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
11 #include <linux/vmalloc.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>
35 #include <linux/uaccess.h>
36 #include <asm/tlbflush.h>
37 #include <asm/shmparam.h>
41 struct vfree_deferred {
42 struct llist_head list;
43 struct work_struct wq;
45 static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
47 static void __vunmap(const void *, int);
49 static void free_work(struct work_struct *w)
51 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
52 struct llist_node *t, *llnode;
54 llist_for_each_safe(llnode, t, llist_del_all(&p->list))
55 __vunmap((void *)llnode, 1);
58 /*** Page table manipulation functions ***/
60 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
64 pte = pte_offset_kernel(pmd, addr);
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);
71 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
76 pmd = pmd_offset(pud, addr);
78 next = pmd_addr_end(addr, end);
79 if (pmd_clear_huge(pmd))
81 if (pmd_none_or_clear_bad(pmd))
83 vunmap_pte_range(pmd, addr, next);
84 } while (pmd++, addr = next, addr != end);
87 static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end)
92 pud = pud_offset(p4d, addr);
94 next = pud_addr_end(addr, end);
95 if (pud_clear_huge(pud))
97 if (pud_none_or_clear_bad(pud))
99 vunmap_pmd_range(pud, addr, next);
100 } while (pud++, addr = next, addr != end);
103 static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end)
108 p4d = p4d_offset(pgd, addr);
110 next = p4d_addr_end(addr, end);
111 if (p4d_clear_huge(p4d))
113 if (p4d_none_or_clear_bad(p4d))
115 vunmap_pud_range(p4d, addr, next);
116 } while (p4d++, addr = next, addr != end);
119 static void vunmap_page_range(unsigned long addr, unsigned long end)
125 pgd = pgd_offset_k(addr);
127 next = pgd_addr_end(addr, end);
128 if (pgd_none_or_clear_bad(pgd))
130 vunmap_p4d_range(pgd, addr, next);
131 } while (pgd++, addr = next, addr != end);
134 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
135 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
140 * nr is a running index into the array which helps higher level
141 * callers keep track of where we're up to.
144 pte = pte_alloc_kernel(pmd, addr);
148 struct page *page = pages[*nr];
150 if (WARN_ON(!pte_none(*pte)))
154 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
156 } while (pte++, addr += PAGE_SIZE, addr != end);
160 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
161 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
166 pmd = pmd_alloc(&init_mm, pud, addr);
170 next = pmd_addr_end(addr, end);
171 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
173 } while (pmd++, addr = next, addr != end);
177 static int vmap_pud_range(p4d_t *p4d, unsigned long addr,
178 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
183 pud = pud_alloc(&init_mm, p4d, addr);
187 next = pud_addr_end(addr, end);
188 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
190 } while (pud++, addr = next, addr != end);
194 static int vmap_p4d_range(pgd_t *pgd, unsigned long addr,
195 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
200 p4d = p4d_alloc(&init_mm, pgd, addr);
204 next = p4d_addr_end(addr, end);
205 if (vmap_pud_range(p4d, addr, next, prot, pages, nr))
207 } while (p4d++, addr = next, addr != end);
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.
215 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
217 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
218 pgprot_t prot, struct page **pages)
222 unsigned long addr = start;
227 pgd = pgd_offset_k(addr);
229 next = pgd_addr_end(addr, end);
230 err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr);
233 } while (pgd++, addr = next, addr != end);
238 static int vmap_page_range(unsigned long start, unsigned long end,
239 pgprot_t prot, struct page **pages)
243 ret = vmap_page_range_noflush(start, end, prot, pages);
244 flush_cache_vmap(start, end);
248 int is_vmalloc_or_module_addr(const void *x)
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.
255 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
256 unsigned long addr = (unsigned long)x;
257 if (addr >= MODULES_VADDR && addr < MODULES_END)
260 return is_vmalloc_addr(x);
264 * Walk a vmap address to the struct page it maps.
266 struct page *vmalloc_to_page(const void *vmalloc_addr)
268 unsigned long addr = (unsigned long) vmalloc_addr;
269 struct page *page = NULL;
270 pgd_t *pgd = pgd_offset_k(addr);
277 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
278 * architectures that do not vmalloc module space
280 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
284 p4d = p4d_offset(pgd, addr);
287 pud = pud_offset(p4d, addr);
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.
297 WARN_ON_ONCE(pud_bad(*pud));
298 if (pud_none(*pud) || pud_bad(*pud))
300 pmd = pmd_offset(pud, addr);
301 WARN_ON_ONCE(pmd_bad(*pmd));
302 if (pmd_none(*pmd) || pmd_bad(*pmd))
305 ptep = pte_offset_map(pmd, addr);
307 if (pte_present(pte))
308 page = pte_page(pte);
312 EXPORT_SYMBOL(vmalloc_to_page);
315 * Map a vmalloc()-space virtual address to the physical page frame number.
317 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
319 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
321 EXPORT_SYMBOL(vmalloc_to_pfn);
324 /*** Global kva allocator ***/
326 #define VM_LAZY_FREE 0x02
327 #define VM_VM_AREA 0x04
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;
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;
341 static unsigned long vmap_area_pcpu_hole;
343 static struct vmap_area *__find_vmap_area(unsigned long addr)
345 struct rb_node *n = vmap_area_root.rb_node;
348 struct vmap_area *va;
350 va = rb_entry(n, struct vmap_area, rb_node);
351 if (addr < va->va_start)
353 else if (addr >= va->va_end)
362 static void __insert_vmap_area(struct vmap_area *va)
364 struct rb_node **p = &vmap_area_root.rb_node;
365 struct rb_node *parent = NULL;
369 struct vmap_area *tmp_va;
372 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
373 if (va->va_start < tmp_va->va_end)
375 else if (va->va_end > tmp_va->va_start)
381 rb_link_node(&va->rb_node, parent, p);
382 rb_insert_color(&va->rb_node, &vmap_area_root);
384 /* address-sort this list */
385 tmp = rb_prev(&va->rb_node);
387 struct vmap_area *prev;
388 prev = rb_entry(tmp, struct vmap_area, rb_node);
389 list_add_rcu(&va->list, &prev->list);
391 list_add_rcu(&va->list, &vmap_area_list);
394 static void purge_vmap_area_lazy(void);
396 static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
399 * Allocate a region of KVA of the specified size and alignment, within the
402 static struct vmap_area *alloc_vmap_area(unsigned long size,
404 unsigned long vstart, unsigned long vend,
405 int node, gfp_t gfp_mask)
407 struct vmap_area *va;
411 struct vmap_area *first;
414 BUG_ON(offset_in_page(size));
415 BUG_ON(!is_power_of_2(align));
419 va = kmalloc_node(sizeof(struct vmap_area),
420 gfp_mask & GFP_RECLAIM_MASK, node);
422 return ERR_PTR(-ENOMEM);
425 * Only scan the relevant parts containing pointers to other objects
426 * to avoid false negatives.
