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
12 #include <linux/module.h>
13 #include <linux/highmem.h>
14 #include <linux/slab.h>
15 #include <linux/spinlock.h>
16 #include <linux/interrupt.h>
17 #include <linux/seq_file.h>
18 #include <linux/debugobjects.h>
19 #include <linux/vmalloc.h>
20 #include <linux/kallsyms.h>
22 #include <asm/uaccess.h>
23 #include <asm/tlbflush.h>
26 DEFINE_RWLOCK(vmlist_lock);
27 struct vm_struct *vmlist;
29 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
30 int node, void *caller);
32 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
36 pte = pte_offset_kernel(pmd, addr);
38 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
39 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
40 } while (pte++, addr += PAGE_SIZE, addr != end);
43 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
49 pmd = pmd_offset(pud, addr);
51 next = pmd_addr_end(addr, end);
52 if (pmd_none_or_clear_bad(pmd))
54 vunmap_pte_range(pmd, addr, next);
55 } while (pmd++, addr = next, addr != end);
58 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
64 pud = pud_offset(pgd, addr);
66 next = pud_addr_end(addr, end);
67 if (pud_none_or_clear_bad(pud))
69 vunmap_pmd_range(pud, addr, next);
70 } while (pud++, addr = next, addr != end);
73 void unmap_kernel_range(unsigned long addr, unsigned long size)
77 unsigned long start = addr;
78 unsigned long end = addr + size;
81 pgd = pgd_offset_k(addr);
82 flush_cache_vunmap(addr, end);
84 next = pgd_addr_end(addr, end);
85 if (pgd_none_or_clear_bad(pgd))
87 vunmap_pud_range(pgd, addr, next);
88 } while (pgd++, addr = next, addr != end);
89 flush_tlb_kernel_range(start, end);
92 static void unmap_vm_area(struct vm_struct *area)
94 unmap_kernel_range((unsigned long)area->addr, area->size);
97 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
98 unsigned long end, pgprot_t prot, struct page ***pages)
102 pte = pte_alloc_kernel(pmd, addr);
106 struct page *page = **pages;
107 WARN_ON(!pte_none(*pte));
110 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
112 } while (pte++, addr += PAGE_SIZE, addr != end);
116 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
117 unsigned long end, pgprot_t prot, struct page ***pages)
122 pmd = pmd_alloc(&init_mm, pud, addr);
126 next = pmd_addr_end(addr, end);
127 if (vmap_pte_range(pmd, addr, next, prot, pages))
129 } while (pmd++, addr = next, addr != end);
133 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
134 unsigned long end, pgprot_t prot, struct page ***pages)
139 pud = pud_alloc(&init_mm, pgd, addr);
143 next = pud_addr_end(addr, end);
144 if (vmap_pmd_range(pud, addr, next, prot, pages))
146 } while (pud++, addr = next, addr != end);
150 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
154 unsigned long addr = (unsigned long) area->addr;
155 unsigned long end = addr + area->size - PAGE_SIZE;
159 pgd = pgd_offset_k(addr);
161 next = pgd_addr_end(addr, end);
162 err = vmap_pud_range(pgd, addr, next, prot, pages);
165 } while (pgd++, addr = next, addr != end);
166 flush_cache_vmap((unsigned long) area->addr, end);
169 EXPORT_SYMBOL_GPL(map_vm_area);
172 * Map a vmalloc()-space virtual address to the physical page.
174 struct page *vmalloc_to_page(const void *vmalloc_addr)
176 unsigned long addr = (unsigned long) vmalloc_addr;
177 struct page *page = NULL;
178 pgd_t *pgd = pgd_offset_k(addr);
184 * XXX we might need to change this if we add VIRTUAL_BUG_ON for
185 * architectures that do not vmalloc module space
187 VIRTUAL_BUG_ON(!is_vmalloc_addr(vmalloc_addr) &&
188 !is_module_address(addr));
190 if (!pgd_none(*pgd)) {
191 pud = pud_offset(pgd, addr);
192 if (!pud_none(*pud)) {
193 pmd = pmd_offset(pud, addr);
194 if (!pmd_none(*pmd)) {
195 ptep = pte_offset_map(pmd, addr);
197 if (pte_present(pte))
198 page = pte_page(pte);
205 EXPORT_SYMBOL(vmalloc_to_page);
208 * Map a vmalloc()-space virtual address to the physical page frame number.
