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 static struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
164 unsigned long start, unsigned long end,
165 int node, gfp_t gfp_mask)
167 struct vm_struct **p, *tmp, *area;
168 unsigned long align = 1;
171 BUG_ON(in_interrupt());
172 if (flags & VM_IOREMAP) {
175 if (bit > IOREMAP_MAX_ORDER)
176 bit = IOREMAP_MAX_ORDER;
177 else if (bit < PAGE_SHIFT)
182 addr = ALIGN(start, align);
183 size = PAGE_ALIGN(size);
185 area = kmalloc_node(sizeof(*area), gfp_mask, node);
189 if (unlikely(!size)) {
195 * We always allocate a guard page.
199 write_lock(&vmlist_lock);
200 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
201 if ((unsigned long)tmp->addr < addr) {
202 if((unsigned long)tmp->addr + tmp->size >= addr)
203 addr = ALIGN(tmp->size +
204 (unsigned long)tmp->addr, align);
207 if ((size + addr) < addr)
209 if (size + addr <= (unsigned long)tmp->addr)
211 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
212 if (addr > end - size)
221 area->addr = (void *)addr;
226 write_unlock(&vmlist_lock);
231 write_unlock(&vmlist_lock);
233 if (printk_ratelimit())
234 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
238 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
239 unsigned long start, unsigned long end)
241 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
245 * get_vm_area - reserve a contingous kernel virtual area
246 * @size: size of the area
247 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
249 * Search an area of @size in the kernel virtual mapping area,
250 * and reserved it for out purposes. Returns the area descriptor
251 * on success or %NULL on failure.
253 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
255 return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
258 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
259 int node, gfp_t gfp_mask)
261 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
265 /* Caller must hold vmlist_lock */
266 static struct vm_struct *__find_vm_area(void *addr)
268 struct vm_struct *tmp;
270 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
271 if (tmp->addr == addr)
278 /* Caller must hold vmlist_lock */
279 static struct vm_struct *__remove_vm_area(void *addr)
281 struct vm_struct **p, *tmp;
283 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
284 if (tmp->addr == addr)
294 * Remove the guard page.
296 tmp->size -= PAGE_SIZE;
301 * remove_vm_area - find and remove a contingous kernel virtual area
302 * @addr: base address
304 * Search for the kernel VM area starting at @addr, and remove it.
305 * This function returns the found VM area, but using it is NOT safe
306 * on SMP machines, except for its size or flags.
308 struct vm_struct *remove_vm_area(void *addr)
311 write_lock(&vmlist_lock);
312 v = __remove_vm_area(addr);
313 write_unlock(&vmlist_lock);
317 void __vunmap(void *addr, int deallocate_pages)
319 struct vm_struct *area;
324 if ((PAGE_SIZE-1) & (unsigned long)addr) {
325 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
330 area = remove_vm_area(addr);
331 if (unlikely(!area)) {
332 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
338 debug_check_no_locks_freed(addr, area->size);
340 if (deallocate_pages) {
343 for (i = 0; i < area->nr_pages; i++) {
344 BUG_ON(!area->pages[i]);
345 __free_page(area->pages[i]);
348 if (area->flags & VM_VPAGES)
359 * vfree - release memory allocated by vmalloc()
360 * @addr: memory base address
362 * Free the virtually contiguous memory area starting at @addr, as
363 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
364 * NULL, no operation is performed.
366 * Must not be called in interrupt context.
368 void vfree(void *addr)
370 BUG_ON(in_interrupt());
373 EXPORT_SYMBOL(vfree);
376 * vunmap - release virtual mapping obtained by vmap()
377 * @addr: memory base address
379 * Free the virtually contiguous memory area starting at @addr,
380 * which was created from the page array passed to vmap().
382 * Must not be called in interrupt context.
