0;
2677         vmi->largest_chunk = 0;
2678 
2679         prev_end = VMALLOC_START;
2680 
2681         rcu_read_lock();
2682 
2683         if (list_empty(&vmap_area_list)) {
2684                 vmi->largest_chunk = VMALLOC_TOTAL;
2685                 goto out;
2686         }
2687 
2688         list_for_each_entry_rcu(va, &vmap_area_list, list) {
2689                 unsigned long addr = va->va_start;
2690 
2691                 /*
2692                  * Some archs keep another range for modules in vmalloc space
2693                  */
2694                 if (addr < VMALLOC_START)
2695                         continue;
2696                 if (addr >= VMALLOC_END)
2697                         break;
2698 
2699                 if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
2700                         continue;
2701 
2702                 vmi->used += (va->va_end - va->va_start);
2703 
2704                 free_area_size = addr - prev_end;
2705                 if (vmi->largest_chunk < free_area_size)
2706                         vmi->largest_chunk = free_area_size;
2707 
2708                 prev_end = va->va_end;
2709         }
2710 
2711         if (VMALLOC_END - prev_end > vmi->largest_chunk)
2712                 vmi->largest_chunk = VMALLOC_END - prev_end;
2713 
2714 out:
2715         rcu_read_unlock();
2716 }
2717 #endif
2718 
2719 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp