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

TOMOYO Linux Cross Reference
Linux/mm/vmalloc.c

Version: ~ [ linux-5.10-rc5 ] ~ [ linux-5.9.10 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.79 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.159 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.208 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.245 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.245 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  linux/mm/vmalloc.c
  3  *
  4  *  Copyright (C) 1993  Linus Torvalds
  5  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
  6  *  SMP-safe vmalloc/vfree/ioremap, Tigran Aivazian <tigran@veritas.com>, May 2000
  7  *  Major rework to support vmap/vunmap, Christoph Hellwig, SGI, August 2002
  8  *  Numa awareness, Christoph Lameter, SGI, June 2005
  9  */
 10 
 11 #include <linux/vmalloc.h>
 12 #include <linux/mm.h>
 13 #include <linux/module.h>
 14 #include <linux/highmem.h>
 15 #include <linux/sched/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         if (addr + size > vend)
502                 goto overflow;
503 
504         va->va_start = addr;
505         va->va_end = addr + size;
506         va->flags = 0;
507         __insert_vmap_area(va);
508         free_vmap_cache = &va->rb_node;
509         spin_unlock(&vmap_area_lock);
510 
511         BUG_ON(!IS_ALIGNED(va->va_start, align));
512         BUG_ON(va->va_start < vstart);
513         BUG_ON(va->va_end > vend);
514 
515         return va;
516 
517 overflow:
518         spin_unlock(&vmap_area_lock);
519         if (!purged) {
520                 purge_vmap_area_lazy();
521                 purged = 1;
522                 goto retry;
523         }
524 
525         if (gfpflags_allow_blocking(gfp_mask)) {
526                 unsigned long freed = 0;
527                 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
528                 if (freed > 0) {
529                         purged = 0;
530                         goto retry;
531                 }
532         }
533 
534         if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
535                 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
536                         size);
537         kfree(va);
538         return ERR_PTR(-EBUSY);
539 }
540 
541 int register_vmap_purge_notifier(struct notifier_block *nb)
542 {
543         return blocking_notifier_chain_register(&vmap_notify_list, nb);
544 }
545 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
546 
547 int unregister_vmap_purge_notifier(struct notifier_block *nb)
548 {
549         return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
550 }
551 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
552 
553 static void __free_vmap_area(struct vmap_area *va)
554 {
555         BUG_ON(RB_EMPTY_NODE(&va->rb_node));
556 
557         if (free_vmap_cache) {
558                 if (va->va_end < cached_vstart) {
559                         free_vmap_cache = NULL;
560                 } else {
561                         struct vmap_area *cache;
562                         cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
563                         if (va->va_start <= cache->va_start) {
564                                 free_vmap_cache = rb_prev(&va->rb_node);
565                                 /*
566                                  * We don't try to update cached_hole_size or
567                                  * cached_align, but it won't go very wrong.
568                                  */
569                         }
570                 }
571         }
572         rb_erase(&va->rb_node, &vmap_area_root);
573         RB_CLEAR_NODE(&va->rb_node);
574         list_del_rcu(&va->list);
575 
576         /*
577          * Track the highest possible candidate for pcpu area
578          * allocation.  Areas outside of vmalloc area can be returned
579          * here too, consider only end addresses which fall inside
580          * vmalloc area proper.
581          */
582         if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
583                 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
584 
585         kfree_rcu(va, rcu_head);
586 }
587 
588 /*
589  * Free a region of KVA allocated by alloc_vmap_area
590  */
591 static void free_vmap_area(struct vmap_area *va)
592 {
593         spin_lock(&vmap_area_lock);
594         __free_vmap_area(va);
595         spin_unlock(&vmap_area_lock);
596 }
597 
598 /*
599  * Clear the pagetable entries of a given vmap_area
600  */
601 static void unmap_vmap_area(struct vmap_area *va)
602 {
603         vunmap_page_range(va->va_start, va->va_end);
604 }
605 
606 static void vmap_debug_free_range(unsigned long start, unsigned long end)
607 {
608         /*
609          * Unmap page tables and force a TLB flush immediately if pagealloc
610          * debugging is enabled.  This catches use after free bugs similarly to
611          * those in linear kernel virtual address space after a page has been
612          * freed.
613          *
614          * All the lazy freeing logic is still retained, in order to minimise
615          * intrusiveness of this debugging feature.
616          *
617          * This is going to be *slow* (linear kernel virtual address debugging
618          * doesn't do a broadcast TLB flush so it is a lot faster).
619          */
620         if (debug_pagealloc_enabled()) {
621                 vunmap_page_range(start, end);
622                 flush_tlb_kernel_range(start, end);
623         }
624 }
625 
626 /*
627  * lazy_max_pages is the maximum amount of virtual address space we gather up
628  * before attempting to purge with a TLB flush.
629  *
630  * There is a tradeoff here: a larger number will cover more kernel page tables
631  * and take slightly longer to purge, but it will linearly reduce the number of
632  * global TLB flushes that must be performed. It would seem natural to scale
633  * this number up linearly with the number of CPUs (because vmapping activity
634  * could also scale linearly with the number of CPUs), however it is likely
635  * that in practice, workloads might be constrained in other ways that mean
636  * vmap activity will not scale linearly with CPUs. Also, I want to be
637  * conservative and not introduce a big latency on huge systems, so go with
638  * a less aggressive log scale. It will still be an improvement over the old
639  * code, and it will be simple to change the scale factor if we find that it
640  * becomes a problem on bigger systems.
641  */
642 static unsigned long lazy_max_pages(void)
643 {
644         unsigned int log;
645 
646         log = fls(num_online_cpus());
647 
648         return log * (32UL * 1024 * 1024 / PAGE_SIZE);
649 }
650 
651 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
652 
653 /*
654  * Serialize vmap purging.  There is no actual criticial section protected
655  * by this look, but we want to avoid concurrent calls for performance
656  * reasons and to make the pcpu_get_vm_areas more deterministic.
657  */
658 static DEFINE_MUTEX(vmap_purge_lock);
659 
660 /* for per-CPU blocks */
661 static void purge_fragmented_blocks_allcpus(void);
662 
663 /*
664  * called before a call to iounmap() if the caller wants vm_area_struct's
665  * immediately freed.
666  */
667 void set_iounmap_nonlazy(void)
668 {
669         atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
670 }
671 
672 /*
673  * Purges all lazily-freed vmap areas.
674  */
675 static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
676 {
677         struct llist_node *valist;
678         struct vmap_area *va;
679         struct vmap_area *n_va;
680         bool do_free = false;
681 
682         lockdep_assert_held(&vmap_purge_lock);
683 
684         valist = llist_del_all(&vmap_purge_list);
685         llist_for_each_entry(va, valist, purge_list) {
686                 if (va->va_start < start)
687                         start = va->va_start;
688                 if (va->va_end > end)
689                         end = va->va_end;
690                 do_free = true;
691         }
692 
693         if (!do_free)
694                 return false;
695 
696         flush_tlb_kernel_range(start, end);
697 
698         spin_lock(&vmap_area_lock);
699         llist_for_each_entry_safe(va, n_va, valist, purge_list) {
700                 int nr = (va->va_end - va->va_start) >> PAGE_SHIFT;
701 
702                 __free_vmap_area(va);
703                 atomic_sub(nr, &vmap_lazy_nr);
704                 cond_resched_lock(&vmap_area_lock);
705         }
706         spin_unlock(&vmap_area_lock);
707         return true;
708 }
709 
710 /*
711  * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
712  * is already purging.
713  */
714 static void try_purge_vmap_area_lazy(void)
715 {
716         if (mutex_trylock(&vmap_purge_lock)) {
717                 __purge_vmap_area_lazy(ULONG_MAX, 0);
718                 mutex_unlock(&vmap_purge_lock);
719         }
720 }
721 
722 /*
723  * Kick off a purge of the outstanding lazy areas.
724  */
725 static void purge_vmap_area_lazy(void)
726 {
727         mutex_lock(&vmap_purge_lock);
728         purge_fragmented_blocks_allcpus();
729         __purge_vmap_area_lazy(ULONG_MAX, 0);
730         mutex_unlock(&vmap_purge_lock);
731 }
732 
733 /*
734  * Free a vmap area, caller ensuring that the area has been unmapped
735  * and flush_cache_vunmap had been called for the correct range
736  * previously.
737  */
738 static void free_vmap_area_noflush(struct vmap_area *va)
739 {
740         int nr_lazy;
741 
742         nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
743                                     &vmap_lazy_nr);
744 
745         /* After this point, we may free va at any time */
746         llist_add(&va->purge_list, &vmap_purge_list);
747 
748         if (unlikely(nr_lazy > lazy_max_pages()))
749                 try_purge_vmap_area_lazy();
750 }
751 
752 /*
753  * Free and unmap a vmap area
754  */
755 static void free_unmap_vmap_area(struct vmap_area *va)
756 {
757         flush_cache_vunmap(va->va_start, va->va_end);
758         unmap_vmap_area(va);
759         free_vmap_area_noflush(va);
760 }
761 
762 static struct vmap_area *find_vmap_area(unsigned long addr)
763 {
764         struct vmap_area *va;
765 
766         spin_lock(&vmap_area_lock);
767         va = __find_vmap_area(addr);
768         spin_unlock(&vmap_area_lock);
769 
770         return va;
771 }
772 
773 /*** Per cpu kva allocator ***/
774 
775 /*
776  * vmap space is limited especially on 32 bit architectures. Ensure there is
777  * room for at least 16 percpu vmap blocks per CPU.
