mm/vmalloc.c

   1 /*
   2  *  linux/mm/vmalloc.c
   3  *
   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
   9  */
  10
  11 #include <linux/mm.h>
  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
  18 #include <linux/vmalloc.h>
  19
  20 #include <asm/uaccess.h>
  21 #include <asm/tlbflush.h>
  22
  23
  24 DEFINE_RWLOCK(vmlist_lock);
  25 struct vm_struct *vmlist;
  26
  27 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
  28                             int node);
  29
  30 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
  31 {
  32         pte_t *pte;
  33
  34         pte = pte_offset_kernel(pmd, addr);
  35         do {
  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);
  39 }
  40
  41 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
  42                                                 unsigned long end)
  43 {
  44         pmd_t *pmd;
  45         unsigned long next;
  46
  47         pmd = pmd_offset(pud, addr);
  48         do {
  49                 next = pmd_addr_end(addr, end);
  50                 if (pmd_none_or_clear_bad(pmd))
  51                         continue;
  52                 vunmap_pte_range(pmd, addr, next);
  53         } while (pmd++, addr = next, addr != end);
  54 }
  55
  56 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
  57                                                 unsigned long end)
  58 {
  59         pud_t *pud;
  60         unsigned long next;
  61
  62         pud = pud_offset(pgd, addr);
  63         do {
  64                 next = pud_addr_end(addr, end);
  65                 if (pud_none_or_clear_bad(pud))
  66                         continue;
  67                 vunmap_pmd_range(pud, addr, next);
  68         } while (pud++, addr = next, addr != end);
  69 }
  70
  71 void unmap_vm_area(struct vm_struct *area)
  72 {
  73         pgd_t *pgd;
  74         unsigned long next;
  75         unsigned long addr = (unsigned long) area->addr;
  76         unsigned long end = addr + area->size;
  77
  78         BUG_ON(addr >= end);
  79         pgd = pgd_offset_k(addr);
  80         flush_cache_vunmap(addr, end);
  81         do {
  82                 next = pgd_addr_end(addr, end);
  83                 if (pgd_none_or_clear_bad(pgd))
  84                         continue;
  85                 vunmap_pud_range(pgd, addr, next);
  86         } while (pgd++, addr = next, addr != end);
  87         flush_tlb_kernel_range((unsigned long) area->addr, end);
  88 }
  89
  90 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  91                         unsigned long end, pgprot_t prot, struct page ***pages)
  92 {
  93         pte_t *pte;
  94
  95         pte = pte_alloc_kernel(pmd, addr);
  96         if (!pte)
  97                 return -ENOMEM;
  98         do {
  99                 struct page *page = **pages;
 100                 WARN_ON(!pte_none(*pte));
 101                 if (!page)
 102                         return -ENOMEM;
 103                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
 104                 (*pages)++;
 105         } while (pte++, addr += PAGE_SIZE, addr != end);
 106         return 0;
 107 }
 108
 109 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
 110                         unsigned long end, pgprot_t prot, struct page ***pages)
 111 {
 112         pmd_t *pmd;
 113         unsigned long next;
 114
 115         pmd = pmd_alloc(&init_mm, pud, addr);
 116         if (!pmd)
 117                 return -ENOMEM;
 118         do {
 119                 next = pmd_addr_end(addr, end);
 120                 if (vmap_pte_range(pmd, addr, next, prot, pages))
 121                         return -ENOMEM;
 122         } while (pmd++, addr = next, addr != end);
 123         return 0;
 124 }
 125
 126 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
 127                         unsigned long end, pgprot_t prot, struct page ***pages)
 128 {
 129         pud_t *pud;
 130         unsigned long next;
 131
 132         pud = pud_alloc(&init_mm, pgd, addr);
 133         if (!pud)
 134                 return -ENOMEM;
 135         do {
 136                 next = pud_addr_end(addr, end);
 137                 if (vmap_pmd_range(pud, addr, next, prot, pages))
 138                         return -ENOMEM;
 139         } while (pud++, addr = next, addr != end);
 140         return 0;
 141 }
 142
 143 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 144 {
 145         pgd_t *pgd;
 146         unsigned long next;
 147         unsigned long addr = (unsigned long) area->addr;
 148         unsigned long end = addr + area->size - PAGE_SIZE;
 149         int err;
 150
 151         BUG_ON(addr >= end);
 152         pgd = pgd_offset_k(addr);
 153         do {
 154                 next = pgd_addr_end(addr, end);
 155                 err = vmap_pud_range(pgd, addr, next, prot, pages);
 156                 if (err)
 157                         break;
 158         } while (pgd++, addr = next, addr != end);
 159         flush_cache_vmap((unsigned long) area->addr, end);
 160         return err;
 161 }
 162
 163 struct vm_struct *__get_vm_area_node(unsigned long size, unsigned long flags,
 164                                 unsigned long start, unsigned long end, int node)
 165 {
 166         struct vm_struct **p, *tmp, *area;
 167         unsigned long align = 1;
 168         unsigned long addr;
 169
 170         if (flags & VM_IOREMAP) {
 171                 int bit = fls(size);
 172
 173                 if (bit > IOREMAP_MAX_ORDER)
 174                         bit = IOREMAP_MAX_ORDER;
 175                 else if (bit < PAGE_SHIFT)
 176                         bit = PAGE_SHIFT;
 177
 178                 align = 1ul << bit;
 179         }
 180         addr = ALIGN(start, align);
 181         size = PAGE_ALIGN(size);
 182
 183         area = kmalloc_node(sizeof(*area), GFP_KERNEL, node);
 184         if (unlikely(!area))
 185                 return NULL;
 186
 187         if (unlikely(!size)) {
 188                 kfree (area);
 189                 return NULL;
 190         }
 191
 192         /*
 193          * We always allocate a guard page.
 194          */
 195         size += PAGE_SIZE;
 196
 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);
 203                         continue;
 204                 }
 205                 if ((size + addr) < addr)
 206                         goto out;
 207                 if (size + addr <= (unsigned long)tmp->addr)
 208                         goto found;
 209                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 210                 if (addr > end - size)
 211                         goto out;
 212         }
 213
 214 found:
 215         area->next = *p;
 216         *p = area;
 217
 218         area->flags = flags;
 219         area->addr = (void *)addr;
 220         area->size = size;
 221         area->pages = NULL;
 222         area->nr_pages = 0;
 223         area->phys_addr = 0;
 224         write_unlock(&vmlist_lock);
 225
 226         return area;
 227
 228 out:
 229         write_unlock(&vmlist_lock);
 230         kfree(area);
 231         if (printk_ratelimit())
 232                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 233         return NULL;
 234 }
 235
 236 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 237                                 unsigned long start, unsigned long end)
 238 {
 239         return __get_vm_area_node(size, flags, start, end, -1);
 240 }
 241
 242 /**
 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
 246  *
 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.
 250  */
 251 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 252 {
 253         return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
 254 }
 255
 256 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, int node)
 257 {
 258         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node);
 259 }
 260
 261 /* Caller must hold vmlist_lock */
 262 static struct vm_struct *__find_vm_area(void *addr)
 263 {
 264         struct vm_struct *tmp;
 265
 266         for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
 267                  if (tmp->addr == addr)
 268                         break;
 269         }
 270
 271         return tmp;
 272 }
 273
 274 /* Caller must hold vmlist_lock */
 275 static struct vm_struct *__remove_vm_area(void *addr)
 276 {
 277         struct vm_struct **p, *tmp;
 278
 279         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 280                  if (tmp->addr == addr)
 281                          goto found;
 282         }
 283         return NULL;
 284
 285 found:
 286         unmap_vm_area(tmp);
 287         *p = tmp->next;
 288
 289         /*
 290          * Remove the guard page.
 291          */
 292         tmp->size -= PAGE_SIZE;
 293         return tmp;
 294 }
 295
 296 /**
 297  *      remove_vm_area  -  find and remove a contingous kernel virtual area
 298  *      @addr:          base address
 299  *
 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.
