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>
34 #include <linux/overflow.h>
36 #include <linux/uaccess.h>
37 #include <asm/tlbflush.h>
38 #include <asm/shmparam.h>
42 struct vfree_deferred {
43 struct llist_head list;
44 struct work_struct wq;
46 static DEFINE_PER_CPU(struct vfree_deferred, vfree_deferred);
48 static void __vunmap(const void *, int);
50 static void free_work(struct work_struct *w)
52 struct vfree_deferred *p = container_of(w, struct vfree_deferred, wq);
53 struct llist_node *t, *llnode;
55 llist_for_each_safe(llnode, t, llist_del_all(&p->list))
56 __vunmap((void *)llnode, 1);
59 /*** Page table manipulation functions ***/
61 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
65 pte = pte_offset_kernel(pmd, addr);
67 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
68 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
69 } while (pte++, addr += PAGE_SIZE, addr != end);
72 static void vunmap_pmd_range(pud_t *pud, unsigned long addr, unsigned long end)
77 pmd = pmd_offset(pud, addr);
79 next = pmd_addr_end(addr, end);
80 if (pmd_clear_huge(pmd))
82 if (pmd_none_or_clear_bad(pmd))
84 vunmap_pte_range(pmd, addr, next);
85 } while (pmd++, addr = next, addr != end);
88 static void vunmap_pud_range(p4d_t *p4d, unsigned long addr, unsigned long end)
93 pud = pud_offset(p4d, addr);
95 next = pud_addr_end(addr, end);
96 if (pud_clear_huge(pud))
98 if (pud_none_or_clear_bad(pud))
100 vunmap_pmd_range(pud, addr, next);
101 } while (pud++, addr = next, addr != end);
104 static void vunmap_p4d_range(pgd_t *pgd, unsigned long addr, unsigned long end)
109 p4d = p4d_offset(pgd, addr);
111 next = p4d_addr_end(addr, end);
112 if (p4d_clear_huge(p4d))
114 if (p4d_none_or_clear_bad(p4d))
116 vunmap_pud_range(p4d, addr, next);
117 } while (p4d++, addr = next, addr != end);
120 static void vunmap_page_range(unsigned long addr, unsigned long end)
126 pgd = pgd_offset_k(addr);
128 next = pgd_addr_end(addr, end);
129 if (pgd_none_or_clear_bad(pgd))
131 vunmap_p4d_range(pgd, addr, next);
132 } while (pgd++, addr = next, addr != end);
135 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
136 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
141 * nr is a running index into the array which helps higher level
142 * callers keep track of where we're up to.
145 pte = pte_alloc_kernel(pmd, addr);
149 struct page *page = pages[*nr];
151 if (WARN_ON(!pte_none(*pte)))
155 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
157 } while (pte++, addr += PAGE_SIZE, addr != end);
161 static int vmap_pmd_range(pud_t *pud, unsigned long addr,
162 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
167 pmd = pmd_alloc(&init_mm, pud, addr);
171 next = pmd_addr_end(addr, end);
172 if (vmap_pte_range(pmd, addr, next, prot, pages, nr))
174 } while (pmd++, addr = next, addr != end);
178 static int vmap_pud_range(p4d_t *p4d, unsigned long addr,
179 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
184 pud = pud_alloc(&init_mm, p4d, addr);
188 next = pud_addr_end(addr, end);
189 if (vmap_pmd_range(pud, addr, next, prot, pages, nr))
191 } while (pud++, addr = next, addr != end);
195 static int vmap_p4d_range(pgd_t *pgd, unsigned long addr,
196 unsigned long end, pgprot_t prot, struct page **pages, int *nr)
201 p4d = p4d_alloc(&init_mm, pgd, addr);
205 next = p4d_addr_end(addr, end);
206 if (vmap_pud_range(p4d, addr, next, prot, pages, nr))
208 } while (p4d++, addr = next, addr != end);
213 * Set up page tables in kva (addr, end). The ptes shall have prot "prot", and
214 * will have pfns corresponding to the "pages" array.
216 * Ie. pte at addr+N*PAGE_SIZE shall point to pfn corresponding to pages[N]
218 static int vmap_page_range_noflush(unsigned long start, unsigned long end,
219 pgprot_t prot, struct page **pages)
223 unsigned long addr = start;
228 pgd = pgd_offset_k(addr);
230 next = pgd_addr_end(addr, end);
231 err = vmap_p4d_range(pgd, addr, next, prot, pages, &nr);
234 } while (pgd++, addr = next, addr != end);
239 static int vmap_page_range(unsigned long start, unsigned long end,
240 pgprot_t prot, struct page **pages)
244 ret = vmap_page_range_noflush(start, end, prot, pages);
245 flush_cache_vmap(start, end);
249 int is_vmalloc_or_module_addr(const void *x)
252 * ARM, x86-64 and sparc64 put modules in a special place,
253 * and fall back on vmalloc() if that fails. Others
254 * just put it in the vmalloc space.
256 #if defined(CONFIG_MODULES) && defined(MODULES_VADDR)
257 unsigned long addr = (unsigned long)x;
258 if (addr >= MODULES_VADDR && addr < MODULES_END)
261 return is_vmalloc_addr(x);
265 * Walk a vmap address to the struct page it maps.
267 struct page *vmalloc_to_page(const void *vmalloc_addr)
269 unsigned long addr = (unsigned long) vmalloc_addr;
270 struct page *page = NULL;
271 pgd_t *pgd = pgd_offset_k(addr);
278 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
279 * architectures that do not vmalloc module space
281 VIRTUAL_BUG_ON(!is_vmalloc_or_module_addr(vmalloc_addr));
285 p4d = p4d_offset(pgd, addr);
288 pud = pud_offset(p4d, addr);
291 * Don't dereference bad PUD or PMD (below) entries. This will also
292 * identify huge mappings, which we may encounter on architectures
293 * that define CONFIG_HAVE_ARCH_HUGE_VMAP=y. Such regions will be
294 * identified as vmalloc addresses by is_vmalloc_addr(), but are
295 * not [unambiguously] associated with a struct page, so there is
296 * no correct value to return for them.
298 WARN_ON_ONCE(pud_bad(*pud));
299 if (pud_none(*pud) || pud_bad(*pud))
301 pmd = pmd_offset(pud, addr);
302 WARN_ON_ONCE(pmd_bad(*pmd));
303 if (pmd_none(*pmd) || pmd_bad(*pmd))
306 ptep = pte_offset_map(pmd, addr);
308 if (pte_present(pte))
309 page = pte_page(pte);
313 EXPORT_SYMBOL(vmalloc_to_page);
316 * Map a vmalloc()-space virtual address to the physical page frame number.
