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>
18 #include <linux/vmalloc.h>
20 #include <asm/uaccess.h>
21 #include <asm/tlbflush.h>
24 DEFINE_RWLOCK(vmlist_lock);
25 struct vm_struct *vmlist;
27 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
30 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
34 pte = pte_offset_kernel(pmd, addr);
36 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
37 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
38 } while (pte++, addr += PAGE_SIZE, addr != end);
41 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
47 pmd = pmd_offset(pud, addr);
49 next = pmd_addr_end(addr, end);
50 if (pmd_none_or_clear_bad(pmd))
52 vunmap_pte_range(pmd, addr, next);
53 } while (pmd++, addr = next, addr != end);
56 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
62 pud = pud_offset(pgd, addr);
64 next = pud_addr_end(addr, end);
65 if (pud_none_or_clear_bad(pud))
67 vunmap_pmd_range(pud, addr, next);
68 } while (pud++, addr = next, addr != end);
71 void unmap_kernel_range(unsigned long addr, unsigned long size)
75 unsigned long start = addr;
76 unsigned long end = addr + size;
79 pgd = pgd_offset_k(addr);
80 flush_cache_vunmap(addr, end);
82 next = pgd_addr_end(addr, end);
83 if (pgd_none_or_clear_bad(pgd))
85 vunmap_pud_range(pgd, addr, next);
86 } while (pgd++, addr = next, addr != end);
87 flush_tlb_kernel_range(start, end);
90 static void unmap_vm_area(struct vm_struct *area)
92 unmap_kernel_range((unsigned long)area->addr, area->size);
95 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
96 unsigned long end, pgprot_t prot, struct page ***pages)
100 pte = pte_alloc_kernel(pmd, addr);
104 struct page *page = **pages;
105 WARN_ON(!pte_none(*pte));
108 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
110 } while (pte++, addr += PAGE_SIZE, addr != end);
114 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
115 unsigned long end, pgprot_t prot, struct page ***pages)
120 pmd = pmd_alloc(&init_mm, pud, addr);
124 next = pmd_addr_end(addr, end);
125 if (vmap_pte_range(pmd, addr, next, prot, pages))
127 } while (pmd++, addr = next, addr != end);
131 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
132 unsigned long end, pgprot_t prot, struct page ***pages)
137 pud = pud_alloc(&init_mm, pgd, addr);
141 next = pud_addr_end(addr, end);
142 if (vmap_pmd_range(pud, addr, next, prot, pages))
144 } while (pud++, addr = next, addr != end);
148 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
152 unsigned long addr = (unsigned long) area->addr;
153 unsigned long end = addr + area->size - PAGE_SIZE;
157 pgd = pgd_offset_k(addr);
159 next = pgd_addr_end(addr, end);
160 err = vmap_pud_range(pgd, addr, next, prot, pages);
163 } while (pgd++, addr = next, addr != end);
164 flush_cache_vmap((unsigned long) area->addr, end);
168 static struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
169 unsigned long start, unsigned long end,
170 int node, gfp_t gfp_mask)
172 struct vm_struct **p, *tmp, *area;
173 unsigned long align = 1;
176 BUG_ON(in_interrupt());
177 if (flags & VM_IOREMAP) {
180 if (bit > IOREMAP_MAX_ORDER)
181 bit = IOREMAP_MAX_ORDER;
182 else if (bit < PAGE_SHIFT)
187 addr = ALIGN(start, align);
188 size = PAGE_ALIGN(size);
192 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_LEVEL_MASK, node);
197 * We always allocate a guard page.
