Linux kernel & device driver programming

Cross-Referenced Linux and Device Driver Code

[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ]
Version: [ 2.6.11.8 ] [ 2.6.25 ] [ 2.6.25.8 ] [ 2.6.31.13 ] Architecture: [ i386 ]
  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/slab.h>
 16 #include <linux/spinlock.h>
 17 #include <linux/interrupt.h>
 18 #include <linux/proc_fs.h>
 19 #include <linux/seq_file.h>
 20 #include <linux/debugobjects.h>
 21 #include <linux/kallsyms.h>
 22 #include <linux/list.h>
 23 #include <linux/rbtree.h>
 24 #include <linux/radix-tree.h>
 25 #include <linux/rcupdate.h>
 26 #include <linux/pfn.h>
 27 #include <linux/kmemleak.h>
 28 
 29 #include <asm/atomic.h>
 30 #include <asm/uaccess.h>
 31 #include <asm/tlbflush.h>
 32 
 33 
 34 /*** Page table manipulation functions ***/
 35 
 36 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
 37 {
 38         pte_t *pte;
 39 
 40         pte = pte_offset_kernel(pmd, addr);
 41         do {
 42                 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
 43                 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
 44         } while (pte++, addr += PAGE_SIZE, addr != end);
 45 }
 46 
 47 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
 48 {
 49         pmd_t *pmd;
 50         unsigned long next;
 51 
 52         pmd = pmd_offset(pud, addr);
 53         do {
 54                 next = pmd_addr_end(addr, end);
 55                 if (pmd_none_or_clear_bad(pmd))
 56                         continue;
 57                 vunmap_pte_range(pmd, addr, next);
 58         } while (pmd++, addr = next, addr != end);
 59 }
 60 
 61 static void vunmap_pud_range(pgd_t *pgd, unsigned long addr, unsigned long end)
 62 {
 63         pud_t *pud;
 64         unsigned long next;
 65 
 66         pud = pud_offset(pgd, addr);
 67         do {
 68                 next = pud_addr_end(addr, end);
 69                 if (pud_none_or_clear_bad(pud))
 70                         continue;
 71                 vunmap_pmd_range(pud, addr, next);
 72         } while (pud++, addr = next, addr != end);
 73 }
 74 
 75 static void vunmap_page_range(unsigned long addr, unsigned long end)
 76 {
 77         pgd_t *pgd;
 78         unsigned long next;
 79 
 80         BUG_ON(addr >= end);
 81         pgd = pgd_offset_k(addr);
 82         do {
 83                 next = pgd_addr_end(addr, end);
 84                 if (pgd_none_or_clear_bad(pgd))
 85                         continue;
 86                 vunmap_pud_range(pgd, addr, next);
 87         } while (pgd++, addr = next, addr != end);
 88 }
 89 
 90 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
 91                 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
 92 {
 93         pte_t *pte;
 94 
 95         /*
 96          * nr is a running index into the array which helps higher level
 97          * callers keep track of where we're up to.
 98          */
 99 
100         pte = pte_alloc_kernel(pmd, addr);
101         if (!pte)
102                 return -ENOMEM;
103         do {
104                 struct page *page = pages[*nr];
105 
106                 if (WARN_ON(!pte_none(*pte)))
107                         return -EBUSY;
108                 if (WARN_ON(!page))
109                         return -ENOMEM;
110                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
111                 (*nr)++;
112         } while (pte++, addr += PAGE_SIZE, addr != end);
113         return 0;
114 }
115 
116 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
117                 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
118 {
119         pmd_t *pmd;
120         unsigned long next;
121 
122         pmd = pmd_alloc(&init_mm, pud, addr);
123         if (!pmd)
124                 return -ENOMEM;
125         do {
126                 next = pmd_addr_end(addr, end);
127                 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
128                         return -ENOMEM;
129         } while (pmd++, addr = next, addr != end);
130         return 0;
131 }
132 
133 static int vmap_pud_range(pgd_t *pgd, unsigned long addr,
134                 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
135 {
136         pud_t *pud;
137         unsigned long next;
138 
139         pud = pud_alloc(&init_mm, pgd, addr);
140         if (!pud)
141                 return -ENOMEM;
142         do {
143                 next = pud_addr_end(addr, end);
144                 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
145                         return -ENOMEM;
146         } while (pud++, addr = next, addr != end);
147         return 0;
148 }
149 
150 /*
151  * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
152  * will have pfns corresponding to the "pages" array.
153  *
154  * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
155  */
156 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
157                                    pgprot_t prot, struct page **pages)
158 {
159         pgd_t *pgd;
160         unsigned long next;
161         unsigned long addr = start;
162         int err = 0;
163         int nr = 0;
164 
165         BUG_ON(addr >= end);
166         pgd = pgd_offset_k(addr);
167         do {
168                 next = pgd_addr_end(addr, end);
169                 err = vmap_pud_range(pgd, addr, next, prot, pages, &nr);
170                 if (err)
171                         break;
172         } while (pgd++, addr = next, addr != end);
173 
174         if (unlikely(err))
175                 return err;
176         return nr;
177 }
178 
179 static int vmap_page_range(unsigned long start, unsigned long end,
180                            pgprot_t prot, struct page **pages)
181 {
182         int ret;
183 
184         ret = vmap_page_range_noflush(start, end, prot, pages);
185         flush_cache_vmap(start, end);
186         return ret;
187 }
188 
189 static inline int is_vmalloc_or_module_addr(const void *x)
190 {
191         /*
192          * ARM, x86-64 and sparc64 put modules in a special place,
193          * and fall back on vmalloc() if that fails. Others
194          * just put it in the vmalloc space.
195          */
196 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
197         unsigned long addr = (unsigned long)x;
198         if (addr >= MODULES_VADDR && addr < MODULES_END)
199                 return 1;
200 #endif
201         return is_vmalloc_addr(x);
202 }
203 
204 /*
205  * Walk a vmap address to the struct page it maps.
206  */
207 struct page *vmalloc_to_page(const void *vmalloc_addr)
208 {
209         unsigned long addr = (unsigned long) vmalloc_addr;
210         struct page *page = NULL;
211         pgd_t *pgd = pgd_offset_k(addr);
212 
213         /*
214          * XXX we might need to change this if we add VIRTUAL_BUG_ON for
215          * architectures that do not vmalloc module space
216          */
217         VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
218 
219         if (!pgd_none(*pgd)) {
220                 pud_t *pud = pud_offset(pgd, addr);
221                 if (!pud_none(*pud)) {
222                         pmd_t *pmd = pmd_offset(pud, addr);
223                         if (!pmd_none(*pmd)) {
224                                 pte_t *ptep, pte;
225 
226                                 ptep = pte_offset_map(pmd, addr);
227                                 pte = *ptep;
228                                 if (pte_present(pte))
229                                         page = pte_page(pte);
230                                 pte_unmap(ptep);
231                         }
232                 }
233         }
234         return page;
235 }
236 EXPORT_SYMBOL(vmalloc_to_page);
237 
238 /*
239  * Map a vmalloc()-space virtual address to the physical page frame number.