210 unsigned long vmalloc_to_pfn(const void *vmalloc_addr)
212 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
214 EXPORT_SYMBOL(vmalloc_to_pfn);
216 static struct vm_struct *
217 __get_vm_area_node(unsigned long size, unsigned long flags, unsigned long start,
218 unsigned long end, int node, gfp_t gfp_mask, void *caller)
220 struct vm_struct **p, *tmp, *area;
221 unsigned long align = 1;
224 BUG_ON(in_interrupt());
225 if (flags & VM_IOREMAP) {
228 if (bit > IOREMAP_MAX_ORDER)
229 bit = IOREMAP_MAX_ORDER;
230 else if (bit < PAGE_SHIFT)
235 addr = ALIGN(start, align);
236 size = PAGE_ALIGN(size);
240 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
246 * We always allocate a guard page.
250 write_lock(&vmlist_lock);
251 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
252 if ((unsigned long)tmp->addr < addr) {
253 if((unsigned long)tmp->addr + tmp->size >= addr)
254 addr = ALIGN(tmp->size +
255 (unsigned long)tmp->addr, align);
258 if ((size + addr) < addr)
260 if (size + addr <= (unsigned long)tmp->addr)
262 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
263 if (addr > end - size)
266 if ((size + addr) < addr)
268 if (addr > end - size)
276 area->addr = (void *)addr;
281 area->caller = caller;
282 write_unlock(&vmlist_lock);
287 write_unlock(&vmlist_lock);
289 if (printk_ratelimit())
290 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
294 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
295 unsigned long start, unsigned long end)
297 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL,
298 __builtin_return_address(0));
300 EXPORT_SYMBOL_GPL(__get_vm_area);
303 * get_vm_area - reserve a contiguous kernel virtual area
304 * @size: size of the area
305 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
307 * Search an area of @size in the kernel virtual mapping area,
308 * and reserved it for out purposes. Returns the area descriptor
309 * on success or %NULL on failure.
311 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
313 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
314 -1, GFP_KERNEL, __builtin_return_address(0));
317 struct vm_struct *get_vm_area_caller(unsigned long size, unsigned long flags,
320 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END,
321 -1, GFP_KERNEL, caller);
324 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
325 int node, gfp_t gfp_mask)
327 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
328 gfp_mask, __builtin_return_address(0));
331 /* Caller must hold vmlist_lock */
332 static struct vm_struct *__find_vm_area(const void *addr)
334 struct vm_struct *tmp;
336 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
337 if (tmp->addr == addr)
344 /* Caller must hold vmlist_lock */
345 static struct vm_struct *__remove_vm_area(const void *addr)
347 struct vm_struct **p, *tmp;
349 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
350 if (tmp->addr == addr)
360 * Remove the guard page.
362 tmp->size -= PAGE_SIZE;
367 * remove_vm_area - find and remove a continuous kernel virtual area
368 * @addr: base address
370 * Search for the kernel VM area starting at @addr, and remove it.
371 * This function returns the found VM area, but using it is NOT safe
372 * on SMP machines, except for its size or flags.
374 struct vm_struct *remove_vm_area(const void *addr)
377 write_lock(&vmlist_lock);
378 v = __remove_vm_area(addr);
379 write_unlock(&vmlist_lock);
383 static void __vunmap(const void *addr, int deallocate_pages)
385 struct vm_struct *area;
390 if ((PAGE_SIZE-1) & (unsigned long)addr) {
391 WARN(1, KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
395 area = remove_vm_area(addr);
396 if (unlikely(!area)) {
397 WARN(1, KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
402 debug_check_no_locks_freed(addr, area->size);
403 debug_check_no_obj_freed(addr, area->size);
405 if (deallocate_pages) {
408 for (i = 0; i < area->nr_pages; i++) {
409 struct page *page = area->pages[i];
415 if (area->flags & VM_VPAGES)
426 * vfree - release memory allocated by vmalloc()
427 * @addr: memory base address
429 * Free the virtually continuous memory area starting at @addr, as
430 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
431 * NULL, no operation is performed.