 303  */
 304 struct vm_struct *remove_vm_area(void *addr)
 305 {
 306         struct vm_struct *v;
 307         write_lock(&vmlist_lock);
 308         v = __remove_vm_area(addr);
 309         write_unlock(&vmlist_lock);
 310         return v;
 311 }
 312
 313 void __vunmap(void *addr, int deallocate_pages)
 314 {
 315         struct vm_struct *area;
 316
 317         if (!addr)
 318                 return;
 319
 320         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 321                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 322                 WARN_ON(1);
 323                 return;
 324         }
 325
 326         area = remove_vm_area(addr);
 327         if (unlikely(!area)) {
 328                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 329                                 addr);
 330                 WARN_ON(1);
 331                 return;
 332         }
 333
 334         debug_check_no_locks_freed(addr, area->size);
 335
 336         if (deallocate_pages) {
 337                 int i;
 338
 339                 for (i = 0; i < area->nr_pages; i++) {
 340                         BUG_ON(!area->pages[i]);
 341                         __free_page(area->pages[i]);
 342                 }
 343
 344                 if (area->flags & VM_VPAGES)
 345                         vfree(area->pages);
 346                 else
 347                         kfree(area->pages);
 348         }
 349
 350         kfree(area);
 351         return;
 352 }
 353
 354 /**
 355  *      vfree  -  release memory allocated by vmalloc()
 356  *      @addr:          memory base address
 357  *
 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.
 361  *
 362  *      Must not be called in interrupt context.
 363  */
 364 void vfree(void *addr)
 365 {
 366         BUG_ON(in_interrupt());
 367         __vunmap(addr, 1);
 368 }
 369 EXPORT_SYMBOL(vfree);
 370
 371 /**
 372  *      vunmap  -  release virtual mapping obtained by vmap()
 373  *      @addr:          memory base address
 374  *
 375  *      Free the virtually contiguous memory area starting at @addr,
 376  *      which was created from the page array passed to vmap().
 377  *
 378  *      Must not be called in interrupt context.
 379  */
 380 void vunmap(void *addr)
 381 {
 382         BUG_ON(in_interrupt());
 383         __vunmap(addr, 0);
 384 }
 385 EXPORT_SYMBOL(vunmap);
 386
 387 /**
 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
 393  *
 394  *      Maps @count pages from @pages into contiguous kernel virtual
 395  *      space.
 396  */
 397 void *vmap(struct page **pages, unsigned int count,
 398                 unsigned long flags, pgprot_t prot)
 399 {
 400         struct vm_struct *area;
 401
 402         if (count > num_physpages)
 403                 return NULL;
 404
 405         area = get_vm_area((count << PAGE_SHIFT), flags);
 406         if (!area)
 407                 return NULL;
 408         if (map_vm_area(area, prot, &pages)) {
 409                 vunmap(area->addr);
 410                 return NULL;
 411         }
 412
 413         return area->addr;
 414 }
 415 EXPORT_SYMBOL(vmap);
 416
 417 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
 418                                 pgprot_t prot, int node)
 419 {
 420         struct page **pages;
 421         unsigned int nr_pages, array_size, i;
 422
 423         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 424         array_size = (nr_pages * sizeof(struct page *));
 425
 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;
 431         } else {
 432                 pages = kmalloc_node(array_size,
 433                                 (gfp_mask & ~(__GFP_HIGHMEM | __GFP_ZERO)),
 434                                 node);
 435         }
 436         area->pages = pages;
 437         if (!area->pages) {
 438                 remove_vm_area(area->addr);
 439                 kfree(area);
 440                 return NULL;
 441         }
 442         memset(area->pages, 0, array_size);
 443
 444         for (i = 0; i < area->nr_pages; i++) {
 445                 if (node < 0)
 446                         area->pages[i] = alloc_page(gfp_mask);
 447                 else
 448                         area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
 449                 if (unlikely(!area->pages[i])) {
 450                         /* Successfully allocated i pages, free them in __vunmap() */
 451                         area->nr_pages = i;
 452                         goto fail;
 453                 }
 454         }
 455
 456         if (map_vm_area(area, prot, &pages))
 457                 goto fail;
 458         return area->addr;
 459
 460 fail:
 461         vfree(area->addr);
 462         return NULL;
 463 }
 464
 465 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
 466 {
 467         return __vmalloc_area_node(area, gfp_mask, prot, -1);
 468 }
 469
 470 /**
 471  *      __vmalloc_node  -  allocate virtually contiguous memory
 472  *      @size:          allocation size
 473  *      @gfp_mask:      flags for the page level allocator
 474  *      @prot:          protection mask for the allocated pages
 475  *      @node:          node to use for allocation or -1
 476  *
 477  *      Allocate enough pages to cover @size from the page level
 478  *      allocator with @gfp_mask flags.  Map them into contiguous
 479  *      kernel virtual space, using a pagetable protection of @prot.