240  */
241 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
242 {
243         return page_to_pfn(vmalloc_to_page(vmalloc_addr));
244 }
245 EXPORT_SYMBOL(vmalloc_to_pfn);
246 
247 
248 /*** Global kva allocator ***/
249 
250 #define VM_LAZY_FREE    0x01
251 #define VM_LAZY_FREEING 0x02
252 #define VM_VM_AREA      0x04
253 
254 struct vmap_area {
255         unsigned long va_start;
256         unsigned long va_end;
257         unsigned long flags;
258         struct rb_node rb_node;         /* address sorted rbtree */
259         struct list_head list;          /* address sorted list */
260         struct list_head purge_list;    /* "lazy purge" list */
261         void *private;
262         struct rcu_head rcu_head;
263 };
264 
265 static DEFINE_SPINLOCK(vmap_area_lock);
266 static struct rb_root vmap_area_root = RB_ROOT;
267 static LIST_HEAD(vmap_area_list);
268 
269 static struct vmap_area *__find_vmap_area(unsigned long addr)
270 {
271         struct rb_node *n = vmap_area_root.rb_node;
272 
273         while (n) {
274                 struct vmap_area *va;
275 
276                 va = rb_entry(n, struct vmap_area, rb_node);
277                 if (addr < va->va_start)
278                         n = n->rb_left;
279                 else if (addr > va->va_start)
280                         n = n->rb_right;
281                 else
282                         return va;
283         }
284 
285         return NULL;
286 }
287 
288 static void __insert_vmap_area(struct vmap_area *va)
289 {
290         struct rb_node **p = &vmap_area_root.rb_node;
291         struct rb_node *parent = NULL;
292         struct rb_node *tmp;
293 
294         while (*p) {
295                 struct vmap_area *tmp;
296 
297                 parent = *p;
298                 tmp = rb_entry(parent, struct vmap_area, rb_node);
299                 if (va->va_start < tmp->va_end)
300                         p = &(*p)->rb_left;
301                 else if (va->va_end > tmp->va_start)
302                         p = &(*p)->rb_right;
303                 else
304                         BUG();
305         }
306 
307         rb_link_node(&va->rb_node, parent, p);
308         rb_insert_color(&va->rb_node, &vmap_area_root);
309 
310         /* address-sort this list so it is usable like the vmlist */
311         tmp = rb_prev(&va->rb_node);
312         if (tmp) {
313                 struct vmap_area *prev;
314                 prev = rb_entry(tmp, struct vmap_area, rb_node);
315                 list_add_rcu(&va->list, &prev->list);
316         } else
317                 list_add_rcu(&va->list, &vmap_area_list);
318 }
319 
320 static void purge_vmap_area_lazy(void);
321 
322 /*
323  * Allocate a region of KVA of the specified size and alignment, within the
324  * vstart and vend.
325  */
326 static struct vmap_area *alloc_vmap_area(unsigned long size,
327                                 unsigned long align,
328                                 unsigned long vstart, unsigned long vend,
329                                 int node, gfp_t gfp_mask)
330 {
331         struct vmap_area *va;
332         struct rb_node *n;
333         unsigned long addr;
334         int purged = 0;
335 
336         BUG_ON(!size);
337         BUG_ON(size & ~PAGE_MASK);
338 
339         va = kmalloc_node(sizeof(struct vmap_area),
340                         gfp_mask & GFP_RECLAIM_MASK, node);
341         if (unlikely(!va))
342                 return ERR_PTR(-ENOMEM);
343 
344 retry:
345         addr = ALIGN(vstart, align);
346 
347         spin_lock(&vmap_area_lock);
348         if (addr + size - 1 < addr)
349                 goto overflow;
350 
351         /* XXX: could have a last_hole cache */
352         n = vmap_area_root.rb_node;
353         if (n) {
354                 struct vmap_area *first = NULL;
355 
356                 do {
357                         struct vmap_area *tmp;
358                         tmp = rb_entry(n, struct vmap_area, rb_node);
359                         if (tmp->va_end >= addr) {
360                                 if (!first && tmp->va_start < addr + size)
361                                         first = tmp;
362                                 n = n->rb_left;
363                         } else {
364                                 first = tmp;
365                                 n = n->rb_right;
366                         }
367                 } while (n);
368 
369                 if (!first)
370                         goto found;
371 
372                 if (first->va_end < addr) {
373                         n = rb_next(&first->rb_node);
374                         if (n)
375                                 first = rb_entry(n, struct vmap_area, rb_node);
376                         else
377                                 goto found;
378                 }
379 
380                 while (addr + size > first->va_start && addr + size <= vend) {
381                         addr = ALIGN(first->va_end + PAGE_SIZE, align);
382                         if (addr + size - 1 < addr)
383                                 goto overflow;
384 
385                         n = rb_next(&first->rb_node);
386                         if (n)
387                                 first = rb_entry(n, struct vmap_area, rb_node);
388                         else
389                                 goto found;
390                 }
391         }
392 found:
393         if (addr + size > vend) {
394 overflow:
395                 spin_unlock(&vmap_area_lock);
396                 if (!purged) {
397                         purge_vmap_area_lazy();
398                         purged = 1;
399                         goto retry;
400                 }
401                 if (printk_ratelimit())
402                         printk(KERN_WARNING
403                                 "vmap allocation for size %lu failed: "
404                                 "use vmalloc=<size> to increase size.\n", size);
405                 kfree(va);
406                 return ERR_PTR(-EBUSY);
407         }
408 
409         BUG_ON(addr & (align-1));
410 
411         va->va_start = addr;
412         va->va_end = addr + size;
413         va->flags = 0;
414         __insert_vmap_area(va);
415         spin_unlock(&vmap_area_lock);
416 
417         return va;
418 }
419 
420 static void rcu_free_va(struct rcu_head *head)
421 {
422         struct vmap_area *va = container_of(head, struct vmap_area, rcu_head);
423 
424         kfree(va);
425 }
426 
427 static void __free_vmap_area(struct vmap_area *va)
428 {
429         BUG_ON(RB_EMPTY_NODE(&va->rb_node));
430         rb_erase(&va->rb_node, &vmap_area_root);
431         RB_CLEAR_NODE(&va->rb_node);
432         list_del_rcu(&va->list);
433 
434         call_rcu(&va->rcu_head, rcu_free_va);
435 }
436 
437 /*
438  * Free a region of KVA allocated by alloc_vmap_area
439  */
440 static void free_vmap_area(struct vmap_area *va)
441 {
442         spin_lock(&vmap_area_lock);
443         __free_vmap_area(va);
444         spin_unlock(&vmap_area_lock);
445 }
446 
447 /*
448  * Clear the pagetable entries of a given vmap_area
449  */
450 static void unmap_vmap_area(struct vmap_area *va)
451 {
452         vunmap_page_range(va->va_start, va->va_end);
453 }
454 
455 static void vmap_debug_free_range(unsigned long start, unsigned long end)
456 {
457         /*
458          * Unmap page tables and force a TLB flush immediately if
459          * CONFIG_DEBUG_PAGEALLOC is set. This catches use after free
460          * bugs similarly to those in linear kernel virtual address
461          * space after a page has been freed.
462          *
463          * All the lazy freeing logic is still retained, in order to
464          * minimise intrusiveness of this debugging feature.
465          *
466          * This is going to be *slow* (linear kernel virtual address
467          * debugging doesn't do a broadcast TLB flush so it is a lot
468          * faster).
469          */
470 #ifdef CONFIG_DEBUG_PAGEALLOC
471         vunmap_page_range(start, end);
472         flush_tlb_kernel_range(start, end);
473 #endif
474 }
475 
476 /*
477  * lazy_max_pages is the maximum amount of virtual address space we gather up
478  * before attempting to purge with a TLB flush.
479  *
480  * There is a tradeoff here: a larger number will cover more kernel page tables
481  * and take slightly longer to purge, but it will linearly reduce the number of
482  * global TLB flushes that must be performed. It would seem natural to scale
483  * this number up linearly with the number of CPUs (because vmapping activity
484  * could also scale linearly with the number of CPUs), however it is likely
485  * that in practice, workloads might be constrained in other ways that mean
486  * vmap activity will not scale linearly with CPUs. Also, I want to be
487  * conservative and not introduce a big latency on huge systems, so go with
488  * a less aggressive log scale. It will still be an improvement over the old
489  * code, and it will be simple to change the scale factor if we find that it
490  * becomes a problem on bigger systems.