433 * Must not be called in interrupt context.
435 void vfree(const void *addr)
437 BUG_ON(in_interrupt());
440 EXPORT_SYMBOL(vfree);
443 * vunmap - release virtual mapping obtained by vmap()
444 * @addr: memory base address
446 * Free the virtually contiguous memory area starting at @addr,
447 * which was created from the page array passed to vmap().
449 * Must not be called in interrupt context.
451 void vunmap(const void *addr)
453 BUG_ON(in_interrupt());
456 EXPORT_SYMBOL(vunmap);
459 * vmap - map an array of pages into virtually contiguous space
460 * @pages: array of page pointers
461 * @count: number of pages to map
462 * @flags: vm_area->flags
463 * @prot: page protection for the mapping
465 * Maps @count pages from @pages into contiguous kernel virtual
468 void *vmap(struct page **pages, unsigned int count,
469 unsigned long flags, pgprot_t prot)
471 struct vm_struct *area;
473 if (count > num_physpages)
476 area = get_vm_area_caller((count << PAGE_SHIFT), flags,
477 __builtin_return_address(0));
481 if (map_vm_area(area, prot, &pages)) {
490 static void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
491 pgprot_t prot, int node, void *caller)
494 unsigned int nr_pages, array_size, i;
496 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
497 array_size = (nr_pages * sizeof(struct page *));
499 area->nr_pages = nr_pages;
500 /* Please note that the recursion is strictly bounded. */
501 if (array_size > PAGE_SIZE) {
502 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
503 PAGE_KERNEL, node, caller);
504 area->flags |= VM_VPAGES;
506 pages = kmalloc_node(array_size,
507 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
511 area->caller = caller;
513 remove_vm_area(area->addr);
518 for (i = 0; i < area->nr_pages; i++) {
522 page = alloc_page(gfp_mask);
524 page = alloc_pages_node(node, gfp_mask, 0);
526 if (unlikely(!page)) {
527 /* Successfully allocated i pages, free them in __vunmap() */
531 area->pages[i] = page;
534 if (map_vm_area(area, prot, &pages))
543 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
545 return __vmalloc_area_node(area, gfp_mask, prot, -1,
546 __builtin_return_address(0));
550 * __vmalloc_node - allocate virtually contiguous memory
551 * @size: allocation size
552 * @gfp_mask: flags for the page level allocator
553 * @prot: protection mask for the allocated pages
554 * @node: node to use for allocation or -1
555 * @caller: caller's return address
557 * Allocate enough pages to cover @size from the page level
558 * allocator with @gfp_mask flags. Map them into contiguous
559 * kernel virtual space, using a pagetable protection of @prot.
561 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
562 int node, void *caller)
564 struct vm_struct *area;
566 size = PAGE_ALIGN(size);
567 if (!size || (size >> PAGE_SHIFT) > num_physpages)
570 area = __get_vm_area_node(size, VM_ALLOC, VMALLOC_START, VMALLOC_END,
571 node, gfp_mask, caller);
576 return __vmalloc_area_node(area, gfp_mask, prot, node, caller);
579 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
581 return __vmalloc_node(size, gfp_mask, prot, -1,
582 __builtin_return_address(0));
584 EXPORT_SYMBOL(__vmalloc);
587 * vmalloc - allocate virtually contiguous memory
588 * @size: allocation size
589 * Allocate enough pages to cover @size from the page level
590 * allocator and map them into contiguous kernel virtual space.
592 * For tight control over page level allocator and protection flags
593 * use __vmalloc() instead.
595 void *vmalloc(unsigned long size)
597 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
598 -1, __builtin_return_address(0));
600 EXPORT_SYMBOL(vmalloc);
603 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
604 * @size: allocation size
606 * The resulting memory area is zeroed so it can be mapped to userspace
607 * without leaking data.
