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_vm_area(struct vm_struct *area)
75 unsigned long addr = (unsigned long) area->addr;
76 unsigned long end = addr + area->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((unsigned long) area->addr, end);
90 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
91 unsigned long end, pgprot_t prot, struct page ***pages)
95 pte = pte_alloc_kernel(pmd, addr);
99 struct page *page = **pages;
100 WARN_ON(!pte_none(*pte));
103 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
105 } while (pte++, addr += PAGE_SIZE, addr != end);
109 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
110 unsigned long end, pgprot_t prot, struct page ***pages)
115 pmd = pmd_alloc(&init_mm, pud, addr);
119 next = pmd_addr_end(addr, end);
120 if (vmap_pte_range(pmd, addr, next, prot, pages))
122 } while (pmd++, addr = next, addr != end);
126 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
127 unsigned long end, pgprot_t prot, struct page ***pages)
132 pud = pud_alloc(&init_mm, pgd, addr);
136 next = pud_addr_end(addr, end);
137 if (vmap_pmd_range(pud, addr, next, prot, pages))
139 } while (pud++, addr = next, addr != end);
143 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
147 unsigned long addr = (unsigned long) area->addr;
148 unsigned long end = addr + area->size - PAGE_SIZE;
152 pgd = pgd_offset_k(addr);
154 next = pgd_addr_end(addr, end);
155 err = vmap_pud_range(pgd, addr, next, prot, pages);
158 } while (pgd++, addr = next, addr != end);
159 flush_cache_vmap((unsigned long) area->addr, end);
163 struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
164 unsigned long start, unsigned long end, int node)
166 struct vm_struct **p, *tmp, *area;
167 unsigned long align = 1;
170 if (flags & VM_IOREMAP) {
173 if (bit > IOREMAP_MAX_ORDER)
174 bit = IOREMAP_MAX_ORDER;
175 else if (bit < PAGE_SHIFT)
180 addr = ALIGN(start, align);
181 size = PAGE_ALIGN(size);
183 area = kmalloc_node(sizeof(*area), GFP_KERNEL, node);
187 if (unlikely(!size)) {
193 * We always allocate a guard page.
197 write_lock(&vmlist_lock);
198 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
199 if ((unsigned long)tmp->addr < addr) {
200 if((unsigned long)tmp->addr + tmp->size >= addr)
201 addr = ALIGN(tmp->size +
202 (unsigned long)tmp->addr, align);
205 if ((size + addr) < addr)
207 if (size + addr <= (unsigned long)tmp->addr)
209 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
210 if (addr > end - size)
219 area->addr = (void *)addr;
224 write_unlock(&vmlist_lock);
229 write_unlock(&vmlist_lock);
231 if (printk_ratelimit())
232 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
236 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
237 unsigned long start, unsigned long end)
239 return __get_vm_area_node(size, flags, start, end, -1);
243 * get_vm_area - reserve a contingous kernel virtual area
244 * @size: size of the area
245 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
247 * Search an area of @size in the kernel virtual mapping area,
248 * and reserved it for out purposes. Returns the area descriptor
249 * on success or %NULL on failure.
251 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
253 return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
256 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, int node)
258 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node);
261 /* Caller must hold vmlist_lock */
262 static struct vm_struct *__find_vm_area(void *addr)
264 struct vm_struct *tmp;
266 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
267 if (tmp->addr == addr)
274 /* Caller must hold vmlist_lock */
275 static struct vm_struct *__remove_vm_area(void *addr)
277 struct vm_struct **p, *tmp;
279 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
280 if (tmp->addr == addr)
290 * Remove the guard page.
292 tmp->size -= PAGE_SIZE;
297 * remove_vm_area - find and remove a contingous kernel virtual area
298 * @addr: base address
300 * Search for the kernel VM area starting at @addr, and remove it.
301 * This function returns the found VM area, but using it is NOT safe
302 * on SMP machines, except for its size or flags.
304 struct vm_struct *remove_vm_area(void *addr)
307 write_lock(&vmlist_lock);
308 v = __remove_vm_area(addr);
309 write_unlock(&vmlist_lock);
313 void __vunmap(void *addr, int deallocate_pages)
315 struct vm_struct *area;
320 if ((PAGE_SIZE-1) & (unsigned long)addr) {
321 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
326 area = remove_vm_area(addr);
327 if (unlikely(!area)) {
328 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
334 debug_check_no_locks_freed(addr, area->size);
336 if (deallocate_pages) {
339 for (i = 0; i < area->nr_pages; i++) {
340 BUG_ON(!area->pages[i]);
341 __free_page(area->pages[i]);
344 if (area->flags & VM_VPAGES)
355 * vfree - release memory allocated by vmalloc()
356 * @addr: memory base address
358 * Free the virtually contiguous memory area starting at @addr, as
359 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
360 * NULL, no operation is performed.