491  */
492 static unsigned long lazy_max_pages(void)
493 {
494         unsigned int log;
495 
496         log = fls(num_online_cpus());
497 
498         return log * (32UL * 1024 * 1024 / PAGE_SIZE);
499 }
500 
501 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
502 
503 /*
504  * Purges all lazily-freed vmap areas.
505  *
506  * If sync is 0 then don't purge if there is already a purge in progress.
507  * If force_flush is 1, then flush kernel TLBs between *start and *end even
508  * if we found no lazy vmap areas to unmap (callers can use this to optimise
509  * their own TLB flushing).
510  * Returns with *start = min(*start, lowest purged address)
511  *              *end = max(*end, highest purged address)
512  */
513 static void __purge_vmap_area_lazy(unsigned long *start, unsigned long *end,
514                                         int sync, int force_flush)
515 {
516         static DEFINE_SPINLOCK(purge_lock);
517         LIST_HEAD(valist);
518         struct vmap_area *va;
519         struct vmap_area *n_va;
520         int nr = 0;
521 
522         /*
523          * If sync is 0 but force_flush is 1, we'll go sync anyway but callers
524          * should not expect such behaviour. This just simplifies locking for
525          * the case that isn't actually used at the moment anyway.
526          */
527         if (!sync && !force_flush) {
528                 if (!spin_trylock(&purge_lock))
529                         return;
530         } else
531                 spin_lock(&purge_lock);
532 
533         rcu_read_lock();
534         list_for_each_entry_rcu(va, &vmap_area_list, list) {
535                 if (va->flags & VM_LAZY_FREE) {
536                         if (va->va_start < *start)
537                                 *start = va->va_start;
538                         if (va->va_end > *end)
539                                 *end = va->va_end;
540                         nr += (va->va_end - va->va_start) >> PAGE_SHIFT;
541                         unmap_vmap_area(va);
542                         list_add_tail(&va->purge_list, &valist);
543                         va->flags |= VM_LAZY_FREEING;
544                         va->flags &= ~VM_LAZY_FREE;
545                 }
546         }
547         rcu_read_unlock();
548 
549         if (nr)
550                 atomic_sub(nr, &vmap_lazy_nr);
551 
552         if (nr || force_flush)
553                 flush_tlb_kernel_range(*start, *end);
554 
555         if (nr) {
556                 spin_lock(&vmap_area_lock);
557                 list_for_each_entry_safe(va, n_va, &valist, purge_list)
558                         __free_vmap_area(va);
559                 spin_unlock(&vmap_area_lock);
560         }
561         spin_unlock(&purge_lock);
562 }
563 
564 /*
565  * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
566  * is already purging.
567  */
568 static void try_purge_vmap_area_lazy(void)
569 {
570         unsigned long start = ULONG_MAX, end = 0;
571 
572         __purge_vmap_area_lazy(&start, &end, 0, 0);
573 }
574 
575 /*
576  * Kick off a purge of the outstanding lazy areas.
577  */
578 static void purge_vmap_area_lazy(void)
579 {
580         unsigned long start = ULONG_MAX, end = 0;
581 
582         __purge_vmap_area_lazy(&start, &end, 1, 0);
583 }
584 
585 /*
586  * Free and unmap a vmap area, caller ensuring flush_cache_vunmap had been
587  * called for the correct range previously.
588  */
589 static void free_unmap_vmap_area_noflush(struct vmap_area *va)
590 {
591         va->flags |= VM_LAZY_FREE;
592         atomic_add((va->va_end - va->va_start) >> PAGE_SHIFT, &vmap_lazy_nr);
593         if (unlikely(atomic_read(&vmap_lazy_nr) > lazy_max_pages()))
594                 try_purge_vmap_area_lazy();
595 }
596 
597 /*
598  * Free and unmap a vmap area
599  */
600 static void free_unmap_vmap_area(struct vmap_area *va)
601 {
602         flush_cache_vunmap(va->va_start, va->va_end);
603         free_unmap_vmap_area_noflush(va);
604 }
605 
606 static struct vmap_area *find_vmap_area(unsigned long addr)
607 {
608         struct vmap_area *va;
609 
610         spin_lock(&vmap_area_lock);
611         va = __find_vmap_area(addr);
612         spin_unlock(&vmap_area_lock);
613 
614         return va;
615 }
616 
617 static void free_unmap_vmap_area_addr(unsigned long addr)
618 {
619         struct vmap_area *va;
620 
621         va = find_vmap_area(addr);
622         BUG_ON(!va);
623         free_unmap_vmap_area(va);
624 }
625 
626 
627 /*** Per cpu kva allocator ***/
628 
629 /*
630  * vmap space is limited especially on 32 bit architectures. Ensure there is
631  * room for at least 16 percpu vmap blocks per CPU.
632  */
633 /*
634  * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
635  * to #define VMALLOC_SPACE             (VMALLOC_END-VMALLOC_START). Guess
636  * instead (we just need a rough idea)
637  */
638 #if BITS_PER_LONG == 32
639 #define VMALLOC_SPACE           (128UL*1024*1024)
640 #else
641 #define VMALLOC_SPACE           (128UL*1024*1024*1024)
642 #endif
643 
644 #define VMALLOC_PAGES           (VMALLOC_SPACE / PAGE_SIZE)
645 #define VMAP_MAX_ALLOC          BITS_PER_LONG   /* 256K with 4K pages */
646 #define VMAP_BBMAP_BITS_MAX     1024    /* 4MB with 4K pages */
647 #define VMAP_BBMAP_BITS_MIN     (VMAP_MAX_ALLOC*2)
648 #define VMAP_MIN(x, y)          ((x) < (y) ? (x) : (y)) /* can't use min() */
649 #define VMAP_MAX(x, y)          ((x) > (y) ? (x) : (y)) /* can't use max() */
650 #define VMAP_BBMAP_BITS         VMAP_MIN(VMAP_BBMAP_BITS_MAX,           \
651                                         VMAP_MAX(VMAP_BBMAP_BITS_MIN,   \
652                                                 VMALLOC_PAGES / NR_CPUS / 16))
653 
654 #define VMAP_BLOCK_SIZE         (VMAP_BBMAP_BITS * PAGE_SIZE)
655 
656 static bool vmap_initialized __read_mostly = false;
657 
658 struct vmap_block_queue {
659         spinlock_t lock;
660         struct list_head free;
661         struct list_head dirty;
662         unsigned int nr_dirty;
663 };
664 
665 struct vmap_block {
666         spinlock_t lock;
667         struct vmap_area *va;
668         struct vmap_block_queue *vbq;
669         unsigned long free, dirty;
670         DECLARE_BITMAP(alloc_map, VMAP_BBMAP_BITS);
671         DECLARE_BITMAP(dirty_map, VMAP_BBMAP_BITS);
672         union {
673                 struct list_head free_list;
674                 struct rcu_head rcu_head;
675         };
676 };
677 
678 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
679 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
680 
681 /*
682  * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
683  * in the free path. Could get rid of this if we change the API to return a
684  * "cookie" from alloc, to be passed to free. But no big deal yet.
685  */
686 static DEFINE_SPINLOCK(vmap_block_tree_lock);
687 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
688 
689 /*
690  * We should probably have a fallback mechanism to allocate virtual memory
691  * out of partially filled vmap blocks. However vmap block sizing should be
692  * fairly reasonable according to the vmalloc size, so it shouldn't be a
693  * big problem.
