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);
167 EXPORT_SYMBOL_GPL(map_vm_area);
169 static struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
170 unsigned long start, unsigned long end,
171 int node, gfp_t gfp_mask)
173 struct vm_struct **p, *tmp, *area;
174 unsigned long align = 1;
177 BUG_ON(in_interrupt());
178 if (flags & VM_IOREMAP) {
181 if (bit > IOREMAP_MAX_ORDER)
182 bit = IOREMAP_MAX_ORDER;
183 else if (bit < PAGE_SHIFT)
188 addr = ALIGN(start, align);
189 size = PAGE_ALIGN(size);
193 area = kmalloc_node(sizeof(*area), gfp_mask & GFP_RECLAIM_MASK, node);
199 * We always allocate a guard page.
203 write_lock(&vmlist_lock);
204 for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
205 if ((unsigned long)tmp->addr < addr) {
206 if((unsigned long)tmp->addr + tmp->size >= addr)
207 addr = ALIGN(tmp->size +
208 (unsigned long)tmp->addr, align);
211 if ((size + addr) < addr)
213 if (size + addr <= (unsigned long)tmp->addr)
215 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
216 if (addr > end - size)
225 area->addr = (void *)addr;
230 write_unlock(&vmlist_lock);
235 write_unlock(&vmlist_lock);
237 if (printk_ratelimit())
238 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
242 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
243 unsigned long start, unsigned long end)
245 return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
247 EXPORT_SYMBOL_GPL(__get_vm_area);
250 * get_vm_area - reserve a contiguous kernel virtual area
251 * @size: size of the area
252 * @flags: %VM_IOREMAP for I/O mappings or VM_ALLOC
254 * Search an area of @size in the kernel virtual mapping area,
255 * and reserved it for out purposes. Returns the area descriptor
256 * on success or %NULL on failure.
258 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
260 return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
263 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
264 int node, gfp_t gfp_mask)
266 return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
270 /* Caller must hold vmlist_lock */
271 static struct vm_struct *__find_vm_area(void *addr)
273 struct vm_struct *tmp;
275 for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
276 if (tmp->addr == addr)
283 /* Caller must hold vmlist_lock */
284 static struct vm_struct *__remove_vm_area(void *addr)
286 struct vm_struct **p, *tmp;
288 for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
289 if (tmp->addr == addr)
299 * Remove the guard page.
301 tmp->size -= PAGE_SIZE;
306 * remove_vm_area - find and remove a continuous kernel virtual area
307 * @addr: base address
309 * Search for the kernel VM area starting at @addr, and remove it.
310 * This function returns the found VM area, but using it is NOT safe
311 * on SMP machines, except for its size or flags.
313 struct vm_struct *remove_vm_area(void *addr)
316 write_lock(&vmlist_lock);
317 v = __remove_vm_area(addr);
318 write_unlock(&vmlist_lock);
322 static void __vunmap(void *addr, int deallocate_pages)
324 struct vm_struct *area;
329 if ((PAGE_SIZE-1) & (unsigned long)addr) {
330 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
335 area = remove_vm_area(addr);
336 if (unlikely(!area)) {
337 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
343 debug_check_no_locks_freed(addr, area->size);
345 if (deallocate_pages) {
348 for (i = 0; i < area->nr_pages; i++) {
349 BUG_ON(!area->pages[i]);
350 __free_page(area->pages[i]);
353 if (area->flags & VM_VPAGES)
364 * vfree - release memory allocated by vmalloc()
365 * @addr: memory base address
367 * Free the virtually continuous memory area starting at @addr, as
368 * obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
369 * NULL, no operation is performed.
371 * Must not be called in interrupt context.
373 void vfree(void *addr)
375 BUG_ON(in_interrupt());
378 EXPORT_SYMBOL(vfree);
381 * vunmap - release virtual mapping obtained by vmap()
382 * @addr: memory base address
384 * Free the virtually contiguous memory area starting at @addr,
385 * which was created from the page array passed to vmap().
387 * Must not be called in interrupt context.