362 * Must not be called in interrupt context.
364 void vfree(void *addr)
366 BUG_ON(in_interrupt());
369 EXPORT_SYMBOL(vfree);
372 * vunmap - release virtual mapping obtained by vmap()
373 * @addr: memory base address
375 * Free the virtually contiguous memory area starting at @addr,
376 * which was created from the page array passed to vmap().
378 * Must not be called in interrupt context.
380 void vunmap(void *addr)
382 BUG_ON(in_interrupt());
385 EXPORT_SYMBOL(vunmap);
388 * vmap - map an array of pages into virtually contiguous space
389 * @pages: array of page pointers
390 * @count: number of pages to map
391 * @flags: vm_area->flags
392 * @prot: page protection for the mapping
394 * Maps @count pages from @pages into contiguous kernel virtual
397 void *vmap(struct page **pages, unsigned int count,
398 unsigned long flags, pgprot_t prot)
400 struct vm_struct *area;
402 if (count > num_physpages)
405 area = get_vm_area((count << PAGE_SHIFT), flags);
408 if (map_vm_area(area, prot, &pages)) {
417 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
418 pgprot_t prot, int node)
421 unsigned int nr_pages, array_size, i;
423 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
424 array_size = (nr_pages * sizeof(struct page *));
426 area->nr_pages = nr_pages;
427 /* Please note that the recursion is strictly bounded. */
428 if (array_size > PAGE_SIZE) {
429 pages = __vmalloc_node(array_size, gfp_mask, PAGE_KERNEL, node);
430 area->flags |= VM_VPAGES;
432 pages = kmalloc_node(array_size, (gfp_mask & ~__GFP_HIGHMEM), node);
435 remove_vm_area(area->addr);
439 memset(area->pages, 0, array_size);
441 for (i = 0; i < area->nr_pages; i++) {
443 area->pages[i] = alloc_page(gfp_mask);
445 area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
446 if (unlikely(!area->pages[i])) {
447 /* Successfully allocated i pages, free them in __vunmap() */
453 if (map_vm_area(area, prot, &pages))
462 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
464 return __vmalloc_area_node(area, gfp_mask, prot, -1);
468 * __vmalloc_node - allocate virtually contiguous memory
469 * @size: allocation size
470 * @gfp_mask: flags for the page level allocator
471 * @prot: protection mask for the allocated pages
472 * @node: node to use for allocation or -1
474 * Allocate enough pages to cover @size from the page level
475 * allocator with @gfp_mask flags. Map them into contiguous
476 * kernel virtual space, using a pagetable protection of @prot.
478 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
481 struct vm_struct *area;
483 size = PAGE_ALIGN(size);
484 if (!size || (size >> PAGE_SHIFT) > num_physpages)
487 area = get_vm_area_node(size, VM_ALLOC, node);
491 return __vmalloc_area_node(area, gfp_mask, prot, node);
494 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
496 return __vmalloc_node(size, gfp_mask, prot, -1);
498 EXPORT_SYMBOL(__vmalloc);
501 * vmalloc - allocate virtually contiguous memory
502 * @size: allocation size
503 * Allocate enough pages to cover @size from the page level
504 * allocator and map them into contiguous kernel virtual space.
506 * For tight control over page level allocator and protection flags
507 * use __vmalloc() instead.
509 void *vmalloc(unsigned long size)
511 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
513 EXPORT_SYMBOL(vmalloc);
516 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
517 * @size: allocation size
519 * The resulting memory area is zeroed so it can be mapped to userspace
520 * without leaking data.
522 void *vmalloc_user(unsigned long size)
524 struct vm_struct *area;
527 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
528 write_lock(&vmlist_lock);
529 area = __find_vm_area(ret);
530 area->flags |= VM_USERMAP;
531 write_unlock(&vmlist_lock);
535 EXPORT_SYMBOL(vmalloc_user);
538 * vmalloc_node - allocate memory on a specific node
539 * @size: allocation size
542 * Allocate enough pages to cover @size from the page level
543 * allocator and map them into contiguous kernel virtual space.