694  */
695 
696 static unsigned long addr_to_vb_idx(unsigned long addr)
697 {
698         addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
699         addr /= VMAP_BLOCK_SIZE;
700         return addr;
701 }
702 
703 static struct vmap_block *new_vmap_block(gfp_t gfp_mask)
704 {
705         struct vmap_block_queue *vbq;
706         struct vmap_block *vb;
707         struct vmap_area *va;
708         unsigned long vb_idx;
709         int node, err;
710 
711         node = numa_node_id();
712 
713         vb = kmalloc_node(sizeof(struct vmap_block),
714                         gfp_mask & GFP_RECLAIM_MASK, node);
715         if (unlikely(!vb))
716                 return ERR_PTR(-ENOMEM);
717 
718         va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
719                                         VMALLOC_START, VMALLOC_END,
720                                         node, gfp_mask);
721         if (unlikely(IS_ERR(va))) {
722                 kfree(vb);
723                 return ERR_PTR(PTR_ERR(va));
724         }
725 
726         err = radix_tree_preload(gfp_mask);
727         if (unlikely(err)) {
728                 kfree(vb);
729                 free_vmap_area(va);
730                 return ERR_PTR(err);
731         }
732 
733         spin_lock_init(&vb->lock);
734         vb->va = va;
735         vb->free = VMAP_BBMAP_BITS;
736         vb->dirty = 0;
737         bitmap_zero(vb->alloc_map, VMAP_BBMAP_BITS);
738         bitmap_zero(vb->dirty_map, VMAP_BBMAP_BITS);
739         INIT_LIST_HEAD(&vb->free_list);
740 
741         vb_idx = addr_to_vb_idx(va->va_start);
742         spin_lock(&vmap_block_tree_lock);
743         err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
744         spin_unlock(&vmap_block_tree_lock);
745         BUG_ON(err);
746         radix_tree_preload_end();
747 
748         vbq = &get_cpu_var(vmap_block_queue);
749         vb->vbq = vbq;
750         spin_lock(&vbq->lock);
751         list_add(&vb->free_list, &vbq->free);
752         spin_unlock(&vbq->lock);
753         put_cpu_var(vmap_cpu_blocks);
754 
755         return vb;
756 }
757 
758 static void rcu_free_vb(struct rcu_head *head)
759 {
760         struct vmap_block *vb = container_of(head, struct vmap_block, rcu_head);
761 
762         kfree(vb);
763 }
764 
765 static void free_vmap_block(struct vmap_block *vb)
766 {
767         struct vmap_block *tmp;
768         unsigned long vb_idx;
769 
770         BUG_ON(!list_empty(&vb->free_list));
771 
772         vb_idx = addr_to_vb_idx(vb->va->va_start);
773         spin_lock(&vmap_block_tree_lock);
774         tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
775         spin_unlock(&vmap_block_tree_lock);
776         BUG_ON(tmp != vb);
777 
778         free_unmap_vmap_area_noflush(vb->va);
779         call_rcu(&vb->rcu_head, rcu_free_vb);
780 }
781 
782 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
783 {
784         struct vmap_block_queue *vbq;
785         struct vmap_block *vb;
786         unsigned long addr = 0;
787         unsigned int order;
788 
789         BUG_ON(size & ~PAGE_MASK);
790         BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
791         order = get_order(size);
792 
793 again:
794         rcu_read_lock();
795         vbq = &get_cpu_var(vmap_block_queue);
796         list_for_each_entry_rcu(vb, &vbq->free, free_list) {
797                 int i;
798 
799                 spin_lock(&vb->lock);
800                 i = bitmap_find_free_region(vb->alloc_map,
801                                                 VMAP_BBMAP_BITS, order);
802 
803                 if (i >= 0) {
804                         addr = vb->va->va_start + (i << PAGE_SHIFT);
805                         BUG_ON(addr_to_vb_idx(addr) !=
806                                         addr_to_vb_idx(vb->va->va_start));
807                         vb->free -= 1UL << order;
808                         if (vb->free == 0) {
809                                 spin_lock(&vbq->lock);
810                                 list_del_init(&vb->free_list);
811                                 spin_unlock(&vbq->lock);
812                         }
813                         spin_unlock(&vb->lock);
814                         break;
815                 }
816                 spin_unlock(&vb->lock);
817         }
818         put_cpu_var(vmap_cpu_blocks);
819         rcu_read_unlock();
820 
821         if (!addr) {
822                 vb = new_vmap_block(gfp_mask);
823                 if (IS_ERR(vb))
824                         return vb;
825                 goto again;
826         }
827 
828         return (void *)addr;
829 }
830 
831 static void vb_free(const void *addr, unsigned long size)
832 {
833         unsigned long offset;
834         unsigned long vb_idx;
835         unsigned int order;
836         struct vmap_block *vb;
837 
838         BUG_ON(size & ~PAGE_MASK);
839         BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
840 
841         flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
842 
843         order = get_order(size);
844 
845         offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
846 
847         vb_idx = addr_to_vb_idx((unsigned long)addr);
848         rcu_read_lock();
849         vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
850         rcu_read_unlock();
851         BUG_ON(!vb);
852 
853         spin_lock(&vb->lock);
854         bitmap_allocate_region(vb->dirty_map, offset >> PAGE_SHIFT, order);
855 
856         vb->dirty += 1UL << order;
857         if (vb->dirty == VMAP_BBMAP_BITS) {
858                 BUG_ON(vb->free || !list_empty(&vb->free_list));
859                 spin_unlock(&vb->lock);
860                 free_vmap_block(vb);
861         } else
862                 spin_unlock(&vb->lock);
863 }
864 
865 /**
866  * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
867  *
868  * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
869  * to amortize TLB flushing overheads. What this means is that any page you
870  * have now, may, in a former life, have been mapped into kernel virtual
871  * address by the vmap layer and so there might be some CPUs with TLB entries
872  * still referencing that page (additional to the regular 1:1 kernel mapping).
873  *
874  * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
875  * be sure that none of the pages we have control over will have any aliases
876  * from the vmap layer.
877  */
878 void vm_unmap_aliases(void)
879 {
880         unsigned long start = ULONG_MAX, end = 0;
881         int cpu;
882         int flush = 0;
883 
884         if (unlikely(!vmap_initialized))
885                 return;
886 
887         for_each_possible_cpu(cpu) {
888                 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
889                 struct vmap_block *vb;
890 
891                 rcu_read_lock();
892                 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
893                         int i;
894 
895                         spin_lock(&vb->lock);
896                         i = find_first_bit(vb->dirty_map, VMAP_BBMAP_BITS);
897                         while (i < VMAP_BBMAP_BITS) {
898                                 unsigned long s, e;
899                                 int j;
900                                 j = find_next_zero_bit(vb->dirty_map,
901                                         VMAP_BBMAP_BITS, i);
902 
903                                 s = vb->va->va_start + (i << PAGE_SHIFT);
904                                 e = vb->va->va_start + (j << PAGE_SHIFT);
905                                 vunmap_page_range(s, e);
906                                 flush = 1;
907 
908                                 if (s < start)
909                                         start = s;
910                                 if (e > end)
911                                         end = e;
912 
913                                 i = j;
914                                 i = find_next_bit(vb->dirty_map,
915                                                         VMAP_BBMAP_BITS, i);
916                         }
917                         spin_unlock(&vb->lock);
918                 }
919                 rcu_read_unlock();
920         }
921 
922         __purge_vmap_area_lazy(&start, &end, 1, flush);
923 }
924 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
925 
926 /**
927  * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
928  * @mem: the pointer returned by vm_map_ram
929  * @count: the count passed to that vm_map_ram call (cannot unmap partial)
930  */
931 void vm_unmap_ram(const void *mem, unsigned int count)
932 {
933         unsigned long size = count << PAGE_SHIFT;
934         unsigned long addr = (unsigned long)mem;
935 
936         BUG_ON(!addr);
937         BUG_ON(addr < VMALLOC_START);
938         BUG_ON(addr > VMALLOC_END);
939         BUG_ON(addr & (PAGE_SIZE-1));
940 
941         debug_check_no_locks_freed(mem, size);
942         vmap_debug_free_range(addr, addr+size);
943 
944         if (likely(count <= VMAP_MAX_ALLOC))
945                 vb_free(mem, size);
946         else
947                 free_unmap_vmap_area_addr(addr);
948 }
949 EXPORT_SYMBOL(vm_unmap_ram);
950 
951 /**
952  * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
953  * @pages: an array of pointers to the pages to be mapped
954  * @count: number of pages
955  * @node: prefer to allocate data structures on this node
956  * @prot: memory protection to use. PAGE_KERNEL for regular RAM
957  *
958  * Returns: a pointer to the address that has been mapped, or %NULL on failure
959  */
960 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
961 {
962         unsigned long size = count << PAGE_SHIFT;
963         unsigned long addr;
964         void *mem;
965 
966         if (likely(count <= VMAP_MAX_ALLOC)) {
967                 mem = vb_alloc(size, GFP_KERNEL);
968                 if (IS_ERR(mem))
969                         return NULL;
970                 addr = (unsigned long)mem;
971         } else {
972                 struct vmap_area *va;
973                 va = alloc_vmap_area(size, PAGE_SIZE,
974                                 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
975                 if (IS_ERR(va))
976                         return NULL;
977 
978                 addr = va->va_start;
979                 mem = (void *)addr;
980         }
981         if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
982                 vm_unmap_ram(mem, count);
983                 return NULL;
984         }
985         return mem;
986 }
987 EXPORT_SYMBOL(vm_map_ram);
988 
989 /**
990  * vm_area_register_early - register vmap area early during boot
991  * @vm: vm_struct to register
992  * @align: requested alignment
993  *
994  * This function is used to register kernel vm area before
995  * vmalloc_init() is called.  @vm->size and @vm->flags should contain
996  * proper values on entry and other fields should be zero.  On return,
997  * vm->addr contains the allocated address.