201 write_lock(&vmlist_lock);
202 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
203 if ((unsigned long)tmp->addr < addr) {
204 if((unsigned long)tmp->addr + tmp->size >= addr)
205 addr = ALIGN(tmp->size +
206 (unsigned long)tmp->addr, align);
209 if ((size + addr) < addr)
211 if (size + addr <= (unsigned long)tmp->addr)
213 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
214 if (addr > end - size)
223 area->addr = (void *)addr;
228 write_unlock(&vmlist_lock);
233 write_unlock(&vmlist_lock);
235 if (printk_ratelimit())
236 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
240 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
241 unsigned long start, unsigned long end)
243 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
247 * get_vm_area - reserve a contingous kernel virtual area
248 * @size: size of the area
249 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
251 * Search an area of @size in the kernel virtual mapping area,
252 * and reserved it for out purposes. Returns the area descriptor
253 * on success or %NULL on failure.
255 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
257 return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
260 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
261 int node, gfp_t gfp_mask)
263 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
267 /* Caller must hold vmlist_lock */
268 static struct vm_struct *__find_vm_area(void *addr)
270 struct vm_struct *tmp;
272 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
273 if (tmp->addr == addr)
280 /* Caller must hold vmlist_lock */
281 static struct vm_struct *__remove_vm_area(void *addr)
283 struct vm_struct **p, *tmp;
285 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
286 if (tmp->addr == addr)
296 * Remove the guard page.
298 tmp->size -= PAGE_SIZE;
303 * remove_vm_area - find and remove a contingous kernel virtual area
304 * @addr: base address
306 * Search for the kernel VM area starting at @addr, and remove it.
307 * This function returns the found VM area, but using it is NOT safe
308 * on SMP machines, except for its size or flags.
310 struct vm_struct *remove_vm_area(void *addr)
313 write_lock(&vmlist_lock);
314 v = __remove_vm_area(addr);
315 write_unlock(&vmlist_lock);
319 static void __vunmap(void *addr, int deallocate_pages)
321 struct vm_struct *area;
326 if ((PAGE_SIZE-1) & (unsigned long)addr) {
327 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
332 area = remove_vm_area(addr);
333 if (unlikely(!area)) {
334 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
340 debug_check_no_locks_freed(addr, area->size);
342 if (deallocate_pages) {
345 for (i = 0; i < area->nr_pages; i++) {
346 BUG_ON(!area->pages[i]);
347 __free_page(area->pages[i]);
350 if (area->flags & VM_VPAGES)
361 * vfree - release memory allocated by vmalloc()
362 * @addr: memory base address
364 * Free the virtually contiguous memory area starting at @addr, as
365 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
366 * NULL, no operation is performed.
368 * Must not be called in interrupt context.
370 void vfree(void *addr)
372 BUG_ON(in_interrupt());
375 EXPORT_SYMBOL(vfree);
378 * vunmap - release virtual mapping obtained by vmap()
379 * @addr: memory base address
381 * Free the virtually contiguous memory area starting at @addr,
382 * which was created from the page array passed to vmap().
384 * Must not be called in interrupt context.
386 void vunmap(void *addr)
388 BUG_ON(in_interrupt());
391 EXPORT_SYMBOL(vunmap);
394 * vmap - map an array of pages into virtually contiguous space
395 * @pages: array of page pointers
396 * @count: number of pages to map
397 * @flags: vm_area->flags
398 * @prot: page protection for the mapping
400 * Maps @count pages from @pages into contiguous kernel virtual
403 void *vmap(struct page **pages, unsigned int count,
404 unsigned long flags, pgprot_t prot)
406 struct vm_struct *area;
408 if (count > num_physpages)
411 area = get_vm_area((count << PAGE_SHIFT), flags);
414 if (map_vm_area(area, prot, &pages)) {
423 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
424 pgprot_t prot, int node)
427 unsigned int nr_pages, array_size, i;
429 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
430 array_size = (nr_pages * sizeof(struct page *));
432 area->nr_pages = nr_pages;
433 /* Please note that the recursion is strictly bounded. */
434 if (array_size > PAGE_SIZE) {
435 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
437 area->flags |= VM_VPAGES;
439 pages = kmalloc_node(array_size,
440 (gfp_mask & GFP_LEVEL_MASK) | __GFP_ZERO,
445 remove_vm_area(area->addr);
450 for (i = 0; i < area->nr_pages; i++) {
452 area->pages[i] = alloc_page(gfp_mask);
454 area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
455 if (unlikely(!area->pages[i])) {
456 /* Successfully allocated i pages, free them in __vunmap() */
462 if (map_vm_area(area, prot, &pages))
471 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
473 return __vmalloc_area_node(area, gfp_mask, prot, -1);
477 * __vmalloc_node - allocate virtually contiguous memory
478 * @size: allocation size
479 * @gfp_mask: flags for the page level allocator
480 * @prot: protection mask for the allocated pages
481 * @node: node to use for allocation or -1
483 * Allocate enough pages to cover @size from the page level
484 * allocator with @gfp_mask flags. Map them into contiguous
485 * kernel virtual space, using a pagetable protection of @prot.