778  */
779 /*
780  * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
781  * to #define VMALLOC_SPACE             (VMALLOC_END-VMALLOC_START). Guess
782  * instead (we just need a rough idea)
783  */
784 #if BITS_PER_LONG == 32
785 #define VMALLOC_SPACE           (128UL*1024*1024)
786 #else
787 #define VMALLOC_SPACE           (128UL*1024*1024*1024)
788 #endif
789 
790 #define VMALLOC_PAGES           (VMALLOC_SPACE / PAGE_SIZE)
791 #define VMAP_MAX_ALLOC          BITS_PER_LONG   /* 256K with 4K pages */
792 #define VMAP_BBMAP_BITS_MAX     1024    /* 4MB with 4K pages */
793 #define VMAP_BBMAP_BITS_MIN     (VMAP_MAX_ALLOC*2)
794 #define VMAP_MIN(x, y)          ((x) < (y) ? (x) : (y)) /* can't use min() */
795 #define VMAP_MAX(x, y)          ((x) > (y) ? (x) : (y)) /* can't use max() */
796 #define VMAP_BBMAP_BITS         \
797                 VMAP_MIN(VMAP_BBMAP_BITS_MAX,   \
798                 VMAP_MAX(VMAP_BBMAP_BITS_MIN,   \
799                         VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
800 
801 #define VMAP_BLOCK_SIZE         (VMAP_BBMAP_BITS * PAGE_SIZE)
802 
803 static bool vmap_initialized __read_mostly = false;
804 
805 struct vmap_block_queue {
806         spinlock_t lock;
807         struct list_head free;
808 };
809 
810 struct vmap_block {
811         spinlock_t lock;
812         struct vmap_area *va;
813         unsigned long free, dirty;
814         unsigned long dirty_min, dirty_max; /*< dirty range */
815         struct list_head free_list;
816         struct rcu_head rcu_head;
817         struct list_head purge;
818 };
819 
820 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
821 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
822 
823 /*
824  * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
825  * in the free path. Could get rid of this if we change the API to return a
826  * "cookie" from alloc, to be passed to free. But no big deal yet.
827  */
828 static DEFINE_SPINLOCK(vmap_block_tree_lock);
829 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
830 
831 /*
832  * We should probably have a fallback mechanism to allocate virtual memory
833  * out of partially filled vmap blocks. However vmap block sizing should be
834  * fairly reasonable according to the vmalloc size, so it shouldn't be a
835  * big problem.
836  */
837 
838 static unsigned long addr_to_vb_idx(unsigned long addr)
839 {
840         addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
841         addr /= VMAP_BLOCK_SIZE;
842         return addr;
843 }
844 
845 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
846 {
847         unsigned long addr;
848 
849         addr = va_start + (pages_off << PAGE_SHIFT);
850         BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
851         return (void *)addr;
852 }
853 
854 /**
855  * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
856  *                  block. Of course pages number can't exceed VMAP_BBMAP_BITS
857  * @order:    how many 2^order pages should be occupied in newly allocated block
858  * @gfp_mask: flags for the page level allocator
859  *
860  * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
861  */
862 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
863 {
864         struct vmap_block_queue *vbq;
865         struct vmap_block *vb;
866         struct vmap_area *va;
867         unsigned long vb_idx;
868         int node, err;
869         void *vaddr;
870 
871         node = numa_node_id();
872 
873         vb = kmalloc_node(sizeof(struct vmap_block),
874                         gfp_mask & GFP_RECLAIM_MASK, node);
875         if (unlikely(!vb))
876                 return ERR_PTR(-ENOMEM);
877 
878         va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
879                                         VMALLOC_START, VMALLOC_END,
880                                         node, gfp_mask);
881         if (IS_ERR(va)) {
882                 kfree(vb);
883                 return ERR_CAST(va);
884         }
885 
886         err = radix_tree_preload(gfp_mask);
887         if (unlikely(err)) {
888                 kfree(vb);
889                 free_vmap_area(va);
890                 return ERR_PTR(err);
891         }
892 
893         vaddr = vmap_block_vaddr(va->va_start, 0);
894         spin_lock_init(&vb->lock);
895         vb->va = va;
896         /* At least something should be left free */
897         BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
898         vb->free = VMAP_BBMAP_BITS - (1UL << order);
899         vb->dirty = 0;
900         vb->dirty_min = VMAP_BBMAP_BITS;
901         vb->dirty_max = 0;
902         INIT_LIST_HEAD(&vb->free_list);
903 
904         vb_idx = addr_to_vb_idx(va->va_start);
905         spin_lock(&vmap_block_tree_lock);
906         err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
907         spin_unlock(&vmap_block_tree_lock);
908         BUG_ON(err);
909         radix_tree_preload_end();
910 
911         vbq = &get_cpu_var(vmap_block_queue);
912         spin_lock(&vbq->lock);
913         list_add_tail_rcu(&vb->free_list, &vbq->free);
914         spin_unlock(&vbq->lock);
915         put_cpu_var(vmap_block_queue);
916 
917         return vaddr;
918 }
919 
920 static void free_vmap_block(struct vmap_block *vb)
921 {
922         struct vmap_block *tmp;
923         unsigned long vb_idx;
924 
925         vb_idx = addr_to_vb_idx(vb->va->va_start);
926         spin_lock(&vmap_block_tree_lock);
927         tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
928         spin_unlock(&vmap_block_tree_lock);
929         BUG_ON(tmp != vb);
930 
931         free_vmap_area_noflush(vb->va);
932         kfree_rcu(vb, rcu_head);
933 }
934 
935 static void purge_fragmented_blocks(int cpu)
936 {
937         LIST_HEAD(purge);
938         struct vmap_block *vb;
939         struct vmap_block *n_vb;
940         struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
941 
942         rcu_read_lock();
943         list_for_each_entry_rcu(vb, &vbq->free, free_list) {
944 
945                 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
946                         continue;
947 
948                 spin_lock(&vb->lock);
949                 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
950                         vb->free = 0; /* prevent further allocs after releasing lock */
951                         vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
952                         vb->dirty_min = 0;
953                         vb->dirty_max = VMAP_BBMAP_BITS;
954                         spin_lock(&vbq->lock);
955                         list_del_rcu(&vb->free_list);
956                         spin_unlock(&vbq->lock);
957                         spin_unlock(&vb->lock);
958                         list_add_tail(&vb->purge, &purge);
959                 } else
960                         spin_unlock(&vb->lock);
961         }
962         rcu_read_unlock();
963 
964         list_for_each_entry_safe(vb, n_vb, &purge, purge) {
965                 list_del(&vb->purge);
966                 free_vmap_block(vb);
967         }
968 }
969 
970 static void purge_fragmented_blocks_allcpus(void)
971 {
972         int cpu;
973 
974         for_each_possible_cpu(cpu)
975                 purge_fragmented_blocks(cpu);
976 }
977 
978 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
979 {
980         struct vmap_block_queue *vbq;
981         struct vmap_block *vb;
982         void *vaddr = NULL;
983         unsigned int order;
984 
985         BUG_ON(offset_in_page(size));
986         BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
987         if (WARN_ON(size == 0)) {
988                 /*
989                  * Allocating 0 bytes isn't what caller wants since
990                  * get_order(0) returns funny result. Just warn and terminate
991                  * early.
992                  */
993                 return NULL;
994         }
995         order = get_order(size);
996 
997         rcu_read_lock();
998         vbq = &get_cpu_var(vmap_block_queue);
999         list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1000                 unsigned long pages_off;
1001 
1002                 spin_lock(&vb->lock);
1003                 if (vb->free < (1UL << order)) {
1004                         spin_unlock(&vb->lock);
1005                         continue;
1006                 }
1007 
1008                 pages_off = VMAP_BBMAP_BITS - vb->free;
1009                 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
1010                 vb->free -= 1UL << order;
1011                 if (vb->free == 0) {
1012                         spin_lock(&vbq->lock);
1013                         list_del_rcu(&vb->free_list);
1014                         spin_unlock(&vbq->lock);
1015                 }
1016 
1017                 spin_unlock(&vb->lock);
1018                 break;
1019         }
1020 
1021         put_cpu_var(vmap_block_queue);
1022         rcu_read_unlock();
1023 
1024         /* Allocate new block if nothing was found */
1025         if (!vaddr)
1026                 vaddr = new_vmap_block(order, gfp_mask);
1027 
1028         return vaddr;
1029 }
1030 
1031 static void vb_free(const void *addr, unsigned long size)
1032 {
1033         unsigned long offset;
1034         unsigned long vb_idx;
1035         unsigned int order;
1036         struct vmap_block *vb;
1037 
1038         BUG_ON(offset_in_page(size));
1039         BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
1040 
1041         flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1042 
1043         order = get_order(size);
1044 
1045         offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
1046         offset >>= PAGE_SHIFT;
1047 
1048         vb_idx = addr_to_vb_idx((unsigned long)addr);
1049         rcu_read_lock();
1050         vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1051         rcu_read_unlock();
1052         BUG_ON(!vb);
1053 
1054         vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1055 
1056         spin_lock(&vb->lock);
1057 
1058         /* Expand dirty range */
1059         vb->dirty_min = min(vb->dirty_min, offset);
1060         vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
1061 
1062         vb->dirty += 1UL << order;
1063         if (vb->dirty == VMAP_BBMAP_BITS) {
1064                 BUG_ON(vb->free);
1065                 spin_unlock(&vb->lock);
1066                 free_vmap_block(vb);
1067         } else
1068                 spin_unlock(&vb->lock);
1069 }
1070 
1071 /**
1072  * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1073  *
1074  * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1075  * to amortize TLB flushing overheads. What this means is that any page you
1076  * have now, may, in a former life, have been mapped into kernel virtual
1077  * address by the vmap layer and so there might be some CPUs with TLB entries
1078  * still referencing that page (additional to the regular 1:1 kernel mapping).