998  *
999  * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1000  */
1001 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1002 {
1003         static size_t vm_init_off __initdata;
1004         unsigned long addr;
1005 
1006         addr = ALIGN(VMALLOC_START + vm_init_off, align);
1007         vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1008 
1009         vm->addr = (void *)addr;
1010 
1011         vm->next = vmlist;
1012         vmlist = vm;
1013 }
1014 
1015 void __init vmalloc_init(void)
1016 {
1017         struct vmap_area *va;
1018         struct vm_struct *tmp;
1019         int i;
1020 
1021         for_each_possible_cpu(i) {
1022                 struct vmap_block_queue *vbq;
1023 
1024                 vbq = &per_cpu(vmap_block_queue, i);
1025                 spin_lock_init(&vbq->lock);
1026                 INIT_LIST_HEAD(&vbq->free);
1027                 INIT_LIST_HEAD(&vbq->dirty);
1028                 vbq->nr_dirty = 0;
1029         }
1030 
1031         /* Import existing vmlist entries. */
1032         for (tmp = vmlist; tmp; tmp = tmp->next) {
1033                 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1034                 va->flags = tmp->flags | VM_VM_AREA;
1035                 va->va_start = (unsigned long)tmp->addr;
1036                 va->va_end = va->va_start + tmp->size;
1037                 __insert_vmap_area(va);
1038         }
1039         vmap_initialized = true;
1040 }
1041 
1042 /**
1043  * map_kernel_range_noflush - map kernel VM area with the specified pages
1044  * @addr: start of the VM area to map
1045  * @size: size of the VM area to map
1046  * @prot: page protection flags to use
1047  * @pages: pages to map
1048  *
1049  * Map PFN_UP(@size) pages at @addr.  The VM area @addr and @size
1050  * specify should have been allocated using get_vm_area() and its
1051  * friends.
1052  *
1053  * NOTE:
1054  * This function does NOT do any cache flushing.  The caller is
1055  * responsible for calling flush_cache_vmap() on to-be-mapped areas
1056  * before calling this function.
1057  *
1058  * RETURNS:
1059  * The number of pages mapped on success, -errno on failure.
1060  */
1061 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1062                              pgprot_t prot, struct page **pages)
1063 {
1064         return vmap_page_range_noflush(addr, addr + size, prot, pages);
1065 }
1066 
1067 /**
1068  * unmap_kernel_range_noflush - unmap kernel VM area
1069  * @addr: start of the VM area to unmap
1070  * @size: size of the VM area to unmap
1071  *
1072  * Unmap PFN_UP(@size) pages at @addr.  The VM area @addr and @size
1073  * specify should have been allocated using get_vm_area() and its
1074  * friends.
1075  *
1076  * NOTE:
1077  * This function does NOT do any cache flushing.  The caller is
1078  * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1079  * before calling this function and flush_tlb_kernel_range() after.
1080  */
1081 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1082 {
1083         vunmap_page_range(addr, addr + size);
1084 }
1085 
1086 /**
1087  * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1088  * @addr: start of the VM area to unmap
1089  * @size: size of the VM area to unmap
1090  *
1091  * Similar to unmap_kernel_range_noflush() but flushes vcache before
1092  * the unmapping and tlb after.
1093  */
1094 void unmap_kernel_range(unsigned long addr, unsigned long size)
1095 {
1096         unsigned long end = addr + size;
1097 
1098         flush_cache_vunmap(addr, end);
1099         vunmap_page_range(addr, end);
1100         flush_tlb_kernel_range(addr, end);
1101 }
1102 
1103 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
1104 {
1105         unsigned long addr = (unsigned long)area->addr;
1106         unsigned long end = addr + area->size - PAGE_SIZE;
1107         int err;
1108 
1109         err = vmap_page_range(addr, end, prot, *pages);
1110         if (err > 0) {
1111                 *pages += err;
1112                 err = 0;
1113         }
1114 
1115         return err;
1116 }
1117 EXPORT_SYMBOL_GPL(map_vm_area);
1118 
1119 /*** Old vmalloc interfaces ***/
1120 DEFINE_RWLOCK(vmlist_lock);
1121 struct vm_struct *vmlist;
1122 
1123 static struct vm_struct *__get_vm_area_node(unsigned long size,
1124                 unsigned long flags, unsigned long start, unsigned long end,
1125                 int node, gfp_t gfp_mask, void *caller)
1126 {
1127         static struct vmap_area *va;
1128         struct vm_struct *area;
1129         struct vm_struct *tmp, **p;
1130         unsigned long align = 1;
1131 
1132         BUG_ON(in_interrupt());
1133         if (flags & VM_IOREMAP) {
1134                 int bit = fls(size);
1135 
1136                 if (bit > IOREMAP_MAX_ORDER)
1137                         bit = IOREMAP_MAX_ORDER;
1138                 else if (bit < PAGE_SHIFT)
1139                         bit = PAGE_SHIFT;
1140 
1141                 align = 1ul << bit;
1142         }
1143 
1144         size = PAGE_ALIGN(size);
1145         if (unlikely(!size))
1146                 return NULL;
1147 
1148         area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1149         if (unlikely(!area))
1150                 return NULL;
1151 
1152         /*
1153          * We always allocate a guard page.
1154          */
1155         size += PAGE_SIZE;
1156 
1157         va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1158         if (IS_ERR(va)) {
1159                 kfree(area);
1160                 return NULL;
1161         }
1162 
1163         area->flags = flags;
1164         area->addr = (void *)va->va_start;
1165         area->size = size;
1166         area->pages = NULL;
1167         area->nr_pages = 0;
1168         area->phys_addr = 0;
1169         area->caller = caller;
1170         va->private = area;
1171         va->flags |= VM_VM_AREA;
1172 
1173         write_lock(&vmlist_lock);
1174         for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1175                 if (tmp->addr >= area->addr)
1176                         break;
1177         }
1178         area->next = *p;
1179         *p = area;
1180         write_unlock(&vmlist_lock);
1181 
1182         return area;
1183 }
1184 
1185 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1186                                 unsigned long start, unsigned long end)
1187 {
1188         return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1189                                                 __builtin_return_address(0));
1190 }
1191 EXPORT_SYMBOL_GPL(__get_vm_area);
1192 
1193 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1194                                        unsigned long start, unsigned long end,
1195                                        void *caller)
1196 {
1197         return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
1198                                   caller);
1199 }
1200 
1201 /**
1202  *      get_vm_area  -  reserve a contiguous kernel virtual area
1203  *      @size:          size of the area
1204  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
1205  *
1206  *      Search an area of @size in the kernel virtual mapping area,
1207  *      and reserved it for out purposes.  Returns the area descriptor
1208  *      on success or %NULL on failure.