487 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
490 struct vm_struct *area;
492 size = PAGE_ALIGN(size);
493 if (!size || (size >> PAGE_SHIFT) > num_physpages)
496 area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
500 return __vmalloc_area_node(area, gfp_mask, prot, node);
503 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
505 return __vmalloc_node(size, gfp_mask, prot, -1);
507 EXPORT_SYMBOL(__vmalloc);
510 * vmalloc - allocate virtually contiguous memory
511 * @size: allocation size
512 * Allocate enough pages to cover @size from the page level
513 * allocator and map them into contiguous kernel virtual space.
515 * For tight control over page level allocator and protection flags
516 * use __vmalloc() instead.
518 void *vmalloc(unsigned long size)
520 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
522 EXPORT_SYMBOL(vmalloc);
525 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
526 * @size: allocation size
528 * The resulting memory area is zeroed so it can be mapped to userspace
529 * without leaking data.
531 void *vmalloc_user(unsigned long size)
533 struct vm_struct *area;
536 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
538 write_lock(&vmlist_lock);
539 area = __find_vm_area(ret);
540 area->flags |= VM_USERMAP;
541 write_unlock(&vmlist_lock);
545 EXPORT_SYMBOL(vmalloc_user);
548 * vmalloc_node - allocate memory on a specific node
549 * @size: allocation size
552 * Allocate enough pages to cover @size from the page level
553 * allocator and map them into contiguous kernel virtual space.
555 * For tight control over page level allocator and protection flags
556 * use __vmalloc() instead.
558 void *vmalloc_node(unsigned long size, int node)
560 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
562 EXPORT_SYMBOL(vmalloc_node);
564 #ifndef PAGE_KERNEL_EXEC
565 # define PAGE_KERNEL_EXEC PAGE_KERNEL
569 * vmalloc_exec - allocate virtually contiguous, executable memory
570 * @size: allocation size
572 * Kernel-internal function to allocate enough pages to cover @size
573 * the page level allocator and map them into contiguous and
574 * executable kernel virtual space.
576 * For tight control over page level allocator and protection flags
577 * use __vmalloc() instead.
580 void *vmalloc_exec(unsigned long size)
582 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
585 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
586 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
587 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
588 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
590 #define GFP_VMALLOC32 GFP_KERNEL
594 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
595 * @size: allocation size
597 * Allocate enough 32bit PA addressable pages to cover @size from the
598 * page level allocator and map them into contiguous kernel virtual space.
600 void *vmalloc_32(unsigned long size)
602 return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
604 EXPORT_SYMBOL(vmalloc_32);
607 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
608 * @size: allocation size
610 * The resulting memory area is 32bit addressable and zeroed so it can be
611 * mapped to userspace without leaking data.
