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

TOMOYO Linux Cross Reference
Linux/mm/vmalloc.c

Version: ~ [ linux-5.6-rc7 ] ~ [ linux-5.5.11 ] ~ [ linux-5.4.27 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.112 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.174 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.217 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.217 ] ~ [ 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-3.16.82 ] ~ [ 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-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ 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/signal.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/notifier.h>
 25 #include <linux/rbtree.h>
 26 #include <linux/radix-tree.h>
 27 #include <linux/rcupdate.h>
 28 #include <linux/pfn.h>
 29 #include <linux/kmemleak.h>
 30 #include <linux/atomic.h>
 31 #include <linux/compiler.h>
 32 #include <linux/llist.h>
 33 #include <linux/bitops.h>
 34 
 35 #include <linux/uaccess.h>
 36 #include <asm/tlbflush.h>
 37 #include <asm/shmparam.h>
 38 
 39 #include "internal.h"
 40 
 41 struct vfree_deferred {
 42         struct llist_head list;
 43         struct work_struct wq;
 44 };
 45 static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
 46 
 47 static void __vunmap(const void *, int);
 48 
 49 static void free_work(struct work_struct *w)
 50 {
 51         struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
 52         struct llist_node *t, *llnode;
 53 
 54         llist_for_each_safe(llnode, t, llist_del_all(&p->list))
 55                 __vunmap((void *)llnode, 1);
 56 }
 57 
 58 /*** Page table manipulation functions ***/
 59 
 60 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
 61 {
 62         pte_t *pte;
 63 
 64         pte = pte_offset_kernel(pmd, addr);
 65         do {
 66                 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
 67                 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
 68         } while (pte++, addr += PAGE_SIZE, addr != end);
 69 }
 70 
 71 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
 72 {
 73         pmd_t *pmd;
 74         unsigned long next;
 75 
 76         pmd = pmd_offset(pud, addr);
 77         do {
 78                 next = pmd_addr_end(addr, end);
 79                 if (pmd_clear_huge(pmd))
 80                         continue;
 81                 if (pmd_none_or_clear_bad(pmd))
 82                         continue;
 83                 vunmap_pte_range(pmd, addr, next);
 84         } while (pmd++, addr = next, addr != end);
 85 }
 86 
 87 static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end)
 88 {
 89         pud_t *pud;
 90         unsigned long next;
 91 
 92         pud = pud_offset(p4d, addr);
 93         do {
 94                 next = pud_addr_end(addr, end);
 95                 if (pud_clear_huge(pud))
 96                         continue;
 97                 if (pud_none_or_clear_bad(pud))
 98                         continue;
 99                 vunmap_pmd_range(pud, addr, next);
100         } while (pud++, addr = next, addr != end);
101 }
102 
103 static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end)
104 {
105         p4d_t *p4d;
106         unsigned long next;
107 
108         p4d = p4d_offset(pgd, addr);
109         do {
110                 next = p4d_addr_end(addr, end);
111                 if (p4d_clear_huge(p4d))
112                         continue;
113                 if (p4d_none_or_clear_bad(p4d))
114                         continue;
115                 vunmap_pud_range(p4d, addr, next);
116         } while (p4d++, addr = next, addr != end);
117 }
118 
119 static void vunmap_page_range(unsigned long addr, unsigned long end)
120 {
121         pgd_t *pgd;
122         unsigned long next;
123 
124         BUG_ON(addr >= end);
125         pgd = pgd_offset_k(addr);
126         do {
127                 next = pgd_addr_end(addr, end);
128                 if (pgd_none_or_clear_bad(pgd))
129                         continue;
130                 vunmap_p4d_range(pgd, addr, next);
131         } while (pgd++, addr = next, addr != end);
132 }
133 
134 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
135                 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
136 {
137         pte_t *pte;
138 
139         /*
140          * nr is a running index into the array which helps higher level
141          * callers keep track of where we're up to.
142          */
143 
144         pte = pte_alloc_kernel(pmd, addr);
145         if (!pte)
146                 return -ENOMEM;
147         do {
148                 struct page *page = pages[*nr];
149 
150                 if (WARN_ON(!pte_none(*pte)))
151                         return -EBUSY;
152                 if (WARN_ON(!page))
153                         return -ENOMEM;
154                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
155                 (*nr)++;
156         } while (pte++, addr += PAGE_SIZE, addr != end);
157         return 0;
158 }
159 
160 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
161                 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
162 {
163         pmd_t *pmd;
164         unsigned long next;
165 
166         pmd = pmd_alloc(&init_mm, pud, addr);
167         if (!pmd)
168                 return -ENOMEM;
169         do {
170                 next = pmd_addr_end(addr, end);
171                 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
172                         return -ENOMEM;
173         } while (pmd++, addr = next, addr != end);
174         return 0;
175 }
176 
177 static int vmap_pud_range(p4d_t *p4d, unsigned long addr,
178                 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
179 {
180         pud_t *pud;
181         unsigned long next;
182 
183         pud = pud_alloc(&init_mm, p4d, addr);
184         if (!pud)
185                 return -ENOMEM;
186         do {
187                 next = pud_addr_end(addr, end);
188                 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
189                         return -ENOMEM;
190         } while (pud++, addr = next, addr != end);
191         return 0;
192 }
193 
194 static int vmap_p4d_range(pgd_t *pgd, unsigned long addr,
195                 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
196 {
197         p4d_t *p4d;
198         unsigned long next;
199 
200         p4d = p4d_alloc(&init_mm, pgd, addr);
201         if (!p4d)
202                 return -ENOMEM;
203         do {
204                 next = p4d_addr_end(addr, end);
205                 if (vmap_pud_range(p4d, addr, next, prot, pages, nr))
206                         return -ENOMEM;
207         } while (p4d++, addr = next, addr != end);
208         return 0;
209 }
210 
211 /*
212  * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
213  * will have pfns corresponding to the "pages" array.
214  *
215  * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
216  */
217 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
218                                    pgprot_t prot, struct page **pages)
219 {
220         pgd_t *pgd;
221         unsigned long next;
222         unsigned long addr = start;
223         int err = 0;
224         int nr = 0;
225 
226         BUG_ON(addr >= end);
227         pgd = pgd_offset_k(addr);
228         do {
229                 next = pgd_addr_end(addr, end);
230                 err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr);
231                 if (err)
232                         return err;
233         } while (pgd++, addr = next, addr != end);
234 
235         return nr;
236 }
237 
238 static int vmap_page_range(unsigned long start, unsigned long end,
239                            pgprot_t prot, struct page **pages)
240 {
241         int ret;
242 
243         ret = vmap_page_range_noflush(start, end, prot, pages);
244         flush_cache_vmap(start, end);
245         return ret;
246 }
247 
248 int is_vmalloc_or_module_addr(const void *x)
249 {
250         /*
251          * ARM, x86-64 and sparc64 put modules in a special place,
252          * and fall back on vmalloc() if that fails. Others
253          * just put it in the vmalloc space.
254          */
255 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
256         unsigned long addr = (unsigned long)x;
257         if (addr >= MODULES_VADDR && addr < MODULES_END)
258                 return 1;
259 #endif
260         return is_vmalloc_addr(x);
261 }
262 
263 /*
264  * Walk a vmap address to the struct page it maps.
265  */
266 struct page *vmalloc_to_page(const void *vmalloc_addr)
267 {
268         unsigned long addr = (unsigned long) vmalloc_addr;
269         struct page *page = NULL;
270         pgd_t *pgd = pgd_offset_k(addr);
271         p4d_t *p4d;
272         pud_t *pud;
273         pmd_t *pmd;
274         pte_t *ptep, pte;
275 
276         /*
277          * XXX we might need to change this if we add VIRTUAL_BUG_ON for
278          * architectures that do not vmalloc module space
279          */
280         VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
281 
282         if (pgd_none(*pgd))
283                 return NULL;
284         p4d = p4d_offset(pgd, addr);
285         if (p4d_none(*p4d))
286                 return NULL;
287         pud = pud_offset(p4d, addr);
288 
289         /*
290          * Don't dereference bad PUD or PMD (below) entries. This will also
291          * identify huge mappings, which we may encounter on architectures
292          * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
293          * identified as vmalloc addresses by is_vmalloc_addr(), but are
294          * not [unambiguously] associated with a struct page, so there is
295          * no correct value to return for them.
296          */
297         WARN_ON_ONCE(pud_bad(*pud));
298         if (pud_none(*pud) || pud_bad(*pud))
299                 return NULL;
300         pmd = pmd_offset(pud, addr);
301         WARN_ON_ONCE(pmd_bad(*pmd));
302         if (pmd_none(*pmd) || pmd_bad(*pmd))
303                 return NULL;
304 
305         ptep = pte_offset_map(pmd, addr);
306         pte = *ptep;
307         if (pte_present(pte))
308                 page = pte_page(pte);
309         pte_unmap(ptep);
310         return page;
311 }
312 EXPORT_SYMBOL(vmalloc_to_page);
313 
314 /*
315  * Map a vmalloc()-space virtual address to the physical page frame number.
316  */
317 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
318 {
319         return page_to_pfn(vmalloc_to_page(vmalloc_addr));
320 }
321 EXPORT_SYMBOL(vmalloc_to_pfn);
322 
323 
324 /*** Global kva allocator ***/
325 
326 #define VM_LAZY_FREE    0x02
327 #define VM_VM_AREA      0x04
328 
329 static DEFINE_SPINLOCK(vmap_area_lock);
330 /* Export for kexec only */
331 LIST_HEAD(vmap_area_list);
332 static LLIST_HEAD(vmap_purge_list);
333 static struct rb_root vmap_area_root = RB_ROOT;
334 
335 /* The vmap cache globals are protected by vmap_area_lock */
336 static struct rb_node *free_vmap_cache;
337 static unsigned long cached_hole_size;
338 static unsigned long cached_vstart;
339 static unsigned long cached_align;
340 
341 static unsigned long vmap_area_pcpu_hole;
342 
343 static struct vmap_area *__find_vmap_area(unsigned long addr)
344 {
345         struct rb_node *n = vmap_area_root.rb_node;
346 
347         while (n) {
348                 struct vmap_area *va;
349 
350                 va = rb_entry(n, struct vmap_area, rb_node);
351                 if (addr < va->va_start)
352                         n = n->rb_left;
353                 else if (addr >= va->va_end)
354                         n = n->rb_right;
355                 else
356                         return va;
357         }
358 
359         return NULL;
360 }
361 
362 static void __insert_vmap_area(struct vmap_area *va)
363 {
364         struct rb_node **p = &vmap_area_root.rb_node;
365         struct rb_node *parent = NULL;
366         struct rb_node *tmp;
367 
368         while (*p) {
369                 struct vmap_area *tmp_va;
370 
371                 parent = *p;
372                 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
373                 if (va->va_start < tmp_va->va_end)
374                         p = &(*p)->rb_left;
375                 else if (va->va_end > tmp_va->va_start)
376                         p = &(*p)->rb_right;
377                 else
378                         BUG();
379         }
380 
381         rb_link_node(&va->rb_node, parent, p);
382         rb_insert_color(&va->rb_node, &vmap_area_root);
383 
384         /* address-sort this list */
385         tmp = rb_prev(&va->rb_node);
386         if (tmp) {
387                 struct vmap_area *prev;
388                 prev = rb_entry(tmp, struct vmap_area, rb_node);
389                 list_add_rcu(&va->list, &prev->list);
390         } else
391                 list_add_rcu(&va->list, &vmap_area_list);
392 }
393 
394 static void purge_vmap_area_lazy(void);
395 
396 static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
397 
398 /*
399  * Allocate a region of KVA of the specified size and alignment, within the
400  * vstart and vend.