609 void *vmalloc_user(unsigned long size)
611 struct vm_struct *area;
614 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
616 write_lock(&vmlist_lock);
617 area = __find_vm_area(ret);
618 area->flags |= VM_USERMAP;
619 write_unlock(&vmlist_lock);
623 EXPORT_SYMBOL(vmalloc_user);
626 * vmalloc_node - allocate memory on a specific node
627 * @size: allocation size
630 * Allocate enough pages to cover @size from the page level
631 * allocator and map them into contiguous kernel virtual space.
633 * For tight control over page level allocator and protection flags
634 * use __vmalloc() instead.
636 void *vmalloc_node(unsigned long size, int node)
638 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL,
639 node, __builtin_return_address(0));
641 EXPORT_SYMBOL(vmalloc_node);
643 #ifndef PAGE_KERNEL_EXEC
644 # define PAGE_KERNEL_EXEC PAGE_KERNEL
648 * vmalloc_exec - allocate virtually contiguous, executable memory
649 * @size: allocation size
651 * Kernel-internal function to allocate enough pages to cover @size
652 * the page level allocator and map them into contiguous and
653 * executable kernel virtual space.
655 * For tight control over page level allocator and protection flags
656 * use __vmalloc() instead.
659 void *vmalloc_exec(unsigned long size)
661 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
664 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
665 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
666 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
667 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
669 #define GFP_VMALLOC32 GFP_KERNEL
673 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
674 * @size: allocation size
676 * Allocate enough 32bit PA addressable pages to cover @size from the
677 * page level allocator and map them into contiguous kernel virtual space.
679 void *vmalloc_32(unsigned long size)
681 return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
683 EXPORT_SYMBOL(vmalloc_32);
686 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
687 * @size: allocation size
689 * The resulting memory area is 32bit addressable and zeroed so it can be
690 * mapped to userspace without leaking data.
692 void *vmalloc_32_user(unsigned long size)
694 struct vm_struct *area;
697 ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
699 write_lock(&vmlist_lock);
700 area = __find_vm_area(ret);
701 area->flags |= VM_USERMAP;
702 write_unlock(&vmlist_lock);
706 EXPORT_SYMBOL(vmalloc_32_user);
708 long vread(char *buf, char *addr, unsigned long count)
710 struct vm_struct *tmp;
711 char *vaddr, *buf_start = buf;
714 /* Don't allow overflow */
715 if ((unsigned long) addr + count < count)
716 count = -(unsigned long) addr;
718 read_lock(&vmlist_lock);
719 for (tmp = vmlist; tmp; tmp = tmp->next) {
720 vaddr = (char *) tmp->addr;
721 if (addr >= vaddr + tmp->size - PAGE_SIZE)
723 while (addr < vaddr) {
731 n = vaddr + tmp->size - PAGE_SIZE - addr;
742 read_unlock(&vmlist_lock);
743 return buf - buf_start;
746 long vwrite(char *buf, char *addr, unsigned long count)
748 struct vm_struct *tmp;
749 char *vaddr, *buf_start = buf;
752 /* Don't allow overflow */
753 if ((unsigned long) addr + count < count)
754 count = -(unsigned long) addr;
756 read_lock(&vmlist_lock);
757 for (tmp = vmlist; tmp; tmp = tmp->next) {
758 vaddr = (char *) tmp->addr;
759 if (addr >= vaddr + tmp->size - PAGE_SIZE)
761 while (addr < vaddr) {
768 n = vaddr + tmp->size - PAGE_SIZE - addr;
779 read_unlock(&vmlist_lock);
780 return buf - buf_start;
784 * remap_vmalloc_range - map vmalloc pages to userspace
785 * @vma: vma to cover (map full range of vma)
786 * @addr: vmalloc memory
787 * @pgoff: number of pages into addr before first page to map
789 * Returns: 0 for success, -Exxx on failure
791 * This function checks that addr is a valid vmalloc'ed area, and
792 * that it is big enough to cover the vma. Will return failure if
793 * that criteria isn't met.