389 void vunmap(void *addr)
391 BUG_ON(in_interrupt());
394 EXPORT_SYMBOL(vunmap);
397 * vmap - map an array of pages into virtually contiguous space
398 * @pages: array of page pointers
399 * @count: number of pages to map
400 * @flags: vm_area->flags
401 * @prot: page protection for the mapping
403 * Maps @count pages from @pages into contiguous kernel virtual
406 void *vmap(struct page **pages, unsigned int count,
407 unsigned long flags, pgprot_t prot)
409 struct vm_struct *area;
411 if (count > num_physpages)
414 area = get_vm_area((count << PAGE_SHIFT), flags);
417 if (map_vm_area(area, prot, &pages)) {
426 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
427 pgprot_t prot, int node)
430 unsigned int nr_pages, array_size, i;
432 nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
433 array_size = (nr_pages * sizeof(struct page *));
435 area->nr_pages = nr_pages;
436 /* Please note that the recursion is strictly bounded. */
437 if (array_size > PAGE_SIZE) {
438 pages = __vmalloc_node(array_size, gfp_mask | __GFP_ZERO,
440 area->flags |= VM_VPAGES;
442 pages = kmalloc_node(array_size,
443 (gfp_mask & GFP_RECLAIM_MASK) | __GFP_ZERO,
448 remove_vm_area(area->addr);
453 for (i = 0; i < area->nr_pages; i++) {
455 area->pages[i] = alloc_page(gfp_mask);
457 area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
458 if (unlikely(!area->pages[i])) {
459 /* Successfully allocated i pages, free them in __vunmap() */
465 if (map_vm_area(area, prot, &pages))
474 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
476 return __vmalloc_area_node(area, gfp_mask, prot, -1);
480 * __vmalloc_node - allocate virtually contiguous memory
481 * @size: allocation size
482 * @gfp_mask: flags for the page level allocator
483 * @prot: protection mask for the allocated pages
484 * @node: node to use for allocation or -1
486 * Allocate enough pages to cover @size from the page level
487 * allocator with @gfp_mask flags. Map them into contiguous
488 * kernel virtual space, using a pagetable protection of @prot.
490 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
493 struct vm_struct *area;
495 size = PAGE_ALIGN(size);
496 if (!size || (size >> PAGE_SHIFT) > num_physpages)
499 area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
503 return __vmalloc_area_node(area, gfp_mask, prot, node);
506 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
508 return __vmalloc_node(size, gfp_mask, prot, -1);
510 EXPORT_SYMBOL(__vmalloc);
513 * vmalloc - allocate virtually contiguous memory
514 * @size: allocation size
515 * Allocate enough pages to cover @size from the page level
516 * allocator and map them into contiguous kernel virtual space.
518 * For tight control over page level allocator and protection flags
519 * use __vmalloc() instead.
521 void *vmalloc(unsigned long size)
523 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
525 EXPORT_SYMBOL(vmalloc);
528 * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
529 * @size: allocation size
531 * The resulting memory area is zeroed so it can be mapped to userspace
532 * without leaking data.
534 void *vmalloc_user(unsigned long size)
536 struct vm_struct *area;
539 ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
541 write_lock(&vmlist_lock);
542 area = __find_vm_area(ret);
543 area->flags |= VM_USERMAP;
544 write_unlock(&vmlist_lock);
548 EXPORT_SYMBOL(vmalloc_user);
551 * vmalloc_node - allocate memory on a specific node
552 * @size: allocation size
555 * Allocate enough pages to cover @size from the page level
556 * allocator and map them into contiguous kernel virtual space.
558 * For tight control over page level allocator and protection flags
559 * use __vmalloc() instead.
561 void *vmalloc_node(unsigned long size, int node)
563 return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
565 EXPORT_SYMBOL(vmalloc_node);
567 #ifndef PAGE_KERNEL_EXEC
568 # define PAGE_KERNEL_EXEC PAGE_KERNEL
572 * vmalloc_exec - allocate virtually contiguous, executable memory
573 * @size: allocation size
575 * Kernel-internal function to allocate enough pages to cover @size
576 * the page level allocator and map them into contiguous and
577 * executable kernel virtual space.
579 * For tight control over page level allocator and protection flags
580 * use __vmalloc() instead.
583 void *vmalloc_exec(unsigned long size)
585 return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
588 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
589 #define GFP_VMALLOC32 GFP_DMA32 | GFP_KERNEL
590 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
591 #define GFP_VMALLOC32 GFP_DMA | GFP_KERNEL
593 #define GFP_VMALLOC32 GFP_KERNEL
597 * vmalloc_32 - allocate virtually contiguous memory (32bit addressable)
598 * @size: allocation size
600 * Allocate enough 32bit PA addressable pages to cover @size from the
601 * page level allocator and map them into contiguous kernel virtual space.
603 void *vmalloc_32(unsigned long size)
605 return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
607 EXPORT_SYMBOL(vmalloc_32);
610 * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
611 * @size: allocation size
613 * The resulting memory area is 32bit addressable and zeroed so it can be
614 * mapped to userspace without leaking data.