401  */
402 static struct vmap_area *alloc_vmap_area(unsigned long size,
403                                 unsigned long align,
404                                 unsigned long vstart, unsigned long vend,
405                                 int node, gfp_t gfp_mask)
406 {
407         struct vmap_area *va;
408         struct rb_node *n;
409         unsigned long addr;
410         int purged = 0;
411         struct vmap_area *first;
412 
413         BUG_ON(!size);
414         BUG_ON(offset_in_page(size));
415         BUG_ON(!is_power_of_2(align));
416 
417         might_sleep();
418 
419         va = kmalloc_node(sizeof(struct vmap_area),
420                         gfp_mask & GFP_RECLAIM_MASK, node);
421         if (unlikely(!va))
422                 return ERR_PTR(-ENOMEM);
423 
424         /*
425          * Only scan the relevant parts containing pointers to other objects
426          * to avoid false negatives.
427          */
428         kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
429 
430 retry:
431         spin_lock(&vmap_area_lock);
432         /*
433          * Invalidate cache if we have more permissive parameters.
434          * cached_hole_size notes the largest hole noticed _below_
435          * the vmap_area cached in free_vmap_cache: if size fits
436          * into that hole, we want to scan from vstart to reuse
437          * the hole instead of allocating above free_vmap_cache.
438          * Note that __free_vmap_area may update free_vmap_cache
439          * without updating cached_hole_size or cached_align.
440          */
441         if (!free_vmap_cache ||
442                         size < cached_hole_size ||
443                         vstart < cached_vstart ||
444                         align < cached_align) {
445 nocache:
446                 cached_hole_size = 0;
447                 free_vmap_cache = NULL;
448         }
449         /* record if we encounter less permissive parameters */
450         cached_vstart = vstart;
451         cached_align = align;
452 
453         /* find starting point for our search */
454         if (free_vmap_cache) {
455                 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
456                 addr = ALIGN(first->va_end, align);
457                 if (addr < vstart)
458                         goto nocache;
459                 if (addr + size < addr)
460                         goto overflow;
461 
462         } else {
463                 addr = ALIGN(vstart, align);
464                 if (addr + size < addr)
465                         goto overflow;
466 
467                 n = vmap_area_root.rb_node;
468                 first = NULL;
469 
470                 while (n) {
471                         struct vmap_area *tmp;
472                         tmp = rb_entry(n, struct vmap_area, rb_node);
473                         if (tmp->va_end >= addr) {
474                                 first = tmp;
475                                 if (tmp->va_start <= addr)
476                                         break;
477                                 n = n->rb_left;
478                         } else
479                                 n = n->rb_right;
480                 }
481 
482                 if (!first)
483                         goto found;
484         }
485 
486         /* from the starting point, walk areas until a suitable hole is found */
487         while (addr + size > first->va_start && addr + size <= vend) {
488                 if (addr + cached_hole_size < first->va_start)
489                         cached_hole_size = first->va_start - addr;
490                 addr = ALIGN(first->va_end, align);
491                 if (addr + size < addr)
492                         goto overflow;
493 
494                 if (list_is_last(&first->list, &vmap_area_list))
495                         goto found;
496 
497                 first = list_next_entry(first, list);
498         }
499 
500 found:
501         /*
502          * Check also calculated address against the vstart,
503          * because it can be 0 because of big align request.
504          */
505         if (addr + size > vend || addr < vstart)
506                 goto overflow;
507 
508         va->va_start = addr;
509         va->va_end = addr + size;
510         va->flags = 0;
511         __insert_vmap_area(va);
512         free_vmap_cache = &va->rb_node;
513         spin_unlock(&vmap_area_lock);
514 
515         BUG_ON(!IS_ALIGNED(va->va_start, align));
516         BUG_ON(va->va_start < vstart);
517         BUG_ON(va->va_end > vend);
518 
519         return va;
520 
521 overflow:
522         spin_unlock(&vmap_area_lock);
523         if (!purged) {
524                 purge_vmap_area_lazy();
525                 purged = 1;
526                 goto retry;
527         }
528 
529         if (gfpflags_allow_blocking(gfp_mask)) {
530                 unsigned long freed = 0;
531                 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
532                 if (freed > 0) {
533                         purged = 0;
534                         goto retry;
535                 }
536         }
537 
538         if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
539                 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
540                         size);
541         kfree(va);
542         return ERR_PTR(-EBUSY);
543 }
544 
545 int register_vmap_purge_notifier(struct notifier_block *nb)
546 {
547         return blocking_notifier_chain_register(&vmap_notify_list, nb);
548 }
549 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
550 
551 int unregister_vmap_purge_notifier(struct notifier_block *nb)
552 {
553         return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
554 }
555 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
556 
557 static void __free_vmap_area(struct vmap_area *va)
558 {
559         BUG_ON(RB_EMPTY_NODE(&va->rb_node));
560 
561         if (free_vmap_cache) {
562                 if (va->va_end < cached_vstart) {
563                         free_vmap_cache = NULL;
564                 } else {
565                         struct vmap_area *cache;
566                         cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
567                         if (va->va_start <= cache->va_start) {
568                                 free_vmap_cache = rb_prev(&va->rb_node);
569                                 /*
570                                  * We don't try to update cached_hole_size or
571                                  * cached_align, but it won't go very wrong.
572                                  */
573                         }
574                 }
575         }
576         rb_erase(&va->rb_node, &vmap_area_root);
577         RB_CLEAR_NODE(&va->rb_node);
578         list_del_rcu(&va->list);
579 
580         /*
581          * Track the highest possible candidate for pcpu area
582          * allocation.  Areas outside of vmalloc area can be returned
583          * here too, consider only end addresses which fall inside
584          * vmalloc area proper.
585          */
586         if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
587                 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
588 
589         kfree_rcu(va, rcu_head);
590 }
591 
592 /*
593  * Free a region of KVA allocated by alloc_vmap_area
594  */
595 static void free_vmap_area(struct vmap_area *va)
596 {
597         spin_lock(&vmap_area_lock);
598         __free_vmap_area(va);
599         spin_unlock(&vmap_area_lock);
600 }
601 
602 /*
603  * Clear the pagetable entries of a given vmap_area
604  */
605 static void unmap_vmap_area(struct vmap_area *va)
606 {
607         vunmap_page_range(va->va_start, va->va_end);
608 }
609 
610 /*
611  * lazy_max_pages is the maximum amount of virtual address space we gather up
612  * before attempting to purge with a TLB flush.
613  *
614  * There is a tradeoff here: a larger number will cover more kernel page tables
615  * and take slightly longer to purge, but it will linearly reduce the number of
616  * global TLB flushes that must be performed. It would seem natural to scale
617  * this number up linearly with the number of CPUs (because vmapping activity
618  * could also scale linearly with the number of CPUs), however it is likely
619  * that in practice, workloads might be constrained in other ways that mean
620  * vmap activity will not scale linearly with CPUs. Also, I want to be
621  * conservative and not introduce a big latency on huge systems, so go with
622  * a less aggressive log scale. It will still be an improvement over the old
623  * code, and it will be simple to change the scale factor if we find that it
624  * becomes a problem on bigger systems.
625  */
626 static unsigned long lazy_max_pages(void)
627 {
628         unsigned int log;
629 
630         log = fls(num_online_cpus());
631 
632         return log * (32UL * 1024 * 1024 / PAGE_SIZE);
633 }
634 
635 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
636 
637 /*
638  * Serialize vmap purging.  There is no actual criticial section protected
639  * by this look, but we want to avoid concurrent calls for performance
640  * reasons and to make the pcpu_get_vm_areas more deterministic.
641  */
642 static DEFINE_MUTEX(vmap_purge_lock);
643 
644 /* for per-CPU blocks */
645 static void purge_fragmented_blocks_allcpus(void);
646 
647 /*
648  * called before a call to iounmap() if the caller wants vm_area_struct's
649  * immediately freed.
650  */
651 void set_iounmap_nonlazy(void)
652 {
653         atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
654 }
655 
656 /*
657  * Purges all lazily-freed vmap areas.
658  */
659 static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
660 {
661         struct llist_node *valist;
662         struct vmap_area *va;
663         struct vmap_area *n_va;
664         bool do_free = false;
665 
666         lockdep_assert_held(&vmap_purge_lock);
667 
668         valist = llist_del_all(&vmap_purge_list);
669         llist_for_each_entry(va, valist, purge_list) {
670                 if (va->va_start < start)
671                         start = va->va_start;
672                 if (va->va_end > end)
673                         end = va->va_end;
674                 do_free = true;
675         }
676 
677         if (!do_free)
678                 return false;
679 
680         flush_tlb_kernel_range(start, end);
681 
682         spin_lock(&vmap_area_lock);
683         llist_for_each_entry_safe(va, n_va, valist, purge_list) {
684                 int nr = (va->va_end - va->va_start) >> PAGE_SHIFT;
685 
686                 __free_vmap_area(va);
687                 atomic_sub(nr, &vmap_lazy_nr);
688                 cond_resched_lock(&vmap_area_lock);
689         }
690         spin_unlock(&vmap_area_lock);
691         return true;
692 }
693 
694 /*
695  * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
696  * is already purging.
697  */
698 static void try_purge_vmap_area_lazy(void)
699 {
700         if (mutex_trylock(&vmap_purge_lock)) {
701                 __purge_vmap_area_lazy(ULONG_MAX, 0);
702                 mutex_unlock(&vmap_purge_lock);
703         }
704 }
705 
706 /*
707  * Kick off a purge of the outstanding lazy areas.
708  */
709 static void purge_vmap_area_lazy(void)
710 {
711         mutex_lock(&vmap_purge_lock);
712         purge_fragmented_blocks_allcpus();
713         __purge_vmap_area_lazy(ULONG_MAX, 0);
714         mutex_unlock(&vmap_purge_lock);
715 }
716 
717 /*
718  * Free a vmap area, caller ensuring that the area has been unmapped
719  * and flush_cache_vunmap had been called for the correct range
720  * previously.