318 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
320 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
322 EXPORT_SYMBOL(vmalloc_to_pfn);
325 /*** Global kva allocator ***/
327 #define VM_LAZY_FREE 0x02
328 #define VM_VM_AREA 0x04
330 static DEFINE_SPINLOCK(vmap_area_lock);
331 /* Export for kexec only */
332 LIST_HEAD(vmap_area_list);
333 static LLIST_HEAD(vmap_purge_list);
334 static struct rb_root vmap_area_root = RB_ROOT;
336 /* The vmap cache globals are protected by vmap_area_lock */
337 static struct rb_node *free_vmap_cache;
338 static unsigned long cached_hole_size;
339 static unsigned long cached_vstart;
340 static unsigned long cached_align;
342 static unsigned long vmap_area_pcpu_hole;
344 static struct vmap_area *__find_vmap_area(unsigned long addr)
346 struct rb_node *n = vmap_area_root.rb_node;
349 struct vmap_area *va;
351 va = rb_entry(n, struct vmap_area, rb_node);
352 if (addr < va->va_start)
354 else if (addr >= va->va_end)
363 static void __insert_vmap_area(struct vmap_area *va)
365 struct rb_node **p = &vmap_area_root.rb_node;
366 struct rb_node *parent = NULL;
370 struct vmap_area *tmp_va;
373 tmp_va = rb_entry(parent, struct vmap_area, rb_node);
374 if (va->va_start < tmp_va->va_end)
376 else if (va->va_end > tmp_va->va_start)
382 rb_link_node(&va->rb_node, parent, p);
383 rb_insert_color(&va->rb_node, &vmap_area_root);
385 /* address-sort this list */
386 tmp = rb_prev(&va->rb_node);
388 struct vmap_area *prev;
389 prev = rb_entry(tmp, struct vmap_area, rb_node);
390 list_add_rcu(&va->list, &prev->list);
392 list_add_rcu(&va->list, &vmap_area_list);
395 static void purge_vmap_area_lazy(void);
397 static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
400 * Allocate a region of KVA of the specified size and alignment, within the
403 static struct vmap_area *alloc_vmap_area(unsigned long size,
405 unsigned long vstart, unsigned long vend,
406 int node, gfp_t gfp_mask)
408 struct vmap_area *va;
412 struct vmap_area *first;
415 BUG_ON(offset_in_page(size));
416 BUG_ON(!is_power_of_2(align));
420 va = kmalloc_node(sizeof(struct vmap_area),
421 gfp_mask & GFP_RECLAIM_MASK, node);
423 return ERR_PTR(-ENOMEM);
426 * Only scan the relevant parts containing pointers to other objects
427 * to avoid false negatives.
429 kmemleak_scan_area(&va->rb_node, SIZE_MAX, gfp_mask & GFP_RECLAIM_MASK);
432 spin_lock(&vmap_area_lock);
434 * Invalidate cache if we have more permissive parameters.
435 * cached_hole_size notes the largest hole noticed _below_
436 * the vmap_area cached in free_vmap_cache: if size fits
437 * into that hole, we want to scan from vstart to reuse
438 * the hole instead of allocating above free_vmap_cache.
439 * Note that __free_vmap_area may update free_vmap_cache
440 * without updating cached_hole_size or cached_align.
442 if (!free_vmap_cache ||
443 size < cached_hole_size ||
444 vstart < cached_vstart ||
445 align < cached_align) {
447 cached_hole_size = 0;
448 free_vmap_cache = NULL;
450 /* record if we encounter less permissive parameters */
451 cached_vstart = vstart;
452 cached_align = align;
454 /* find starting point for our search */
455 if (free_vmap_cache) {
456 first = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
457 addr = ALIGN(first->va_end, align);
460 if (addr + size < addr)
464 addr = ALIGN(vstart, align);
465 if (addr + size < addr)
468 n = vmap_area_root.rb_node;
472 struct vmap_area *tmp;
473 tmp = rb_entry(n, struct vmap_area, rb_node);
474 if (tmp->va_end >= addr) {
476 if (tmp->va_start <= addr)
487 /* from the starting point, walk areas until a suitable hole is found */
488 while (addr + size > first->va_start && addr + size <= vend) {
489 if (addr + cached_hole_size < first->va_start)
490 cached_hole_size = first->va_start - addr;
491 addr = ALIGN(first->va_end, align);
492 if (addr + size < addr)
495 if (list_is_last(&first->list, &vmap_area_list))
498 first = list_next_entry(first, list);
503 * Check also calculated address against the vstart,
504 * because it can be 0 because of big align request.
506 if (addr + size > vend || addr < vstart)
510 va->va_end = addr + size;
512 __insert_vmap_area(va);
513 free_vmap_cache = &va->rb_node;
514 spin_unlock(&vmap_area_lock);
516 BUG_ON(!IS_ALIGNED(va->va_start, align));
517 BUG_ON(va->va_start < vstart);
518 BUG_ON(va->va_end > vend);
523 spin_unlock(&vmap_area_lock);
525 purge_vmap_area_lazy();
530 if (gfpflags_allow_blocking(gfp_mask)) {
531 unsigned long freed = 0;
532 blocking_notifier_call_chain(&vmap_notify_list, 0, &freed);
539 if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
540 pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
543 return ERR_PTR(-EBUSY);
546 int register_vmap_purge_notifier(struct notifier_block *nb)
548 return blocking_notifier_chain_register(&vmap_notify_list, nb);
550 EXPORT_SYMBOL_GPL(register_vmap_purge_notifier);
552 int unregister_vmap_purge_notifier(struct notifier_block *nb)
554 return blocking_notifier_chain_unregister(&vmap_notify_list, nb);
556 EXPORT_SYMBOL_GPL(unregister_vmap_purge_notifier);
558 static void __free_vmap_area(struct vmap_area *va)
560 BUG_ON(RB_EMPTY_NODE(&va->rb_node));
562 if (free_vmap_cache) {
563 if (va->va_end < cached_vstart) {
564 free_vmap_cache = NULL;
566 struct vmap_area *cache;
567 cache = rb_entry(free_vmap_cache, struct vmap_area, rb_node);
568 if (va->va_start <= cache->va_start) {
569 free_vmap_cache = rb_prev(&va->rb_node);
571 * We don't try to update cached_hole_size or
572 * cached_align, but it won't go very wrong.
577 rb_erase(&va->rb_node, &vmap_area_root);
578 RB_CLEAR_NODE(&va->rb_node);
579 list_del_rcu(&va->list);
582 * Track the highest possible candidate for pcpu area
583 * allocation. Areas outside of vmalloc area can be returned
584 * here too, consider only end addresses which fall inside
585 * vmalloc area proper.
587 if (va->va_end > VMALLOC_START && va->va_end <= VMALLOC_END)
588 vmap_area_pcpu_hole = max(vmap_area_pcpu_hole, va->va_end);
590 kfree_rcu(va, rcu_head);
594 * Free a region of KVA allocated by alloc_vmap_area
596 static void free_vmap_area(struct vmap_area *va)
598 spin_lock(&vmap_area_lock);
599 __free_vmap_area(va);
600 spin_unlock(&vmap_area_lock);
604 * Clear the pagetable entries of a given vmap_area
606 static void unmap_vmap_area(struct vmap_area *va)
608 vunmap_page_range(va->va_start, va->va_end);
612 * lazy_max_pages is the maximum amount of virtual address space we gather up
613 * before attempting to purge with a TLB flush.
615 * There is a tradeoff here: a larger number will cover more kernel page tables
616 * and take slightly longer to purge, but it will linearly reduce the number of
617 * global TLB flushes that must be performed. It would seem natural to scale
618 * this number up linearly with the number of CPUs (because vmapping activity
619 * could also scale linearly with the number of CPUs), however it is likely
620 * that in practice, workloads might be constrained in other ways that mean
621 * vmap activity will not scale linearly with CPUs. Also, I want to be
622 * conservative and not introduce a big latency on huge systems, so go with
623 * a less aggressive log scale. It will still be an improvement over the old
624 * code, and it will be simple to change the scale factor if we find that it
625 * becomes a problem on bigger systems.
627 static unsigned long lazy_max_pages(void)
631 log = fls(num_online_cpus());
633 return log * (32UL * 1024 * 1024 / PAGE_SIZE);
636 static atomic_t vmap_lazy_nr = ATOMIC_INIT(0);
639 * Serialize vmap purging. There is no actual criticial section protected
640 * by this look, but we want to avoid concurrent calls for performance
641 * reasons and to make the pcpu_get_vm_areas more deterministic.
643 static DEFINE_MUTEX(vmap_purge_lock);
645 /* for per-CPU blocks */
646 static void purge_fragmented_blocks_allcpus(void);
649 * called before a call to iounmap() if the caller wants vm_area_struct's
652 void set_iounmap_nonlazy(void)
654 atomic_set(&vmap_lazy_nr, lazy_max_pages()+1);
658 * Purges all lazily-freed vmap areas.