384 void vunmap(void *addr)
386 BUG_ON(in_interrupt());
389 EXPORT_SYMBOL(vunmap);
392 * vmap - map an array of pages into virtually contiguous space
393 * @pages: array of page pointers
394 * @count: number of pages to map
395 * @flags: vm_area->flags
396 * @prot: page protection for the mapping
398 * Maps @count pages from @pages into contiguous kernel virtual
401 void *vmap(struct page **pages, unsigned int count,
402 unsigned long flags, pgprot_t prot)
404 struct vm_struct *area;
406 if (count > num_physpages)
409 area = get_vm_area((count << PAGE_SHIFT), flags);
412 if (map_vm_area(area, prot, &pages)) {
421 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
422 pgprot_t prot, int node)
425 unsigned int nr_pages, array_size, i;
427 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
428 array_size = (nr_pages * sizeof(struct page *));
430 area->nr_pages = nr_pages;
431 /* Please note that the recursion is strictly bounded. */
432 if (array_size > PAGE_SIZE) {
433 pages = __vmalloc_node(array_size, gfp_mask, PAGE_KERNEL, node);
434 area->flags |= VM_VPAGES;
436 pages = kmalloc_node(array_size,
437 (gfp_mask & ~(__GFP_HIGHMEM | __GFP_ZERO)),
442 remove_vm_area(area->addr);
446 memset(area->pages, 0, array_size);
448 for (i = 0; i < area->nr_pages; i++) {
450 area->pages[i] = alloc_page(gfp_mask);
452 area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
453 if (unlikely(!area->pages[i])) {
454 /* Successfully allocated i pages, free them in __vunmap() */
460 if (map_vm_area(area, prot, &pages))
469 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
471 return __vmalloc_area_node(area, gfp_mask, prot, -1);
475 * __vmalloc_node - allocate virtually contiguous memory
476 * @size: allocation size
477 * @gfp_mask: flags for the page level allocator
478 * @prot: protection mask for the allocated pages
479 * @node: node to use for allocation or -1
481 * Allocate enough pages to cover @size from the page level
482 * allocator with @gfp_mask flags. Map them into contiguous
483 * kernel virtual space, using a pagetable protection of @prot.
485 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
488 struct vm_struct *area;
490 size = PAGE_ALIGN(size);
491 if (!size || (size >> PAGE_SHIFT) > num_physpages)
494 area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
498 return __vmalloc_area_node(area, gfp_mask, prot, node);
501 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
503 return __vmalloc_node(size, gfp_mask, prot, -1);
505 EXPORT_SYMBOL(__vmalloc);
508 * vmalloc - allocate virtually contiguous memory
509 * @size: allocation size
510 * Allocate enough pages to cover @size from the page level
511 * allocator and map them into contiguous kernel virtual space.
513 * For tight control over page level allocator and protection flags
514 * use __vmalloc() instead.
516 void *vmalloc(unsigned long size)
518 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
520 EXPORT_SYMBOL(vmalloc);
523 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
524 * @size: allocation size
526 * The resulting memory area is zeroed so it can be mapped to userspace
527 * without leaking data.
529 void *vmalloc_user(unsigned long size)
531 struct vm_struct *area;
534 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
535 write_lock(&vmlist_lock);
536 area = __find_vm_area(ret);
537 area->flags |= VM_USERMAP;
538 write_unlock(&vmlist_lock);
542 EXPORT_SYMBOL(vmalloc_user);
545 * vmalloc_node - allocate memory on a specific node
546 * @size: allocation size
549 * Allocate enough pages to cover @size from the page level
550 * allocator and map them into contiguous kernel virtual space.
552 * For tight control over page level allocator and protection flags
553 * use __vmalloc() instead.
555 void *vmalloc_node(unsigned long size, int node)
557 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
559 EXPORT_SYMBOL(vmalloc_node);
561 #ifndef PAGE_KERNEL_EXEC
562 # define PAGE_KERNEL_EXEC PAGE_KERNEL
566 * vmalloc_exec - allocate virtually contiguous, executable memory
567 * @size: allocation size
569 * Kernel-internal function to allocate enough pages to cover @size
570 * the page level allocator and map them into contiguous and
571 * executable kernel virtual space.