721  */
722 static void free_vmap_area_noflush(struct vmap_area *va)
723 {
724         int nr_lazy;
725 
726         nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
727                                     &vmap_lazy_nr);
728 
729         /* After this point, we may free va at any time */
730         llist_add(&va->purge_list, &vmap_purge_list);
731 
732         if (unlikely(nr_lazy > lazy_max_pages()))
733                 try_purge_vmap_area_lazy();
734 }
735 
736 /*
737  * Free and unmap a vmap area
738  */
739 static void free_unmap_vmap_area(struct vmap_area *va)
740 {
741         flush_cache_vunmap(va->va_start, va->va_end);
742         unmap_vmap_area(va);
743         if (debug_pagealloc_enabled())
744                 flush_tlb_kernel_range(va->va_start, va->va_end);
745 
746         free_vmap_area_noflush(va);
747 }
748 
749 static struct vmap_area *find_vmap_area(unsigned long addr)
750 {
751         struct vmap_area *va;
752 
753         spin_lock(&vmap_area_lock);
754         va = __find_vmap_area(addr);
755         spin_unlock(&vmap_area_lock);
756 
757         return va;
758 }
759 
760 /*** Per cpu kva allocator ***/
761 
762 /*
763  * vmap space is limited especially on 32 bit architectures. Ensure there is
764  * room for at least 16 percpu vmap blocks per CPU.
765  */
766 /*
767  * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
768  * to #define VMALLOC_SPACE             (VMALLOC_END-VMALLOC_START). Guess
769  * instead (we just need a rough idea)
770  */
771 #if BITS_PER_LONG == 32
772 #define VMALLOC_SPACE           (128UL*1024*1024)
773 #else
774 #define VMALLOC_SPACE           (128UL*1024*1024*1024)
775 #endif
776 
777 #define VMALLOC_PAGES           (VMALLOC_SPACE / PAGE_SIZE)
778 #define VMAP_MAX_ALLOC          BITS_PER_LONG   /* 256K with 4K pages */
779 #define VMAP_BBMAP_BITS_MAX     1024    /* 4MB with 4K pages */
780 #define VMAP_BBMAP_BITS_MIN     (VMAP_MAX_ALLOC*2)
781 #define VMAP_MIN(x, y)          ((x) < (y) ? (x) : (y)) /* can't use min() */
782 #define VMAP_MAX(x, y)          ((x) > (y) ? (x) : (y)) /* can't use max() */
783 #define VMAP_BBMAP_BITS         \
784                 VMAP_MIN(VMAP_BBMAP_BITS_MAX,   \
785                 VMAP_MAX(VMAP_BBMAP_BITS_MIN,   \
786                         VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
787 
788 #define VMAP_BLOCK_SIZE         (VMAP_BBMAP_BITS * PAGE_SIZE)
789 
790 static bool vmap_initialized __read_mostly = false;
791 
792 struct vmap_block_queue {
793         spinlock_t lock;
794         struct list_head free;
795 };
796 
797 struct vmap_block {
798         spinlock_t lock;
799         struct vmap_area *va;
800         unsigned long free, dirty;
801         unsigned long dirty_min, dirty_max; /*< dirty range */
802         struct list_head free_list;
803         struct rcu_head rcu_head;
804         struct list_head purge;
805 };
806 
807 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
808 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
809 
810 /*
811  * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
812  * in the free path. Could get rid of this if we change the API to return a
813  * "cookie" from alloc, to be passed to free. But no big deal yet.
814  */
815 static DEFINE_SPINLOCK(vmap_block_tree_lock);
816 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
817 
818 /*
819  * We should probably have a fallback mechanism to allocate virtual memory
820  * out of partially filled vmap blocks. However vmap block sizing should be
821  * fairly reasonable according to the vmalloc size, so it shouldn't be a
822  * big problem.
823  */
824 
825 static unsigned long addr_to_vb_idx(unsigned long addr)
826 {
827         addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
828         addr /= VMAP_BLOCK_SIZE;
829         return addr;
830 }
831 
832 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
833 {
834         unsigned long addr;
835 
836         addr = va_start + (pages_off << PAGE_SHIFT);
837         BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
838         return (void *)addr;
839 }
840 
841 /**
842  * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
843  *                  block. Of course pages number can't exceed VMAP_BBMAP_BITS
844  * @order:    how many 2^order pages should be occupied in newly allocated block
845  * @gfp_mask: flags for the page level allocator
846  *
847  * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
848  */
849 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
850 {
851         struct vmap_block_queue *vbq;
852         struct vmap_block *vb;
853         struct vmap_area *va;
854         unsigned long vb_idx;
855         int node, err;
856         void *vaddr;
857 
858         node = numa_node_id();
859 
860         vb = kmalloc_node(sizeof(struct vmap_block),
861                         gfp_mask & GFP_RECLAIM_MASK, node);
862         if (unlikely(!vb))
863                 return ERR_PTR(-ENOMEM);
864 
865         va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
866                                         VMALLOC_START, VMALLOC_END,
867                                         node, gfp_mask);
868         if (IS_ERR(va)) {
869                 kfree(vb);
870                 return ERR_CAST(va);
871         }
872 
873         err = radix_tree_preload(gfp_mask);
874         if (unlikely(err)) {
875                 kfree(vb);
876                 free_vmap_area(va);
877                 return ERR_PTR(err);
878         }
879 
880         vaddr = vmap_block_vaddr(va->va_start, 0);
881         spin_lock_init(&vb->lock);
882         vb->va = va;
883         /* At least something should be left free */
884         BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
885         vb->free = VMAP_BBMAP_BITS - (1UL << order);
886         vb->dirty = 0;
887         vb->dirty_min = VMAP_BBMAP_BITS;
888         vb->dirty_max = 0;
889         INIT_LIST_HEAD(&vb->free_list);
890 
891         vb_idx = addr_to_vb_idx(va->va_start);
892         spin_lock(&vmap_block_tree_lock);
893         err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
894         spin_unlock(&vmap_block_tree_lock);
895         BUG_ON(err);
896         radix_tree_preload_end();
897 
898         vbq = &get_cpu_var(vmap_block_queue);
899         spin_lock(&vbq->lock);
900         list_add_tail_rcu(&vb->free_list, &vbq->free);
901         spin_unlock(&vbq->lock);
902         put_cpu_var(vmap_block_queue);
903 
904         return vaddr;
905 }
906 
907 static void free_vmap_block(struct vmap_block *vb)
908 {
909         struct vmap_block *tmp;
910         unsigned long vb_idx;
911 
912         vb_idx = addr_to_vb_idx(vb->va->va_start);
913         spin_lock(&vmap_block_tree_lock);
914         tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
915         spin_unlock(&vmap_block_tree_lock);
916         BUG_ON(tmp != vb);
917 
918         free_vmap_area_noflush(vb->va);
919         kfree_rcu(vb, rcu_head);
920 }
921 
922 static void purge_fragmented_blocks(int cpu)
923 {
924         LIST_HEAD(purge);
925         struct vmap_block *vb;
926         struct vmap_block *n_vb;
927         struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
928 
929         rcu_read_lock();
930         list_for_each_entry_rcu(vb, &vbq->free, free_list) {
931 
932                 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
933                         continue;
934 
935                 spin_lock(&vb->lock);
936                 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
937                         vb->free = 0; /* prevent further allocs after releasing lock */
938                         vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
939                         vb->dirty_min = 0;
940                         vb->dirty_max = VMAP_BBMAP_BITS;
941                         spin_lock(&vbq->lock);
942                         list_del_rcu(&vb->free_list);
943                         spin_unlock(&vbq->lock);
944                         spin_unlock(&vb->lock);
945                         list_add_tail(&vb->purge, &purge);
946                 } else
947                         spin_unlock(&vb->lock);
948         }
949         rcu_read_unlock();
950 
951         list_for_each_entry_safe(vb, n_vb, &purge, purge) {
952                 list_del(&vb->purge);
953                 free_vmap_block(vb);
954         }
955 }
956 
957 static void purge_fragmented_blocks_allcpus(void)
958 {
959         int cpu;
960 
961         for_each_possible_cpu(cpu)
962                 purge_fragmented_blocks(cpu);
963 }
964 
965 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
966 {
967         struct vmap_block_queue *vbq;
968         struct vmap_block *vb;
969         void *vaddr = NULL;
970         unsigned int order;
971 
972         BUG_ON(offset_in_page(size));
973         BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
974         if (WARN_ON(size == 0)) {
975                 /*
976                  * Allocating 0 bytes isn't what caller wants since
977                  * get_order(0) returns funny result. Just warn and terminate
978                  * early.
979                  */
980                 return NULL;
981         }
982         order = get_order(size);
983 
984         rcu_read_lock();
985         vbq = &get_cpu_var(vmap_block_queue);
986         list_for_each_entry_rcu(vb, &vbq->free, free_list) {
987                 unsigned long pages_off;
988 
989                 spin_lock(&vb->lock);
990                 if (vb->free < (1UL << order)) {
991                         spin_unlock(&vb->lock);
992                         continue;
993                 }
994 
995                 pages_off = VMAP_BBMAP_BITS - vb->free;
996                 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
997                 vb->free -= 1UL << order;
998                 if (vb->free == 0) {
999                         spin_lock(&vbq->lock);
1000                         list_del_rcu(&vb->free_list);
1001                         spin_unlock(&vbq->lock);
1002                 }
1003 
1004                 spin_unlock(&vb->lock);
1005                 break;
1006         }
1007 
1008         put_cpu_var(vmap_block_queue);
1009         rcu_read_unlock();
1010 
1011         /* Allocate new block if nothing was found */
1012         if (!vaddr)
1013                 vaddr = new_vmap_block(order, gfp_mask);
1014 
1015         return vaddr;
1016 }
1017 
1018 static void vb_free(const void *addr, unsigned long size)
1019 {
1020         unsigned long offset;
1021         unsigned long vb_idx;
1022         unsigned int order;
1023         struct vmap_block *vb;
1024 
1025         BUG_ON(offset_in_page(size));
1026         BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
1027 
1028         flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1029 
1030         order = get_order(size);
1031 
1032         offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
1033         offset >>= PAGE_SHIFT;
1034 
1035         vb_idx = addr_to_vb_idx((unsigned long)addr);
1036         rcu_read_lock();
1037         vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1038         rcu_read_unlock();
1039         BUG_ON(!vb);
1040 
1041         vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1042 
1043         if (debug_pagealloc_enabled())
1044                 flush_tlb_kernel_range((unsigned long)addr,
1045                                         (unsigned long)addr + size);
1046 
1047         spin_lock(&vb->lock);
1048 
1049         /* Expand dirty range */
1050         vb->dirty_min = min(vb->dirty_min, offset);
1051         vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
1052 
1053         vb->dirty += 1UL << order;
1054         if (vb->dirty == VMAP_BBMAP_BITS) {
1055                 BUG_ON(vb->free);
1056                 spin_unlock(&vb->lock);
1057                 free_vmap_block(vb);
1058         } else
1059                 spin_unlock(&vb->lock);
1060 }
1061 
1062 /**
1063  * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1064  *
1065  * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1066  * to amortize TLB flushing overheads. What this means is that any page you
1067  * have now, may, in a former life, have been mapped into kernel virtual
1068  * address by the vmap layer and so there might be some CPUs with TLB entries
1069  * still referencing that page (additional to the regular 1:1 kernel mapping).