660 static bool __purge_vmap_area_lazy(unsigned long start, unsigned long end)
662 struct llist_node *valist;
663 struct vmap_area *va;
664 struct vmap_area *n_va;
665 bool do_free = false;
667 lockdep_assert_held(&vmap_purge_lock);
669 valist = llist_del_all(&vmap_purge_list);
670 llist_for_each_entry(va, valist, purge_list) {
671 if (va->va_start < start)
672 start = va->va_start;
673 if (va->va_end > end)
681 flush_tlb_kernel_range(start, end);
683 spin_lock(&vmap_area_lock);
684 llist_for_each_entry_safe(va, n_va, valist, purge_list) {
685 int nr = (va->va_end - va->va_start) >> PAGE_SHIFT;
687 __free_vmap_area(va);
688 atomic_sub(nr, &vmap_lazy_nr);
689 cond_resched_lock(&vmap_area_lock);
691 spin_unlock(&vmap_area_lock);
696 * Kick off a purge of the outstanding lazy areas. Don't bother if somebody
697 * is already purging.
699 static void try_purge_vmap_area_lazy(void)
701 if (mutex_trylock(&vmap_purge_lock)) {
702 __purge_vmap_area_lazy(ULONG_MAX, 0);
703 mutex_unlock(&vmap_purge_lock);
708 * Kick off a purge of the outstanding lazy areas.
710 static void purge_vmap_area_lazy(void)
712 mutex_lock(&vmap_purge_lock);
713 purge_fragmented_blocks_allcpus();
714 __purge_vmap_area_lazy(ULONG_MAX, 0);
715 mutex_unlock(&vmap_purge_lock);
719 * Free a vmap area, caller ensuring that the area has been unmapped
720 * and flush_cache_vunmap had been called for the correct range
723 static void free_vmap_area_noflush(struct vmap_area *va)
727 nr_lazy = atomic_add_return((va->va_end - va->va_start) >> PAGE_SHIFT,
730 /* After this point, we may free va at any time */
731 llist_add(&va->purge_list, &vmap_purge_list);
733 if (unlikely(nr_lazy > lazy_max_pages()))
734 try_purge_vmap_area_lazy();
738 * Free and unmap a vmap area
740 static void free_unmap_vmap_area(struct vmap_area *va)
742 flush_cache_vunmap(va->va_start, va->va_end);
744 if (debug_pagealloc_enabled())
745 flush_tlb_kernel_range(va->va_start, va->va_end);
747 free_vmap_area_noflush(va);
750 static struct vmap_area *find_vmap_area(unsigned long addr)
752 struct vmap_area *va;
754 spin_lock(&vmap_area_lock);
755 va = __find_vmap_area(addr);
756 spin_unlock(&vmap_area_lock);
761 /*** Per cpu kva allocator ***/
764 * vmap space is limited especially on 32 bit architectures. Ensure there is
765 * room for at least 16 percpu vmap blocks per CPU.
768 * If we had a constant VMALLOC_START and VMALLOC_END, we'd like to be able
769 * to #define VMALLOC_SPACE (VMALLOC_END-VMALLOC_START). Guess
770 * instead (we just need a rough idea)
772 #if BITS_PER_LONG == 32
773 #define VMALLOC_SPACE (128UL*1024*1024)
775 #define VMALLOC_SPACE (128UL*1024*1024*1024)
778 #define VMALLOC_PAGES (VMALLOC_SPACE / PAGE_SIZE)
779 #define VMAP_MAX_ALLOC BITS_PER_LONG /* 256K with 4K pages */
780 #define VMAP_BBMAP_BITS_MAX 1024 /* 4MB with 4K pages */
781 #define VMAP_BBMAP_BITS_MIN (VMAP_MAX_ALLOC*2)
782 #define VMAP_MIN(x, y) ((x) < (y) ? (x) : (y)) /* can't use min() */
783 #define VMAP_MAX(x, y) ((x) > (y) ? (x) : (y)) /* can't use max() */
784 #define VMAP_BBMAP_BITS \
785 VMAP_MIN(VMAP_BBMAP_BITS_MAX, \
786 VMAP_MAX(VMAP_BBMAP_BITS_MIN, \
787 VMALLOC_PAGES / roundup_pow_of_two(NR_CPUS) / 16))
789 #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
791 static bool vmap_initialized __read_mostly = false;
793 struct vmap_block_queue {
795 struct list_head free;
800 struct vmap_area *va;
801 unsigned long free, dirty;
802 unsigned long dirty_min, dirty_max; /*< dirty range */
803 struct list_head free_list;
804 struct rcu_head rcu_head;
805 struct list_head purge;
808 /* Queue of free and dirty vmap blocks, for allocation and flushing purposes */
809 static DEFINE_PER_CPU(struct vmap_block_queue, vmap_block_queue);
812 * Radix tree of vmap blocks, indexed by address, to quickly find a vmap block
813 * in the free path. Could get rid of this if we change the API to return a
814 * "cookie" from alloc, to be passed to free. But no big deal yet.
816 static DEFINE_SPINLOCK(vmap_block_tree_lock);
817 static RADIX_TREE(vmap_block_tree, GFP_ATOMIC);
820 * We should probably have a fallback mechanism to allocate virtual memory
821 * out of partially filled vmap blocks. However vmap block sizing should be
822 * fairly reasonable according to the vmalloc size, so it shouldn't be a
826 static unsigned long addr_to_vb_idx(unsigned long addr)
828 addr -= VMALLOC_START & ~(VMAP_BLOCK_SIZE-1);
829 addr /= VMAP_BLOCK_SIZE;
833 static void *vmap_block_vaddr(unsigned long va_start, unsigned long pages_off)
837 addr = va_start + (pages_off << PAGE_SHIFT);
838 BUG_ON(addr_to_vb_idx(addr) != addr_to_vb_idx(va_start));
843 * new_vmap_block - allocates new vmap_block and occupies 2^order pages in this
844 * block. Of course pages number can't exceed VMAP_BBMAP_BITS
845 * @order: how many 2^order pages should be occupied in newly allocated block
846 * @gfp_mask: flags for the page level allocator
848 * Returns: virtual address in a newly allocated block or ERR_PTR(-errno)
850 static void *new_vmap_block(unsigned int order, gfp_t gfp_mask)
852 struct vmap_block_queue *vbq;
853 struct vmap_block *vb;
854 struct vmap_area *va;
855 unsigned long vb_idx;
859 node = numa_node_id();
861 vb = kmalloc_node(sizeof(struct vmap_block),
862 gfp_mask & GFP_RECLAIM_MASK, node);
864 return ERR_PTR(-ENOMEM);
866 va = alloc_vmap_area(VMAP_BLOCK_SIZE, VMAP_BLOCK_SIZE,
867 VMALLOC_START, VMALLOC_END,
874 err = radix_tree_preload(gfp_mask);
881 vaddr = vmap_block_vaddr(va->va_start, 0);
882 spin_lock_init(&vb->lock);
884 /* At least something should be left free */
885 BUG_ON(VMAP_BBMAP_BITS <= (1UL << order));
886 vb->free = VMAP_BBMAP_BITS - (1UL << order);
888 vb->dirty_min = VMAP_BBMAP_BITS;
890 INIT_LIST_HEAD(&vb->free_list);
892 vb_idx = addr_to_vb_idx(va->va_start);
893 spin_lock(&vmap_block_tree_lock);
894 err = radix_tree_insert(&vmap_block_tree, vb_idx, vb);
895 spin_unlock(&vmap_block_tree_lock);
897 radix_tree_preload_end();
899 vbq = &get_cpu_var(vmap_block_queue);
900 spin_lock(&vbq->lock);
901 list_add_tail_rcu(&vb->free_list, &vbq->free);
902 spin_unlock(&vbq->lock);
903 put_cpu_var(vmap_block_queue);
908 static void free_vmap_block(struct vmap_block *vb)
910 struct vmap_block *tmp;