428 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
431 spin_lock(&vmap_area_lock);
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.
441 if (!free_vmap_cache ||
442 size < cached_hole_size ||
443 vstart < cached_vstart ||
444 align < cached_align) {
446 cached_hole_size = 0;
447 free_vmap_cache = NULL;
449 /* record if we encounter less permissive parameters */
450 cached_vstart = vstart;
451 cached_align = align;
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);
459 if (addr + size < addr)
463 addr = ALIGN(vstart, align);
464 if (addr + size < addr)
467 n = vmap_area_root.rb_node;
471 struct vmap_area *tmp;
472 tmp = rb_entry(n, struct vmap_area, rb_node);
473 if (tmp->va_end >= addr) {
475 if (tmp->va_start <= addr)
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)
494 if (list_is_last(&first->list, &vmap_area_list))
497 first = list_next_entry(first, list);
501 if (addr + size > vend)
505 va->va_end = addr + size;
507 __insert_vmap_area(va);
508 free_vmap_cache = &va->rb_node;
509 spin_unlock(&vmap_area_lock);
511 BUG_ON(!IS_ALIGNED(va->va_start, align));
512 BUG_ON(va->va_start < vstart);
513 BUG_ON(va->va_end > vend);
518 spin_unlock(&vmap_area_lock);
520 purge_vmap_area_lazy();
525 if (gfpflags_allow_blocking(gfp_mask)) {
526 unsigned long freed = 0;
527 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
534 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
535 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
538 return ERR_PTR(-EBUSY);
541 int register_vmap_purge_notifier(struct notifier_block *nb)
543 return blocking_notifier_chain_register(&vmap_notify_list, nb);
545 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
547 int unregister_vmap_purge_notifier(struct notifier_block *nb)
549 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
551 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
553 static void __free_vmap_area(struct vmap_area *va)
555 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
557 if (free_vmap_cache) {
558 if (va->va_end < cached_vstart) {
559 free_vmap_cache = NULL;
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);
566 * We don't try to update cached_hole_size or
567 * cached_align, but it won't go very wrong.
572 rb_erase(&va->rb_node, &vmap_area_root);
573 RB_CLEAR_NODE(&va->rb_node);
574 list_del_rcu(&va->list);
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.
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);
585 kfree_rcu(va, rcu_head);
589 * Free a region of KVA allocated by alloc_vmap_area
591 static void free_vmap_area(struct vmap_area *va)
593 spin_lock(&vmap_area_lock);
594 __free_vmap_area(va);
595 spin_unlock(&vmap_area_lock);
599 * Clear the pagetable entries of a given vmap_area
601 static void unmap_vmap_area(struct vmap_area *va)
603 vunmap_page_range(va->va_start, va->va_end);
606 static void vmap_debug_free_range(unsigned long start, unsigned long end)
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
614 * All the lazy freeing logic is still retained, in order to minimise
615 * intrusiveness of this debugging feature.
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).
620 if (debug_pagealloc_enabled()) {
621 vunmap_page_range(start, end);
622 flush_tlb_kernel_range(start, end);
627 * lazy_max_pages is the maximum amount of virtual address space we gather up
628 * before attempting to purge with a TLB flush.
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.
642 static unsigned long lazy_max_pages(void)
646 log = fls(num_online_cpus());
648 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
651 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
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.
658 static DEFINE_MUTEX(vmap_purge_lock);
660 /* for per-CPU blocks */
661 static void purge_fragmented_blocks_allcpus(void);
664 * called before a call to iounmap() if the caller wants vm_area_struct's
667 void set_iounmap_nonlazy(void)
669 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
673 * Purges all lazily-freed vmap areas.
675 static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
677 struct llist_node *valist;
678 struct vmap_area *va;
679 struct vmap_area *n_va;
680 bool do_free = false;
682 lockdep_assert_held(&vmap_purge_lock);
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)
696 flush_tlb_kernel_range(start, end);
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;
702 __free_vmap_area(va);
703 atomic_sub(nr, &vmap_lazy_nr);
704 cond_resched_lock(&vmap_area_lock);
706 spin_unlock(&vmap_area_lock);
711 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
712 * is already purging.
714 static void try_purge_vmap_area_lazy(void)
716 if (mutex_trylock(&vmap_purge_lock)) {
717 __purge_vmap_area_lazy(ULONG_MAX, 0);
718 mutex_unlock(&vmap_purge_lock);
723 * Kick off a purge of the outstanding lazy areas.
725 static void purge_vmap_area_lazy(void)
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);
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
738 static void free_vmap_area_noflush(struct vmap_area *va)
742 nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
745 /* After this point, we may free va at any time */
746 llist_add(&va->purge_list, &vmap_purge_list);
748 if (unlikely(nr_lazy > lazy_max_pages()))
749 try_purge_vmap_area_lazy();
753 * Free and unmap a vmap area
755 static void free_unmap_vmap_area(struct vmap_area *va)
757 flush_cache_vunmap(va->va_start, va->va_end);
759 free_vmap_area_noflush(va);
762 static struct vmap_area *find_vmap_area(unsigned long addr)
764 struct vmap_area *va;
766 spin_lock(&vmap_area_lock);
767 va = __find_vmap_area(addr);
768 spin_unlock(&vmap_area_lock);
773 /*** Per cpu kva allocator ***/
776 * vmap space is limited especially on 32 bit architectures. Ensure there is
777 * room for at least 16 percpu vmap blocks per CPU.
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)
784 #if BITS_PER_LONG == 32
785 #define VMALLOC_SPACE (128UL*1024*1024)
787 #define VMALLOC_SPACE (128UL*1024*1024*1024)
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))
801 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
803 static bool vmap_initialized __read_mostly = false;
805 struct vmap_block_queue {
807 struct list_head free;
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;
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);
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.