1070  *
1071  * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1072  * be sure that none of the pages we have control over will have any aliases
1073  * from the vmap layer.
1074  */
1075 void vm_unmap_aliases(void)
1076 {
1077         unsigned long start = ULONG_MAX, end = 0;
1078         int cpu;
1079         int flush = 0;
1080 
1081         if (unlikely(!vmap_initialized))
1082                 return;
1083 
1084         might_sleep();
1085 
1086         for_each_possible_cpu(cpu) {
1087                 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1088                 struct vmap_block *vb;
1089 
1090                 rcu_read_lock();
1091                 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1092                         spin_lock(&vb->lock);
1093                         if (vb->dirty) {
1094                                 unsigned long va_start = vb->va->va_start;
1095                                 unsigned long s, e;
1096 
1097                                 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1098                                 e = va_start + (vb->dirty_max << PAGE_SHIFT);
1099 
1100                                 start = min(s, start);
1101                                 end   = max(e, end);
1102 
1103                                 flush = 1;
1104                         }
1105                         spin_unlock(&vb->lock);
1106                 }
1107                 rcu_read_unlock();
1108         }
1109 
1110         mutex_lock(&vmap_purge_lock);
1111         purge_fragmented_blocks_allcpus();
1112         if (!__purge_vmap_area_lazy(start, end) && flush)
1113                 flush_tlb_kernel_range(start, end);
1114         mutex_unlock(&vmap_purge_lock);
1115 }
1116 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1117 
1118 /**
1119  * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1120  * @mem: the pointer returned by vm_map_ram
1121  * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1122  */
1123 void vm_unmap_ram(const void *mem, unsigned int count)
1124 {
1125         unsigned long size = (unsigned long)count << PAGE_SHIFT;
1126         unsigned long addr = (unsigned long)mem;
1127         struct vmap_area *va;
1128 
1129         might_sleep();
1130         BUG_ON(!addr);
1131         BUG_ON(addr < VMALLOC_START);
1132         BUG_ON(addr > VMALLOC_END);
1133         BUG_ON(!PAGE_ALIGNED(addr));
1134 
1135         if (likely(count <= VMAP_MAX_ALLOC)) {
1136                 debug_check_no_locks_freed(mem, size);
1137                 vb_free(mem, size);
1138                 return;
1139         }
1140 
1141         va = find_vmap_area(addr);
1142         BUG_ON(!va);
1143         debug_check_no_locks_freed((void *)va->va_start,
1144                                     (va->va_end - va->va_start));
1145         free_unmap_vmap_area(va);
1146 }
1147 EXPORT_SYMBOL(vm_unmap_ram);
1148 
1149 /**
1150  * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1151  * @pages: an array of pointers to the pages to be mapped
1152  * @count: number of pages
1153  * @node: prefer to allocate data structures on this node
1154  * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1155  *
1156  * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1157  * faster than vmap so it's good.  But if you mix long-life and short-life
1158  * objects with vm_map_ram(), it could consume lots of address space through
1159  * fragmentation (especially on a 32bit machine).  You could see failures in
1160  * the end.  Please use this function for short-lived objects.
1161  *
1162  * Returns: a pointer to the address that has been mapped, or %NULL on failure
1163  */
1164 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1165 {
1166         unsigned long size = (unsigned long)count << PAGE_SHIFT;
1167         unsigned long addr;
1168         void *mem;
1169 
1170         if (likely(count <= VMAP_MAX_ALLOC)) {
1171                 mem = vb_alloc(size, GFP_KERNEL);
1172                 if (IS_ERR(mem))
1173                         return NULL;
1174                 addr = (unsigned long)mem;
1175         } else {
1176                 struct vmap_area *va;
1177                 va = alloc_vmap_area(size, PAGE_SIZE,
1178                                 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1179                 if (IS_ERR(va))
1180                         return NULL;
1181 
1182                 addr = va->va_start;
1183                 mem = (void *)addr;
1184         }
1185         if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1186                 vm_unmap_ram(mem, count);
1187                 return NULL;
1188         }
1189         return mem;
1190 }
1191 EXPORT_SYMBOL(vm_map_ram);
1192 
1193 static struct vm_struct *vmlist __initdata;
1194 /**
1195  * vm_area_add_early - add vmap area early during boot
1196  * @vm: vm_struct to add
1197  *
1198  * This function is used to add fixed kernel vm area to vmlist before
1199  * vmalloc_init() is called.  @vm->addr, @vm->size, and @vm->flags
1200  * should contain proper values and the other fields should be zero.
1201  *
1202  * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1203  */
1204 void __init vm_area_add_early(struct vm_struct *vm)
1205 {
1206         struct vm_struct *tmp, **p;
1207 
1208         BUG_ON(vmap_initialized);
1209         for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1210                 if (tmp->addr >= vm->addr) {
1211                         BUG_ON(tmp->addr < vm->addr + vm->size);
1212                         break;
1213                 } else
1214                         BUG_ON(tmp->addr + tmp->size > vm->addr);
1215         }
1216         vm->next = *p;
1217         *p = vm;
1218 }
1219 
1220 /**
1221  * vm_area_register_early - register vmap area early during boot
1222  * @vm: vm_struct to register
1223  * @align: requested alignment
1224  *
1225  * This function is used to register kernel vm area before
1226  * vmalloc_init() is called.  @vm->size and @vm->flags should contain
1227  * proper values on entry and other fields should be zero.  On return,
1228  * vm->addr contains the allocated address.
1229  *
1230  * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1231  */
1232 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1233 {
1234         static size_t vm_init_off __initdata;
1235         unsigned long addr;
1236 
1237         addr = ALIGN(VMALLOC_START + vm_init_off, align);
1238         vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1239 
1240         vm->addr = (void *)addr;
1241 
1242         vm_area_add_early(vm);
1243 }
1244 
1245 void __init vmalloc_init(void)
1246 {
1247         struct vmap_area *va;
1248         struct vm_struct *tmp;
1249         int i;
1250 
1251         for_each_possible_cpu(i) {
1252                 struct vmap_block_queue *vbq;
1253                 struct vfree_deferred *p;
1254 
1255                 vbq = &per_cpu(vmap_block_queue, i);
1256                 spin_lock_init(&vbq->lock);
1257                 INIT_LIST_HEAD(&vbq->free);
1258                 p = &per_cpu(vfree_deferred, i);
1259                 init_llist_head(&p->list);
1260                 INIT_WORK(&p->wq, free_work);
1261         }
1262 
1263         /* Import existing vmlist entries. */
1264         for (tmp = vmlist; tmp; tmp = tmp->next) {
1265                 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1266                 va->flags = VM_VM_AREA;
1267                 va->va_start = (unsigned long)tmp->addr;
1268                 va->va_end = va->va_start + tmp->size;
1269                 va->vm = tmp;
1270                 __insert_vmap_area(va);
1271         }
1272 
1273         vmap_area_pcpu_hole = VMALLOC_END;
1274 
1275         vmap_initialized = true;
1276 }
1277 
1278 /**
1279  * map_kernel_range_noflush - map kernel VM area with the specified pages
1280  * @addr: start of the VM area to map
1281  * @size: size of the VM area to map
1282  * @prot: page protection flags to use
1283  * @pages: pages to map
1284  *
1285  * Map PFN_UP(@size) pages at @addr.  The VM area @addr and @size
1286  * specify should have been allocated using get_vm_area() and its
1287  * friends.
1288  *
1289  * NOTE:
1290  * This function does NOT do any cache flushing.  The caller is
1291  * responsible for calling flush_cache_vmap() on to-be-mapped areas
1292  * before calling this function.
1293  *
1294  * RETURNS:
1295  * The number of pages mapped on success, -errno on failure.
1296  */
1297 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1298                              pgprot_t prot, struct page **pages)
1299 {
1300         return vmap_page_range_noflush(addr, addr + size, prot, pages);
1301 }
1302 
1303 /**
1304  * unmap_kernel_range_noflush - unmap kernel VM area
1305  * @addr: start of the VM area to unmap
1306  * @size: size of the VM area to unmap
1307  *
1308  * Unmap PFN_UP(@size) pages at @addr.  The VM area @addr and @size
1309  * specify should have been allocated using get_vm_area() and its
1310  * friends.
1311  *
1312  * NOTE:
1313  * This function does NOT do any cache flushing.  The caller is
1314  * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1315  * before calling this function and flush_tlb_kernel_range() after.
1316  */
1317 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1318 {
1319         vunmap_page_range(addr, addr + size);
1320 }
1321 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1322 
1323 /**
1324  * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1325  * @addr: start of the VM area to unmap
1326  * @size: size of the VM area to unmap
1327  *
1328  * Similar to unmap_kernel_range_noflush() but flushes vcache before
1329  * the unmapping and tlb after.
1330  */
1331 void unmap_kernel_range(unsigned long addr, unsigned long size)
1332 {
1333         unsigned long end = addr + size;
1334 
1335         flush_cache_vunmap(addr, end);
1336         vunmap_page_range(addr, end);
1337         flush_tlb_kernel_range(addr, end);
1338 }
1339 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1340 
1341 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1342 {
1343         unsigned long addr = (unsigned long)area->addr;
1344         unsigned long end = addr + get_vm_area_size(area);
1345         int err;
1346 
1347         err = vmap_page_range(addr, end, prot, pages);
1348 
1349         return err > 0 ? 0 : err;
1350 }
1351 EXPORT_SYMBOL_GPL(map_vm_area);
1352 
1353 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1354                               unsigned long flags, const void *caller)
1355 {
1356         spin_lock(&vmap_area_lock);
1357         vm->flags = flags;
1358         vm->addr = (void *)va->va_start;
1359         vm->size = va->va_end - va->va_start;
1360         vm->caller = caller;
1361         va->vm = vm;
1362         va->flags |= VM_VM_AREA;
1363         spin_unlock(&vmap_area_lock);
1364 }
1365 
1366 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1367 {
1368         /*
1369          * Before removing VM_UNINITIALIZED,
1370          * we should make sure that vm has proper values.
1371          * Pair with smp_rmb() in show_numa_info().
1372          */
1373         smp_wmb();
1374         vm->flags &= ~VM_UNINITIALIZED;
1375 }
1376 
1377 static struct vm_struct *__get_vm_area_node(unsigned long size,
1378                 unsigned long align, unsigned long flags, unsigned long start,
1379                 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1380 {
1381         struct vmap_area *va;
1382         struct vm_struct *area;
1383 
1384         BUG_ON(in_interrupt());
1385         size = PAGE_ALIGN(size);
1386         if (unlikely(!size))
1387                 return NULL;
1388 
1389         if (flags & VM_IOREMAP)
1390                 align = 1ul << clamp_t(int, get_count_order_long(size),
1391                                        PAGE_SHIFT, IOREMAP_MAX_ORDER);
1392 
1393         area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1394         if (unlikely(!area))
1395                 return NULL;
1396 
1397         if (!(flags & VM_NO_GUARD))
1398                 size += PAGE_SIZE;
1399 
1400         va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1401         if (IS_ERR(va)) {
1402                 kfree(area);
1403                 return NULL;
1404         }
1405 
1406         setup_vmalloc_vm(area, va, flags, caller);
1407 
1408         return area;
1409 }
1410 
1411 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1412                                 unsigned long start, unsigned long end)
1413 {
1414         return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1415                                   GFP_KERNEL, __builtin_return_address(0));
1416 }
1417 EXPORT_SYMBOL_GPL(__get_vm_area);
1418 
1419 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1420                                        unsigned long start, unsigned long end,
1421                                        const void *caller)
1422 {
1423         return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1424                                   GFP_KERNEL, caller);
1425 }
1426 
1427 /**
1428  *      get_vm_area  -  reserve a contiguous kernel virtual area
1429  *      @size:          size of the area
1430  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
1431  *
1432  *      Search an area of @size in the kernel virtual mapping area,
1433  *      and reserved it for out purposes.  Returns the area descriptor
1434  *      on success or %NULL on failure.