911 unsigned long vb_idx;
913 vb_idx = addr_to_vb_idx(vb->va->va_start);
914 spin_lock(&vmap_block_tree_lock);
915 tmp = radix_tree_delete(&vmap_block_tree, vb_idx);
916 spin_unlock(&vmap_block_tree_lock);
919 free_vmap_area_noflush(vb->va);
920 kfree_rcu(vb, rcu_head);
923 static void purge_fragmented_blocks(int cpu)
926 struct vmap_block *vb;
927 struct vmap_block *n_vb;
928 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
931 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
933 if (!(vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS))
936 spin_lock(&vb->lock);
937 if (vb->free + vb->dirty == VMAP_BBMAP_BITS && vb->dirty != VMAP_BBMAP_BITS) {
938 vb->free = 0; /* prevent further allocs after releasing lock */
939 vb->dirty = VMAP_BBMAP_BITS; /* prevent purging it again */
941 vb->dirty_max = VMAP_BBMAP_BITS;
942 spin_lock(&vbq->lock);
943 list_del_rcu(&vb->free_list);
944 spin_unlock(&vbq->lock);
945 spin_unlock(&vb->lock);
946 list_add_tail(&vb->purge, &purge);
948 spin_unlock(&vb->lock);
952 list_for_each_entry_safe(vb, n_vb, &purge, purge) {
953 list_del(&vb->purge);
958 static void purge_fragmented_blocks_allcpus(void)
962 for_each_possible_cpu(cpu)
963 purge_fragmented_blocks(cpu);
966 static void *vb_alloc(unsigned long size, gfp_t gfp_mask)
968 struct vmap_block_queue *vbq;
969 struct vmap_block *vb;
973 BUG_ON(offset_in_page(size));
974 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
975 if (WARN_ON(size == 0)) {
977 * Allocating 0 bytes isn't what caller wants since
978 * get_order(0) returns funny result. Just warn and terminate
983 order = get_order(size);
986 vbq = &get_cpu_var(vmap_block_queue);
987 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
988 unsigned long pages_off;
990 spin_lock(&vb->lock);
991 if (vb->free < (1UL << order)) {
992 spin_unlock(&vb->lock);
996 pages_off = VMAP_BBMAP_BITS - vb->free;
997 vaddr = vmap_block_vaddr(vb->va->va_start, pages_off);
998 vb->free -= 1UL << order;
1000 spin_lock(&vbq->lock);
1001 list_del_rcu(&vb->free_list);
1002 spin_unlock(&vbq->lock);
1005 spin_unlock(&vb->lock);
1009 put_cpu_var(vmap_block_queue);
1012 /* Allocate new block if nothing was found */
1014 vaddr = new_vmap_block(order, gfp_mask);
1019 static void vb_free(const void *addr, unsigned long size)
1021 unsigned long offset;
1022 unsigned long vb_idx;
1024 struct vmap_block *vb;
1026 BUG_ON(offset_in_page(size));
1027 BUG_ON(size > PAGE_SIZE*VMAP_MAX_ALLOC);
1029 flush_cache_vunmap((unsigned long)addr, (unsigned long)addr + size);
1031 order = get_order(size);
1033 offset = (unsigned long)addr & (VMAP_BLOCK_SIZE - 1);
1034 offset >>= PAGE_SHIFT;
1036 vb_idx = addr_to_vb_idx((unsigned long)addr);
1038 vb = radix_tree_lookup(&vmap_block_tree, vb_idx);
1042 vunmap_page_range((unsigned long)addr, (unsigned long)addr + size);
1044 if (debug_pagealloc_enabled())
1045 flush_tlb_kernel_range((unsigned long)addr,
1046 (unsigned long)addr + size);
1048 spin_lock(&vb->lock);
1050 /* Expand dirty range */
1051 vb->dirty_min = min(vb->dirty_min, offset);
1052 vb->dirty_max = max(vb->dirty_max, offset + (1UL << order));
1054 vb->dirty += 1UL << order;
1055 if (vb->dirty == VMAP_BBMAP_BITS) {
1057 spin_unlock(&vb->lock);
1058 free_vmap_block(vb);
1060 spin_unlock(&vb->lock);
1064 * vm_unmap_aliases - unmap outstanding lazy aliases in the vmap layer
1066 * The vmap/vmalloc layer lazily flushes kernel virtual mappings primarily
1067 * to amortize TLB flushing overheads. What this means is that any page you
1068 * have now, may, in a former life, have been mapped into kernel virtual
1069 * address by the vmap layer and so there might be some CPUs with TLB entries
1070 * still referencing that page (additional to the regular 1:1 kernel mapping).
1072 * vm_unmap_aliases flushes all such lazy mappings. After it returns, we can
1073 * be sure that none of the pages we have control over will have any aliases
1074 * from the vmap layer.
1076 void vm_unmap_aliases(void)
1078 unsigned long start = ULONG_MAX, end = 0;
1082 if (unlikely(!vmap_initialized))
1087 for_each_possible_cpu(cpu) {
1088 struct vmap_block_queue *vbq = &per_cpu(vmap_block_queue, cpu);
1089 struct vmap_block *vb;
1092 list_for_each_entry_rcu(vb, &vbq->free, free_list) {
1093 spin_lock(&vb->lock);
1095 unsigned long va_start = vb->va->va_start;
1098 s = va_start + (vb->dirty_min << PAGE_SHIFT);
1099 e = va_start + (vb->dirty_max << PAGE_SHIFT);
1101 start = min(s, start);
1106 spin_unlock(&vb->lock);
1111 mutex_lock(&vmap_purge_lock);
1112 purge_fragmented_blocks_allcpus();
1113 if (!__purge_vmap_area_lazy(start, end) && flush)
1114 flush_tlb_kernel_range(start, end);
1115 mutex_unlock(&vmap_purge_lock);
1117 EXPORT_SYMBOL_GPL(vm_unmap_aliases);
1120 * vm_unmap_ram - unmap linear kernel address space set up by vm_map_ram
1121 * @mem: the pointer returned by vm_map_ram
1122 * @count: the count passed to that vm_map_ram call (cannot unmap partial)
1124 void vm_unmap_ram(const void *mem, unsigned int count)
1126 unsigned long size = (unsigned long)count << PAGE_SHIFT;
1127 unsigned long addr = (unsigned long)mem;
1128 struct vmap_area *va;
1132 BUG_ON(addr < VMALLOC_START);
1133 BUG_ON(addr > VMALLOC_END);
1134 BUG_ON(!PAGE_ALIGNED(addr));
1136 if (likely(count <= VMAP_MAX_ALLOC)) {
1137 debug_check_no_locks_freed(mem, size);
1142 va = find_vmap_area(addr);
1144 debug_check_no_locks_freed((void *)va->va_start,
1145 (va->va_end - va->va_start));
1146 free_unmap_vmap_area(va);
1148 EXPORT_SYMBOL(vm_unmap_ram);
1151 * vm_map_ram - map pages linearly into kernel virtual address (vmalloc space)
1152 * @pages: an array of pointers to the pages to be mapped
1153 * @count: number of pages
1154 * @node: prefer to allocate data structures on this node
1155 * @prot: memory protection to use. PAGE_KERNEL for regular RAM
1157 * If you use this function for less than VMAP_MAX_ALLOC pages, it could be
1158 * faster than vmap so it's good. But if you mix long-life and short-life
1159 * objects with vm_map_ram(), it could consume lots of address space through
1160 * fragmentation (especially on a 32bit machine). You could see failures in
1161 * the end. Please use this function for short-lived objects.