828 static DEFINE_SPINLOCK(vmap_block_tree_lock);
829 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
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
838 static unsigned long addr_to_vb_idx(unsigned long addr)
840 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
841 addr /= VMAP_BLOCK_SIZE;
845 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
849 addr = va_start + (pages_off << PAGE_SHIFT);
850 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
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
860 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
862 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
864 struct vmap_block_queue *vbq;
865 struct vmap_block *vb;
866 struct vmap_area *va;
867 unsigned long vb_idx;
871 node = numa_node_id();
873 vb = kmalloc_node(sizeof(struct vmap_block),
874 gfp_mask & GFP_RECLAIM_MASK, node);
876 return ERR_PTR(-ENOMEM);
878 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
879 VMALLOC_START, VMALLOC_END,
886 err = radix_tree_preload(gfp_mask);
893 vaddr = vmap_block_vaddr(va->va_start, 0);
894 spin_lock_init(&vb->lock);
896 /* At least something should be left free */
897 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
898 vb->free = VMAP_BBMAP_BITS - (1UL << order);
900 vb->dirty_min = VMAP_BBMAP_BITS;
902 INIT_LIST_HEAD(&vb->free_list);
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);
909 radix_tree_preload_end();
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);
920 static void free_vmap_block(struct vmap_block *vb)
922 struct vmap_block *tmp;
923 unsigned long vb_idx;
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);
931 free_vmap_area_noflush(vb->va);
932 kfree_rcu(vb, rcu_head);
935 static void purge_fragmented_blocks(int cpu)
938 struct vmap_block *vb;
939 struct vmap_block *n_vb;
940 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
943 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
945 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
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 */
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);
960 spin_unlock(&vb->lock);
964 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
965 list_del(&vb->purge);
970 static void purge_fragmented_blocks_allcpus(void)
974 for_each_possible_cpu(cpu)
975 purge_fragmented_blocks(cpu);
978 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
980 struct vmap_block_queue *vbq;
981 struct vmap_block *vb;
985 BUG_ON(offset_in_page(size));
986 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
987 if (WARN_ON(size == 0)) {
989 * Allocating 0 bytes isn't what caller wants since
990 * get_order(0) returns funny result. Just warn and terminate
995 order = get_order(size);
998 vbq = &get_cpu_var(vmap_block_queue);
999 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1000 unsigned long pages_off;
1002 spin_lock(&vb->lock);
1003 if (vb->free < (1UL << order)) {
1004 spin_unlock(&vb->lock);
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);
1017 spin_unlock(&vb->lock);
1021 put_cpu_var(vmap_block_queue);
1024 /* Allocate new block if nothing was found */
1026 vaddr = new_vmap_block(order, gfp_mask);
1031 static void vb_free(const void *addr, unsigned long size)
1033 unsigned long offset;
1034 unsigned long vb_idx;
1036 struct vmap_block *vb;
1038 BUG_ON(offset_in_page(size));
1039 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
1041 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1043 order = get_order(size);
1045 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
1046 offset >>= PAGE_SHIFT;
1048 vb_idx = addr_to_vb_idx((unsigned long)addr);
1050 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1054 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1056 spin_lock(&vb->lock);
1058 /* Expand dirty range */
1059 vb->dirty_min = min(vb->dirty_min, offset);
1060 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
1062 vb->dirty += 1UL << order;
1063 if (vb->dirty == VMAP_BBMAP_BITS) {
1065 spin_unlock(&vb->lock);
1066 free_vmap_block(vb);
1068 spin_unlock(&vb->lock);
1072 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
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).
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.
1084 void vm_unmap_aliases(void)
1086 unsigned long start = ULONG_MAX, end = 0;
1090 if (unlikely(!vmap_initialized))
1095 for_each_possible_cpu(cpu) {
1096 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1097 struct vmap_block *vb;
1100 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1101 spin_lock(&vb->lock);
1103 unsigned long va_start = vb->va->va_start;
1106 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1107 e = va_start + (vb->dirty_max << PAGE_SHIFT);
1109 start = min(s, start);
1114 spin_unlock(&vb->lock);
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);
1125 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
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)
1132 void vm_unmap_ram(const void *mem, unsigned int count)
1134 unsigned long size = (unsigned long)count << PAGE_SHIFT;
1135 unsigned long addr = (unsigned long)mem;
1136 struct vmap_area *va;
1140 BUG_ON(addr < VMALLOC_START);
1141 BUG_ON(addr > VMALLOC_END);
1142 BUG_ON(!PAGE_ALIGNED(addr));
1144 debug_check_no_locks_freed(mem, size);
1145 vmap_debug_free_range(addr, addr+size);
1147 if (likely(count <= VMAP_MAX_ALLOC)) {
1152 va = find_vmap_area(addr);
1154 free_unmap_vmap_area(va);
1156 EXPORT_SYMBOL(vm_unmap_ram);
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
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.
1171 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1173 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1175 unsigned long size = (unsigned long)count << PAGE_SHIFT;
1179 if (likely(count <= VMAP_MAX_ALLOC)) {
1180 mem = vb_alloc(size, GFP_KERNEL);
1183 addr = (unsigned long)mem;
1185 struct vmap_area *va;
1186 va = alloc_vmap_area(size, PAGE_SIZE,
1187 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1191 addr = va->va_start;
1194 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1195 vm_unmap_ram(mem, count);
1200 EXPORT_SYMBOL(vm_map_ram);
1202 static struct vm_struct *vmlist __initdata;
1204 * vm_area_add_early - add vmap area early during boot
1205 * @vm: vm_struct to add
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.
1211 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1213 void __init vm_area_add_early(struct vm_struct *vm)
1215 struct vm_struct *tmp, **p;
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);
1223 BUG_ON(tmp->addr + tmp->size > vm->addr);
1230 * vm_area_register_early - register vmap area early during boot
1231 * @vm: vm_struct to register
1232 * @align: requested alignment
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.
1239 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1241 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1243 static size_t vm_init_off __initdata;
1246 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1247 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1249 vm->addr = (void *)addr;
1251 vm_area_add_early(vm);
1254 void __init vmalloc_init(void)
1256 struct vmap_area *va;
1257 struct vm_struct *tmp;
1260 for_each_possible_cpu(i) {
1261 struct vmap_block_queue *vbq;
1262 struct vfree_deferred *p;
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);
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;
1279 __insert_vmap_area(va);
1282 vmap_area_pcpu_hole = VMALLOC_END;
1284 vmap_initialized = true;
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
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
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.
1304 * The number of pages mapped on success, -errno on failure.
1306 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1307 pgprot_t prot, struct page **pages)
1309 return vmap_page_range_noflush(addr, addr + size, prot, pages);
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
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
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.
1326 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1328 vunmap_page_range(addr, addr + size);
1330 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
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
1337 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1338 * the unmapping and tlb after.