1435  */
1436 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1437 {
1438         return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1439                                   NUMA_NO_NODE, GFP_KERNEL,
1440                                   __builtin_return_address(0));
1441 }
1442 
1443 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1444                                 const void *caller)
1445 {
1446         return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1447                                   NUMA_NO_NODE, GFP_KERNEL, caller);
1448 }
1449 
1450 /**
1451  *      find_vm_area  -  find a continuous kernel virtual area
1452  *      @addr:          base address
1453  *
1454  *      Search for the kernel VM area starting at @addr, and return it.
1455  *      It is up to the caller to do all required locking to keep the returned
1456  *      pointer valid.
1457  */
1458 struct vm_struct *find_vm_area(const void *addr)
1459 {
1460         struct vmap_area *va;
1461 
1462         va = find_vmap_area((unsigned long)addr);
1463         if (va && va->flags & VM_VM_AREA)
1464                 return va->vm;
1465 
1466         return NULL;
1467 }
1468 
1469 /**
1470  *      remove_vm_area  -  find and remove a continuous kernel virtual area
1471  *      @addr:          base address
1472  *
1473  *      Search for the kernel VM area starting at @addr, and remove it.
1474  *      This function returns the found VM area, but using it is NOT safe
1475  *      on SMP machines, except for its size or flags.
1476  */
1477 struct vm_struct *remove_vm_area(const void *addr)
1478 {
1479         struct vmap_area *va;
1480 
1481         might_sleep();
1482 
1483         va = find_vmap_area((unsigned long)addr);
1484         if (va && va->flags & VM_VM_AREA) {
1485                 struct vm_struct *vm = va->vm;
1486 
1487                 spin_lock(&vmap_area_lock);
1488                 va->vm = NULL;
1489                 va->flags &= ~VM_VM_AREA;
1490                 va->flags |= VM_LAZY_FREE;
1491                 spin_unlock(&vmap_area_lock);
1492 
1493                 kasan_free_shadow(vm);
1494                 free_unmap_vmap_area(va);
1495 
1496                 return vm;
1497         }
1498         return NULL;
1499 }
1500 
1501 static void __vunmap(const void *addr, int deallocate_pages)
1502 {
1503         struct vm_struct *area;
1504 
1505         if (!addr)
1506                 return;
1507 
1508         if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1509                         addr))
1510                 return;
1511 
1512         area = find_vmap_area((unsigned long)addr)->vm;
1513         if (unlikely(!area)) {
1514                 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1515                                 addr);
1516                 return;
1517         }
1518 
1519         debug_check_no_locks_freed(area->addr, get_vm_area_size(area));
1520         debug_check_no_obj_freed(area->addr, get_vm_area_size(area));
1521 
1522         remove_vm_area(addr);
1523         if (deallocate_pages) {
1524                 int i;
1525 
1526                 for (i = 0; i < area->nr_pages; i++) {
1527                         struct page *page = area->pages[i];
1528 
1529                         BUG_ON(!page);
1530                         __free_pages(page, 0);
1531                 }
1532 
1533                 kvfree(area->pages);
1534         }
1535 
1536         kfree(area);
1537         return;
1538 }
1539 
1540 static inline void __vfree_deferred(const void *addr)
1541 {
1542         /*
1543          * Use raw_cpu_ptr() because this can be called from preemptible
1544          * context. Preemption is absolutely fine here, because the llist_add()
1545          * implementation is lockless, so it works even if we are adding to
1546          * nother cpu's list.  schedule_work() should be fine with this too.
1547          */
1548         struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
1549 
1550         if (llist_add((struct llist_node *)addr, &p->list))
1551                 schedule_work(&p->wq);
1552 }
1553 
1554 /**
1555  *      vfree_atomic  -  release memory allocated by vmalloc()
1556  *      @addr:          memory base address
1557  *
1558  *      This one is just like vfree() but can be called in any atomic context
1559  *      except NMIs.
1560  */
1561 void vfree_atomic(const void *addr)
1562 {
1563         BUG_ON(in_nmi());
1564 
1565         kmemleak_free(addr);
1566 
1567         if (!addr)
1568                 return;
1569         __vfree_deferred(addr);
1570 }
1571 
1572 /**
1573  *      vfree  -  release memory allocated by vmalloc()
1574  *      @addr:          memory base address
1575  *
1576  *      Free the virtually continuous memory area starting at @addr, as
1577  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1578  *      NULL, no operation is performed.
1579  *
1580  *      Must not be called in NMI context (strictly speaking, only if we don't
1581  *      have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1582  *      conventions for vfree() arch-depenedent would be a really bad idea)
1583  *
1584  *      May sleep if called *not* from interrupt context.
1585  *
1586  *      NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
1587  */
1588 void vfree(const void *addr)
1589 {
1590         BUG_ON(in_nmi());
1591 
1592         kmemleak_free(addr);
1593 
1594         might_sleep_if(!in_interrupt());
1595 
1596         if (!addr)
1597                 return;
1598         if (unlikely(in_interrupt()))
1599                 __vfree_deferred(addr);
1600         else
1601                 __vunmap(addr, 1);
1602 }
1603 EXPORT_SYMBOL(vfree);
1604 
1605 /**
1606  *      vunmap  -  release virtual mapping obtained by vmap()
1607  *      @addr:          memory base address
1608  *
1609  *      Free the virtually contiguous memory area starting at @addr,
1610  *      which was created from the page array passed to vmap().
1611  *
1612  *      Must not be called in interrupt context.
1613  */
1614 void vunmap(const void *addr)
1615 {
1616         BUG_ON(in_interrupt());
1617         might_sleep();
1618         if (addr)
1619                 __vunmap(addr, 0);
1620 }
1621 EXPORT_SYMBOL(vunmap);
1622 
1623 /**
1624  *      vmap  -  map an array of pages into virtually contiguous space
1625  *      @pages:         array of page pointers
1626  *      @count:         number of pages to map
1627  *      @flags:         vm_area->flags
1628  *      @prot:          page protection for the mapping
1629  *
1630  *      Maps @count pages from @pages into contiguous kernel virtual
1631  *      space.
1632  */
1633 void *vmap(struct page **pages, unsigned int count,
1634                 unsigned long flags, pgprot_t prot)
1635 {
1636         struct vm_struct *area;
1637         unsigned long size;             /* In bytes */
1638 
1639         might_sleep();
1640 
1641         if (count > totalram_pages())
1642                 return NULL;
1643 
1644         size = (unsigned long)count << PAGE_SHIFT;
1645         area = get_vm_area_caller(size, flags, __builtin_return_address(0));
1646         if (!area)
1647                 return NULL;
1648 
1649         if (map_vm_area(area, prot, pages)) {
1650                 vunmap(area->addr);
1651                 return NULL;
1652         }
1653 
1654         return area->addr;
1655 }
1656 EXPORT_SYMBOL(vmap);
1657 
1658 static void *__vmalloc_node(unsigned long size, unsigned long align,
1659                             gfp_t gfp_mask, pgprot_t prot,
1660                             int node, const void *caller);
1661 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1662                                  pgprot_t prot, int node)
1663 {
1664         struct page **pages;
1665         unsigned int nr_pages, array_size, i;
1666         const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1667         const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1668         const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ?
1669                                         0 :
1670                                         __GFP_HIGHMEM;
1671 
1672         nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1673         array_size = (nr_pages * sizeof(struct page *));
1674 
1675         area->nr_pages = nr_pages;
1676         /* Please note that the recursion is strictly bounded. */
1677         if (array_size > PAGE_SIZE) {
1678                 pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask,
1679                                 PAGE_KERNEL, node, area->caller);
1680         } else {
1681                 pages = kmalloc_node(array_size, nested_gfp, node);
1682         }
1683         area->pages = pages;
1684         if (!area->pages) {
1685                 remove_vm_area(area->addr);
1686                 kfree(area);
1687                 return NULL;
1688         }
1689 
1690         for (i = 0; i < area->nr_pages; i++) {
1691                 struct page *page;
1692 
1693                 if (node == NUMA_NO_NODE)
1694                         page = alloc_page(alloc_mask|highmem_mask);
1695                 else
1696                         page = alloc_pages_node(node, alloc_mask|highmem_mask, 0);
1697 
1698                 if (unlikely(!page)) {
1699                         /* Successfully allocated i pages, free them in __vunmap() */
1700                         area->nr_pages = i;
1701                         goto fail;
1702                 }
1703                 area->pages[i] = page;
1704                 if (gfpflags_allow_blocking(gfp_mask|highmem_mask))
1705                         cond_resched();
1706         }
1707 
1708         if (map_vm_area(area, prot, pages))
1709                 goto fail;
1710         return area->addr;
1711 
1712 fail:
1713         warn_alloc(gfp_mask, NULL,
1714                           "vmalloc: allocation failure, allocated %ld of %ld bytes",
1715                           (area->nr_pages*PAGE_SIZE), area->size);
1716         vfree(area->addr);
1717         return NULL;
1718 }
1719 
1720 /**
1721  *      __vmalloc_node_range  -  allocate virtually contiguous memory
1722  *      @size:          allocation size
1723  *      @align:         desired alignment
1724  *      @start:         vm area range start
1725  *      @end:           vm area range end
1726  *      @gfp_mask:      flags for the page level allocator
1727  *      @prot:          protection mask for the allocated pages
1728  *      @vm_flags:      additional vm area flags (e.g. %VM_NO_GUARD)
1729  *      @node:          node to use for allocation or NUMA_NO_NODE
1730  *      @caller:        caller's return address
1731  *
1732  *      Allocate enough pages to cover @size from the page level
1733  *      allocator with @gfp_mask flags.  Map them into contiguous
1734  *      kernel virtual space, using a pagetable protection of @prot.