1209  */
1210 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1211 {
1212         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1213                                 -1, GFP_KERNEL, __builtin_return_address(0));
1214 }
1215 
1216 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1217                                 void *caller)
1218 {
1219         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
1220                                                 -1, GFP_KERNEL, caller);
1221 }
1222 
1223 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
1224                                    int node, gfp_t gfp_mask)
1225 {
1226         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
1227                                   gfp_mask, __builtin_return_address(0));
1228 }
1229 
1230 static struct vm_struct *find_vm_area(const void *addr)
1231 {
1232         struct vmap_area *va;
1233 
1234         va = find_vmap_area((unsigned long)addr);
1235         if (va && va->flags & VM_VM_AREA)
1236                 return va->private;
1237 
1238         return NULL;
1239 }
1240 
1241 /**
1242  *      remove_vm_area  -  find and remove a continuous kernel virtual area
1243  *      @addr:          base address
1244  *
1245  *      Search for the kernel VM area starting at @addr, and remove it.
1246  *      This function returns the found VM area, but using it is NOT safe
1247  *      on SMP machines, except for its size or flags.
1248  */
1249 struct vm_struct *remove_vm_area(const void *addr)
1250 {
1251         struct vmap_area *va;
1252 
1253         va = find_vmap_area((unsigned long)addr);
1254         if (va && va->flags & VM_VM_AREA) {
1255                 struct vm_struct *vm = va->private;
1256                 struct vm_struct *tmp, **p;
1257 
1258                 vmap_debug_free_range(va->va_start, va->va_end);
1259                 free_unmap_vmap_area(va);
1260                 vm->size -= PAGE_SIZE;
1261 
1262                 write_lock(&vmlist_lock);
1263                 for (p = &vmlist; (tmp = *p) != vm; p = &tmp->next)
1264                         ;
1265                 *p = tmp->next;
1266                 write_unlock(&vmlist_lock);
1267 
1268                 return vm;
1269         }
1270         return NULL;
1271 }
1272 
1273 static void __vunmap(const void *addr, int deallocate_pages)
1274 {
1275         struct vm_struct *area;
1276 
1277         if (!addr)
1278                 return;
1279 
1280         if ((PAGE_SIZE-1) & (unsigned long)addr) {
1281                 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
1282                 return;
1283         }
1284 
1285         area = remove_vm_area(addr);
1286         if (unlikely(!area)) {
1287                 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1288                                 addr);
1289                 return;
1290         }
1291 
1292         debug_check_no_locks_freed(addr, area->size);
1293         debug_check_no_obj_freed(addr, area->size);
1294 
1295         if (deallocate_pages) {
1296                 int i;
1297 
1298                 for (i = 0; i < area->nr_pages; i++) {
1299                         struct page *page = area->pages[i];
1300 
1301                         BUG_ON(!page);
1302                         __free_page(page);
1303                 }
1304 
1305                 if (area->flags & VM_VPAGES)
1306                         vfree(area->pages);
1307                 else
1308                         kfree(area->pages);
1309         }
1310 
1311         kfree(area);
1312         return;
1313 }
1314 
1315 /**
1316  *      vfree  -  release memory allocated by vmalloc()
1317  *      @addr:          memory base address
1318  *
1319  *      Free the virtually continuous memory area starting at @addr, as
1320  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1321  *      NULL, no operation is performed.
1322  *
1323  *      Must not be called in interrupt context.
1324  */
1325 void vfree(const void *addr)
1326 {
1327         BUG_ON(in_interrupt());
1328 
1329         kmemleak_free(addr);
1330 
1331         __vunmap(addr, 1);
1332 }
1333 EXPORT_SYMBOL(vfree);
1334 
1335 /**
1336  *      vunmap  -  release virtual mapping obtained by vmap()
1337  *      @addr:          memory base address
1338  *
1339  *      Free the virtually contiguous memory area starting at @addr,
1340  *      which was created from the page array passed to vmap().
1341  *
1342  *      Must not be called in interrupt context.
1343  */
1344 void vunmap(const void *addr)
1345 {
1346         BUG_ON(in_interrupt());
1347         might_sleep();
1348         __vunmap(addr, 0);
1349 }
1350 EXPORT_SYMBOL(vunmap);
1351 
1352 /**
1353  *      vmap  -  map an array of pages into virtually contiguous space
1354  *      @pages:         array of page pointers
1355  *      @count:         number of pages to map
1356  *      @flags:         vm_area->flags
1357  *      @prot:          page protection for the mapping
1358  *
1359  *      Maps @count pages from @pages into contiguous kernel virtual
1360  *      space.
1361  */
1362 void *vmap(struct page **pages, unsigned int count,
1363                 unsigned long flags, pgprot_t prot)
1364 {
1365         struct vm_struct *area;
1366 
1367         might_sleep();
1368 
1369         if (count > totalram_pages)
1370                 return NULL;
1371 
1372         area = get_vm_area_caller((count << PAGE_SHIFT), flags,
1373                                         __builtin_return_address(0));
1374         if (!area)
1375                 return NULL;
1376 
1377         if (map_vm_area(area, prot, &pages)) {
1378                 vunmap(area->addr);
1379                 return NULL;
1380         }
1381 
1382         return area->addr;
1383 }
1384 EXPORT_SYMBOL(vmap);
1385 
1386 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1387                             int node, void *caller);
1388 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1389                                  pgprot_t prot, int node, void *caller)
1390 {
1391         struct page **pages;
1392         unsigned int nr_pages, array_size, i;
1393 
1394         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
1395         array_size = (nr_pages * sizeof(struct page *));
1396 
1397         area->nr_pages = nr_pages;
1398         /* Please note that the recursion is strictly bounded. */
1399         if (array_size > PAGE_SIZE) {
1400                 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
1401                                 PAGE_KERNEL, node, caller);
1402                 area->flags |= VM_VPAGES;
1403         } else {
1404                 pages = kmalloc_node(array_size,
1405                                 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
1406                                 node);
1407         }
1408         area->pages = pages;
1409         area->caller = caller;
1410         if (!area->pages) {
1411                 remove_vm_area(area->addr);
1412                 kfree(area);
1413                 return NULL;
1414         }
1415 
1416         for (i = 0; i < area->nr_pages; i++) {
1417                 struct page *page;
1418 
1419                 if (node < 0)
1420                         page = alloc_page(gfp_mask);
1421                 else
1422                         page = alloc_pages_node(node, gfp_mask, 0);
1423 
1424                 if (unlikely(!page)) {
1425                         /* Successfully allocated i pages, free them in __vunmap() */
1426                         area->nr_pages = i;
1427                         goto fail;
1428                 }
1429                 area->pages[i] = page;
1430         }
1431 
1432         if (map_vm_area(area, prot, &pages))
1433                 goto fail;
1434         return area->addr;
1435 
1436 fail:
1437         vfree(area->addr);
1438         return NULL;
1439 }
1440 
1441 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
1442 {
1443         void *addr = __vmalloc_area_node(area, gfp_mask, prot, -1,
1444                                          __builtin_return_address(0));
1445 
1446         /*
1447          * A ref_count = 3 is needed because the vm_struct and vmap_area
1448          * structures allocated in the __get_vm_area_node() function contain
1449          * references to the virtual address of the vmalloc'ed block.