1163 * Returns: a pointer to the address that has been mapped, or %NULL on failure
1165 void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot)
1167 unsigned long size = (unsigned long)count << PAGE_SHIFT;
1171 if (likely(count <= VMAP_MAX_ALLOC)) {
1172 mem = vb_alloc(size, GFP_KERNEL);
1175 addr = (unsigned long)mem;
1177 struct vmap_area *va;
1178 va = alloc_vmap_area(size, PAGE_SIZE,
1179 VMALLOC_START, VMALLOC_END, node, GFP_KERNEL);
1183 addr = va->va_start;
1186 if (vmap_page_range(addr, addr + size, prot, pages) < 0) {
1187 vm_unmap_ram(mem, count);
1192 EXPORT_SYMBOL(vm_map_ram);
1194 static struct vm_struct *vmlist __initdata;
1196 * vm_area_add_early - add vmap area early during boot
1197 * @vm: vm_struct to add
1199 * This function is used to add fixed kernel vm area to vmlist before
1200 * vmalloc_init() is called. @vm->addr, @vm->size, and @vm->flags
1201 * should contain proper values and the other fields should be zero.
1203 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1205 void __init vm_area_add_early(struct vm_struct *vm)
1207 struct vm_struct *tmp, **p;
1209 BUG_ON(vmap_initialized);
1210 for (p = &vmlist; (tmp = *p) != NULL; p = &tmp->next) {
1211 if (tmp->addr >= vm->addr) {
1212 BUG_ON(tmp->addr < vm->addr + vm->size);
1215 BUG_ON(tmp->addr + tmp->size > vm->addr);
1222 * vm_area_register_early - register vmap area early during boot
1223 * @vm: vm_struct to register
1224 * @align: requested alignment
1226 * This function is used to register kernel vm area before
1227 * vmalloc_init() is called. @vm->size and @vm->flags should contain
1228 * proper values on entry and other fields should be zero. On return,
1229 * vm->addr contains the allocated address.
1231 * DO NOT USE THIS FUNCTION UNLESS YOU KNOW WHAT YOU'RE DOING.
1233 void __init vm_area_register_early(struct vm_struct *vm, size_t align)
1235 static size_t vm_init_off __initdata;
1238 addr = ALIGN(VMALLOC_START + vm_init_off, align);
1239 vm_init_off = PFN_ALIGN(addr + vm->size) - VMALLOC_START;
1241 vm->addr = (void *)addr;
1243 vm_area_add_early(vm);
1246 void __init vmalloc_init(void)
1248 struct vmap_area *va;
1249 struct vm_struct *tmp;
1252 for_each_possible_cpu(i) {
1253 struct vmap_block_queue *vbq;
1254 struct vfree_deferred *p;
1256 vbq = &per_cpu(vmap_block_queue, i);
1257 spin_lock_init(&vbq->lock);
1258 INIT_LIST_HEAD(&vbq->free);
1259 p = &per_cpu(vfree_deferred, i);
1260 init_llist_head(&p->list);
1261 INIT_WORK(&p->wq, free_work);
1264 /* Import existing vmlist entries. */
1265 for (tmp = vmlist; tmp; tmp = tmp->next) {
1266 va = kzalloc(sizeof(struct vmap_area), GFP_NOWAIT);
1267 va->flags = VM_VM_AREA;
1268 va->va_start = (unsigned long)tmp->addr;
1269 va->va_end = va->va_start + tmp->size;
1271 __insert_vmap_area(va);
1274 vmap_area_pcpu_hole = VMALLOC_END;
1276 vmap_initialized = true;
1280 * map_kernel_range_noflush - map kernel VM area with the specified pages
1281 * @addr: start of the VM area to map
1282 * @size: size of the VM area to map
1283 * @prot: page protection flags to use
1284 * @pages: pages to map
1286 * Map PFN_UP(@size) pages at @addr. The VM area @addr and @size
1287 * specify should have been allocated using get_vm_area() and its
1291 * This function does NOT do any cache flushing. The caller is
1292 * responsible for calling flush_cache_vmap() on to-be-mapped areas
1293 * before calling this function.
1296 * The number of pages mapped on success, -errno on failure.
1298 int map_kernel_range_noflush(unsigned long addr, unsigned long size,
1299 pgprot_t prot, struct page **pages)
1301 return vmap_page_range_noflush(addr, addr + size, prot, pages);
1305 * unmap_kernel_range_noflush - unmap kernel VM area
1306 * @addr: start of the VM area to unmap
1307 * @size: size of the VM area to unmap
1309 * Unmap PFN_UP(@size) pages at @addr. The VM area @addr and @size
1310 * specify should have been allocated using get_vm_area() and its
1314 * This function does NOT do any cache flushing. The caller is
1315 * responsible for calling flush_cache_vunmap() on to-be-mapped areas
1316 * before calling this function and flush_tlb_kernel_range() after.
1318 void unmap_kernel_range_noflush(unsigned long addr, unsigned long size)
1320 vunmap_page_range(addr, addr + size);
1322 EXPORT_SYMBOL_GPL(unmap_kernel_range_noflush);
1325 * unmap_kernel_range - unmap kernel VM area and flush cache and TLB
1326 * @addr: start of the VM area to unmap
1327 * @size: size of the VM area to unmap
1329 * Similar to unmap_kernel_range_noflush() but flushes vcache before
1330 * the unmapping and tlb after.
1332 void unmap_kernel_range(unsigned long addr, unsigned long size)
1334 unsigned long end = addr + size;
1336 flush_cache_vunmap(addr, end);
1337 vunmap_page_range(addr, end);
1338 flush_tlb_kernel_range(addr, end);
1340 EXPORT_SYMBOL_GPL(unmap_kernel_range);
1342 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page **pages)
1344 unsigned long addr = (unsigned long)area->addr;
1345 unsigned long end = addr + get_vm_area_size(area);
1348 err = vmap_page_range(addr, end, prot, pages);
1350 return err > 0 ? 0 : err;
1352 EXPORT_SYMBOL_GPL(map_vm_area);
1354 static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
1355 unsigned long flags, const void *caller)
1357 spin_lock(&vmap_area_lock);
1359 vm->addr = (void *)va->va_start;
1360 vm->size = va->va_end - va->va_start;
1361 vm->caller = caller;
1363 va->flags |= VM_VM_AREA;
1364 spin_unlock(&vmap_area_lock);
1367 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
1370 * Before removing VM_UNINITIALIZED,
1371 * we should make sure that vm has proper values.
1372 * Pair with smp_rmb() in show_numa_info().
1375 vm->flags &= ~VM_UNINITIALIZED;
1378 static struct vm_struct *__get_vm_area_node(unsigned long size,
1379 unsigned long align, unsigned long flags, unsigned long start,
1380 unsigned long end, int node, gfp_t gfp_mask, const void *caller)
1382 struct vmap_area *va;
1383 struct vm_struct *area;
1385 BUG_ON(in_interrupt());
1386 size = PAGE_ALIGN(size);
1387 if (unlikely(!size))
1390 if (flags & VM_IOREMAP)
1391 align = 1ul << clamp_t(int, get_count_order_long(size),
1392 PAGE_SHIFT, IOREMAP_MAX_ORDER);
1394 area = kzalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
1395 if (unlikely(!area))
1398 if (!(flags & VM_NO_GUARD))
1401 va = alloc_vmap_area(size, align, start, end, node, gfp_mask);
1407 setup_vmalloc_vm(area, va, flags, caller);
1412 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
1413 unsigned long start, unsigned long end)
1415 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1416 GFP_KERNEL, __builtin_return_address(0));
1418 EXPORT_SYMBOL_GPL(__get_vm_area);
1420 struct vm_struct *__get_vm_area_caller(unsigned long size, unsigned long flags,
1421 unsigned long start, unsigned long end,
1424 return __get_vm_area_node(size, 1, flags, start, end, NUMA_NO_NODE,
1425 GFP_KERNEL, caller);
1429 * get_vm_area - reserve a contiguous kernel virtual area
1430 * @size: size of the area
1431 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
1433 * Search an area of @size in the kernel virtual mapping area,
1434 * and reserved it for out purposes. Returns the area descriptor
1435 * on success or %NULL on failure.