1340 void unmap_kernel_range(unsigned long addr, unsigned long size)
1342 unsigned long end = addr + size;
1344 flush_cache_vunmap(addr, end);
1345 vunmap_page_range(addr, end);
1346 flush_tlb_kernel_range(addr, end);
1348 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1350 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1352 unsigned long addr = (unsigned long)area->addr;
1353 unsigned long end = addr + get_vm_area_size(area);
1356 err = vmap_page_range(addr, end, prot, pages);
1358 return err > 0 ? 0 : err;
1360 EXPORT_SYMBOL_GPL(map_vm_area);
1362 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1363 unsigned long flags, const void *caller)
1365 spin_lock(&vmap_area_lock);
1367 vm->addr = (void *)va->va_start;
1368 vm->size = va->va_end - va->va_start;
1369 vm->caller = caller;
1371 va->flags |= VM_VM_AREA;
1372 spin_unlock(&vmap_area_lock);
1375 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1378 * Before removing VM_UNINITIALIZED,
1379 * we should make sure that vm has proper values.
1380 * Pair with smp_rmb() in show_numa_info().
1383 vm->flags &= ~VM_UNINITIALIZED;
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)
1390 struct vmap_area *va;
1391 struct vm_struct *area;
1393 BUG_ON(in_interrupt());
1394 size = PAGE_ALIGN(size);
1395 if (unlikely(!size))
1398 if (flags & VM_IOREMAP)
1399 align = 1ul << clamp_t(int, get_count_order_long(size),
1400 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1402 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1403 if (unlikely(!area))
1406 if (!(flags & VM_NO_GUARD))
1409 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1415 setup_vmalloc_vm(area, va, flags, caller);
1420 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1421 unsigned long start, unsigned long end)
1423 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1424 GFP_KERNEL, __builtin_return_address(0));
1426 EXPORT_SYMBOL_GPL(__get_vm_area);
1428 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1429 unsigned long start, unsigned long end,
1432 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1433 GFP_KERNEL, caller);
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
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.
1445 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1447 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1448 NUMA_NO_NODE, GFP_KERNEL,
1449 __builtin_return_address(0));
1452 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
1455 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1456 NUMA_NO_NODE, GFP_KERNEL, caller);
1460 * find_vm_area - find a continuous kernel virtual area
1461 * @addr: base address
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
1467 struct vm_struct *find_vm_area(const void *addr)
1469 struct vmap_area *va;
1471 va = find_vmap_area((unsigned long)addr);
1472 if (va && va->flags & VM_VM_AREA)
1479 * remove_vm_area - find and remove a continuous kernel virtual area
1480 * @addr: base address
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.
1486 struct vm_struct *remove_vm_area(const void *addr)
1488 struct vmap_area *va;
1492 va = find_vmap_area((unsigned long)addr);
1493 if (va && va->flags & VM_VM_AREA) {
1494 struct vm_struct *vm = va->vm;
1496 spin_lock(&vmap_area_lock);
1498 va->flags &= ~VM_VM_AREA;
1499 va->flags |= VM_LAZY_FREE;
1500 spin_unlock(&vmap_area_lock);
1502 vmap_debug_free_range(va->va_start, va->va_end);
1503 kasan_free_shadow(vm);
1504 free_unmap_vmap_area(va);
1511 static void __vunmap(const void *addr, int deallocate_pages)
1513 struct vm_struct *area;
1518 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1522 area = remove_vm_area(addr);
1523 if (unlikely(!area)) {
1524 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1529 debug_check_no_locks_freed(addr, get_vm_area_size(area));
1530 debug_check_no_obj_freed(addr, get_vm_area_size(area));
1532 if (deallocate_pages) {
1535 for (i = 0; i < area->nr_pages; i++) {
1536 struct page *page = area->pages[i];
1539 __free_pages(page, 0);
1542 kvfree(area->pages);
1549 static inline void __vfree_deferred(const void *addr)
1552 * Use raw_cpu_ptr() because this can be called from preemptible
1553 * context. Preemption is absolutely fine here, because the llist_add()
1554 * implementation is lockless, so it works even if we are adding to
1555 * nother cpu's list. schedule_work() should be fine with this too.
1557 struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
1559 if (llist_add((struct llist_node *)addr, &p->list))
1560 schedule_work(&p->wq);
1564 * vfree_atomic - release memory allocated by vmalloc()
1565 * @addr: memory base address
1567 * This one is just like vfree() but can be called in any atomic context
1570 void vfree_atomic(const void *addr)
1574 kmemleak_free(addr);
1578 __vfree_deferred(addr);
1582 * vfree - release memory allocated by vmalloc()
1583 * @addr: memory base address
1585 * Free the virtually continuous memory area starting at @addr, as
1586 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1587 * NULL, no operation is performed.
1589 * Must not be called in NMI context (strictly speaking, only if we don't
1590 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1591 * conventions for vfree() arch-depenedent would be a really bad idea)
1593 * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
1595 void vfree(const void *addr)
1599 kmemleak_free(addr);
1603 if (unlikely(in_interrupt()))
1604 __vfree_deferred(addr);
1608 EXPORT_SYMBOL(vfree);
1611 * vunmap - release virtual mapping obtained by vmap()
1612 * @addr: memory base address
1614 * Free the virtually contiguous memory area starting at @addr,
1615 * which was created from the page array passed to vmap().
1617 * Must not be called in interrupt context.
1619 void vunmap(const void *addr)
1621 BUG_ON(in_interrupt());
1626 EXPORT_SYMBOL(vunmap);
1629 * vmap - map an array of pages into virtually contiguous space
1630 * @pages: array of page pointers
1631 * @count: number of pages to map
1632 * @flags: vm_area->flags
1633 * @prot: page protection for the mapping
1635 * Maps @count pages from @pages into contiguous kernel virtual
1638 void *vmap(struct page **pages, unsigned int count,
1639 unsigned long flags, pgprot_t prot)
1641 struct vm_struct *area;
1642 unsigned long size; /* In bytes */
1646 if (count > totalram_pages)
1649 size = (unsigned long)count << PAGE_SHIFT;
1650 area = get_vm_area_caller(size, flags, __builtin_return_address(0));
1654 if (map_vm_area(area, prot, pages)) {
1661 EXPORT_SYMBOL(vmap);
1663 static void *__vmalloc_node(unsigned long size, unsigned long align,
1664 gfp_t gfp_mask, pgprot_t prot,
1665 int node, const void *caller);
1666 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1667 pgprot_t prot, int node)
1669 struct page **pages;
1670 unsigned int nr_pages, array_size, i;
1671 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1672 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1673 const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ?