613 void *vmalloc_32_user(unsigned long size)
615 struct vm_struct *area;
618 ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
620 write_lock(&vmlist_lock);
621 area = __find_vm_area(ret);
622 area->flags |= VM_USERMAP;
623 write_unlock(&vmlist_lock);
627 EXPORT_SYMBOL(vmalloc_32_user);
629 long vread(char *buf, char *addr, unsigned long count)
631 struct vm_struct *tmp;
632 char *vaddr, *buf_start = buf;
635 /* Don't allow overflow */
636 if ((unsigned long) addr + count < count)
637 count = -(unsigned long) addr;
639 read_lock(&vmlist_lock);
640 for (tmp = vmlist; tmp; tmp = tmp->next) {
641 vaddr = (char *) tmp->addr;
642 if (addr >= vaddr + tmp->size - PAGE_SIZE)
644 while (addr < vaddr) {
652 n = vaddr + tmp->size - PAGE_SIZE - addr;
663 read_unlock(&vmlist_lock);
664 return buf - buf_start;
667 long vwrite(char *buf, char *addr, unsigned long count)
669 struct vm_struct *tmp;
670 char *vaddr, *buf_start = buf;
673 /* Don't allow overflow */
674 if ((unsigned long) addr + count < count)
675 count = -(unsigned long) addr;
677 read_lock(&vmlist_lock);
678 for (tmp = vmlist; tmp; tmp = tmp->next) {
679 vaddr = (char *) tmp->addr;
680 if (addr >= vaddr + tmp->size - PAGE_SIZE)
682 while (addr < vaddr) {
689 n = vaddr + tmp->size - PAGE_SIZE - addr;
700 read_unlock(&vmlist_lock);
701 return buf - buf_start;
705 * remap_vmalloc_range - map vmalloc pages to userspace
706 * @vma: vma to cover (map full range of vma)
707 * @addr: vmalloc memory
708 * @pgoff: number of pages into addr before first page to map
709 * @returns: 0 for success, -Exxx on failure
711 * This function checks that addr is a valid vmalloc'ed area, and
712 * that it is big enough to cover the vma. Will return failure if
713 * that criteria isn't met.
715 * Similar to remap_pfn_range() (see mm/memory.c)
717 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
720 struct vm_struct *area;
721 unsigned long uaddr = vma->vm_start;
722 unsigned long usize = vma->vm_end - vma->vm_start;
725 if ((PAGE_SIZE-1) & (unsigned long)addr)
728 read_lock(&vmlist_lock);
729 area = __find_vm_area(addr);
731 goto out_einval_locked;
733 if (!(area->flags & VM_USERMAP))
734 goto out_einval_locked;
736 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
737 goto out_einval_locked;
738 read_unlock(&vmlist_lock);
740 addr += pgoff << PAGE_SHIFT;
742 struct page *page = vmalloc_to_page(addr);
743 ret = vm_insert_page(vma, uaddr, page);
752 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
753 vma->vm_flags |= VM_RESERVED;
758 read_unlock(&vmlist_lock);
761 EXPORT_SYMBOL(remap_vmalloc_range);
764 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
767 void __attribute__((weak)) vmalloc_sync_all(void)
772 static int f(pte_t *pte, struct page *pmd_page, unsigned long addr, void *data)
774 /* apply_to_page_range() does all the hard work. */
779 * alloc_vm_area - allocate a range of kernel address space
780 * @size: size of the area
781 * @returns: NULL on failure, vm_struct on success
783 * This function reserves a range of kernel address space, and
784 * allocates pagetables to map that range. No actual mappings
785 * are created. If the kernel address space is not shared
786 * between processes, it syncs the pagetable across all
789 struct vm_struct *alloc_vm_area(size_t size)
791 struct vm_struct *area;
793 area = get_vm_area(size, VM_IOREMAP);
798 * This ensures that page tables are constructed for this region
799 * of kernel virtual address space and mapped into init_mm.
801 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
802 area->size, f, NULL)) {
807 /* Make sure the pagetables are constructed in process kernel
813 EXPORT_SYMBOL_GPL(alloc_vm_area);
815 void free_vm_area(struct vm_struct *area)
817 struct vm_struct *ret;
818 ret = remove_vm_area(area->addr);
822 EXPORT_SYMBOL_GPL(free_vm_area);