1079  *
1080  * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1081  * be sure that none of the pages we have control over will have any aliases
1082  * from the vmap layer.
1083  */
1084 void vm_unmap_aliases(void)
1085 {
1086         unsigned long start = ULONG_MAX, end = 0;
1087         int cpu;
1088         int flush = 0;
1089 
1090         if (unlikely(!vmap_initialized))
1091                 return;
1092 
1093         might_sleep();
1094 
1095         for_each_possible_cpu(cpu) {
1096                 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1097                 struct vmap_block *vb;
1098 
1099                 rcu_read_lock();
1100                 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1101                         spin_lock(&vb->lock);
1102                         if (vb->dirty) {
1103                                 unsigned long va_start = vb->va->va_start;
1104                                 unsigned long s, e;
1105 
1106                                 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1107                                 e = va_start + (vb->dirty_max << PAGE_SHIFT);
1108 
1109                                 start = min(s, start);
1110                                 end   = max(e, end);
1111 
1112                                 flush = 1;
1113                         }
1114                         spin_unlock(&vb->lock);
1115                 }
1116                 rcu_read_unlock();
1117         }
1118 
1119         mutex_lock(&vmap_purge_lock);
1120         purge_fragmented_blocks_allcpus();
1121         if (!__purge_vmap_area_lazy(start, end) && flush)
1122                 flush_tlb_kernel_range(start, end);
1123         mutex_unlock(&vmap_purge_lock);
1124 }
1125 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1126 
1127 /**
1128  * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1129  * @mem: the pointer returned by vm_map_ram
1130  * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1131  */
1132 void vm_unmap_ram(const void *mem, unsigned int count)
1133 {
1134         unsigned long size = (unsigned long)count << PAGE_SHIFT;
1135         unsigned long addr = (unsigned long)mem;
1136         struct vmap_area *va;
1137 
1138         might_sleep();
1139         BUG_ON(!addr);
1140         BUG_ON(addr < VMALLOC_START);
1141         BUG_ON(addr > VMALLOC_END);
1142         BUG_ON(!PAGE_ALIGNED(addr));
1143 
1144         debug_check_no_locks_freed(mem, size);
1145         vmap_debug_free_range(addr, addr+size);
1146 
1147         if (likely(count <= VMAP_MAX_ALLOC)) {
1148                 vb_free(mem, size);
1149                 return;
1150         }
1151 
1152         va = find_vmap_area(addr);
1153         BUG_ON(!va);
1154         free_unmap_vmap_area(va);
1155 }
1156 EXPORT_SYMBOL(vm_unmap_ram);
1157 
1158 /**
1159  * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1160  * @pages: an array of pointers to the pages to be mapped
1161  * @count: number of pages
1162  * @node: prefer to allocate data structures on this node
1163  * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1164  *
1165  * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1166  * faster than vmap so it's good.  But if you mix long-life and short-life
1167  * objects with vm_map_ram(), it could consume lots of address space through
1168  * fragmentation (especially on a 32bit machine).  You could see failures in
1169  * the end.  Please use this function for short-lived objects.
1170  *
1171  * Returns: a pointer to the address that has been mapped, or %NULL on failure
1172  */
1173 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1174 {
1175         unsigned long size = (unsigned long)count << PAGE_SHIFT;
1176         unsigned long addr;
1177         void *mem;
1178 
1179         if (likely(count <= VMAP_MAX_ALLOC)) {
1180                 mem = vb_alloc(size, GFP_KERNEL);
1181                 if (IS_ERR(mem))
1182                         return NULL;
1183                 addr = (unsigned long)mem;
1184         } else {
1185                 struct vmap_area *va;
1186                 va = alloc_vmap_area(size, PAGE_SIZE,
1187                                 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1188                 if (IS_ERR(va))
1189                         return NULL;
1190 
1191                 addr = va->va_start;
1192                 mem = (void *)addr;
1193         }
1194         if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1195                 vm_unmap_ram(mem, count);
1196                 return NULL;
1197         }
1198         return mem;
1199 }
1200 EXPORT_SYMBOL(vm_map_ram);
1201 
1202 static struct vm_struct *vmlist __initdata;
1203 /**
1204  * vm_area_add_early - add vmap area early during boot
1205  * @vm: vm_struct to add
1206  *
1207  * This function is used to add fixed kernel vm area to vmlist before
1208  * vmalloc_init() is called.  @vm->addr, @vm->size, and @vm->flags
1209  * should contain proper values and the other fields should be zero.
1210  *
1211  * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1212  */
1213 void __init vm_area_add_early(struct vm_struct *vm)
1214 {
1215         struct vm_struct *tmp, **p;
1216 
1217         BUG_ON(vmap_initialized);
1218         for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1219                 if (tmp->addr >= vm->addr) {
1220                         BUG_ON(tmp->addr < vm->addr + vm->size);
1221                         break;
1222                 } else
1223                         BUG_ON(tmp->addr + tmp->size > vm->addr);
1224         }
1225         vm->next = *p;
1226         *p = vm;
1227 }
1228 
1229 /**
1230  * vm_area_register_early - register vmap area early during boot
1231  * @vm: vm_struct to register
1232  * @align: requested alignment
1233  *
1234  * This function is used to register kernel vm area before
1235  * vmalloc_init() is called.  @vm->size and @vm->flags should contain
1236  * proper values on entry and other fields should be zero.  On return,
1237  * vm->addr contains the allocated address.
1238  *
1239  * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1240  */
1241 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1242 {
1243         static size_t vm_init_off __initdata;
1244         unsigned long addr;
1245 
1246         addr = ALIGN(VMALLOC_START + vm_init_off, align);
1247         vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1248 
1249         vm->addr = (void *)addr;
1250 
1251         vm_area_add_early(vm);
1252 }
1253 
1254 void __init vmalloc_init(void)
1255 {
1256         struct vmap_area *va;
1257         struct vm_struct *tmp;
1258         int i;
1259 
1260         for_each_possible_cpu(i) {
1261                 struct vmap_block_queue *vbq;
1262                 struct vfree_deferred *p;
1263 
1264                 vbq = &per_cpu(vmap_block_queue, i);
1265                 spin_lock_init(&vbq->lock);
1266                 INIT_LIST_HEAD(&vbq->free);
1267                 p = &per_cpu(vfree_deferred, i);
1268                 init_llist_head(&p->list);
1269                 INIT_WORK(&p->wq, free_work);
1270         }
1271 
1272         /* Import existing vmlist entries. */
1273         for (tmp = vmlist; tmp; tmp = tmp->next) {
1274                 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1275                 va->flags = VM_VM_AREA;
1276                 va->va_start = (unsigned long)tmp->addr;
1277                 va->va_end = va->va_start + tmp->size;
1278                 va->vm = tmp;
1279                 __insert_vmap_area(va);
1280         }
1281 
1282         vmap_area_pcpu_hole = VMALLOC_END;
1283 
1284         vmap_initialized = true;
1285 }
1286 
1287 /**
1288  * map_kernel_range_noflush - map kernel VM area with the specified pages
1289  * @addr: start of the VM area to map
1290  * @size: size of the VM area to map
1291  * @prot: page protection flags to use
1292  * @pages: pages to map
1293  *
1294  * Map PFN_UP(@size) pages at @addr.  The VM area @addr and @size
1295  * specify should have been allocated using get_vm_area() and its
1296  * friends.
1297  *
1298  * NOTE:
1299  * This function does NOT do any cache flushing.  The caller is
1300  * responsible for calling flush_cache_vmap() on to-be-mapped areas
1301  * before calling this function.
1302  *
1303  * RETURNS:
1304  * The number of pages mapped on success, -errno on failure.
1305  */
1306 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1307                              pgprot_t prot, struct page **pages)
1308 {
1309         return vmap_page_range_noflush(addr, addr + size, prot, pages);
1310 }
1311 
1312 /**
1313  * unmap_kernel_range_noflush - unmap kernel VM area
1314  * @addr: start of the VM area to unmap
1315  * @size: size of the VM area to unmap
1316  *
1317  * Unmap PFN_UP(@size) pages at @addr.  The VM area @addr and @size
1318  * specify should have been allocated using get_vm_area() and its
1319  * friends.
1320  *
1321  * NOTE:
1322  * This function does NOT do any cache flushing.  The caller is
1323  * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1324  * before calling this function and flush_tlb_kernel_range() after.
1325  */
1326 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1327 {
1328         vunmap_page_range(addr, addr + size);
1329 }
1330 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1331 
1332 /**
1333  * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1334  * @addr: start of the VM area to unmap
1335  * @size: size of the VM area to unmap
1336  *
1337  * Similar to unmap_kernel_range_noflush() but flushes vcache before
1338  * the unmapping and tlb after.