1437 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
1439 return __get_vm_area_node(size, 1, flags, VMALLOC_START, VMALLOC_END,
1440 NUMA_NO_NODE, GFP_KERNEL,
1441 __builtin_return_address(0));
1444 struct vm_struct *get_vm_area_caller(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, caller);
1452 * find_vm_area - find a continuous kernel virtual area
1453 * @addr: base address
1455 * Search for the kernel VM area starting at @addr, and return it.
1456 * It is up to the caller to do all required locking to keep the returned
1459 struct vm_struct *find_vm_area(const void *addr)
1461 struct vmap_area *va;
1463 va = find_vmap_area((unsigned long)addr);
1464 if (va && va->flags & VM_VM_AREA)
1471 * remove_vm_area - find and remove a continuous kernel virtual area
1472 * @addr: base address
1474 * Search for the kernel VM area starting at @addr, and remove it.
1475 * This function returns the found VM area, but using it is NOT safe
1476 * on SMP machines, except for its size or flags.
1478 struct vm_struct *remove_vm_area(const void *addr)
1480 struct vmap_area *va;
1484 va = find_vmap_area((unsigned long)addr);
1485 if (va && va->flags & VM_VM_AREA) {
1486 struct vm_struct *vm = va->vm;
1488 spin_lock(&vmap_area_lock);
1490 va->flags &= ~VM_VM_AREA;
1491 va->flags |= VM_LAZY_FREE;
1492 spin_unlock(&vmap_area_lock);
1494 kasan_free_shadow(vm);
1495 free_unmap_vmap_area(va);
1502 static void __vunmap(const void *addr, int deallocate_pages)
1504 struct vm_struct *area;
1509 if (WARN(!PAGE_ALIGNED(addr), "Trying to vfree() bad address (%p)\n",
1513 area = find_vm_area(addr);
1514 if (unlikely(!area)) {
1515 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
1520 debug_check_no_locks_freed(area->addr, get_vm_area_size(area));
1521 debug_check_no_obj_freed(area->addr, get_vm_area_size(area));
1523 remove_vm_area(addr);
1524 if (deallocate_pages) {
1527 for (i = 0; i < area->nr_pages; i++) {
1528 struct page *page = area->pages[i];
1531 __free_pages(page, 0);
1534 kvfree(area->pages);
1541 static inline void __vfree_deferred(const void *addr)
1544 * Use raw_cpu_ptr() because this can be called from preemptible
1545 * context. Preemption is absolutely fine here, because the llist_add()
1546 * implementation is lockless, so it works even if we are adding to
1547 * nother cpu's list. schedule_work() should be fine with this too.
1549 struct vfree_deferred *p = raw_cpu_ptr(&vfree_deferred);
1551 if (llist_add((struct llist_node *)addr, &p->list))
1552 schedule_work(&p->wq);
1556 * vfree_atomic - release memory allocated by vmalloc()
1557 * @addr: memory base address
1559 * This one is just like vfree() but can be called in any atomic context
1562 void vfree_atomic(const void *addr)
1566 kmemleak_free(addr);
1570 __vfree_deferred(addr);
1574 * vfree - release memory allocated by vmalloc()
1575 * @addr: memory base address
1577 * Free the virtually continuous memory area starting at @addr, as
1578 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
1579 * NULL, no operation is performed.
1581 * Must not be called in NMI context (strictly speaking, only if we don't
1582 * have CONFIG_ARCH_HAVE_NMI_SAFE_CMPXCHG, but making the calling
1583 * conventions for vfree() arch-depenedent would be a really bad idea)
1585 * NOTE: assumes that the object at @addr has a size >= sizeof(llist_node)
1587 void vfree(const void *addr)
1591 kmemleak_free(addr);
1595 if (unlikely(in_interrupt()))
1596 __vfree_deferred(addr);
1600 EXPORT_SYMBOL(vfree);
1603 * vunmap - release virtual mapping obtained by vmap()
1604 * @addr: memory base address
1606 * Free the virtually contiguous memory area starting at @addr,
1607 * which was created from the page array passed to vmap().
1609 * Must not be called in interrupt context.
1611 void vunmap(const void *addr)
1613 BUG_ON(in_interrupt());
1618 EXPORT_SYMBOL(vunmap);
1621 * vmap - map an array of pages into virtually contiguous space
1622 * @pages: array of page pointers
1623 * @count: number of pages to map
1624 * @flags: vm_area->flags
1625 * @prot: page protection for the mapping
1627 * Maps @count pages from @pages into contiguous kernel virtual
1630 void *vmap(struct page **pages, unsigned int count,
1631 unsigned long flags, pgprot_t prot)
1633 struct vm_struct *area;
1634 unsigned long size; /* In bytes */
1638 if (count > totalram_pages)
1641 size = (unsigned long)count << PAGE_SHIFT;
1642 area = get_vm_area_caller(size, flags, __builtin_return_address(0));
1646 if (map_vm_area(area, prot, pages)) {
1653 EXPORT_SYMBOL(vmap);
1655 static void *__vmalloc_node(unsigned long size, unsigned long align,
1656 gfp_t gfp_mask, pgprot_t prot,
1657 int node, const void *caller);
1658 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
1659 pgprot_t prot, int node)
1661 struct page **pages;
1662 unsigned int nr_pages, array_size, i;
1663 const gfp_t nested_gfp = (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO;
1664 const gfp_t alloc_mask = gfp_mask | __GFP_NOWARN;
1665 const gfp_t highmem_mask = (gfp_mask & (GFP_DMA | GFP_DMA32)) ?
1669 nr_pages = get_vm_area_size(area) >> PAGE_SHIFT;
1670 array_size = (nr_pages * sizeof(struct page *));
1672 /* Please note that the recursion is strictly bounded. */
1673 if (array_size > PAGE_SIZE) {
1674 pages = __vmalloc_node(array_size, 1, nested_gfp|highmem_mask,
1675 PAGE_KERNEL, node, area->caller);
1677 pages = kmalloc_node(array_size, nested_gfp, node);
1681 remove_vm_area(area->addr);
1686 area->pages = pages;
1687 area->nr_pages = nr_pages;
1689 for (i = 0; i < area->nr_pages; i++) {
1692 if (node == NUMA_NO_NODE)
1693 page = alloc_page(alloc_mask|highmem_mask);
1695 page = alloc_pages_node(node, alloc_mask|highmem_mask, 0);
1697 if (unlikely(!page)) {
1698 /* Successfully allocated i pages, free them in __vunmap() */
1702 area->pages[i] = page;
1703 if (gfpflags_allow_blocking(gfp_mask|highmem_mask))
1707 if (map_vm_area(area, prot, pages))
1712 warn_alloc(gfp_mask, NULL,
1713 "vmalloc: allocation failure, allocated %ld of %ld bytes",
1714 (area->nr_pages*PAGE_SIZE), area->size);
1720 * __vmalloc_node_range - allocate virtually contiguous memory
1721 * @size: allocation size
1722 * @align: desired alignment
1723 * @start: vm area range start
1724 * @end: vm area range end
1725 * @gfp_mask: flags for the page level allocator
1726 * @prot: protection mask for the allocated pages
1727 * @vm_flags: additional vm area flags (e.g. %VM_NO_GUARD)
1728 * @node: node to use for allocation or NUMA_NO_NODE
1729 * @caller: caller's return address
1731 * Allocate enough pages to cover @size from the page level
1732 * allocator with @gfp_mask flags. Map them into contiguous
1733 * kernel virtual space, using a pagetable protection of @prot.