616 void *vmalloc_32_user(unsigned long size)
618 struct vm_struct *area;
621 ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
623 write_lock(&vmlist_lock);
624 area = __find_vm_area(ret);
625 area->flags |= VM_USERMAP;
626 write_unlock(&vmlist_lock);
630 EXPORT_SYMBOL(vmalloc_32_user);
632 long vread(char *buf, char *addr, unsigned long count)
634 struct vm_struct *tmp;
635 char *vaddr, *buf_start = buf;
638 /* Don't allow overflow */
639 if ((unsigned long) addr + count < count)
640 count = -(unsigned long) addr;
642 read_lock(&vmlist_lock);
643 for (tmp = vmlist; tmp; tmp = tmp->next) {
644 vaddr = (char *) tmp->addr;
645 if (addr >= vaddr + tmp->size - PAGE_SIZE)
647 while (addr < vaddr) {
655 n = vaddr + tmp->size - PAGE_SIZE - addr;
666 read_unlock(&vmlist_lock);
667 return buf - buf_start;
670 long vwrite(char *buf, char *addr, unsigned long count)
672 struct vm_struct *tmp;
673 char *vaddr, *buf_start = buf;
676 /* Don't allow overflow */
677 if ((unsigned long) addr + count < count)
678 count = -(unsigned long) addr;
680 read_lock(&vmlist_lock);
681 for (tmp = vmlist; tmp; tmp = tmp->next) {
682 vaddr = (char *) tmp->addr;
683 if (addr >= vaddr + tmp->size - PAGE_SIZE)
685 while (addr < vaddr) {
692 n = vaddr + tmp->size - PAGE_SIZE - addr;
703 read_unlock(&vmlist_lock);
704 return buf - buf_start;
708 * remap_vmalloc_range - map vmalloc pages to userspace
709 * @vma: vma to cover (map full range of vma)
710 * @addr: vmalloc memory
711 * @pgoff: number of pages into addr before first page to map
712 * @returns: 0 for success, -Exxx on failure
714 * This function checks that addr is a valid vmalloc'ed area, and
715 * that it is big enough to cover the vma. Will return failure if
716 * that criteria isn't met.
718 * Similar to remap_pfn_range() (see mm/memory.c)
720 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
723 struct vm_struct *area;
724 unsigned long uaddr = vma->vm_start;
725 unsigned long usize = vma->vm_end - vma->vm_start;
728 if ((PAGE_SIZE-1) & (unsigned long)addr)
731 read_lock(&vmlist_lock);
732 area = __find_vm_area(addr);
734 goto out_einval_locked;
736 if (!(area->flags & VM_USERMAP))
737 goto out_einval_locked;
739 if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
740 goto out_einval_locked;
741 read_unlock(&vmlist_lock);
743 addr += pgoff << PAGE_SHIFT;
745 struct page *page = vmalloc_to_page(addr);
746 ret = vm_insert_page(vma, uaddr, page);
755 /* Prevent "things" like memory migration? VM_flags need a cleanup... */
756 vma->vm_flags |= VM_RESERVED;
761 read_unlock(&vmlist_lock);
764 EXPORT_SYMBOL(remap_vmalloc_range);
767 * Implement a stub for vmalloc_sync_all() if the architecture chose not to
770 void __attribute__((weak)) vmalloc_sync_all(void)
775 static int f(pte_t *pte, struct page *pmd_page, unsigned long addr, void *data)
777 /* apply_to_page_range() does all the hard work. */
782 * alloc_vm_area - allocate a range of kernel address space
783 * @size: size of the area
784 * @returns: NULL on failure, vm_struct on success
786 * This function reserves a range of kernel address space, and
787 * allocates pagetables to map that range. No actual mappings
788 * are created. If the kernel address space is not shared
789 * between processes, it syncs the pagetable across all
792 struct vm_struct *alloc_vm_area(size_t size)
794 struct vm_struct *area;
796 area = get_vm_area(size, VM_IOREMAP);
801 * This ensures that page tables are constructed for this region
802 * of kernel virtual address space and mapped into init_mm.
804 if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
805 area->size, f, NULL)) {
810 /* Make sure the pagetables are constructed in process kernel
816 EXPORT_SYMBOL_GPL(alloc_vm_area);
818 void free_vm_area(struct vm_struct *area)
820 struct vm_struct *ret;
821 ret = remove_vm_area(area->addr);
825 EXPORT_SYMBOL_GPL(free_vm_area);