1339  */
1340 void unmap_kernel_range(unsigned long addr, unsigned long size)
1341 {
1342         unsigned long end = addr + size;
1343 
1344         flush_cache_vunmap(addr, end);
1345         vunmap_page_range(addr, end);
1346         flush_tlb_kernel_range(addr, end);
1347 }
1348 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1349 
1350 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1351 {
1352         unsigned long addr = (unsigned long)area->addr;
1353         unsigned long end = addr + get_vm_area_size(area);
1354         int err;
1355 
1356         err = vmap_page_range(addr, end, prot, pages);
1357 
1358         return err > 0 ? 0 : err;
1359 }
1360 EXPORT_SYMBOL_GPL(map_vm_area);
1361 
1362 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1363                               unsigned long flags, const void *caller)
1364 {
1365         spin_lock(&vmap_area_lock);
1366         vm->flags = flags;
1367         vm->addr = (void *)va->va_start;
1368         vm->size = va->va_end - va->va_start;
1369         vm->caller = caller;
1370         va->vm = vm;
1371         va->flags |= VM_VM_AREA;
1372         spin_unlock(&vmap_area_lock);
1373 }
1374 
1375 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1376 {
1377         /*
1378          * Before removing VM_UNINITIALIZED,
1379          * we should make sure that vm has proper values.
1380          * Pair with smp_rmb() in show_numa_info().
1381          */
1382         smp_wmb();
1383         vm->flags &= ~VM_UNINITIALIZED;
1384 }
1385 
1386 static struct vm_struct *__get_vm_area_node(unsigned long size,
1387                 unsigned long align, unsigned long flags, unsigned long start,
1388                 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1389 {
1390         struct vmap_area *va;
1391         struct vm_struct *area;
1392 
1393         BUG_ON(in_interrupt());
1394         size = PAGE_ALIGN(size);
1395         if (unlikely(!size))
1396                 return NULL;
1397 
1398         if (flags & VM_IOREMAP)
1399                 align = 1ul << clamp_t(int, get_count_order_long(size),
1400                                        PAGE_SHIFT, IOREMAP_MAX_ORDER);
1401 
1402         area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1403         if (unlikely(!area))
1404                 return NULL;
1405 
1406         if (!(flags & VM_NO_GUARD))
1407                 size += PAGE_SIZE;
1408 
1409         va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1410         if (IS_ERR(va)) {
1411                 kfree(area);
1412                 return NULL;
1413         }
1414 
1415         setup_vmalloc_vm(area, va, flags, caller);
1416 
1417         return area;
1418 }
1419 
1420 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1421                                 unsigned long start, unsigned long end)
1422 {
1423         return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1424                                   GFP_KERNEL, __builtin_return_address(0));
1425 }
1426 EXPORT_SYMBOL_GPL(__get_vm_area);
1427 
1428 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1429                                        unsigned long start, unsigned long end,
1430                                        const void *caller)
1431 {
1432         return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1433                                   GFP_KERNEL, caller);
1434 }
1435 
1436 /**
1437  *      get_vm_area  -  reserve a contiguous kernel virtual area
1438  *      @size:          size of the area
1439  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
1440  *
1441  *      Search an area of @size in the kernel virtual mapping area,
1442  *      and reserved it for out purposes.  Returns the area descriptor
1443  *      on success or %NULL on failure.
1444  */
1445 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1446 {
1447         return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1448                                   NUMA_NO_NODE, GFP_KERNEL,
1449                                   __builtin_return_address(0));
1450 }
1451 
1452 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1453                                 const void *caller)
1454 {
1455         return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1456                                   NUMA_NO_NODE, GFP_KERNEL, caller);
1457 }
1458 
1459 /**
1460  *      find_vm_area  -  find a continuous kernel virtual area
1461  *      @addr:          base address
1462  *
1463  *      Search for the kernel VM area starting at @addr, and return it.
1464  *      It is up to the caller to do all required locking to keep the returned
1465  *      pointer valid.
1466  */
1467 struct vm_struct *find_vm_area(const void *addr)
1468 {
1469         struct vmap_area *va;
1470 
1471         va = find_vmap_area((unsigned long)addr);
1472         if (va && va->flags & VM_VM_AREA)
1473                 return va->vm;
1474 
1475         return NULL;
1476 }
1477 
1478 /**
1479  *      remove_vm_area  -  find and remove a continuous kernel virtual area
1480  *      @addr:          base address
1481  *
1482  *      Search for the kernel VM area starting at @addr, and remove it.
1483  *      This function returns the found VM area, but using it is NOT safe
1484  *      on SMP machines, except for its size or flags.
1485  */
1486 struct vm_struct *remove_vm_area(const void *addr)
1487 {
1488         struct vmap_area *va;
1489 
1490         might_sleep();
1491 
1492         va = find_vmap_area((unsigned long)addr);
1493         if (va && va->flags & VM_VM_AREA) {
1494                 struct vm_struct *vm = va->vm;
1495 
1496                 spin_lock(&vmap_area_lock);
1497                 va->vm = NULL;
1498                 va->flags &= ~VM_VM_AREA;
1499                 va->flags |= VM_LAZY_FREE;
1500                 spin_unlock(&vmap_area_lock);
1501 
1502                 vmap_debug_free_range(va->va_start, va->va_end);
1503                 kasan_free_shadow(vm);
1504                 free_unmap_vmap_area(va);
1505 
1506                 return vm;
1507         }
1508         return NULL;
1509 }
1510 
1511 static void __vunmap(const void *addr, int deallocate_pages)
1512 {
1513         struct vm_struct *area;
1514 
1515         if (!addr)
1516                 return;
1517 
1518         if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1519                         addr))
1520                 return;
1521 
1522         area = find_vmap_area((unsigned long)addr)->vm;
1523         if (unlikely(!area)) {
1524                 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1525                                 addr);
1526                 return;
1527         }
1528 
1529         debug_check_no_locks_freed(addr, get_vm_area_size(area));
1530         debug_check_no_obj_freed(addr, get_vm_area_size(area));
1531 
1532         remove_vm_area(addr);
1533         if (deallocate_pages) {
1534                 int i;
1535 
1536                 for (i = 0; i < area->nr_pages; i++) {
1537                         struct page *page = area->pages[i];
1538 
1539                         BUG_ON(!page);
1540                         __free_pages(page, 0);
1541                 }
1542 
1543                 kvfree(area->pages);
1544         }
1545 
1546         kfree(area);
1547         return;
1548 }
1549 
1550 static inline void __vfree_deferred(const void *addr)
1551 {
1552         /*
1553          * Use raw_cpu_ptr() because this can be called from preemptible
1554          * context. Preemption is absolutely fine here, because the llist_add()
1555          * implementation is lockless, so it works even if we are adding to
1556          * nother cpu's list.  schedule_work() should be fine with this too.
1557          */
1558         struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
1559 
1560         if (llist_add((struct llist_node *)addr, &p->list))
1561                 schedule_work(&p->wq);
1562 }
1563 
1564 /**
1565  *      vfree_atomic  -  release memory allocated by vmalloc()
1566  *      @addr:          memory base address
1567  *
1568  *      This one is just like vfree() but can be called in any atomic context
1569  *      except NMIs.
1570  */
1571 void vfree_atomic(const void *addr)
1572 {
1573         BUG_ON(in_nmi());
1574 
1575         kmemleak_free(addr);
1576 
1577         if (!addr)
1578                 return;
1579         __vfree_deferred(addr);
1580 }
1581 
1582 /**
1583  *      vfree  -  release memory allocated by vmalloc()
1584  *      @addr:          memory base address
1585  *
1586  *      Free the virtually continuous memory area starting at @addr, as
1587  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1588  *      NULL, no operation is performed.
1589  *
1590  *      Must not be called in NMI context (strictly speaking, only if we don't
1591  *      have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1592  *      conventions for vfree() arch-depenedent would be a really bad idea)
1593  *
1594  *      NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
1595  */
1596 void vfree(const void *addr)
1597 {
1598         BUG_ON(in_nmi());
1599 
1600         kmemleak_free(addr);
1601 
1602         if (!addr)
1603                 return;
1604         if (unlikely(in_interrupt()))
1605                 __vfree_deferred(addr);
1606         else
1607                 __vunmap(addr, 1);
1608 }
1609 EXPORT_SYMBOL(vfree);
1610 
1611 /**
1612  *      vunmap  -  release virtual mapping obtained by vmap()
1613  *      @addr:          memory base address
1614  *
1615  *      Free the virtually contiguous memory area starting at @addr,
1616  *      which was created from the page array passed to vmap().
1617  *
1618  *      Must not be called in interrupt context.
1619  */
1620 void vunmap(const void *addr)
1621 {
1622         BUG_ON(in_interrupt());
1623         might_sleep();
1624         if (addr)
1625                 __vunmap(addr, 0);
1626 }
1627 EXPORT_SYMBOL(vunmap);
1628 
1629 /**
1630  *      vmap  -  map an array of pages into virtually contiguous space
1631  *      @pages:         array of page pointers
1632  *      @count:         number of pages to map
1633  *      @flags:         vm_area->flags
1634  *      @prot:          page protection for the mapping
1635  *
1636  *      Maps @count pages from @pages into contiguous kernel virtual
1637  *      space.