1735 void *__vmalloc_node_range(unsigned long size, unsigned long align,
1736 unsigned long start, unsigned long end, gfp_t gfp_mask,
1737 pgprot_t prot, unsigned long vm_flags, int node,
1740 struct vm_struct *area;
1742 unsigned long real_size = size;
1744 size = PAGE_ALIGN(size);
1745 if (!size || (size >> PAGE_SHIFT) > totalram_pages)
1748 area = __get_vm_area_node(size, align, VM_ALLOC | VM_UNINITIALIZED |
1749 vm_flags, start, end, node, gfp_mask, caller);
1753 addr = __vmalloc_area_node(area, gfp_mask, prot, node);
1758 * First make sure the mappings are removed from all page-tables
1759 * before they are freed.
1761 vmalloc_sync_unmappings();
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);
1924 * vmalloc_exec - allocate virtually contiguous, executable memory
1925 * @size: allocation size
1927 * Kernel-internal function to allocate enough pages to cover @size
1928 * the page level allocator and map them into contiguous and
1929 * executable kernel virtual space.
1931 * For tight control over page level allocator and protection flags
1932 * use __vmalloc() instead.
1935 void *vmalloc_exec(unsigned long size)
1937 return __vmalloc_node(size, 1, GFP_KERNEL, PAGE_KERNEL_EXEC,
1938 NUMA_NO_NODE, __builtin_return_address(0));
1941 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
1942 #define GFP_VMALLOC32 (GFP_DMA32 | GFP_KERNEL)
1943 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
1944 #define GFP_VMALLOC32 (GFP_DMA | GFP_KERNEL)
1947 * 64b systems should always have either DMA or DMA32 zones. For others
1948 * GFP_DMA32 should do the right thing and use the normal zone.
1950 #define GFP_VMALLOC32 GFP_DMA32 | 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 * @pgoff: offset from @kaddr to start at
2233 * @size: size of map area
2235 * Returns: 0 for success, -Exxx on failure
2237 * This function checks that @kaddr is a valid vmalloc'ed area,
2238 * and that it is big enough to cover the range starting at
2239 * @uaddr in @vma. Will return failure if that criteria isn't
2242 * Similar to remap_pfn_range() (see mm/memory.c)
2244 int remap_vmalloc_range_partial(struct vm_area_struct *vma, unsigned long uaddr,
2245 void *kaddr, unsigned long pgoff,
2248 struct vm_struct *area;
2250 unsigned long end_index;
2252 if (check_shl_overflow(pgoff, PAGE_SHIFT, &off))
2255 size = PAGE_ALIGN(size);
2257 if (!PAGE_ALIGNED(uaddr) || !PAGE_ALIGNED(kaddr))
2260 area = find_vm_area(kaddr);
2264 if (!(area->flags & VM_USERMAP))
2267 if (check_add_overflow(size, off, &end_index) ||
2268 end_index > get_vm_area_size(area))
2273 struct page *page = vmalloc_to_page(kaddr);
2276 ret = vm_insert_page(vma, uaddr, page);
2285 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
2289 EXPORT_SYMBOL(remap_vmalloc_range_partial);
2292 * remap_vmalloc_range - map vmalloc pages to userspace
2293 * @vma: vma to cover (map full range of vma)
2294 * @addr: vmalloc memory
2295 * @pgoff: number of pages into addr before first page to map
2297 * Returns: 0 for success, -Exxx on failure
2299 * This function checks that addr is a valid vmalloc'ed area, and
2300 * that it is big enough to cover the vma. Will return failure if
2301 * that criteria isn't met.
2303 * Similar to remap_pfn_range() (see mm/memory.c)
2305 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
2306 unsigned long pgoff)
2308 return remap_vmalloc_range_partial(vma, vma->vm_start,
2310 vma->vm_end - vma->vm_start);
2312 EXPORT_SYMBOL(remap_vmalloc_range);
2315 * Implement stubs for vmalloc_sync_[un]mappings () if the architecture chose
2318 * The purpose of this function is to make sure the vmalloc area
2319 * mappings are identical in all page-tables in the system.
2321 void __weak vmalloc_sync_mappings(void)
2325 void __weak vmalloc_sync_unmappings(void)
2329 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
2341 * alloc_vm_area - allocate a range of kernel address space
2342 * @size: size of the area
2343 * @ptes: returns the PTEs for the address space
2345 * Returns: NULL on failure, vm_struct on success
2347 * This function reserves a range of kernel address space, and
2348 * allocates pagetables to map that range. No actual mappings
2351 * If @ptes is non-NULL, pointers to the PTEs (in init_mm)
2352 * allocated for the VM area are returned.
2354 struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes)
2356 struct vm_struct *area;
2358 area = get_vm_area_caller(size, VM_IOREMAP,
2359 __builtin_return_address(0));
2364 * This ensures that page tables are constructed for this region
2365 * of kernel virtual address space and mapped into init_mm.
2367 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
2368 size, f, ptes ? &ptes : NULL)) {
2375 EXPORT_SYMBOL_GPL(alloc_vm_area);
2377 void free_vm_area(struct vm_struct *area)
2379 struct vm_struct *ret;
2380 ret = remove_vm_area(area->addr);
2381 BUG_ON(ret != area);
2384 EXPORT_SYMBOL_GPL(free_vm_area);
2387 static struct vmap_area *node_to_va(struct rb_node *n)
2389 return rb_entry_safe(n, struct vmap_area, rb_node);
2393 * pvm_find_next_prev - find the next and prev vmap_area surrounding @end
2394 * @end: target address
2395 * @pnext: out arg for the next vmap_area
2396 * @pprev: out arg for the previous vmap_area
2398 * Returns: %true if either or both of next and prev are found,
2399 * %false if no vmap_area exists
2401 * Find vmap_areas end addresses of which enclose @end. ie. if not
2402 * NULL, *pnext->va_end > @end and *pprev->va_end <= @end.
2404 static bool pvm_find_next_prev(unsigned long end,
2405 struct vmap_area **pnext,
2406 struct vmap_area **pprev)
2408 struct rb_node *n = vmap_area_root.rb_node;
2409 struct vmap_area *va = NULL;
2412 va = rb_entry(n, struct vmap_area, rb_node);
2413 if (end < va->va_end)
2415 else if (end > va->va_end)
2424 if (va->va_end > end) {
2426 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2429 *pnext = node_to_va(rb_next(&(*pprev)->rb_node));
2435 * pvm_determine_end - find the highest aligned address between two vmap_areas
2436 * @pnext: in/out arg for the next vmap_area
2437 * @pprev: in/out arg for the previous vmap_area
2440 * Returns: determined end address
2442 * Find the highest aligned address between *@pnext and *@pprev below
2443 * VMALLOC_END. *@pnext and *@pprev are adjusted so that the aligned
2444 * down address is between the end addresses of the two vmap_areas.
2446 * Please note that the address returned by this function may fall
2447 * inside *@pnext vmap_area. The caller is responsible for checking
2450 static unsigned long pvm_determine_end(struct vmap_area **pnext,
2451 struct vmap_area **pprev,
2452 unsigned long align)
2454 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2458 addr = min((*pnext)->va_start & ~(align - 1), vmalloc_end);
2462 while (*pprev && (*pprev)->va_end > addr) {
2464 *pprev = node_to_va(rb_prev(&(*pnext)->rb_node));
2471 * pcpu_get_vm_areas - allocate vmalloc areas for percpu allocator
2472 * @offsets: array containing offset of each area
2473 * @sizes: array containing size of each area
2474 * @nr_vms: the number of areas to allocate
2475 * @align: alignment, all entries in @offsets and @sizes must be aligned to this
2477 * Returns: kmalloc'd vm_struct pointer array pointing to allocated
2478 * vm_structs on success, %NULL on failure
2480 * Percpu allocator wants to use congruent vm areas so that it can
2481 * maintain the offsets among percpu areas. This function allocates
2482 * congruent vmalloc areas for it with GFP_KERNEL. These areas tend to
2483 * be scattered pretty far, distance between two areas easily going up
2484 * to gigabytes. To avoid interacting with regular vmallocs, these
2485 * areas are allocated from top.