1677 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1678 array_size = (nr_pages * sizeof(struct page *));
1680 area->nr_pages = nr_pages;
1681 /* Please note that the recursion is strictly bounded. */
1682 if (array_size > PAGE_SIZE) {
1683 pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask,
1684 PAGE_KERNEL, node, area->caller);
1686 pages = kmalloc_node(array_size, nested_gfp, node);
1688 area->pages = pages;
1690 remove_vm_area(area->addr);
1695 for (i = 0; i < area->nr_pages; i++) {
1698 if (node == NUMA_NO_NODE)
1699 page = alloc_page(alloc_mask|highmem_mask);
1701 page = alloc_pages_node(node, alloc_mask|highmem_mask, 0);
1703 if (unlikely(!page)) {
1704 /* Successfully allocated i pages, free them in __vunmap() */
1708 area->pages[i] = page;
1709 if (gfpflags_allow_blocking(gfp_mask|highmem_mask))
1713 if (map_vm_area(area, prot, pages))
1718 warn_alloc(gfp_mask, NULL,
1719 "vmalloc: allocation failure, allocated %ld of %ld bytes",
1720 (area->nr_pages*PAGE_SIZE), area->size);
1726 * __vmalloc_node_range - allocate virtually contiguous memory
1727 * @size: allocation size
1728 * @align: desired alignment
1729 * @start: vm area range start
1730 * @end: vm area range end
1731 * @gfp_mask: flags for the page level allocator
1732 * @prot: protection mask for the allocated pages
1733 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1734 * @node: node to use for allocation or NUMA_NO_NODE
1735 * @caller: caller's return address
1737 * Allocate enough pages to cover @size from the page level
1738 * allocator with @gfp_mask flags. Map them into contiguous
1739 * kernel virtual space, using a pagetable protection of @prot.
1741 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1742 unsigned long start, unsigned long end, gfp_t gfp_mask,
1743 pgprot_t prot, unsigned long vm_flags, int node,
1746 struct vm_struct *area;
1748 unsigned long real_size = size;
1750 size = PAGE_ALIGN(size);
1751 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1754 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1755 vm_flags, start, end, node, gfp_mask, caller);
1759 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1764 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
1765 * flag. It means that vm_struct is not fully initialized.
1766 * Now, it is fully initialized, so remove this flag here.
1768 clear_vm_uninitialized_flag(area);
1770 kmemleak_vmalloc(area, size, gfp_mask);
1775 warn_alloc(gfp_mask, NULL,
1776 "vmalloc: allocation failure: %lu bytes", real_size);
1781 * __vmalloc_node - allocate virtually contiguous memory
1782 * @size: allocation size
1783 * @align: desired alignment
1784 * @gfp_mask: flags for the page level allocator
1785 * @prot: protection mask for the allocated pages
1786 * @node: node to use for allocation or NUMA_NO_NODE
1787 * @caller: caller's return address
1789 * Allocate enough pages to cover @size from the page level
1790 * allocator with @gfp_mask flags. Map them into contiguous
1791 * kernel virtual space, using a pagetable protection of @prot.
1793 * Reclaim modifiers in @gfp_mask - __GFP_NORETRY, __GFP_RETRY_MAYFAIL
1794 * and __GFP_NOFAIL are not supported
1796 * Any use of gfp flags outside of GFP_KERNEL should be consulted
1800 static void *__vmalloc_node(unsigned long size, unsigned long align,
1801 gfp_t gfp_mask, pgprot_t prot,
1802 int node, const void *caller)
1804 return __vmalloc_node_range(size, align, VMALLOC_START, VMALLOC_END,
1805 gfp_mask, prot, 0, node, caller);
1808 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
1810 return __vmalloc_node(size, 1, gfp_mask, prot, NUMA_NO_NODE,
1811 __builtin_return_address(0));
1813 EXPORT_SYMBOL(__vmalloc);
1815 static inline void *__vmalloc_node_flags(unsigned long size,
1816 int node, gfp_t flags)
1818 return __vmalloc_node(size, 1, flags, PAGE_KERNEL,
1819 node, __builtin_return_address(0));
1823 void *__vmalloc_node_flags_caller(unsigned long size, int node, gfp_t flags,
1826 return __vmalloc_node(size, 1, flags, PAGE_KERNEL, node, caller);
1830 * vmalloc - allocate virtually contiguous memory
1831 * @size: allocation size
1832 * Allocate enough pages to cover @size from the page level
1833 * allocator and map them into contiguous kernel virtual space.
1835 * For tight control over page level allocator and protection flags
1836 * use __vmalloc() instead.
1838 void *vmalloc(unsigned long size)
1840 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1843 EXPORT_SYMBOL(vmalloc);
1846 * vzalloc - allocate virtually contiguous memory with zero fill
1847 * @size: allocation size
1848 * Allocate enough pages to cover @size from the page level
1849 * allocator and map them into contiguous kernel virtual space.
1850 * The memory allocated is set to zero.
1852 * For tight control over page level allocator and protection flags
1853 * use __vmalloc() instead.
1855 void *vzalloc(unsigned long size)
1857 return __vmalloc_node_flags(size, NUMA_NO_NODE,
1858 GFP_KERNEL | __GFP_ZERO);
1860 EXPORT_SYMBOL(vzalloc);
1863 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
1864 * @size: allocation size
1866 * The resulting memory area is zeroed so it can be mapped to userspace
1867 * without leaking data.
1869 void *vmalloc_user(unsigned long size)
1871 struct vm_struct *area;
1874 ret = __vmalloc_node(size, SHMLBA,
1875 GFP_KERNEL | __GFP_ZERO,
1876 PAGE_KERNEL, NUMA_NO_NODE,
1877 __builtin_return_address(0));
1879 area = find_vm_area(ret);
1880 area->flags |= VM_USERMAP;
1884 EXPORT_SYMBOL(vmalloc_user);
1887 * vmalloc_node - allocate memory on a specific node
1888 * @size: allocation size
1891 * Allocate enough pages to cover @size from the page level
1892 * allocator and map them into contiguous kernel virtual space.
1894 * For tight control over page level allocator and protection flags
1895 * use __vmalloc() instead.
1897 void *vmalloc_node(unsigned long size, int node)
1899 return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL,
1900 node, __builtin_return_address(0));
1902 EXPORT_SYMBOL(vmalloc_node);
1905 * vzalloc_node - allocate memory on a specific node with zero fill
1906 * @size: allocation size
1909 * Allocate enough pages to cover @size from the page level
1910 * allocator and map them into contiguous kernel virtual space.
1911 * The memory allocated is set to zero.
1913 * For tight control over page level allocator and protection flags
1914 * use __vmalloc_node() instead.