1638  */
1639 void *vmap(struct page **pages, unsigned int count,
1640                 unsigned long flags, pgprot_t prot)
1641 {
1642         struct vm_struct *area;
1643         unsigned long size;             /* In bytes */
1644 
1645         might_sleep();
1646 
1647         if (count > totalram_pages)
1648                 return NULL;
1649 
1650         size = (unsigned long)count << PAGE_SHIFT;
1651         area = get_vm_area_caller(size, flags, __builtin_return_address(0));
1652         if (!area)
1653                 return NULL;
1654 
1655         if (map_vm_area(area, prot, pages)) {
1656                 vunmap(area->addr);
1657                 return NULL;
1658         }
1659 
1660         return area->addr;
1661 }
1662 EXPORT_SYMBOL(vmap);
1663 
1664 static void *__vmalloc_node(unsigned long size, unsigned long align,
1665                             gfp_t gfp_mask, pgprot_t prot,
1666                             int node, const void *caller);
1667 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1668                                  pgprot_t prot, int node)
1669 {
1670         struct page **pages;
1671         unsigned int nr_pages, array_size, i;
1672         const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1673         const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1674         const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ?
1675                                         0 :
1676                                         __GFP_HIGHMEM;
1677 
1678         nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1679         array_size = (nr_pages * sizeof(struct page *));
1680 
1681         area->nr_pages = nr_pages;
1682         /* Please note that the recursion is strictly bounded. */
1683         if (array_size > PAGE_SIZE) {
1684                 pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask,
1685                                 PAGE_KERNEL, node, area->caller);
1686         } else {
1687                 pages = kmalloc_node(array_size, nested_gfp, node);
1688         }
1689         area->pages = pages;
1690         if (!area->pages) {
1691                 remove_vm_area(area->addr);
1692                 kfree(area);
1693                 return NULL;
1694         }
1695 
1696         for (i = 0; i < area->nr_pages; i++) {
1697                 struct page *page;
1698 
1699                 if (node == NUMA_NO_NODE)
1700                         page = alloc_page(alloc_mask|highmem_mask);
1701                 else
1702                         page = alloc_pages_node(node, alloc_mask|highmem_mask, 0);
1703 
1704                 if (unlikely(!page)) {
1705                         /* Successfully allocated i pages, free them in __vunmap() */
1706                         area->nr_pages = i;
1707                         goto fail;
1708                 }
1709                 area->pages[i] = page;
1710                 if (gfpflags_allow_blocking(gfp_mask|highmem_mask))
1711                         cond_resched();
1712         }
1713 
1714         if (map_vm_area(area, prot, pages))
1715                 goto fail;
1716         return area->addr;
1717 
1718 fail:
1719         warn_alloc(gfp_mask, NULL,
1720                           "vmalloc: allocation failure, allocated %ld of %ld bytes",
1721                           (area->nr_pages*PAGE_SIZE), area->size);
1722         vfree(area->addr);
1723         return NULL;
1724 }
1725 
1726 /**
1727  *      __vmalloc_node_range  -  allocate virtually contiguous memory
1728  *      @size:          allocation size
1729  *      @align:         desired alignment
1730  *      @start:         vm area range start
1731  *      @end:           vm area range end
1732  *      @gfp_mask:      flags for the page level allocator
1733  *      @prot:          protection mask for the allocated pages
1734  *      @vm_flags:      additional vm area flags (e.g. %VM_NO_GUARD)
1735  *      @node:          node to use for allocation or NUMA_NO_NODE
1736  *      @caller:        caller's return address
1737  *
1738  *      Allocate enough pages to cover @size from the page level
1739  *      allocator with @gfp_mask flags.  Map them into contiguous
1740  *      kernel virtual space, using a pagetable protection of @prot.
1741  */
1742 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1743                         unsigned long start, unsigned long end, gfp_t gfp_mask,
1744                         pgprot_t prot, unsigned long vm_flags, int node,
1745                         const void *caller)
1746 {
1747         struct vm_struct *area;
1748         void *addr;
1749         unsigned long real_size = size;
1750 
1751         size = PAGE_ALIGN(size);
1752         if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1753                 goto fail;
1754 
1755         area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1756                                 vm_flags, start, end, node, gfp_mask, caller);
1757         if (!area)
1758                 goto fail;
1759 
1760         addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1761         if (!addr)
1762                 return NULL;
1763 
1764         /*
1765          * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1766          * flag. It means that vm_struct is not fully initialized.
1767          * Now, it is fully initialized, so remove this flag here.
1768          */
1769         clear_vm_uninitialized_flag(area);
1770 
1771         kmemleak_vmalloc(area, size, gfp_mask);
1772 
1773         return addr;
1774 
1775 fail:
1776         warn_alloc(gfp_mask, NULL,
1777                           "vmalloc: allocation failure: %lu bytes", real_size);
1778         return NULL;
1779 }
1780 
1781 /**
1782  *      __vmalloc_node  -  allocate virtually contiguous memory
1783  *      @size:          allocation size
1784  *      @align:         desired alignment
1785  *      @gfp_mask:      flags for the page level allocator
1786  *      @prot:          protection mask for the allocated pages
1787  *      @node:          node to use for allocation or NUMA_NO_NODE
1788  *      @caller:        caller's return address
1789  *
1790  *      Allocate enough pages to cover @size from the page level
1791  *      allocator with @gfp_mask flags.  Map them into contiguous
1792  *      kernel virtual space, using a pagetable protection of @prot.
1793  *
1794  *      Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL
1795  *      and __GFP_NOFAIL are not supported
1796  *
1797  *      Any use of gfp flags outside of GFP_KERNEL should be consulted
1798  *      with mm people.
1799  *
1800  */
1801 static void *__vmalloc_node(unsigned long size, unsigned long align,
1802                             gfp_t gfp_mask, pgprot_t prot,
1803                             int node, const void *caller)
1804 {
1805         return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1806                                 gfp_mask, prot, 0, node, caller);
1807 }
1808 
1809 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1810 {
1811         return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1812                                 __builtin_return_address(0));
1813 }
1814 EXPORT_SYMBOL(__vmalloc);
1815 
1816 static inline void *__vmalloc_node_flags(unsigned long size,
1817                                         int node, gfp_t flags)
1818 {
1819         return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1820                                         node, __builtin_return_address(0));
1821 }
1822 
1823 
1824 void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags,
1825                                   void *caller)
1826 {
1827         return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller);
1828 }
1829 
1830 /**
1831  *      vmalloc  -  allocate virtually contiguous memory
1832  *      @size:          allocation size
1833  *      Allocate enough pages to cover @size from the page level
1834  *      allocator and map them into contiguous kernel virtual space.
1835  *
1836  *      For tight control over page level allocator and protection flags
1837  *      use __vmalloc() instead.
1838  */
1839 void *vmalloc(unsigned long size)
1840 {
1841         return __vmalloc_node_flags(size, NUMA_NO_NODE,
1842                                     GFP_KERNEL);
1843 }
1844 EXPORT_SYMBOL(vmalloc);
1845 
1846 /**
1847  *      vzalloc - allocate virtually contiguous memory with zero fill
1848  *      @size:  allocation size
1849  *      Allocate enough pages to cover @size from the page level
1850  *      allocator and map them into contiguous kernel virtual space.
1851  *      The memory allocated is set to zero.
1852  *
1853  *      For tight control over page level allocator and protection flags
1854  *      use __vmalloc() instead.
1855  */
1856 void *vzalloc(unsigned long size)
1857 {
1858         return __vmalloc_node_flags(size, NUMA_NO_NODE,
1859                                 GFP_KERNEL | __GFP_ZERO);
1860 }
1861 EXPORT_SYMBOL(vzalloc);
1862 
1863 /**
1864  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1865  * @size: allocation size
1866  *
1867  * The resulting memory area is zeroed so it can be mapped to userspace
1868  * without leaking data.
1869  */
1870 void *vmalloc_user(unsigned long size)
1871 {
1872         struct vm_struct *area;
1873         void *ret;
1874 
1875         ret = __vmalloc_node(size, SHMLBA,
1876                              GFP_KERNEL | __GFP_ZERO,
1877                              PAGE_KERNEL, NUMA_NO_NODE,
1878                              __builtin_return_address(0));
1879         if (ret) {
1880                 area = find_vm_area(ret);
1881                 area->flags |= VM_USERMAP;
1882         }
1883         return ret;
1884 }
1885 EXPORT_SYMBOL(vmalloc_user);
1886 
1887 /**
1888  *      vmalloc_node  -  allocate memory on a specific node
1889  *      @size:          allocation size
1890  *      @node:          numa node
1891  *
1892  *      Allocate enough pages to cover @size from the page level
1893  *      allocator and map them into contiguous kernel virtual space.
1894  *
1895  *      For tight control over page level allocator and protection flags
1896  *      use __vmalloc() instead.
1897  */
1898 void *vmalloc_node(unsigned long size, int node)
1899 {
1900         return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL,
1901                                         node, __builtin_return_address(0));
1902 }
1903 EXPORT_SYMBOL(vmalloc_node);
1904 
1905 /**
1906  * vzalloc_node - allocate memory on a specific node with zero fill
1907  * @size:       allocation size
1908  * @node:       numa node
1909  *
1910  * Allocate enough pages to cover @size from the page level
1911  * allocator and map them into contiguous kernel virtual space.
1912  * The memory allocated is set to zero.
1913  *
1914  * For tight control over page level allocator and protection flags
1915  * use __vmalloc_node() instead.