1735  */
1736 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1737                         unsigned long start, unsigned long end, gfp_t gfp_mask,
1738                         pgprot_t prot, unsigned long vm_flags, int node,
1739                         const void *caller)
1740 {
1741         struct vm_struct *area;
1742         void *addr;
1743         unsigned long real_size = size;
1744 
1745         size = PAGE_ALIGN(size);
1746         if (!size || (size >> PAGE_SHIFT) > totalram_pages())
1747                 goto fail;
1748 
1749         area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1750                                 vm_flags, start, end, node, gfp_mask, caller);
1751         if (!area)
1752                 goto fail;
1753 
1754         addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1755         if (!addr)
1756                 return NULL;
1757 
1758         /*
1759          * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1760          * flag. It means that vm_struct is not fully initialized.
1761          * Now, it is fully initialized, so remove this flag here.
1762          */
1763         clear_vm_uninitialized_flag(area);
1764 
1765         kmemleak_vmalloc(area, size, gfp_mask);
1766 
1767         return addr;
1768 
1769 fail:
1770         warn_alloc(gfp_mask, NULL,
1771                           "vmalloc: allocation failure: %lu bytes", real_size);
1772         return NULL;
1773 }
1774 
1775 /**
1776  *      __vmalloc_node  -  allocate virtually contiguous memory
1777  *      @size:          allocation size
1778  *      @align:         desired alignment
1779  *      @gfp_mask:      flags for the page level allocator
1780  *      @prot:          protection mask for the allocated pages
1781  *      @node:          node to use for allocation or NUMA_NO_NODE
1782  *      @caller:        caller's return address
1783  *
1784  *      Allocate enough pages to cover @size from the page level
1785  *      allocator with @gfp_mask flags.  Map them into contiguous
1786  *      kernel virtual space, using a pagetable protection of @prot.
1787  *
1788  *      Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL
1789  *      and __GFP_NOFAIL are not supported
1790  *
1791  *      Any use of gfp flags outside of GFP_KERNEL should be consulted
1792  *      with mm people.
1793  *
1794  */
1795 static void *__vmalloc_node(unsigned long size, unsigned long align,
1796                             gfp_t gfp_mask, pgprot_t prot,
1797                             int node, const void *caller)
1798 {
1799         return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1800                                 gfp_mask, prot, 0, node, caller);
1801 }
1802 
1803 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1804 {
1805         return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1806                                 __builtin_return_address(0));
1807 }
1808 EXPORT_SYMBOL(__vmalloc);
1809 
1810 static inline void *__vmalloc_node_flags(unsigned long size,
1811                                         int node, gfp_t flags)
1812 {
1813         return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1814                                         node, __builtin_return_address(0));
1815 }
1816 
1817 
1818 void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags,
1819                                   void *caller)
1820 {
1821         return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller);
1822 }
1823 
1824 /**
1825  *      vmalloc  -  allocate virtually contiguous memory
1826  *      @size:          allocation size
1827  *      Allocate enough pages to cover @size from the page level
1828  *      allocator and map them into contiguous kernel virtual space.
1829  *
1830  *      For tight control over page level allocator and protection flags
1831  *      use __vmalloc() instead.
1832  */
1833 void *vmalloc(unsigned long size)
1834 {
1835         return __vmalloc_node_flags(size, NUMA_NO_NODE,
1836                                     GFP_KERNEL);
1837 }
1838 EXPORT_SYMBOL(vmalloc);
1839 
1840 /**
1841  *      vzalloc - allocate virtually contiguous memory with zero fill
1842  *      @size:  allocation size
1843  *      Allocate enough pages to cover @size from the page level
1844  *      allocator and map them into contiguous kernel virtual space.
1845  *      The memory allocated is set to zero.
1846  *
1847  *      For tight control over page level allocator and protection flags
1848  *      use __vmalloc() instead.
1849  */
1850 void *vzalloc(unsigned long size)
1851 {
1852         return __vmalloc_node_flags(size, NUMA_NO_NODE,
1853                                 GFP_KERNEL | __GFP_ZERO);
1854 }
1855 EXPORT_SYMBOL(vzalloc);
1856 
1857 /**
1858  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1859  * @size: allocation size
1860  *
1861  * The resulting memory area is zeroed so it can be mapped to userspace
1862  * without leaking data.
1863  */
1864 void *vmalloc_user(unsigned long size)
1865 {
1866         struct vm_struct *area;
1867         void *ret;
1868 
1869         ret = __vmalloc_node(size, SHMLBA,
1870                              GFP_KERNEL | __GFP_ZERO,
1871                              PAGE_KERNEL, NUMA_NO_NODE,
1872                              __builtin_return_address(0));
1873         if (ret) {
1874                 area = find_vm_area(ret);
1875                 area->flags |= VM_USERMAP;
1876         }
1877         return ret;
1878 }
1879 EXPORT_SYMBOL(vmalloc_user);
1880 
1881 /**
1882  *      vmalloc_node  -  allocate memory on a specific node
1883  *      @size:          allocation size
1884  *      @node:          numa node
1885  *
1886  *      Allocate enough pages to cover @size from the page level
1887  *      allocator and map them into contiguous kernel virtual space.
1888  *
1889  *      For tight control over page level allocator and protection flags
1890  *      use __vmalloc() instead.
1891  */
1892 void *vmalloc_node(unsigned long size, int node)
1893 {
1894         return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL,
1895                                         node, __builtin_return_address(0));
1896 }
1897 EXPORT_SYMBOL(vmalloc_node);
1898 
1899 /**
1900  * vzalloc_node - allocate memory on a specific node with zero fill
1901  * @size:       allocation size
1902  * @node:       numa node
1903  *
1904  * Allocate enough pages to cover @size from the page level
1905  * allocator and map them into contiguous kernel virtual space.
1906  * The memory allocated is set to zero.
1907  *
1908  * For tight control over page level allocator and protection flags
1909  * use __vmalloc_node() instead.
1910  */
1911 void *vzalloc_node(unsigned long size, int node)
1912 {
1913         return __vmalloc_node_flags(size, node,
1914                          GFP_KERNEL | __GFP_ZERO);
1915 }
1916 EXPORT_SYMBOL(vzalloc_node);
1917 
1918 /**
1919  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
1920  *      @size:          allocation size
1921  *
1922  *      Kernel-internal function to allocate enough pages to cover @size
1923  *      the page level allocator and map them into contiguous and
1924  *      executable kernel virtual space.
1925  *
1926  *      For tight control over page level allocator and protection flags
1927  *      use __vmalloc() instead.
1928  */
1929 
1930 void *vmalloc_exec(unsigned long size)
1931 {
1932         return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL_EXEC,
1933                               NUMA_NO_NODE, __builtin_return_address(0));
1934 }
1935 
1936 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1937 #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL)
1938 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1939 #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL)
1940 #else
1941 /*
1942  * 64b systems should always have either DMA or DMA32 zones. For others
1943  * GFP_DMA32 should do the right thing and use the normal zone.
1944  */
1945 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1946 #endif
1947 
1948 /**
1949  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
1950  *      @size:          allocation size
1951  *
1952  *      Allocate enough 32bit PA addressable pages to cover @size from the
1953  *      page level allocator and map them into contiguous kernel virtual space.
1954  */
1955 void *vmalloc_32(unsigned long size)
1956 {
1957         return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1958                               NUMA_NO_NODE, __builtin_return_address(0));
1959 }
1960 EXPORT_SYMBOL(vmalloc_32);
1961 
1962 /**
1963  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1964  *      @size:          allocation size
1965  *
1966  * The resulting memory area is 32bit addressable and zeroed so it can be
1967  * mapped to userspace without leaking data.
1968  */
1969 void *vmalloc_32_user(unsigned long size)
1970 {
1971         struct vm_struct *area;
1972         void *ret;
1973 
1974         ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1975                              NUMA_NO_NODE, __builtin_return_address(0));
1976         if (ret) {
1977                 area = find_vm_area(ret);
1978                 area->flags |= VM_USERMAP;
1979         }
1980         return ret;
1981 }
1982 EXPORT_SYMBOL(vmalloc_32_user);
1983 
1984 /*
1985  * small helper routine , copy contents to buf from addr.
1986  * If the page is not present, fill zero.
1987  */
1988 
1989 static int aligned_vread(char *buf, char *addr, unsigned long count)
1990 {
1991         struct page *p;
1992         int copied = 0;
1993 
1994         while (count) {
1995                 unsigned long offset, length;
1996 
1997                 offset = offset_in_page(addr);
1998                 length = PAGE_SIZE - offset;
1999                 if (length > count)
2000                         length = count;
2001                 p = vmalloc_to_page(addr);
2002                 /*
2003                  * To do safe access to this _mapped_ area, we need
2004                  * lock. But adding lock here means that we need to add
2005                  * overhead of vmalloc()/vfree() calles for this _debug_
2006                  * interface, rarely used. Instead of that, we'll use
2007                  * kmap() and get small overhead in this access function.
2008                  */
2009                 if (p) {
2010                         /*
2011                          * we can expect USER0 is not used (see vread/vwrite's
2012                          * function description)
2013                          */
2014                         void *map = kmap_atomic(p);
2015                         memcpy(buf, map + offset, length);
2016                         kunmap_atomic(map);
2017                 } else
2018                         memset(buf, 0, length);
2019 
2020                 addr += length;
2021                 buf += length;
2022                 copied += length;
2023                 count -= length;
2024         }
2025         return copied;
2026 }
2027 
2028 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
2029 {
2030         struct page *p;
2031         int copied = 0;
2032 
2033         while (count) {
2034                 unsigned long offset, length;
2035 
2036                 offset = offset_in_page(addr);
2037                 length = PAGE_SIZE - offset;
2038                 if (length > count)
2039                         length = count;
2040                 p = vmalloc_to_page(addr);
2041                 /*
2042                  * To do safe access to this _mapped_ area, we need
2043                  * lock. But adding lock here means that we need to add
2044                  * overhead of vmalloc()/vfree() calles for this _debug_
2045                  * interface, rarely used. Instead of that, we'll use
2046                  * kmap() and get small overhead in this access function.
2047                  */
2048                 if (p) {
2049                         /*
2050                          * we can expect USER0 is not used (see vread/vwrite's
2051                          * function description)
2052                          */
2053                         void *map = kmap_atomic(p);
2054                         memcpy(map + offset, buf, length);
2055                         kunmap_atomic(map);
2056                 }
2057                 addr += length;
2058                 buf += length;
2059                 copied += length;
2060                 count -= length;
2061         }
2062         return copied;
2063 }
2064 
2065 /**
2066  *      vread() -  read vmalloc area in a safe way.
2067  *      @buf:           buffer for reading data
2068  *      @addr:          vm address.
2069  *      @count:         number of bytes to be read.
2070  *
2071  *      Returns # of bytes which addr and buf should be increased.
2072  *      (same number to @count). Returns 0 if [addr...addr+count) doesn't
2073  *      includes any intersect with alive vmalloc area.
2074  *
2075  *      This function checks that addr is a valid vmalloc'ed area, and
2076  *      copy data from that area to a given buffer. If the given memory range
2077  *      of [addr...addr+count) includes some valid address, data is copied to
2078  *      proper area of @buf. If there are memory holes, they'll be zero-filled.
2079  *      IOREMAP area is treated as memory hole and no copy is done.