2487 * Despite its complicated look, this allocator is rather simple. It
2488 * does everything top-down and scans areas from the end looking for
2489 * matching slot. While scanning, if any of the areas overlaps with
2490 * existing vmap_area, the base address is pulled down to fit the
2491 * area. Scanning is repeated till all the areas fit and then all
2492 * necessary data structures are inserted and the result is returned.
2494 struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
2495 const size_t *sizes, int nr_vms,
2498 const unsigned long vmalloc_start = ALIGN(VMALLOC_START, align);
2499 const unsigned long vmalloc_end = VMALLOC_END & ~(align - 1);
2500 struct vmap_area **vas, *prev, *next;
2501 struct vm_struct **vms;
2502 int area, area2, last_area, term_area;
2503 unsigned long base, start, end, last_end;
2504 bool purged = false;
2506 /* verify parameters and allocate data structures */
2507 BUG_ON(offset_in_page(align) || !is_power_of_2(align));
2508 for (last_area = 0, area = 0; area < nr_vms; area++) {
2509 start = offsets[area];
2510 end = start + sizes[area];
2512 /* is everything aligned properly? */
2513 BUG_ON(!IS_ALIGNED(offsets[area], align));
2514 BUG_ON(!IS_ALIGNED(sizes[area], align));
2516 /* detect the area with the highest address */
2517 if (start > offsets[last_area])
2520 for (area2 = area + 1; area2 < nr_vms; area2++) {
2521 unsigned long start2 = offsets[area2];
2522 unsigned long end2 = start2 + sizes[area2];
2524 BUG_ON(start2 < end && start < end2);
2527 last_end = offsets[last_area] + sizes[last_area];
2529 if (vmalloc_end - vmalloc_start < last_end) {
2534 vms = kcalloc(nr_vms, sizeof(vms[0]), GFP_KERNEL);
2535 vas = kcalloc(nr_vms, sizeof(vas[0]), GFP_KERNEL);
2539 for (area = 0; area < nr_vms; area++) {
2540 vas[area] = kzalloc(sizeof(struct vmap_area), GFP_KERNEL);
2541 vms[area] = kzalloc(sizeof(struct vm_struct), GFP_KERNEL);
2542 if (!vas[area] || !vms[area])
2546 spin_lock(&vmap_area_lock);
2548 /* start scanning - we scan from the top, begin with the last area */
2549 area = term_area = last_area;
2550 start = offsets[area];
2551 end = start + sizes[area];
2553 if (!pvm_find_next_prev(vmap_area_pcpu_hole, &next, &prev)) {
2554 base = vmalloc_end - last_end;
2557 base = pvm_determine_end(&next, &prev, align) - end;
2560 BUG_ON(next && next->va_end <= base + end);
2561 BUG_ON(prev && prev->va_end > base + end);
2564 * base might have underflowed, add last_end before
2567 if (base + last_end < vmalloc_start + last_end) {
2568 spin_unlock(&vmap_area_lock);
2570 purge_vmap_area_lazy();
2578 * If next overlaps, move base downwards so that it's
2579 * right below next and then recheck.
2581 if (next && next->va_start < base + end) {
2582 base = pvm_determine_end(&next, &prev, align) - end;
2588 * If prev overlaps, shift down next and prev and move
2589 * base so that it's right below new next and then
2592 if (prev && prev->va_end > base + start) {
2594 prev = node_to_va(rb_prev(&next->rb_node));
2595 base = pvm_determine_end(&next, &prev, align) - end;
2601 * This area fits, move on to the previous one. If
2602 * the previous one is the terminal one, we're done.
2604 area = (area + nr_vms - 1) % nr_vms;
2605 if (area == term_area)
2607 start = offsets[area];
2608 end = start + sizes[area];
2609 pvm_find_next_prev(base + end, &next, &prev);
2612 /* we've found a fitting base, insert all va's */
2613 for (area = 0; area < nr_vms; area++) {
2614 struct vmap_area *va = vas[area];
2616 va->va_start = base + offsets[area];
2617 va->va_end = va->va_start + sizes[area];
2618 __insert_vmap_area(va);
2621 vmap_area_pcpu_hole = base + offsets[last_area];
2623 spin_unlock(&vmap_area_lock);
2625 /* insert all vm's */
2626 for (area = 0; area < nr_vms; area++)
2627 setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
2634 for (area = 0; area < nr_vms; area++) {
2645 * pcpu_free_vm_areas - free vmalloc areas for percpu allocator
2646 * @vms: vm_struct pointer array returned by pcpu_get_vm_areas()
2647 * @nr_vms: the number of allocated areas
2649 * Free vm_structs and the array allocated by pcpu_get_vm_areas().
2651 void pcpu_free_vm_areas(struct vm_struct **vms, int nr_vms)
2655 for (i = 0; i < nr_vms; i++)
2656 free_vm_area(vms[i]);
2659 #endif /* CONFIG_SMP */
2661 #ifdef CONFIG_PROC_FS
2662 static void *s_start(struct seq_file *m, loff_t *pos)
2663 __acquires(&vmap_area_lock)
2665 spin_lock(&vmap_area_lock);
2666 return seq_list_start(&vmap_area_list, *pos);
2669 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
2671 return seq_list_next(p, &vmap_area_list, pos);
2674 static void s_stop(struct seq_file *m, void *p)
2675 __releases(&vmap_area_lock)
2677 spin_unlock(&vmap_area_lock);
2680 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
2682 if (IS_ENABLED(CONFIG_NUMA)) {
2683 unsigned int nr, *counters = m->private;
2688 if (v->flags & VM_UNINITIALIZED)
2690 /* Pair with smp_wmb() in clear_vm_uninitialized_flag() */
2693 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
2695 for (nr = 0; nr < v->nr_pages; nr++)
2696 counters[page_to_nid(v->pages[nr])]++;
2698 for_each_node_state(nr, N_HIGH_MEMORY)
2700 seq_printf(m, " N%u=%u", nr, counters[nr]);
2704 static int s_show(struct seq_file *m, void *p)
2706 struct vmap_area *va;
2707 struct vm_struct *v;
2709 va = list_entry(p, struct vmap_area, list);
2712 * s_show can encounter race with remove_vm_area, !VM_VM_AREA on
2713 * behalf of vmap area is being tear down or vm_map_ram allocation.
2715 if (!(va->flags & VM_VM_AREA)) {
2716 seq_printf(m, "0x%pK-0x%pK %7ld %s\n",
2717 (void *)va->va_start, (void *)va->va_end,
2718 va->va_end - va->va_start,
2719 va->flags & VM_LAZY_FREE ? "unpurged vm_area" : "vm_map_ram");
2726 seq_printf(m, "0x%pK-0x%pK %7ld",
2727 v->addr, v->addr + v->size, v->size);
2730 seq_printf(m, " %pS", v->caller);
2733 seq_printf(m, " pages=%d", v->nr_pages);
2736 seq_printf(m, " phys=%pa", &v->phys_addr);
2738 if (v->flags & VM_IOREMAP)
2739 seq_puts(m, " ioremap");
2741 if (v->flags & VM_ALLOC)
2742 seq_puts(m, " vmalloc");
2744 if (v->flags & VM_MAP)
2745 seq_puts(m, " vmap");
2747 if (v->flags & VM_USERMAP)
2748 seq_puts(m, " user");
2750 if (is_vmalloc_addr(v->pages))
2751 seq_puts(m, " vpages");
2753 show_numa_info(m, v);
2758 static const struct seq_operations vmalloc_op = {
2765 static int __init proc_vmalloc_init(void)
2767 if (IS_ENABLED(CONFIG_NUMA))
2768 proc_create_seq_private("vmallocinfo", 0400, NULL,
2770 nr_node_ids * sizeof(unsigned int), NULL);
2772 proc_create_seq("vmallocinfo", 0400, NULL, &vmalloc_op);
2775 module_init(proc_vmalloc_init);