1916  */
1917 void *vzalloc_node(unsigned long size, int node)
1918 {
1919         return __vmalloc_node_flags(size, node,
1920                          GFP_KERNEL | __GFP_ZERO);
1921 }
1922 EXPORT_SYMBOL(vzalloc_node);
1923 
1924 #ifndef PAGE_KERNEL_EXEC
1925 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1926 #endif
1927 
1928 /**
1929  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
1930  *      @size:          allocation size
1931  *
1932  *      Kernel-internal function to allocate enough pages to cover @size
1933  *      the page level allocator and map them into contiguous and
1934  *      executable kernel virtual space.
1935  *
1936  *      For tight control over page level allocator and protection flags
1937  *      use __vmalloc() instead.
1938  */
1939 
1940 void *vmalloc_exec(unsigned long size)
1941 {
1942         return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL_EXEC,
1943                               NUMA_NO_NODE, __builtin_return_address(0));
1944 }
1945 
1946 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1947 #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL)
1948 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1949 #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL)
1950 #else
1951 /*
1952  * 64b systems should always have either DMA or DMA32 zones. For others
1953  * GFP_DMA32 should do the right thing and use the normal zone.
1954  */
1955 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1956 #endif
1957 
1958 /**
1959  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
1960  *      @size:          allocation size
1961  *
1962  *      Allocate enough 32bit PA addressable pages to cover @size from the
1963  *      page level allocator and map them into contiguous kernel virtual space.
1964  */
1965 void *vmalloc_32(unsigned long size)
1966 {
1967         return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1968                               NUMA_NO_NODE, __builtin_return_address(0));
1969 }
1970 EXPORT_SYMBOL(vmalloc_32);
1971 
1972 /**
1973  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1974  *      @size:          allocation size
1975  *
1976  * The resulting memory area is 32bit addressable and zeroed so it can be
1977  * mapped to userspace without leaking data.
1978  */
1979 void *vmalloc_32_user(unsigned long size)
1980 {
1981         struct vm_struct *area;
1982         void *ret;
1983 
1984         ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1985                              NUMA_NO_NODE, __builtin_return_address(0));
1986         if (ret) {
1987                 area = find_vm_area(ret);
1988                 area->flags |= VM_USERMAP;
1989         }
1990         return ret;
1991 }
1992 EXPORT_SYMBOL(vmalloc_32_user);
1993 
1994 /*
1995  * small helper routine , copy contents to buf from addr.
1996  * If the page is not present, fill zero.
1997  */
1998 
1999 static int aligned_vread(char *buf, char *addr, unsigned long count)
2000 {
2001         struct page *p;
2002         int copied = 0;
2003 
2004         while (count) {
2005                 unsigned long offset, length;
2006 
2007                 offset = offset_in_page(addr);
2008                 length = PAGE_SIZE - offset;
2009                 if (length > count)
2010                         length = count;
2011                 p = vmalloc_to_page(addr);
2012                 /*
2013                  * To do safe access to this _mapped_ area, we need
2014                  * lock. But adding lock here means that we need to add
2015                  * overhead of vmalloc()/vfree() calles for this _debug_
2016                  * interface, rarely used. Instead of that, we'll use
2017                  * kmap() and get small overhead in this access function.
2018                  */
2019                 if (p) {
2020                         /*
2021                          * we can expect USER0 is not used (see vread/vwrite's
2022                          * function description)
2023                          */
2024                         void *map = kmap_atomic(p);
2025                         memcpy(buf, map + offset, length);
2026                         kunmap_atomic(map);
2027                 } else
2028                         memset(buf, 0, length);
2029 
2030                 addr += length;
2031                 buf += length;
2032                 copied += length;
2033                 count -= length;
2034         }
2035         return copied;
2036 }
2037 
2038 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
2039 {
2040         struct page *p;
2041         int copied = 0;
2042 
2043         while (count) {
2044                 unsigned long offset, length;
2045 
2046                 offset = offset_in_page(addr);
2047                 length = PAGE_SIZE - offset;
2048                 if (length > count)
2049                         length = count;
2050                 p = vmalloc_to_page(addr);
2051                 /*
2052                  * To do safe access to this _mapped_ area, we need
2053                  * lock. But adding lock here means that we need to add
2054                  * overhead of vmalloc()/vfree() calles for this _debug_
2055                  * interface, rarely used. Instead of that, we'll use
2056                  * kmap() and get small overhead in this access function.
2057                  */
2058                 if (p) {
2059                         /*
2060                          * we can expect USER0 is not used (see vread/vwrite's
2061                          * function description)
2062                          */
2063                         void *map = kmap_atomic(p);
2064                         memcpy(map + offset, buf, length);
2065                         kunmap_atomic(map);
2066                 }
2067                 addr += length;
2068                 buf += length;
2069                 copied += length;
2070                 count -= length;
2071         }
2072         return copied;
2073 }
2074 
2075 /**
2076  *      vread() -  read vmalloc area in a safe way.
2077  *      @buf:           buffer for reading data
2078  *      @addr:          vm address.
2079  *      @count:         number of bytes to be read.
2080  *
2081  *      Returns # of bytes which addr and buf should be increased.
2082  *      (same number to @count). Returns 0 if [addr...addr+count) doesn't
2083  *      includes any intersect with alive vmalloc area.
2084  *
2085  *      This function checks that addr is a valid vmalloc'ed area, and
2086  *      copy data from that area to a given buffer. If the given memory range
2087  *      of [addr...addr+count) includes some valid address, data is copied to
2088  *      proper area of @buf. If there are memory holes, they'll be zero-filled.
2089  *      IOREMAP area is treated as memory hole and no copy is done.
2090  *
2091  *      If [addr...addr+count) doesn't includes any intersects with alive
2092  *      vm_struct area, returns 0. @buf should be kernel's buffer.
2093  *
2094  *      Note: In usual ops, vread() is never necessary because the caller
2095  *      should know vmalloc() area is valid and can use memcpy().
2096  *      This is for routines which have to access vmalloc area without
2097  *      any informaion, as /dev/kmem.
2098  *
2099  */
2100 
2101 long vread(char *buf, char *addr, unsigned long count)
2102 {
2103         struct vmap_area *va;
2104         struct vm_struct *vm;
2105         char *vaddr, *buf_start = buf;
2106         unsigned long buflen = count;
2107         unsigned long n;
2108 
2109         /* Don't allow overflow */
2110         if ((unsigned long) addr + count < count)
2111                 count = -(unsigned long) addr;
2112 
2113         spin_lock(&vmap_area_lock);
2114         list_for_each_entry(va, &vmap_area_list, list) {
2115                 if (!count)
2116                         break;
2117 
2118                 if (!(va->flags & VM_VM_AREA))
2119                         continue;
2120 
2121                 vm = va->vm;
2122                 vaddr = (char *) vm->addr;
2123                 if (addr >= vaddr + get_vm_area_size(vm))
2124                         continue;
2125                 while (addr < vaddr) {
2126                         if (count == 0)
2127                                 goto finished;
2128                         *buf = '\0';
2129                         buf++;
2130                         addr++;
2131                         count--;
2132                 }
2133                 n = vaddr + get_vm_area_size(vm) - addr;
2134                 if (n > count)
2135                         n = count;
2136                 if (!(vm->flags & VM_IOREMAP))
2137                         aligned_vread(buf, addr, n);
2138                 else /* IOREMAP area is treated as memory hole */
2139                         memset(buf, 0, n);
2140                 buf += n;
2141                 addr += n;
2142                 count -= n;
2143         }
2144 finished:
2145         spin_unlock(&vmap_area_lock);
2146 
2147         if (buf == buf_start)
2148                 return 0;
2149         /* zero-fill memory holes */
2150         if (buf != buf_start + buflen)
2151                 memset(buf, 0, buflen - (buf - buf_start));
2152 
2153         return buflen;
2154 }
2155 
2156 /**
2157  *      vwrite() -  write vmalloc area in a safe way.
2158  *      @buf:           buffer for source data
2159  *      @addr:          vm address.
2160  *      @count:         number of bytes to be read.
2161  *
2162  *      Returns # of bytes which addr and buf should be incresed.
2163  *      (same number to @count).
2164  *      If [addr...addr+count) doesn't includes any intersect with valid
2165  *      vmalloc area, returns 0.
2166  *
2167  *      This function checks that addr is a valid vmalloc'ed area, and
2168  *      copy data from a buffer to the given addr. If specified range of
2169  *      [addr...addr+count) includes some valid address, data is copied from
2170  *      proper area of @buf. If there are memory holes, no copy to hole.
2171  *      IOREMAP area is treated as memory hole and no copy is done.
2172  *
2173  *      If [addr...addr+count) doesn't includes any intersects with alive
2174  *      vm_struct area, returns 0. @buf should be kernel's buffer.
2175  *
2176  *      Note: In usual ops, vwrite() is never necessary because the caller
2177  *      should know vmalloc() area is valid and can use memcpy().
2178  *      This is for routines which have to access vmalloc area without
2179  *      any informaion, as /dev/kmem.