1450          */
1451         kmemleak_alloc(addr, area->size - PAGE_SIZE, 3, gfp_mask);
1452 
1453         return addr;
1454 }
1455 
1456 /**
1457  *      __vmalloc_node  -  allocate virtually contiguous memory
1458  *      @size:          allocation size
1459  *      @gfp_mask:      flags for the page level allocator
1460  *      @prot:          protection mask for the allocated pages
1461  *      @node:          node to use for allocation or -1
1462  *      @caller:        caller's return address
1463  *
1464  *      Allocate enough pages to cover @size from the page level
1465  *      allocator with @gfp_mask flags.  Map them into contiguous
1466  *      kernel virtual space, using a pagetable protection of @prot.
1467  */
1468 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
1469                                                 int node, void *caller)
1470 {
1471         struct vm_struct *area;
1472         void *addr;
1473         unsigned long real_size = size;
1474 
1475         size = PAGE_ALIGN(size);
1476         if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1477                 return NULL;
1478 
1479         area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
1480                                                 node, gfp_mask, caller);
1481 
1482         if (!area)
1483                 return NULL;
1484 
1485         addr = __vmalloc_area_node(area, gfp_mask, prot, node, caller);
1486 
1487         /*
1488          * A ref_count = 3 is needed because the vm_struct and vmap_area
1489          * structures allocated in the __get_vm_area_node() function contain
1490          * references to the virtual address of the vmalloc'ed block.
1491          */
1492         kmemleak_alloc(addr, real_size, 3, gfp_mask);
1493 
1494         return addr;
1495 }
1496 
1497 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1498 {
1499         return __vmalloc_node(size, gfp_mask, prot, -1,
1500                                 __builtin_return_address(0));
1501 }
1502 EXPORT_SYMBOL(__vmalloc);
1503 
1504 /**
1505  *      vmalloc  -  allocate virtually contiguous memory
1506  *      @size:          allocation size
1507  *      Allocate enough pages to cover @size from the page level
1508  *      allocator and map them into contiguous kernel virtual space.
1509  *
1510  *      For tight control over page level allocator and protection flags
1511  *      use __vmalloc() instead.
1512  */
1513 void *vmalloc(unsigned long size)
1514 {
1515         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1516                                         -1, __builtin_return_address(0));
1517 }
1518 EXPORT_SYMBOL(vmalloc);
1519 
1520 /**
1521  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1522  * @size: allocation size
1523  *
1524  * The resulting memory area is zeroed so it can be mapped to userspace
1525  * without leaking data.
1526  */
1527 void *vmalloc_user(unsigned long size)
1528 {
1529         struct vm_struct *area;
1530         void *ret;
1531 
1532         ret = __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO,
1533                              PAGE_KERNEL, -1, __builtin_return_address(0));
1534         if (ret) {
1535                 area = find_vm_area(ret);
1536                 area->flags |= VM_USERMAP;
1537         }
1538         return ret;
1539 }
1540 EXPORT_SYMBOL(vmalloc_user);
1541 
1542 /**
1543  *      vmalloc_node  -  allocate memory on a specific node
1544  *      @size:          allocation size
1545  *      @node:          numa node
1546  *
1547  *      Allocate enough pages to cover @size from the page level
1548  *      allocator and map them into contiguous kernel virtual space.
1549  *
1550  *      For tight control over page level allocator and protection flags
1551  *      use __vmalloc() instead.
1552  */
1553 void *vmalloc_node(unsigned long size, int node)
1554 {
1555         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
1556                                         node, __builtin_return_address(0));
1557 }
1558 EXPORT_SYMBOL(vmalloc_node);
1559 
1560 #ifndef PAGE_KERNEL_EXEC
1561 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1562 #endif
1563 
1564 /**
1565  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
1566  *      @size:          allocation size
1567  *
1568  *      Kernel-internal function to allocate enough pages to cover @size
1569  *      the page level allocator and map them into contiguous and
1570  *      executable kernel virtual space.
1571  *
1572  *      For tight control over page level allocator and protection flags
1573  *      use __vmalloc() instead.
1574  */
1575 
1576 void *vmalloc_exec(unsigned long size)
1577 {
1578         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC,
1579                               -1, __builtin_return_address(0));
1580 }
1581 
1582 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1583 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1584 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1585 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1586 #else
1587 #define GFP_VMALLOC32 GFP_KERNEL
1588 #endif
1589 
1590 /**
1591  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
1592  *      @size:          allocation size
1593  *
1594  *      Allocate enough 32bit PA addressable pages to cover @size from the
1595  *      page level allocator and map them into contiguous kernel virtual space.
1596  */
1597 void *vmalloc_32(unsigned long size)
1598 {
1599         return __vmalloc_node(size, GFP_VMALLOC32, PAGE_KERNEL,
1600                               -1, __builtin_return_address(0));
1601 }
1602 EXPORT_SYMBOL(vmalloc_32);
1603 
1604 /**
1605  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1606  *      @size:          allocation size
1607  *
1608  * The resulting memory area is 32bit addressable and zeroed so it can be
1609  * mapped to userspace without leaking data.
1610  */
1611 void *vmalloc_32_user(unsigned long size)
1612 {
1613         struct vm_struct *area;
1614         void *ret;
1615 
1616         ret = __vmalloc_node(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1617                              -1, __builtin_return_address(0));
1618         if (ret) {
1619                 area = find_vm_area(ret);
1620                 area->flags |= VM_USERMAP;
1621         }
1622         return ret;
1623 }
1624 EXPORT_SYMBOL(vmalloc_32_user);
1625 
1626 long vread(char *buf, char *addr, unsigned long count)
1627 {
1628         struct vm_struct *tmp;
1629         char *vaddr, *buf_start = buf;
1630         unsigned long n;
1631 
1632         /* Don't allow overflow */
1633         if ((unsigned long) addr + count < count)
1634                 count = -(unsigned long) addr;
1635 
1636         read_lock(&vmlist_lock);
1637         for (tmp = vmlist; tmp; tmp = tmp->next) {
1638                 vaddr = (char *) tmp->addr;
1639                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1640                         continue;
1641                 while (addr < vaddr) {
1642                         if (count == 0)
1643                                 goto finished;
1644                         *buf = '\0';
1645                         buf++;
1646                         addr++;
1647                         count--;
1648                 }
1649                 n = vaddr + tmp->size - PAGE_SIZE - addr;
1650                 do {
1651                         if (count == 0)
1652                                 goto finished;
1653                         *buf = *addr;
1654                         buf++;
1655                         addr++;
1656                         count--;
1657                 } while (--n > 0);
1658         }
1659 finished:
1660         read_unlock(&vmlist_lock);
1661         return buf - buf_start;
1662 }
1663 
1664 long vwrite(char *buf, char *addr, unsigned long count)
1665 {
1666         struct vm_struct *tmp;
1667         char *vaddr, *buf_start = buf;
1668         unsigned long n;
1669 
1670         /* Don't allow overflow */
1671         if ((unsigned long) addr + count < count)
1672                 count = -(unsigned long) addr;
1673 
1674         read_lock(&vmlist_lock);
1675         for (tmp = vmlist; tmp; tmp = tmp->next) {
1676                 vaddr = (char *) tmp->addr;
1677                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
1678                         continue;
1679                 while (addr < vaddr) {
1680                         if (count == 0)
1681                                 goto finished;
1682                         buf++;
1683                         addr++;
1684                         count--;
1685                 }
1686                 n = vaddr + tmp->size - PAGE_SIZE - addr;
1687                 do {
1688                         if (count == 0)
1689                                 goto finished;
1690                         *addr = *buf;
1691                         buf++;
1692                         addr++;
1693                         count--;
1694                 } while (--n > 0);
1695         }
1696 finished:
1697         read_unlock(&vmlist_lock);
1698         return buf - buf_start;
1699 }
1700 
1701 /**
1702  *      remap_vmalloc_range  -  map vmalloc pages to userspace
1703  *      @vma:           vma to cover (map full range of vma)
1704  *      @addr:          vmalloc memory
1705  *      @pgoff:         number of pages into addr before first page to map
1706  *
1707  *      Returns:        0 for success, -Exxx on failure
1708  *
1709  *      This function checks that addr is a valid vmalloc'ed area, and
1710  *      that it is big enough to cover the vma. Will return failure if
1711  *      that criteria isn't met.