1916 void *vzalloc_node(unsigned long size, int node)
1918 return __vmalloc_node_flags(size, node,
1919 GFP_KERNEL | __GFP_ZERO);
1921 EXPORT_SYMBOL(vzalloc_node);
1923 #ifndef PAGE_KERNEL_EXEC
1924 # define PAGE_KERNEL_EXEC PAGE_KERNEL
1928 * vmalloc_exec - allocate virtually contiguous, executable memory
1929 * @size: allocation size
1931 * Kernel-internal function to allocate enough pages to cover @size
1932 * the page level allocator and map them into contiguous and
1933 * executable kernel virtual space.
1935 * For tight control over page level allocator and protection flags
1936 * use __vmalloc() instead.
1939 void *vmalloc_exec(unsigned long size)
1941 return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL_EXEC,
1942 NUMA_NO_NODE, __builtin_return_address(0));
1945 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1946 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
1947 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1948 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
1950 #define GFP_VMALLOC32 GFP_KERNEL
1954 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
1955 * @size: allocation size
1957 * Allocate enough 32bit PA addressable pages to cover @size from the
1958 * page level allocator and map them into contiguous kernel virtual space.
1960 void *vmalloc_32(unsigned long size)
1962 return __vmalloc_node(size, 1, GFP_VMALLOC32, PAGE_KERNEL,
1963 NUMA_NO_NODE, __builtin_return_address(0));
1965 EXPORT_SYMBOL(vmalloc_32);
1968 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
1969 * @size: allocation size
1971 * The resulting memory area is 32bit addressable and zeroed so it can be
1972 * mapped to userspace without leaking data.
1974 void *vmalloc_32_user(unsigned long size)
1976 struct vm_struct *area;
1979 ret = __vmalloc_node(size, 1, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL,
1980 NUMA_NO_NODE, __builtin_return_address(0));
1982 area = find_vm_area(ret);
1983 area->flags |= VM_USERMAP;
1987 EXPORT_SYMBOL(vmalloc_32_user);
1990 * small helper routine , copy contents to buf from addr.
1991 * If the page is not present, fill zero.
1994 static int aligned_vread(char *buf, char *addr, unsigned long count)
2000 unsigned long offset, length;
2002 offset = offset_in_page(addr);
2003 length = PAGE_SIZE - offset;
2006 p = vmalloc_to_page(addr);
2008 * To do safe access to this _mapped_ area, we need
2009 * lock. But adding lock here means that we need to add
2010 * overhead of vmalloc()/vfree() calles for this _debug_
2011 * interface, rarely used. Instead of that, we'll use
2012 * kmap() and get small overhead in this access function.
2016 * we can expect USER0 is not used (see vread/vwrite's
2017 * function description)
2019 void *map = kmap_atomic(p);
2020 memcpy(buf, map + offset, length);
2023 memset(buf, 0, length);
2033 static int aligned_vwrite(char *buf, char *addr, unsigned long count)
2039 unsigned long offset, length;
2041 offset = offset_in_page(addr);
2042 length = PAGE_SIZE - offset;
2045 p = vmalloc_to_page(addr);
2047 * To do safe access to this _mapped_ area, we need
2048 * lock. But adding lock here means that we need to add
2049 * overhead of vmalloc()/vfree() calles for this _debug_
2050 * interface, rarely used. Instead of that, we'll use
2051 * kmap() and get small overhead in this access function.
2055 * we can expect USER0 is not used (see vread/vwrite's
2056 * function description)
2058 void *map = kmap_atomic(p);
2059 memcpy(map + offset, buf, length);
2071 * vread() - read vmalloc area in a safe way.
2072 * @buf: buffer for reading data
2073 * @addr: vm address.
2074 * @count: number of bytes to be read.
2076 * Returns # of bytes which addr and buf should be increased.
2077 * (same number to @count). Returns 0 if [addr...addr+count) doesn't
2078 * includes any intersect with alive vmalloc area.
2080 * This function checks that addr is a valid vmalloc'ed area, and
2081 * copy data from that area to a given buffer. If the given memory range
2082 * of [addr...addr+count) includes some valid address, data is copied to
2083 * proper area of @buf. If there are memory holes, they'll be zero-filled.
2084 * IOREMAP area is treated as memory hole and no copy is done.
2086 * If [addr...addr+count) doesn't includes any intersects with alive
2087 * vm_struct area, returns 0. @buf should be kernel's buffer.
2089 * Note: In usual ops, vread() is never necessary because the caller
2090 * should know vmalloc() area is valid and can use memcpy().
2091 * This is for routines which have to access vmalloc area without
2092 * any informaion, as /dev/kmem.
2096 long vread(char *buf, char *addr, unsigned long count)
2098 struct vmap_area *va;
2099 struct vm_struct *vm;
2100 char *vaddr, *buf_start = buf;
2101 unsigned long buflen = count;
2104 /* Don't allow overflow */
2105 if ((unsigned long) addr + count < count)
2106 count = -(unsigned long) addr;
2108 spin_lock(&vmap_area_lock);
2109 list_for_each_entry(va, &vmap_area_list, list) {
2113 if (!(va->flags & VM_VM_AREA))
2117 vaddr = (char *) vm->addr;
2118 if (addr >= vaddr + get_vm_area_size(vm))
2120 while (addr < vaddr) {
2128 n = vaddr + get_vm_area_size(vm) - addr;
2131 if (!(vm->flags & VM_IOREMAP))
2132 aligned_vread(buf, addr, n);
2133 else /* IOREMAP area is treated as memory hole */
2140 spin_unlock(&vmap_area_lock);
2142 if (buf == buf_start)
2144 /* zero-fill memory holes */
2145 if (buf != buf_start + buflen)
2146 memset(buf, 0, buflen - (buf - buf_start));
2152 * vwrite() - write vmalloc area in a safe way.
2153 * @buf: buffer for source data
2154 * @addr: vm address.
2155 * @count: number of bytes to be read.
2157 * Returns # of bytes which addr and buf should be incresed.
2158 * (same number to @count).
2159 * If [addr...addr+count) doesn't includes any intersect with valid
2160 * vmalloc area, returns 0.
2162 * This function checks that addr is a valid vmalloc'ed area, and
2163 * copy data from a buffer to the given addr. If specified range of
2164 * [addr...addr+count) includes some valid address, data is copied from
2165 * proper area of @buf. If there are memory holes, no copy to hole.
2166 * IOREMAP area is treated as memory hole and no copy is done.
2168 * If [addr...addr+count) doesn't includes any intersects with alive
2169 * vm_struct area, returns 0. @buf should be kernel's buffer.