2180  */
2181 
2182 long vwrite(char *buf, char *addr, unsigned long count)
2183 {
2184         struct vmap_area *va;
2185         struct vm_struct *vm;
2186         char *vaddr;
2187         unsigned long n, buflen;
2188         int copied = 0;
2189 
2190         /* Don't allow overflow */
2191         if ((unsigned long) addr + count < count)
2192                 count = -(unsigned long) addr;
2193         buflen = count;
2194 
2195         spin_lock(&vmap_area_lock);
2196         list_for_each_entry(va, &vmap_area_list, list) {
2197                 if (!count)
2198                         break;
2199 
2200                 if (!(va->flags & VM_VM_AREA))
2201                         continue;
2202 
2203                 vm = va->vm;
2204                 vaddr = (char *) vm->addr;
2205                 if (addr >= vaddr + get_vm_area_size(vm))
2206                         continue;
2207                 while (addr < vaddr) {
2208                         if (count == 0)
2209                                 goto finished;
2210                         buf++;
2211                         addr++;
2212                         count--;
2213                 }
2214                 n = vaddr + get_vm_area_size(vm) - addr;
2215                 if (n > count)
2216                         n = count;
2217                 if (!(vm->flags & VM_IOREMAP)) {
2218                         aligned_vwrite(buf, addr, n);
2219                         copied++;
2220                 }
2221                 buf += n;
2222                 addr += n;
2223                 count -= n;
2224         }
2225 finished:
2226         spin_unlock(&vmap_area_lock);
2227         if (!copied)
2228                 return 0;
2229         return buflen;
2230 }
2231 
2232 /**
2233  *      remap_vmalloc_range_partial  -  map vmalloc pages to userspace
2234  *      @vma:           vma to cover
2235  *      @uaddr:         target user address to start at
2236  *      @kaddr:         virtual address of vmalloc kernel memory
2237  *      @size:          size of map area
2238  *
2239  *      Returns:        0 for success, -Exxx on failure
2240  *
2241  *      This function checks that @kaddr is a valid vmalloc'ed area,
2242  *      and that it is big enough to cover the range starting at
2243  *      @uaddr in @vma. Will return failure if that criteria isn't
2244  *      met.
2245  *
2246  *      Similar to remap_pfn_range() (see mm/memory.c)
2247  */
2248 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2249                                 void *kaddr, unsigned long size)
2250 {
2251         struct vm_struct *area;
2252 
2253         size = PAGE_ALIGN(size);
2254 
2255         if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2256                 return -EINVAL;
2257 
2258         area = find_vm_area(kaddr);
2259         if (!area)
2260                 return -EINVAL;
2261 
2262         if (!(area->flags & VM_USERMAP))
2263                 return -EINVAL;
2264 
2265         if (kaddr + size > area->addr + area->size)
2266                 return -EINVAL;
2267 
2268         do {
2269                 struct page *page = vmalloc_to_page(kaddr);
2270                 int ret;
2271 
2272                 ret = vm_insert_page(vma, uaddr, page);
2273                 if (ret)
2274                         return ret;
2275 
2276                 uaddr += PAGE_SIZE;
2277                 kaddr += PAGE_SIZE;
2278                 size -= PAGE_SIZE;
2279         } while (size > 0);
2280 
2281         vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2282 
2283         return 0;
2284 }
2285 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2286 
2287 /**
2288  *      remap_vmalloc_range  -  map vmalloc pages to userspace
2289  *      @vma:           vma to cover (map full range of vma)
2290  *      @addr:          vmalloc memory
2291  *      @pgoff:         number of pages into addr before first page to map
2292  *
2293  *      Returns:        0 for success, -Exxx on failure
2294  *
2295  *      This function checks that addr is a valid vmalloc'ed area, and
2296  *      that it is big enough to cover the vma. Will return failure if
2297  *      that criteria isn't met.
2298  *
2299  *      Similar to remap_pfn_range() (see mm/memory.c)
2300  */
2301 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2302                                                 unsigned long pgoff)
2303 {
2304         return remap_vmalloc_range_partial(vma, vma->vm_start,
2305                                            addr + (pgoff << PAGE_SHIFT),
2306                                            vma->vm_end - vma->vm_start);
2307 }
2308 EXPORT_SYMBOL(remap_vmalloc_range);
2309 
2310 /*
2311  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2312  * have one.
2313  */
2314 void __weak vmalloc_sync_all(void)
2315 {
2316 }
2317 
2318 
2319 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2320 {
2321         pte_t ***p = data;
2322 
2323         if (p) {
2324                 *(*p) = pte;
2325                 (*p)++;
2326         }
2327         return 0;
2328 }
2329 
2330 /**
2331  *      alloc_vm_area - allocate a range of kernel address space
2332  *      @size:          size of the area
2333  *      @ptes:          returns the PTEs for the address space
2334  *
2335  *      Returns:        NULL on failure, vm_struct on success
2336  *
2337  *      This function reserves a range of kernel address space, and
2338  *      allocates pagetables to map that range.  No actual mappings
2339  *      are created.
2340  *
2341  *      If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2342  *      allocated for the VM area are returned.
2343  */
2344 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2345 {
2346         struct vm_struct *area;
2347 
2348         area = get_vm_area_caller(size, VM_IOREMAP,
2349                                 __builtin_return_address(0));
2350         if (area == NULL)
2351                 return NULL;
2352 
2353         /*
2354          * This ensures that page tables are constructed for this region
2355          * of kernel virtual address space and mapped into init_mm.
2356          */
2357         if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2358                                 size, f, ptes ? &ptes : NULL)) {
2359                 free_vm_area(area);
2360                 return NULL;
2361         }
2362 
2363         return area;
2364 }
2365 EXPORT_SYMBOL_GPL(alloc_vm_area);
2366 
2367 void free_vm_area(struct vm_struct *area)
2368 {
2369         struct vm_struct *ret;
2370         ret = remove_vm_area(area->addr);
2371         BUG_ON(ret != area);
2372         kfree(area);
2373 }
2374 EXPORT_SYMBOL_GPL(free_vm_area);
2375 
2376 #ifdef CONFIG_SMP
2377 static struct vmap_area *node_to_va(struct rb_node *n)
2378 {
2379         return rb_entry_safe(n, struct vmap_area, rb_node);
2380 }
2381 
2382 /**
2383  * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2384  * @end: target address
2385  * @pnext: out arg for the next vmap_area
2386  * @pprev: out arg for the previous vmap_area
2387  *
2388  * Returns: %true if either or both of next and prev are found,
2389  *          %false if no vmap_area exists
2390  *
2391  * Find vmap_areas end addresses of which enclose @end.  ie. if not
2392  * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2393  */
2394 static bool pvm_find_next_prev(unsigned long end,
2395                                struct vmap_area **pnext,
2396                                struct vmap_area **pprev)
2397 {
2398         struct rb_node *n = vmap_area_root.rb_node;
2399         struct vmap_area *va = NULL;
2400 
2401         while (n) {
2402                 va = rb_entry(n, struct vmap_area, rb_node);
2403                 if (end < va->va_end)
2404                         n = n->rb_left;
2405                 else if (end > va->va_end)
2406                         n = n->rb_right;
2407                 else
2408                         break;
2409         }
2410 
2411         if (!va)
2412                 return false;
2413 
2414         if (va->va_end > end) {
2415                 *pnext = va;
2416                 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2417         } else {
2418                 *pprev = va;
2419                 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2420         }
2421         return true;
2422 }
2423 
2424 /**
2425  * pvm_determine_end - find the highest aligned address between two vmap_areas
2426  * @pnext: in/out arg for the next vmap_area
2427  * @pprev: in/out arg for the previous vmap_area
2428  * @align: alignment
2429  *
2430  * Returns: determined end address
2431  *
2432  * Find the highest aligned address between *@pnext and *@pprev below
2433  * VMALLOC_END.  *@pnext and *@pprev are adjusted so that the aligned
2434  * down address is between the end addresses of the two vmap_areas.
2435  *
2436  * Please note that the address returned by this function may fall
2437  * inside *@pnext vmap_area.  The caller is responsible for checking
2438  * that.
2439  */
2440 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2441                                        struct vmap_area **pprev,
2442                                        unsigned long align)
2443 {
2444         const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2445         unsigned long addr;
2446 
2447         if (*pnext)
2448                 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2449         else
2450                 addr = vmalloc_end;
2451 
2452         while (*pprev && (*pprev)->va_end > addr) {
2453                 *pnext = *pprev;
2454                 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2455         }
2456 
2457         return addr;
2458 }
2459 
2460 /**
2461  * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2462  * @offsets: array containing offset of each area
2463  * @sizes: array containing size of each area
2464  * @nr_vms: the number of areas to allocate
2465  * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2466  *
2467  * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2468  *          vm_structs on success, %NULL on failure
2469  *
2470  * Percpu allocator wants to use congruent vm areas so that it can
2471  * maintain the offsets among percpu areas.  This function allocates
2472  * congruent vmalloc areas for it with GFP_KERNEL.  These areas tend to
2473  * be scattered pretty far, distance between two areas easily going up
2474  * to gigabytes.  To avoid interacting with regular vmallocs, these
2475  * areas are allocated from top.
2476  *
2477  * Despite its complicated look, this allocator is rather simple.  It
2478  * does everything top-down and scans areas from the end looking for
2479  * matching slot.  While scanning, if any of the areas overlaps with
2480  * existing vmap_area, the base address is pulled down to fit the
2481  * area.  Scanning is repeated till all the areas fit and then all
2482  * necessary data structures are inserted and the result is returned.