2080  *
2081  *      If [addr...addr+count) doesn't includes any intersects with alive
2082  *      vm_struct area, returns 0. @buf should be kernel's buffer.
2083  *
2084  *      Note: In usual ops, vread() is never necessary because the caller
2085  *      should know vmalloc() area is valid and can use memcpy().
2086  *      This is for routines which have to access vmalloc area without
2087  *      any informaion, as /dev/kmem.
2088  *
2089  */
2090 
2091 long vread(char *buf, char *addr, unsigned long count)
2092 {
2093         struct vmap_area *va;
2094         struct vm_struct *vm;
2095         char *vaddr, *buf_start = buf;
2096         unsigned long buflen = count;
2097         unsigned long n;
2098 
2099         /* Don't allow overflow */
2100         if ((unsigned long) addr + count < count)
2101                 count = -(unsigned long) addr;
2102 
2103         spin_lock(&vmap_area_lock);
2104         list_for_each_entry(va, &vmap_area_list, list) {
2105                 if (!count)
2106                         break;
2107 
2108                 if (!(va->flags & VM_VM_AREA))
2109                         continue;
2110 
2111                 vm = va->vm;
2112                 vaddr = (char *) vm->addr;
2113                 if (addr >= vaddr + get_vm_area_size(vm))
2114                         continue;
2115                 while (addr < vaddr) {
2116                         if (count == 0)
2117                                 goto finished;
2118                         *buf = '\0';
2119                         buf++;
2120                         addr++;
2121                         count--;
2122                 }
2123                 n = vaddr + get_vm_area_size(vm) - addr;
2124                 if (n > count)
2125                         n = count;
2126                 if (!(vm->flags & VM_IOREMAP))
2127                         aligned_vread(buf, addr, n);
2128                 else /* IOREMAP area is treated as memory hole */
2129                         memset(buf, 0, n);
2130                 buf += n;
2131                 addr += n;
2132                 count -= n;
2133         }
2134 finished:
2135         spin_unlock(&vmap_area_lock);
2136 
2137         if (buf == buf_start)
2138                 return 0;
2139         /* zero-fill memory holes */
2140         if (buf != buf_start + buflen)
2141                 memset(buf, 0, buflen - (buf - buf_start));
2142 
2143         return buflen;
2144 }
2145 
2146 /**
2147  *      vwrite() -  write vmalloc area in a safe way.
2148  *      @buf:           buffer for source data
2149  *      @addr:          vm address.
2150  *      @count:         number of bytes to be read.
2151  *
2152  *      Returns # of bytes which addr and buf should be incresed.
2153  *      (same number to @count).
2154  *      If [addr...addr+count) doesn't includes any intersect with valid
2155  *      vmalloc area, returns 0.
2156  *
2157  *      This function checks that addr is a valid vmalloc'ed area, and
2158  *      copy data from a buffer to the given addr. If specified range of
2159  *      [addr...addr+count) includes some valid address, data is copied from
2160  *      proper area of @buf. If there are memory holes, no copy to hole.
2161  *      IOREMAP area is treated as memory hole and no copy is done.
2162  *
2163  *      If [addr...addr+count) doesn't includes any intersects with alive
2164  *      vm_struct area, returns 0. @buf should be kernel's buffer.
2165  *
2166  *      Note: In usual ops, vwrite() is never necessary because the caller
2167  *      should know vmalloc() area is valid and can use memcpy().
2168  *      This is for routines which have to access vmalloc area without
2169  *      any informaion, as /dev/kmem.
2170  */
2171 
2172 long vwrite(char *buf, char *addr, unsigned long count)
2173 {
2174         struct vmap_area *va;
2175         struct vm_struct *vm;
2176         char *vaddr;
2177         unsigned long n, buflen;
2178         int copied = 0;
2179 
2180         /* Don't allow overflow */
2181         if ((unsigned long) addr + count < count)
2182                 count = -(unsigned long) addr;
2183         buflen = count;
2184 
2185         spin_lock(&vmap_area_lock);
2186         list_for_each_entry(va, &vmap_area_list, list) {
2187                 if (!count)
2188                         break;
2189 
2190                 if (!(va->flags & VM_VM_AREA))
2191                         continue;
2192 
2193                 vm = va->vm;
2194                 vaddr = (char *) vm->addr;
2195                 if (addr >= vaddr + get_vm_area_size(vm))
2196                         continue;
2197                 while (addr < vaddr) {
2198                         if (count == 0)
2199                                 goto finished;
2200                         buf++;
2201                         addr++;
2202                         count--;
2203                 }
2204                 n = vaddr + get_vm_area_size(vm) - addr;
2205                 if (n > count)
2206                         n = count;
2207                 if (!(vm->flags & VM_IOREMAP)) {
2208                         aligned_vwrite(buf, addr, n);
2209                         copied++;
2210                 }
2211                 buf += n;
2212                 addr += n;
2213                 count -= n;
2214         }
2215 finished:
2216         spin_unlock(&vmap_area_lock);
2217         if (!copied)
2218                 return 0;
2219         return buflen;
2220 }
2221 
2222 /**
2223  *      remap_vmalloc_range_partial  -  map vmalloc pages to userspace
2224  *      @vma:           vma to cover
2225  *      @uaddr:         target user address to start at
2226  *      @kaddr:         virtual address of vmalloc kernel memory
2227  *      @size:          size of map area
2228  *
2229  *      Returns:        0 for success, -Exxx on failure
2230  *
2231  *      This function checks that @kaddr is a valid vmalloc'ed area,
2232  *      and that it is big enough to cover the range starting at
2233  *      @uaddr in @vma. Will return failure if that criteria isn't
2234  *      met.
2235  *
2236  *      Similar to remap_pfn_range() (see mm/memory.c)
2237  */
2238 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2239                                 void *kaddr, unsigned long size)
2240 {
2241         struct vm_struct *area;
2242 
2243         size = PAGE_ALIGN(size);
2244 
2245         if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2246                 return -EINVAL;
2247 
2248         area = find_vm_area(kaddr);
2249         if (!area)
2250                 return -EINVAL;
2251 
2252         if (!(area->flags & VM_USERMAP))
2253                 return -EINVAL;
2254 
2255         if (kaddr + size > area->addr + get_vm_area_size(area))
2256                 return -EINVAL;
2257 
2258         do {
2259                 struct page *page = vmalloc_to_page(kaddr);
2260                 int ret;
2261 
2262                 ret = vm_insert_page(vma, uaddr, page);
2263                 if (ret)
2264                         return ret;
2265 
2266                 uaddr += PAGE_SIZE;
2267                 kaddr += PAGE_SIZE;
2268                 size -= PAGE_SIZE;
2269         } while (size > 0);
2270 
2271         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2272 
2273         return 0;
2274 }
2275 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2276 
2277 /**
2278  *      remap_vmalloc_range  -  map vmalloc pages to userspace
2279  *      @vma:           vma to cover (map full range of vma)
2280  *      @addr:          vmalloc memory
2281  *      @pgoff:         number of pages into addr before first page to map
2282  *
2283  *      Returns:        0 for success, -Exxx on failure
2284  *
2285  *      This function checks that addr is a valid vmalloc'ed area, and
2286  *      that it is big enough to cover the vma. Will return failure if
2287  *      that criteria isn't met.
2288  *
2289  *      Similar to remap_pfn_range() (see mm/memory.c)
2290  */
2291 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2292                                                 unsigned long pgoff)
2293 {
2294         return remap_vmalloc_range_partial(vma, vma->vm_start,
2295                                            addr + (pgoff << PAGE_SHIFT),
2296                                            vma->vm_end - vma->vm_start);
2297 }
2298 EXPORT_SYMBOL(remap_vmalloc_range);
2299 
2300 /*
2301  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2302  * have one.
2303  */
2304 void __weak vmalloc_sync_all(void)
2305 {
2306 }
2307 
2308 
2309 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2310 {
2311         pte_t ***p = data;
2312 
2313         if (p) {
2314                 *(*p) = pte;
2315                 (*p)++;
2316         }
2317         return 0;
2318 }
2319 
2320 /**
2321  *      alloc_vm_area - allocate a range of kernel address space
2322  *      @size:          size of the area
2323  *      @ptes:          returns the PTEs for the address space
2324  *
2325  *      Returns:        NULL on failure, vm_struct on success
2326  *
2327  *      This function reserves a range of kernel address space, and
2328  *      allocates pagetables to map that range.  No actual mappings
2329  *      are created.
2330  *
2331  *      If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2332  *      allocated for the VM area are returned.
2333  */
2334 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2335 {
2336         struct vm_struct *area;
2337 
2338         area = get_vm_area_caller(size, VM_IOREMAP,
2339                                 __builtin_return_address(0));
2340         if (area == NULL)
2341                 return NULL;
2342 
2343         /*
2344          * This ensures that page tables are constructed for this region
2345          * of kernel virtual address space and mapped into init_mm.
2346          */
2347         if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2348                                 size, f, ptes ? &ptes : NULL)) {
2349                 free_vm_area(area);
2350                 return NULL;
2351         }
2352 
2353         return area;
2354 }
2355 EXPORT_SYMBOL_GPL(alloc_vm_area);
2356 
2357 void free_vm_area(struct vm_struct *area)
2358 {
2359         struct vm_struct *ret;
2360         ret = remove_vm_area(area->addr);
2361         BUG_ON(ret != area);
2362         kfree(area);
2363 }
2364 EXPORT_SYMBOL_GPL(free_vm_area);
2365 
2366 #ifdef CONFIG_SMP
2367 static struct vmap_area *node_to_va(struct rb_node *n)
2368 {
2369         return rb_entry_safe(n, struct vmap_area, rb_node);
2370 }
2371 
2372 /**
2373  * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2374  * @end: target address
2375  * @pnext: out arg for the next vmap_area
2376  * @pprev: out arg for the previous vmap_area
2377  *
2378  * Returns: %true if either or both of next and prev are found,
2379  *          %false if no vmap_area exists
2380  *
2381  * Find vmap_areas end addresses of which enclose @end.  ie. if not
2382  * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2383  */
2384 static bool pvm_find_next_prev(unsigned long end,
2385                                struct vmap_area **pnext,
2386                                struct vmap_area **pprev)
2387 {
2388         struct rb_node *n = vmap_area_root.rb_node;
2389         struct vmap_area *va = NULL;
2390 
2391         while (n) {
2392                 va = rb_entry(n, struct vmap_area, rb_node);
2393                 if (end < va->va_end)
2394                         n = n->rb_left;
2395                 else if (end > va->va_end)
2396                         n = n->rb_right;
2397                 else
2398                         break;
2399         }
2400 
2401         if (!va)
2402                 return false;
2403 
2404         if (va->va_end > end) {
2405                 *pnext = va;
2406                 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2407         } else {
2408                 *pprev = va;
2409                 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2410         }
2411         return true;
2412 }
2413 
2414 /**
2415  * pvm_determine_end - find the highest aligned address between two vmap_areas
2416  * @pnext: in/out arg for the next vmap_area
2417  * @pprev: in/out arg for the previous vmap_area
2418  * @align: alignment
2419  *
2420  * Returns: determined end address
2421  *
2422  * Find the highest aligned address between *@pnext and *@pprev below
2423  * VMALLOC_END.  *@pnext and *@pprev are adjusted so that the aligned
2424  * down address is between the end addresses of the two vmap_areas.