795 * Similar to remap_pfn_range() (see mm/memory.c)
797 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
800 struct vm_struct *area;
801 unsigned long uaddr = vma->vm_start;
802 unsigned long usize = vma->vm_end - vma->vm_start;
805 if ((PAGE_SIZE-1) & (unsigned long)addr)
808 read_lock(&vmlist_lock);
809 area = __find_vm_area(addr);
811 goto out_einval_locked;
813 if (!(area->flags & VM_USERMAP))
814 goto out_einval_locked;
816 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
817 goto out_einval_locked;
818 read_unlock(&vmlist_lock);
820 addr += pgoff << PAGE_SHIFT;
822 struct page *page = vmalloc_to_page(addr);
823 ret = vm_insert_page(vma, uaddr, page);
832 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
833 vma->vm_flags |= VM_RESERVED;
838 read_unlock(&vmlist_lock);
841 EXPORT_SYMBOL(remap_vmalloc_range);
844 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
847 void __attribute__((weak)) vmalloc_sync_all(void)
852 static int f(pte_t *pte, pgtable_t table, unsigned long addr, void *data)
854 /* apply_to_page_range() does all the hard work. */
859 * alloc_vm_area - allocate a range of kernel address space
860 * @size: size of the area
862 * Returns: NULL on failure, vm_struct on success
864 * This function reserves a range of kernel address space, and
865 * allocates pagetables to map that range. No actual mappings
866 * are created. If the kernel address space is not shared
867 * between processes, it syncs the pagetable across all
870 struct vm_struct *alloc_vm_area(size_t size)
872 struct vm_struct *area;
874 area = get_vm_area_caller(size, VM_IOREMAP,
875 __builtin_return_address(0));
880 * This ensures that page tables are constructed for this region
881 * of kernel virtual address space and mapped into init_mm.
883 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
884 area->size, f, NULL)) {
889 /* Make sure the pagetables are constructed in process kernel
895 EXPORT_SYMBOL_GPL(alloc_vm_area);
897 void free_vm_area(struct vm_struct *area)
899 struct vm_struct *ret;
900 ret = remove_vm_area(area->addr);
904 EXPORT_SYMBOL_GPL(free_vm_area);
907 #ifdef CONFIG_PROC_FS
908 static void *s_start(struct seq_file *m, loff_t *pos)
913 read_lock(&vmlist_lock);
926 static void *s_next(struct seq_file *m, void *p, loff_t *pos)
928 struct vm_struct *v = p;
934 static void s_stop(struct seq_file *m, void *p)
936 read_unlock(&vmlist_lock);
939 static void show_numa_info(struct seq_file *m, struct vm_struct *v)
942 unsigned int nr, *counters = m->private;
947 memset(counters, 0, nr_node_ids * sizeof(unsigned int));
949 for (nr = 0; nr < v->nr_pages; nr++)
950 counters[page_to_nid(v->pages[nr])]++;
952 for_each_node_state(nr, N_HIGH_MEMORY)
954 seq_printf(m, " N%u=%u", nr, counters[nr]);
958 static int s_show(struct seq_file *m, void *p)
960 struct vm_struct *v = p;
962 seq_printf(m, "0x%p-0x%p %7ld",
963 v->addr, v->addr + v->size, v->size);
966 char buff[2 * KSYM_NAME_LEN];
969 sprint_symbol(buff, (unsigned long)v->caller);
974 seq_printf(m, " pages=%d", v->nr_pages);
977 seq_printf(m, " phys=%lx", v->phys_addr);
979 if (v->flags & VM_IOREMAP)
980 seq_printf(m, " ioremap");
982 if (v->flags & VM_ALLOC)
983 seq_printf(m, " vmalloc");
985 if (v->flags & VM_MAP)
986 seq_printf(m, " vmap");
988 if (v->flags & VM_USERMAP)
989 seq_printf(m, " user");
991 if (v->flags & VM_VPAGES)
992 seq_printf(m, " vpages");
994 show_numa_info(m, v);
999 const struct seq_operations vmalloc_op = {