2483  */
2484 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2485                                      const size_t *sizes, int nr_vms,
2486                                      size_t align)
2487 {
2488         const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2489         const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2490         struct vmap_area **vas, *prev, *next;
2491         struct vm_struct **vms;
2492         int area, area2, last_area, term_area;
2493         unsigned long base, start, end, last_end;
2494         bool purged = false;
2495 
2496         /* verify parameters and allocate data structures */
2497         BUG_ON(offset_in_page(align) || !is_power_of_2(align));
2498         for (last_area = 0, area = 0; area < nr_vms; area++) {
2499                 start = offsets[area];
2500                 end = start + sizes[area];
2501 
2502                 /* is everything aligned properly? */
2503                 BUG_ON(!IS_ALIGNED(offsets[area], align));
2504                 BUG_ON(!IS_ALIGNED(sizes[area], align));
2505 
2506                 /* detect the area with the highest address */
2507                 if (start > offsets[last_area])
2508                         last_area = area;
2509 
2510                 for (area2 = area + 1; area2 < nr_vms; area2++) {
2511                         unsigned long start2 = offsets[area2];
2512                         unsigned long end2 = start2 + sizes[area2];
2513 
2514                         BUG_ON(start2 < end && start < end2);
2515                 }
2516         }
2517         last_end = offsets[last_area] + sizes[last_area];
2518 
2519         if (vmalloc_end - vmalloc_start < last_end) {
2520                 WARN_ON(true);
2521                 return NULL;
2522         }
2523 
2524         vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2525         vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2526         if (!vas || !vms)
2527                 goto err_free2;
2528 
2529         for (area = 0; area < nr_vms; area++) {
2530                 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2531                 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2532                 if (!vas[area] || !vms[area])
2533                         goto err_free;
2534         }
2535 retry:
2536         spin_lock(&vmap_area_lock);
2537 
2538         /* start scanning - we scan from the top, begin with the last area */
2539         area = term_area = last_area;
2540         start = offsets[area];
2541         end = start + sizes[area];
2542 
2543         if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2544                 base = vmalloc_end - last_end;
2545                 goto found;
2546         }
2547         base = pvm_determine_end(&next, &prev, align) - end;
2548 
2549         while (true) {
2550                 BUG_ON(next && next->va_end <= base + end);
2551                 BUG_ON(prev && prev->va_end > base + end);
2552 
2553                 /*
2554                  * base might have underflowed, add last_end before
2555                  * comparing.
2556                  */
2557                 if (base + last_end < vmalloc_start + last_end) {
2558                         spin_unlock(&vmap_area_lock);
2559                         if (!purged) {
2560                                 purge_vmap_area_lazy();
2561                                 purged = true;
2562                                 goto retry;
2563                         }
2564                         goto err_free;
2565                 }
2566 
2567                 /*
2568                  * If next overlaps, move base downwards so that it's
2569                  * right below next and then recheck.
2570                  */
2571                 if (next && next->va_start < base + end) {
2572                         base = pvm_determine_end(&next, &prev, align) - end;
2573                         term_area = area;
2574                         continue;
2575                 }
2576 
2577                 /*
2578                  * If prev overlaps, shift down next and prev and move
2579                  * base so that it's right below new next and then
2580                  * recheck.
2581                  */
2582                 if (prev && prev->va_end > base + start)  {
2583                         next = prev;
2584                         prev = node_to_va(rb_prev(&next->rb_node));
2585                         base = pvm_determine_end(&next, &prev, align) - end;
2586                         term_area = area;
2587                         continue;
2588                 }
2589 
2590                 /*
2591                  * This area fits, move on to the previous one.  If
2592                  * the previous one is the terminal one, we're done.
2593                  */
2594                 area = (area + nr_vms - 1) % nr_vms;
2595                 if (area == term_area)
2596                         break;
2597                 start = offsets[area];
2598                 end = start + sizes[area];
2599                 pvm_find_next_prev(base + end, &next, &prev);
2600         }
2601 found:
2602         /* we've found a fitting base, insert all va's */
2603         for (area = 0; area < nr_vms; area++) {
2604                 struct vmap_area *va = vas[area];
2605 
2606                 va->va_start = base + offsets[area];
2607                 va->va_end = va->va_start + sizes[area];
2608                 __insert_vmap_area(va);
2609         }
2610 
2611         vmap_area_pcpu_hole = base + offsets[last_area];
2612 
2613         spin_unlock(&vmap_area_lock);
2614 
2615         /* insert all vm's */
2616         for (area = 0; area < nr_vms; area++)
2617                 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2618                                  pcpu_get_vm_areas);
2619 
2620         kfree(vas);
2621         return vms;
2622 
2623 err_free:
2624         for (area = 0; area < nr_vms; area++) {
2625                 kfree(vas[area]);
2626                 kfree(vms[area]);
2627         }
2628 err_free2:
2629         kfree(vas);
2630         kfree(vms);
2631         return NULL;
2632 }
2633 
2634 /**
2635  * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2636  * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2637  * @nr_vms: the number of allocated areas
2638  *
2639  * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2640  */
2641 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2642 {
2643         int i;
2644 
2645         for (i = 0; i < nr_vms; i++)
2646                 free_vm_area(vms[i]);
2647         kfree(vms);
2648 }
2649 #endif  /* CONFIG_SMP */
2650 
2651 #ifdef CONFIG_PROC_FS
2652 static void *s_start(struct seq_file *m, loff_t *pos)
2653         __acquires(&vmap_area_lock)
2654 {
2655         spin_lock(&vmap_area_lock);
2656         return seq_list_start(&vmap_area_list, *pos);
2657 }
2658 
2659 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2660 {
2661         return seq_list_next(p, &vmap_area_list, pos);
2662 }
2663 
2664 static void s_stop(struct seq_file *m, void *p)
2665         __releases(&vmap_area_lock)
2666 {
2667         spin_unlock(&vmap_area_lock);
2668 }
2669 
2670 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2671 {
2672         if (IS_ENABLED(CONFIG_NUMA)) {
2673                 unsigned int nr, *counters = m->private;
2674 
2675                 if (!counters)
2676                         return;
2677 
2678                 if (v->flags & VM_UNINITIALIZED)
2679                         return;
2680                 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2681                 smp_rmb();
2682 
2683                 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2684 
2685                 for (nr = 0; nr < v->nr_pages; nr++)
2686                         counters[page_to_nid(v->pages[nr])]++;
2687 
2688                 for_each_node_state(nr, N_HIGH_MEMORY)
2689                         if (counters[nr])
2690                                 seq_printf(m, " N%u=%u", nr, counters[nr]);
2691         }
2692 }
2693 
2694 static int s_show(struct seq_file *m, void *p)
2695 {
2696         struct vmap_area *va;
2697         struct vm_struct *v;
2698 
2699         va = list_entry(p, struct vmap_area, list);
2700 
2701         /*
2702          * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2703          * behalf of vmap area is being tear down or vm_map_ram allocation.
2704          */
2705         if (!(va->flags & VM_VM_AREA)) {
2706                 seq_printf(m, "0x%pK-0x%pK %7ld %s\n",
2707                         (void *)va->va_start, (void *)va->va_end,
2708                         va->va_end - va->va_start,
2709                         va->flags & VM_LAZY_FREE ? "unpurged vm_area" : "vm_map_ram");
2710 
2711                 return 0;
2712         }
2713 
2714         v = va->vm;
2715 
2716         seq_printf(m, "0x%pK-0x%pK %7ld",
2717                 v->addr, v->addr + v->size, v->size);
2718 
2719         if (v->caller)
2720                 seq_printf(m, " %pS", v->caller);
2721 
2722         if (v->nr_pages)
2723                 seq_printf(m, " pages=%d", v->nr_pages);
2724 
2725         if (v->phys_addr)
2726                 seq_printf(m, " phys=%pa", &v->phys_addr);
2727 
2728         if (v->flags & VM_IOREMAP)
2729                 seq_puts(m, " ioremap");
2730 
2731         if (v->flags & VM_ALLOC)
2732                 seq_puts(m, " vmalloc");
2733 
2734         if (v->flags & VM_MAP)
2735                 seq_puts(m, " vmap");
2736 
2737         if (v->flags & VM_USERMAP)
2738                 seq_puts(m, " user");
2739 
2740         if (is_vmalloc_addr(v->pages))
2741                 seq_puts(m, " vpages");
2742 
2743         show_numa_info(m, v);
2744         seq_putc(m, '\n');
2745         return 0;
2746 }
2747 
2748 static const struct seq_operations vmalloc_op = {
2749         .start = s_start,
2750         .next = s_next,
2751         .stop = s_stop,
2752         .show = s_show,
2753 };
2754 
2755 static int vmalloc_open(struct inode *inode, struct file *file)
2756 {
2757         if (IS_ENABLED(CONFIG_NUMA))
2758                 return seq_open_private(file, &vmalloc_op,
2759                                         nr_node_ids * sizeof(unsigned int));
2760         else
2761                 return seq_open(file, &vmalloc_op);
2762 }
2763 
2764 static const struct file_operations proc_vmalloc_operations = {
2765         .open           = vmalloc_open,
2766         .read           = seq_read,
2767         .llseek         = seq_lseek,
2768         .release        = seq_release_private,
2769 };
2770 
2771 static int __init proc_vmalloc_init(void)
2772 {
2773         proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2774         return 0;
2775 }
2776 module_init(proc_vmalloc_init);
2777 
2778 #endif
2779 
2780 

~ [ 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