2425  *
2426  * Please note that the address returned by this function may fall
2427  * inside *@pnext vmap_area.  The caller is responsible for checking
2428  * that.
2429  */
2430 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2431                                        struct vmap_area **pprev,
2432                                        unsigned long align)
2433 {
2434         const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2435         unsigned long addr;
2436 
2437         if (*pnext)
2438                 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2439         else
2440                 addr = vmalloc_end;
2441 
2442         while (*pprev && (*pprev)->va_end > addr) {
2443                 *pnext = *pprev;
2444                 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2445         }
2446 
2447         return addr;
2448 }
2449 
2450 /**
2451  * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2452  * @offsets: array containing offset of each area
2453  * @sizes: array containing size of each area
2454  * @nr_vms: the number of areas to allocate
2455  * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2456  *
2457  * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2458  *          vm_structs on success, %NULL on failure
2459  *
2460  * Percpu allocator wants to use congruent vm areas so that it can
2461  * maintain the offsets among percpu areas.  This function allocates
2462  * congruent vmalloc areas for it with GFP_KERNEL.  These areas tend to
2463  * be scattered pretty far, distance between two areas easily going up
2464  * to gigabytes.  To avoid interacting with regular vmallocs, these
2465  * areas are allocated from top.
2466  *
2467  * Despite its complicated look, this allocator is rather simple.  It
2468  * does everything top-down and scans areas from the end looking for
2469  * matching slot.  While scanning, if any of the areas overlaps with
2470  * existing vmap_area, the base address is pulled down to fit the
2471  * area.  Scanning is repeated till all the areas fit and then all
2472  * necessary data structures are inserted and the result is returned.
2473  */
2474 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2475                                      const size_t *sizes, int nr_vms,
2476                                      size_t align)
2477 {
2478         const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2479         const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2480         struct vmap_area **vas, *prev, *next;
2481         struct vm_struct **vms;
2482         int area, area2, last_area, term_area;
2483         unsigned long base, start, end, last_end;
2484         bool purged = false;
2485 
2486         /* verify parameters and allocate data structures */
2487         BUG_ON(offset_in_page(align) || !is_power_of_2(align));
2488         for (last_area = 0, area = 0; area < nr_vms; area++) {
2489                 start = offsets[area];
2490                 end = start + sizes[area];
2491 
2492                 /* is everything aligned properly? */
2493                 BUG_ON(!IS_ALIGNED(offsets[area], align));
2494                 BUG_ON(!IS_ALIGNED(sizes[area], align));
2495 
2496                 /* detect the area with the highest address */
2497                 if (start > offsets[last_area])
2498                         last_area = area;
2499 
2500                 for (area2 = area + 1; area2 < nr_vms; area2++) {
2501                         unsigned long start2 = offsets[area2];
2502                         unsigned long end2 = start2 + sizes[area2];
2503 
2504                         BUG_ON(start2 < end && start < end2);
2505                 }
2506         }
2507         last_end = offsets[last_area] + sizes[last_area];
2508 
2509         if (vmalloc_end - vmalloc_start < last_end) {
2510                 WARN_ON(true);
2511                 return NULL;
2512         }
2513 
2514         vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2515         vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2516         if (!vas || !vms)
2517                 goto err_free2;
2518 
2519         for (area = 0; area < nr_vms; area++) {
2520                 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2521                 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2522                 if (!vas[area] || !vms[area])
2523                         goto err_free;
2524         }
2525 retry:
2526         spin_lock(&vmap_area_lock);
2527 
2528         /* start scanning - we scan from the top, begin with the last area */
2529         area = term_area = last_area;
2530         start = offsets[area];
2531         end = start + sizes[area];
2532 
2533         if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2534                 base = vmalloc_end - last_end;
2535                 goto found;
2536         }
2537         base = pvm_determine_end(&next, &prev, align) - end;
2538 
2539         while (true) {
2540                 BUG_ON(next && next->va_end <= base + end);
2541                 BUG_ON(prev && prev->va_end > base + end);
2542 
2543                 /*
2544                  * base might have underflowed, add last_end before
2545                  * comparing.
2546                  */
2547                 if (base + last_end < vmalloc_start + last_end) {
2548                         spin_unlock(&vmap_area_lock);
2549                         if (!purged) {
2550                                 purge_vmap_area_lazy();
2551                                 purged = true;
2552                                 goto retry;
2553                         }
2554                         goto err_free;
2555                 }
2556 
2557                 /*
2558                  * If next overlaps, move base downwards so that it's
2559                  * right below next and then recheck.
2560                  */
2561                 if (next && next->va_start < base + end) {
2562                         base = pvm_determine_end(&next, &prev, align) - end;
2563                         term_area = area;
2564                         continue;
2565                 }
2566 
2567                 /*
2568                  * If prev overlaps, shift down next and prev and move
2569                  * base so that it's right below new next and then
2570                  * recheck.
2571                  */
2572                 if (prev && prev->va_end > base + start)  {
2573                         next = prev;
2574                         prev = node_to_va(rb_prev(&next->rb_node));
2575                         base = pvm_determine_end(&next, &prev, align) - end;
2576                         term_area = area;
2577                         continue;
2578                 }
2579 
2580                 /*
2581                  * This area fits, move on to the previous one.  If
2582                  * the previous one is the terminal one, we're done.
2583                  */
2584                 area = (area + nr_vms - 1) % nr_vms;
2585                 if (area == term_area)
2586                         break;
2587                 start = offsets[area];
2588                 end = start + sizes[area];
2589                 pvm_find_next_prev(base + end, &next, &prev);
2590         }
2591 found:
2592         /* we've found a fitting base, insert all va's */
2593         for (area = 0; area < nr_vms; area++) {
2594                 struct vmap_area *va = vas[area];
2595 
2596                 va->va_start = base + offsets[area];
2597                 va->va_end = va->va_start + sizes[area];
2598                 __insert_vmap_area(va);
2599         }
2600 
2601         vmap_area_pcpu_hole = base + offsets[last_area];
2602 
2603         spin_unlock(&vmap_area_lock);
2604 
2605         /* insert all vm's */
2606         for (area = 0; area < nr_vms; area++)
2607                 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2608                                  pcpu_get_vm_areas);
2609 
2610         kfree(vas);
2611         return vms;
2612 
2613 err_free:
2614         for (area = 0; area < nr_vms; area++) {
2615                 kfree(vas[area]);
2616                 kfree(vms[area]);
2617         }
2618 err_free2:
2619         kfree(vas);
2620         kfree(vms);
2621         return NULL;
2622 }
2623 
2624 /**
2625  * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2626  * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2627  * @nr_vms: the number of allocated areas
2628  *
2629  * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2630  */
2631 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2632 {
2633         int i;
2634 
2635         for (i = 0; i < nr_vms; i++)
2636                 free_vm_area(vms[i]);
2637         kfree(vms);
2638 }
2639 #endif  /* CONFIG_SMP */
2640 
2641 #ifdef CONFIG_PROC_FS
2642 static void *s_start(struct seq_file *m, loff_t *pos)
2643         __acquires(&vmap_area_lock)
2644 {
2645         spin_lock(&vmap_area_lock);
2646         return seq_list_start(&vmap_area_list, *pos);
2647 }
2648 
2649 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2650 {
2651         return seq_list_next(p, &vmap_area_list, pos);
2652 }
2653 
2654 static void s_stop(struct seq_file *m, void *p)
2655         __releases(&vmap_area_lock)
2656 {
2657         spin_unlock(&vmap_area_lock);
2658 }
2659 
2660 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2661 {
2662         if (IS_ENABLED(CONFIG_NUMA)) {
2663                 unsigned int nr, *counters = m->private;
2664 
2665                 if (!counters)
2666                         return;
2667 
2668                 if (v->flags & VM_UNINITIALIZED)
2669                         return;
2670                 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2671                 smp_rmb();
2672 
2673                 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2674 
2675                 for (nr = 0; nr < v->nr_pages; nr++)
2676                         counters[page_to_nid(v->pages[nr])]++;
2677 
2678                 for_each_node_state(nr, N_HIGH_MEMORY)
2679                         if (counters[nr])
2680                                 seq_printf(m, " N%u=%u", nr, counters[nr]);
2681         }
2682 }
2683 
2684 static int s_show(struct seq_file *m, void *p)
2685 {
2686         struct vmap_area *va;
2687         struct vm_struct *v;
2688 
2689         va = list_entry(p, struct vmap_area, list);
2690 
2691         /*
2692          * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2693          * behalf of vmap area is being tear down or vm_map_ram allocation.
2694          */
2695         if (!(va->flags & VM_VM_AREA)) {
2696                 seq_printf(m, "0x%pK-0x%pK %7ld %s\n",
2697                         (void *)va->va_start, (void *)va->va_end,
2698                         va->va_end - va->va_start,
2699                         va->flags & VM_LAZY_FREE ? "unpurged vm_area" : "vm_map_ram");
2700 
2701                 return 0;
2702         }
2703 
2704         v = va->vm;
2705 
2706         seq_printf(m, "0x%pK-0x%pK %7ld",
2707                 v->addr, v->addr + v->size, v->size);
2708 
2709         if (v->caller)
2710                 seq_printf(m, " %pS", v->caller);
2711 
2712         if (v->nr_pages)
2713                 seq_printf(m, " pages=%d", v->nr_pages);
2714 
2715         if (v->phys_addr)
2716                 seq_printf(m, " phys=%pa", &v->phys_addr);
2717 
2718         if (v->flags & VM_IOREMAP)
2719                 seq_puts(m, " ioremap");
2720 
2721         if (v->flags & VM_ALLOC)
2722                 seq_puts(m, " vmalloc");
2723 
2724         if (v->flags & VM_MAP)
2725                 seq_puts(m, " vmap");
2726 
2727         if (v->flags & VM_USERMAP)
2728                 seq_puts(m, " user");
2729 
2730         if (is_vmalloc_addr(v->pages))
2731                 seq_puts(m, " vpages");
2732 
2733         show_numa_info(m, v);
2734         seq_putc(m, '\n');
2735         return 0;
2736 }
2737 
2738 static const struct seq_operations vmalloc_op = {
2739         .start = s_start,
2740         .next = s_next,
2741         .stop = s_stop,
2742         .show = s_show,
2743 };
2744 
2745 static int __init proc_vmalloc_init(void)
2746 {
2747         if (IS_ENABLED(CONFIG_NUMA))
2748                 proc_create_seq_private("vmallocinfo", 0400, NULL,
2749                                 &vmalloc_op,
2750                                 nr_node_ids * sizeof(unsigned int), NULL);
2751         else
2752                 proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op);
2753         return 0;
2754 }
2755 module_init(proc_vmalloc_init);
2756 
2757 #endif
2758 
2759 

~ [ 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