1712  *
1713  *      Similar to remap_pfn_range() (see mm/memory.c)
1714  */
1715 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
1716                                                 unsigned long pgoff)
1717 {
1718         struct vm_struct *area;
1719         unsigned long uaddr = vma->vm_start;
1720         unsigned long usize = vma->vm_end - vma->vm_start;
1721 
1722         if ((PAGE_SIZE-1) & (unsigned long)addr)
1723                 return -EINVAL;
1724 
1725         area = find_vm_area(addr);
1726         if (!area)
1727                 return -EINVAL;
1728 
1729         if (!(area->flags & VM_USERMAP))
1730                 return -EINVAL;
1731 
1732         if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
1733                 return -EINVAL;
1734 
1735         addr += pgoff << PAGE_SHIFT;
1736         do {
1737                 struct page *page = vmalloc_to_page(addr);
1738                 int ret;
1739 
1740                 ret = vm_insert_page(vma, uaddr, page);
1741                 if (ret)
1742                         return ret;
1743 
1744                 uaddr += PAGE_SIZE;
1745                 addr += PAGE_SIZE;
1746                 usize -= PAGE_SIZE;
1747         } while (usize > 0);
1748 
1749         /* Prevent "things" like memory migration? VM_flags need a cleanup... */
1750         vma->vm_flags |= VM_RESERVED;
1751 
1752         return 0;
1753 }
1754 EXPORT_SYMBOL(remap_vmalloc_range);
1755 
1756 /*
1757  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
1758  * have one.
1759  */
1760 void  __attribute__((weak)) vmalloc_sync_all(void)
1761 {
1762 }
1763 
1764 
1765 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
1766 {
1767         /* apply_to_page_range() does all the hard work. */
1768         return 0;
1769 }
1770 
1771 /**
1772  *      alloc_vm_area - allocate a range of kernel address space
1773  *      @size:          size of the area
1774  *
1775  *      Returns:        NULL on failure, vm_struct on success
1776  *
1777  *      This function reserves a range of kernel address space, and
1778  *      allocates pagetables to map that range.  No actual mappings
1779  *      are created.  If the kernel address space is not shared
1780  *      between processes, it syncs the pagetable across all
1781  *      processes.
1782  */
1783 struct vm_struct *alloc_vm_area(size_t size)
1784 {
1785         struct vm_struct *area;
1786 
1787         area = get_vm_area_caller(size, VM_IOREMAP,
1788                                 __builtin_return_address(0));
1789         if (area == NULL)
1790                 return NULL;
1791 
1792         /*
1793          * This ensures that page tables are constructed for this region
1794          * of kernel virtual address space and mapped into init_mm.
1795          */
1796         if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
1797                                 area->size, f, NULL)) {
1798                 free_vm_area(area);
1799                 return NULL;
1800         }
1801 
1802         /* Make sure the pagetables are constructed in process kernel
1803            mappings */
1804         vmalloc_sync_all();
1805 
1806         return area;
1807 }
1808 EXPORT_SYMBOL_GPL(alloc_vm_area);
1809 
1810 void free_vm_area(struct vm_struct *area)
1811 {
1812         struct vm_struct *ret;
1813         ret = remove_vm_area(area->addr);
1814         BUG_ON(ret != area);
1815         kfree(area);
1816 }
1817 EXPORT_SYMBOL_GPL(free_vm_area);
1818 
1819 
1820 #ifdef CONFIG_PROC_FS
1821 static void *s_start(struct seq_file *m, loff_t *pos)
1822 {
1823         loff_t n = *pos;
1824         struct vm_struct *v;
1825 
1826         read_lock(&vmlist_lock);
1827         v = vmlist;
1828         while (n > 0 && v) {
1829                 n--;
1830                 v = v->next;
1831         }
1832         if (!n)
1833                 return v;
1834 
1835         return NULL;
1836 
1837 }
1838 
1839 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
1840 {
1841         struct vm_struct *v = p;
1842 
1843         ++*pos;
1844         return v->next;
1845 }
1846 
1847 static void s_stop(struct seq_file *m, void *p)
1848 {
1849         read_unlock(&vmlist_lock);
1850 }
1851 
1852 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
1853 {
1854         if (NUMA_BUILD) {
1855                 unsigned int nr, *counters = m->private;
1856 
1857                 if (!counters)
1858                         return;
1859 
1860                 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
1861 
1862                 for (nr = 0; nr < v->nr_pages; nr++)
1863                         counters[page_to_nid(v->pages[nr])]++;
1864 
1865                 for_each_node_state(nr, N_HIGH_MEMORY)
1866                         if (counters[nr])
1867                                 seq_printf(m, " N%u=%u", nr, counters[nr]);
1868         }
1869 }
1870 
1871 static int s_show(struct seq_file *m, void *p)
1872 {
1873         struct vm_struct *v = p;
1874 
1875         seq_printf(m, "0x%p-0x%p %7ld",
1876                 v->addr, v->addr + v->size, v->size);
1877 
1878         if (v->caller) {
1879                 char buff[KSYM_SYMBOL_LEN];
1880 
1881                 seq_putc(m, ' ');
1882                 sprint_symbol(buff, (unsigned long)v->caller);
1883                 seq_puts(m, buff);
1884         }
1885 
1886         if (v->nr_pages)
1887                 seq_printf(m, " pages=%d", v->nr_pages);
1888 
1889         if (v->phys_addr)
1890                 seq_printf(m, " phys=%lx", v->phys_addr);
1891 
1892         if (v->flags & VM_IOREMAP)
1893                 seq_printf(m, " ioremap");
1894 
1895         if (v->flags & VM_ALLOC)
1896                 seq_printf(m, " vmalloc");
1897 
1898         if (v->flags & VM_MAP)
1899                 seq_printf(m, " vmap");
1900 
1901         if (v->flags & VM_USERMAP)
1902                 seq_printf(m, " user");
1903 
1904         if (v->flags & VM_VPAGES)
1905                 seq_printf(m, " vpages");
1906 
1907         show_numa_info(m, v);
1908         seq_putc(m, '\n');
1909         return 0;
1910 }
1911 
1912 static const struct seq_operations vmalloc_op = {
1913         .start = s_start,
1914         .next = s_next,
1915         .stop = s_stop,
1916         .show = s_show,
1917 };
1918 
1919 static int vmalloc_open(struct inode *inode, struct file *file)
1920 {
1921         unsigned int *ptr = NULL;
1922         int ret;
1923 
1924         if (NUMA_BUILD)
1925                 ptr = kmalloc(nr_node_ids * sizeof(unsigned int), GFP_KERNEL);
1926         ret = seq_open(file, &vmalloc_op);
1927         if (!ret) {
1928                 struct seq_file *m = file->private_data;
1929                 m->private = ptr;
1930         } else
1931                 kfree(ptr);
1932         return ret;
1933 }
1934 
1935 static const struct file_operations proc_vmalloc_operations = {
1936         .open           = vmalloc_open,
1937         .read           = seq_read,
1938         .llseek         = seq_lseek,
1939         .release        = seq_release_private,
1940 };
1941 
1942 static int __init proc_vmalloc_init(void)
1943 {
1944         proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
1945         return 0;
1946 }
1947 module_init(proc_vmalloc_init);
1948 #endif
1949 
1950 
  This page was automatically generated by the LXR engine.