573 * For tight control over page level allocator and protection flags
574 * use __vmalloc() instead.
577 void *vmalloc_exec(unsigned long size)
579 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
583 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
584 * @size: allocation size
586 * Allocate enough 32bit PA addressable pages to cover @size from the
587 * page level allocator and map them into contiguous kernel virtual space.
589 void *vmalloc_32(unsigned long size)
591 return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
593 EXPORT_SYMBOL(vmalloc_32);
596 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
597 * @size: allocation size
599 * The resulting memory area is 32bit addressable and zeroed so it can be
600 * mapped to userspace without leaking data.
602 void *vmalloc_32_user(unsigned long size)
604 struct vm_struct *area;
607 ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
608 write_lock(&vmlist_lock);
609 area = __find_vm_area(ret);
610 area->flags |= VM_USERMAP;
611 write_unlock(&vmlist_lock);
615 EXPORT_SYMBOL(vmalloc_32_user);
617 long vread(char *buf, char *addr, unsigned long count)
619 struct vm_struct *tmp;
620 char *vaddr, *buf_start = buf;
623 /* Don't allow overflow */
624 if ((unsigned long) addr + count < count)
625 count = -(unsigned long) addr;
627 read_lock(&vmlist_lock);
628 for (tmp = vmlist; tmp; tmp = tmp->next) {
629 vaddr = (char *) tmp->addr;
630 if (addr >= vaddr + tmp->size - PAGE_SIZE)
632 while (addr < vaddr) {
640 n = vaddr + tmp->size - PAGE_SIZE - addr;
651 read_unlock(&vmlist_lock);
652 return buf - buf_start;
655 long vwrite(char *buf, char *addr, unsigned long count)
657 struct vm_struct *tmp;
658 char *vaddr, *buf_start = buf;
661 /* Don't allow overflow */
662 if ((unsigned long) addr + count < count)
663 count = -(unsigned long) addr;
665 read_lock(&vmlist_lock);
666 for (tmp = vmlist; tmp; tmp = tmp->next) {
667 vaddr = (char *) tmp->addr;
668 if (addr >= vaddr + tmp->size - PAGE_SIZE)
670 while (addr < vaddr) {
677 n = vaddr + tmp->size - PAGE_SIZE - addr;
688 read_unlock(&vmlist_lock);
689 return buf - buf_start;
693 * remap_vmalloc_range - map vmalloc pages to userspace
694 * @vma: vma to cover (map full range of vma)
695 * @addr: vmalloc memory
696 * @pgoff: number of pages into addr before first page to map
697 * @returns: 0 for success, -Exxx on failure
699 * This function checks that addr is a valid vmalloc'ed area, and
700 * that it is big enough to cover the vma. Will return failure if
701 * that criteria isn't met.
703 * Similar to remap_pfn_range (see mm/memory.c)
705 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
708 struct vm_struct *area;
709 unsigned long uaddr = vma->vm_start;
710 unsigned long usize = vma->vm_end - vma->vm_start;
713 if ((PAGE_SIZE-1) & (unsigned long)addr)
716 read_lock(&vmlist_lock);
717 area = __find_vm_area(addr);
719 goto out_einval_locked;
721 if (!(area->flags & VM_USERMAP))
722 goto out_einval_locked;
724 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
725 goto out_einval_locked;
726 read_unlock(&vmlist_lock);
728 addr += pgoff << PAGE_SHIFT;
730 struct page *page = vmalloc_to_page(addr);
731 ret = vm_insert_page(vma, uaddr, page);
740 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
741 vma->vm_flags |= VM_RESERVED;
746 read_unlock(&vmlist_lock);
749 EXPORT_SYMBOL(remap_vmalloc_range);