545 * For tight control over page level allocator and protection flags
546 * use __vmalloc() instead.
548 void *vmalloc_node(unsigned long size, int node)
550 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
552 EXPORT_SYMBOL(vmalloc_node);
554 #ifndef PAGE_KERNEL_EXEC
555 # define PAGE_KERNEL_EXEC PAGE_KERNEL
559 * vmalloc_exec - allocate virtually contiguous, executable memory
560 * @size: allocation size
562 * Kernel-internal function to allocate enough pages to cover @size
563 * the page level allocator and map them into contiguous and
564 * executable kernel virtual space.
566 * For tight control over page level allocator and protection flags
567 * use __vmalloc() instead.
570 void *vmalloc_exec(unsigned long size)
572 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
576 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
577 * @size: allocation size
579 * Allocate enough 32bit PA addressable pages to cover @size from the
580 * page level allocator and map them into contiguous kernel virtual space.
582 void *vmalloc_32(unsigned long size)
584 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
586 EXPORT_SYMBOL(vmalloc_32);
589 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
590 * @size: allocation size
592 * The resulting memory area is 32bit addressable and zeroed so it can be
593 * mapped to userspace without leaking data.
595 void *vmalloc_32_user(unsigned long size)
597 struct vm_struct *area;
600 ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
601 write_lock(&vmlist_lock);
602 area = __find_vm_area(ret);
603 area->flags |= VM_USERMAP;
604 write_unlock(&vmlist_lock);
608 EXPORT_SYMBOL(vmalloc_32_user);
610 long vread(char *buf, char *addr, unsigned long count)
612 struct vm_struct *tmp;
613 char *vaddr, *buf_start = buf;
616 /* Don't allow overflow */
617 if ((unsigned long) addr + count < count)
618 count = -(unsigned long) addr;
620 read_lock(&vmlist_lock);
621 for (tmp = vmlist; tmp; tmp = tmp->next) {
622 vaddr = (char *) tmp->addr;
623 if (addr >= vaddr + tmp->size - PAGE_SIZE)
625 while (addr < vaddr) {
633 n = vaddr + tmp->size - PAGE_SIZE - addr;
644 read_unlock(&vmlist_lock);
645 return buf - buf_start;
648 long vwrite(char *buf, char *addr, unsigned long count)
650 struct vm_struct *tmp;
651 char *vaddr, *buf_start = buf;
654 /* Don't allow overflow */
655 if ((unsigned long) addr + count < count)
656 count = -(unsigned long) addr;
658 read_lock(&vmlist_lock);
659 for (tmp = vmlist; tmp; tmp = tmp->next) {
660 vaddr = (char *) tmp->addr;
661 if (addr >= vaddr + tmp->size - PAGE_SIZE)
663 while (addr < vaddr) {
670 n = vaddr + tmp->size - PAGE_SIZE - addr;
681 read_unlock(&vmlist_lock);
682 return buf - buf_start;
686 * remap_vmalloc_range - map vmalloc pages to userspace
687 * @vma: vma to cover (map full range of vma)
688 * @addr: vmalloc memory
689 * @pgoff: number of pages into addr before first page to map
690 * @returns: 0 for success, -Exxx on failure
692 * This function checks that addr is a valid vmalloc'ed area, and
693 * that it is big enough to cover the vma. Will return failure if
694 * that criteria isn't met.
696 * Similar to remap_pfn_range (see mm/memory.c)
698 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
701 struct vm_struct *area;
702 unsigned long uaddr = vma->vm_start;
703 unsigned long usize = vma->vm_end - vma->vm_start;
706 if ((PAGE_SIZE-1) & (unsigned long)addr)
709 read_lock(&vmlist_lock);
710 area = __find_vm_area(addr);
712 goto out_einval_locked;
714 if (!(area->flags & VM_USERMAP))
715 goto out_einval_locked;
717 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
718 goto out_einval_locked;
719 read_unlock(&vmlist_lock);
721 addr += pgoff << PAGE_SHIFT;
723 struct page *page = vmalloc_to_page(addr);
724 ret = vm_insert_page(vma, uaddr, page);
733 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
734 vma->vm_flags |= VM_RESERVED;
739 read_unlock(&vmlist_lock);
742 EXPORT_SYMBOL(remap_vmalloc_range);