2171 * Note: In usual ops, vwrite() is never necessary because the caller
2172 * should know vmalloc() area is valid and can use memcpy().
2173 * This is for routines which have to access vmalloc area without
2174 * any informaion, as /dev/kmem.
2177 long vwrite(char *buf, char *addr, unsigned long count)
2179 struct vmap_area *va;
2180 struct vm_struct *vm;
2182 unsigned long n, buflen;
2185 /* Don't allow overflow */
2186 if ((unsigned long) addr + count < count)
2187 count = -(unsigned long) addr;
2190 spin_lock(&vmap_area_lock);
2191 list_for_each_entry(va, &vmap_area_list, list) {
2195 if (!(va->flags & VM_VM_AREA))
2199 vaddr = (char *) vm->addr;
2200 if (addr >= vaddr + get_vm_area_size(vm))
2202 while (addr < vaddr) {
2209 n = vaddr + get_vm_area_size(vm) - addr;
2212 if (!(vm->flags & VM_IOREMAP)) {
2213 aligned_vwrite(buf, addr, n);
2221 spin_unlock(&vmap_area_lock);
2228 * remap_vmalloc_range_partial - map vmalloc pages to userspace
2229 * @vma: vma to cover
2230 * @uaddr: target user address to start at
2231 * @kaddr: virtual address of vmalloc kernel memory
2232 * @size: size of map area
2234 * Returns: 0 for success, -Exxx on failure
2236 * This function checks that @kaddr is a valid vmalloc'ed area,
2237 * and that it is big enough to cover the range starting at
2238 * @uaddr in @vma. Will return failure if that criteria isn't
2241 * Similar to remap_pfn_range() (see mm/memory.c)
2243 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2244 void *kaddr, unsigned long size)
2246 struct vm_struct *area;
2248 size = PAGE_ALIGN(size);
2250 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2253 area = find_vm_area(kaddr);
2257 if (!(area->flags & VM_USERMAP))
2260 if (kaddr + size > area->addr + area->size)
2264 struct page *page = vmalloc_to_page(kaddr);
2267 ret = vm_insert_page(vma, uaddr, page);
2276 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2280 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2283 * remap_vmalloc_range - map vmalloc pages to userspace
2284 * @vma: vma to cover (map full range of vma)
2285 * @addr: vmalloc memory
2286 * @pgoff: number of pages into addr before first page to map
2288 * Returns: 0 for success, -Exxx on failure
2290 * This function checks that addr is a valid vmalloc'ed area, and
2291 * that it is big enough to cover the vma. Will return failure if
2292 * that criteria isn't met.
2294 * Similar to remap_pfn_range() (see mm/memory.c)
2296 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2297 unsigned long pgoff)
2299 return remap_vmalloc_range_partial(vma, vma->vm_start,
2300 addr + (pgoff << PAGE_SHIFT),
2301 vma->vm_end - vma->vm_start);
2303 EXPORT_SYMBOL(remap_vmalloc_range);
2306 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
2309 void __weak vmalloc_sync_all(void)
2314 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2326 * alloc_vm_area - allocate a range of kernel address space
2327 * @size: size of the area
2328 * @ptes: returns the PTEs for the address space
2330 * Returns: NULL on failure, vm_struct on success
2332 * This function reserves a range of kernel address space, and
2333 * allocates pagetables to map that range. No actual mappings
2336 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2337 * allocated for the VM area are returned.
2339 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2341 struct vm_struct *area;
2343 area = get_vm_area_caller(size, VM_IOREMAP,
2344 __builtin_return_address(0));
2349 * This ensures that page tables are constructed for this region
2350 * of kernel virtual address space and mapped into init_mm.
2352 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2353 size, f, ptes ? &ptes : NULL)) {
2360 EXPORT_SYMBOL_GPL(alloc_vm_area);
2362 void free_vm_area(struct vm_struct *area)
2364 struct vm_struct *ret;
2365 ret = remove_vm_area(area->addr);
2366 BUG_ON(ret != area);
2369 EXPORT_SYMBOL_GPL(free_vm_area);
2372 static struct vmap_area *node_to_va(struct rb_node *n)
2374 return rb_entry_safe(n, struct vmap_area, rb_node);
2378 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2379 * @end: target address
2380 * @pnext: out arg for the next vmap_area
2381 * @pprev: out arg for the previous vmap_area
2383 * Returns: %true if either or both of next and prev are found,
2384 * %false if no vmap_area exists
2386 * Find vmap_areas end addresses of which enclose @end. ie. if not
2387 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2389 static bool pvm_find_next_prev(unsigned long end,
2390 struct vmap_area **pnext,
2391 struct vmap_area **pprev)
2393 struct rb_node *n = vmap_area_root.rb_node;
2394 struct vmap_area *va = NULL;
2397 va = rb_entry(n, struct vmap_area, rb_node);
2398 if (end < va->va_end)
2400 else if (end > va->va_end)
2409 if (va->va_end > end) {
2411 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2414 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2420 * pvm_determine_end - find the highest aligned address between two vmap_areas
2421 * @pnext: in/out arg for the next vmap_area
2422 * @pprev: in/out arg for the previous vmap_area
2425 * Returns: determined end address
2427 * Find the highest aligned address between *@pnext and *@pprev below
2428 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2429 * down address is between the end addresses of the two vmap_areas.
2431 * Please note that the address returned by this function may fall
2432 * inside *@pnext vmap_area. The caller is responsible for checking
2435 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2436 struct vmap_area **pprev,
2437 unsigned long align)
2439 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2443 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2447 while (*pprev && (*pprev)->va_end > addr) {
2449 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2456 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2457 * @offsets: array containing offset of each area
2458 * @sizes: array containing size of each area
2459 * @nr_vms: the number of areas to allocate
2460 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2462 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2463 * vm_structs on success, %NULL on failure
2465 * Percpu allocator wants to use congruent vm areas so that it can
2466 * maintain the offsets among percpu areas. This function allocates
2467 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2468 * be scattered pretty far, distance between two areas easily going up
2469 * to gigabytes. To avoid interacting with regular vmallocs, these
2470 * areas are allocated from top.
2472 * Despite its complicated look, this allocator is rather simple. It
2473 * does everything top-down and scans areas from the end looking for
2474 * matching slot. While scanning, if any of the areas overlaps with
2475 * existing vmap_area, the base address is pulled down to fit the
2476 * area. Scanning is repeated till all the areas fit and then all
2477 * necessary data structures are inserted and the result is returned.
2479 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2480 const size_t *sizes, int nr_vms,
2483 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2484 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2485 struct vmap_area **vas, *prev, *next;
2486 struct vm_struct **vms;
2487 int area, area2, last_area, term_area;
2488 unsigned long base, start, end, last_end;
2489 bool purged = false;
2491 /* verify parameters and allocate data structures */
2492 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
2493 for (last_area = 0, area = 0; area < nr_vms; area++) {
2494 start = offsets[area];
2495 end = start + sizes[area];
2497 /* is everything aligned properly? */
2498 BUG_ON(!IS_ALIGNED(offsets[area], align));
2499 BUG_ON(!IS_ALIGNED(sizes[area], align));
2501 /* detect the area with the highest address */
2502 if (start > offsets[last_area])
2505 for (area2 = area + 1; area2 < nr_vms; area2++) {
2506 unsigned long start2 = offsets[area2];
2507 unsigned long end2 = start2 + sizes[area2];
2509 BUG_ON(start2 < end && start < end2);
2512 last_end = offsets[last_area] + sizes[last_area];
2514 if (vmalloc_end - vmalloc_start < last_end) {
2519 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2520 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2524 for (area = 0; area < nr_vms; area++) {
2525 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2526 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2527 if (!vas[area] || !vms[area])
2531 spin_lock(&vmap_area_lock);
2533 /* start scanning - we scan from the top, begin with the last area */
2534 area = term_area = last_area;
2535 start = offsets[area];
2536 end = start + sizes[area];
2538 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2539 base = vmalloc_end - last_end;
2542 base = pvm_determine_end(&next, &prev, align) - end;
2545 BUG_ON(next && next->va_end <= base + end);
2546 BUG_ON(prev && prev->va_end > base + end);
2549 * base might have underflowed, add last_end before
2552 if (base + last_end < vmalloc_start + last_end) {
2553 spin_unlock(&vmap_area_lock);
2555 purge_vmap_area_lazy();
2563 * If next overlaps, move base downwards so that it's
2564 * right below next and then recheck.
2566 if (next && next->va_start < base + end) {
2567 base = pvm_determine_end(&next, &prev, align) - end;
2573 * If prev overlaps, shift down next and prev and move
2574 * base so that it's right below new next and then
2577 if (prev && prev->va_end > base + start) {
2579 prev = node_to_va(rb_prev(&next->rb_node));
2580 base = pvm_determine_end(&next, &prev, align) - end;
2586 * This area fits, move on to the previous one. If
2587 * the previous one is the terminal one, we're done.
2589 area = (area + nr_vms - 1) % nr_vms;
2590 if (area == term_area)
2592 start = offsets[area];
2593 end = start + sizes[area];
2594 pvm_find_next_prev(base + end, &next, &prev);
2597 /* we've found a fitting base, insert all va's */
2598 for (area = 0; area < nr_vms; area++) {
2599 struct vmap_area *va = vas[area];
2601 va->va_start = base + offsets[area];
2602 va->va_end = va->va_start + sizes[area];
2603 __insert_vmap_area(va);
2606 vmap_area_pcpu_hole = base + offsets[last_area];
2608 spin_unlock(&vmap_area_lock);
2610 /* insert all vm's */
2611 for (area = 0; area < nr_vms; area++)
2612 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2619 for (area = 0; area < nr_vms; area++) {
2630 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2631 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2632 * @nr_vms: the number of allocated areas
2634 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2636 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2640 for (i = 0; i < nr_vms; i++)
2641 free_vm_area(vms[i]);
2644 #endif /* CONFIG_SMP */
2646 #ifdef CONFIG_PROC_FS
2647 static void *s_start(struct seq_file *m, loff_t *pos)
2648 __acquires(&vmap_area_lock)
2650 spin_lock(&vmap_area_lock);
2651 return seq_list_start(&vmap_area_list, *pos);
2654 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2656 return seq_list_next(p, &vmap_area_list, pos);
2659 static void s_stop(struct seq_file *m, void *p)
2660 __releases(&vmap_area_lock)
2662 spin_unlock(&vmap_area_lock);
2665 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2667 if (IS_ENABLED(CONFIG_NUMA)) {
2668 unsigned int nr, *counters = m->private;
2673 if (v->flags & VM_UNINITIALIZED)
2675 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2678 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2680 for (nr = 0; nr < v->nr_pages; nr++)
2681 counters[page_to_nid(v->pages[nr])]++;
2683 for_each_node_state(nr, N_HIGH_MEMORY)
2685 seq_printf(m, " N%u=%u", nr, counters[nr]);
2689 static int s_show(struct seq_file *m, void *p)
2691 struct vmap_area *va;
2692 struct vm_struct *v;
2694 va = list_entry(p, struct vmap_area, list);
2697 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2698 * behalf of vmap area is being tear down or vm_map_ram allocation.
2700 if (!(va->flags & VM_VM_AREA)) {
2701 seq_printf(m, "0x%pK-0x%pK %7ld %s\n",
2702 (void *)va->va_start, (void *)va->va_end,
2703 va->va_end - va->va_start,
2704 va->flags & VM_LAZY_FREE ? "unpurged vm_area" : "vm_map_ram");
2711 seq_printf(m, "0x%pK-0x%pK %7ld",
2712 v->addr, v->addr + v->size, v->size);
2715 seq_printf(m, " %pS", v->caller);
2718 seq_printf(m, " pages=%d", v->nr_pages);
2721 seq_printf(m, " phys=%pa", &v->phys_addr);
2723 if (v->flags & VM_IOREMAP)
2724 seq_puts(m, " ioremap");
2726 if (v->flags & VM_ALLOC)
2727 seq_puts(m, " vmalloc");
2729 if (v->flags & VM_MAP)
2730 seq_puts(m, " vmap");
2732 if (v->flags & VM_USERMAP)
2733 seq_puts(m, " user");
2735 if (is_vmalloc_addr(v->pages))
2736 seq_puts(m, " vpages");
2738 show_numa_info(m, v);
2743 static const struct seq_operations vmalloc_op = {
2750 static int vmalloc_open(struct inode *inode, struct file *file)
2752 if (IS_ENABLED(CONFIG_NUMA))
2753 return seq_open_private(file, &vmalloc_op,
2754 nr_node_ids * sizeof(unsigned int));
2756 return seq_open(file, &vmalloc_op);
2759 static const struct file_operations proc_vmalloc_operations = {
2760 .open = vmalloc_open,
2762 .llseek = seq_lseek,
2763 .release = seq_release_private,
2766 static int __init proc_vmalloc_init(void)
2768 proc_create("vmallocinfo", S_IRUSR, NULL, &proc_vmalloc_operations);
2771 module_init(proc_vmalloc_init);