 480  */
 481 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
 482                             int node)
 483 {
 484         struct vm_struct *area;
 485
 486         size = PAGE_ALIGN(size);
 487         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 488                 return NULL;
 489
 490         area = get_vm_area_node(size, VM_ALLOC, node);
 491         if (!area)
 492                 return NULL;
 493
 494         return __vmalloc_area_node(area, gfp_mask, prot, node);
 495 }
 496
 497 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 498 {
 499         return __vmalloc_node(size, gfp_mask, prot, -1);
 500 }
 501 EXPORT_SYMBOL(__vmalloc);
 502
 503 /**
 504  *      vmalloc  -  allocate virtually contiguous memory
 505  *      @size:          allocation size
 506  *      Allocate enough pages to cover @size from the page level
 507  *      allocator and map them into contiguous kernel virtual space.
 508  *
 509  *      For tight control over page level allocator and protection flags
 510  *      use __vmalloc() instead.
 511  */
 512 void *vmalloc(unsigned long size)
 513 {
 514         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 515 }
 516 EXPORT_SYMBOL(vmalloc);
 517
 518 /**
 519  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
 520  * @size: allocation size
 521  *
 522  * The resulting memory area is zeroed so it can be mapped to userspace
 523  * without leaking data.
 524  */
 525 void *vmalloc_user(unsigned long size)
 526 {
 527         struct vm_struct *area;
 528         void *ret;
 529
 530         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
 531         write_lock(&vmlist_lock);
 532         area = __find_vm_area(ret);
 533         area->flags |= VM_USERMAP;
 534         write_unlock(&vmlist_lock);
 535
 536         return ret;
 537 }
 538 EXPORT_SYMBOL(vmalloc_user);
 539
 540 /**
 541  *      vmalloc_node  -  allocate memory on a specific node
 542  *      @size:          allocation size
 543  *      @node:          numa node
 544  *
 545  *      Allocate enough pages to cover @size from the page level
 546  *      allocator and map them into contiguous kernel virtual space.
 547  *
 548  *      For tight control over page level allocator and protection flags
 549  *      use __vmalloc() instead.
 550  */
 551 void *vmalloc_node(unsigned long size, int node)
 552 {
 553         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
 554 }
 555 EXPORT_SYMBOL(vmalloc_node);
 556
 557 #ifndef PAGE_KERNEL_EXEC
 558 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 559 #endif
 560
 561 /**
 562  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 563  *      @size:          allocation size
 564  *
 565  *      Kernel-internal function to allocate enough pages to cover @size
 566  *      the page level allocator and map them into contiguous and
 567  *      executable kernel virtual space.
 568  *
 569  *      For tight control over page level allocator and protection flags
 570  *      use __vmalloc() instead.
 571  */
 572
 573 void *vmalloc_exec(unsigned long size)
 574 {
 575         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 576 }
 577
 578 /**
 579  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 580  *      @size:          allocation size
 581  *
 582  *      Allocate enough 32bit PA addressable pages to cover @size from the
 583  *      page level allocator and map them into contiguous kernel virtual space.
 584  */
 585 void *vmalloc_32(unsigned long size)
 586 {
 587         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 588 }
 589 EXPORT_SYMBOL(vmalloc_32);
 590
 591 /**
 592  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 593  *      @size:          allocation size
 594  *
 595  * The resulting memory area is 32bit addressable and zeroed so it can be
 596  * mapped to userspace without leaking data.
 597  */
 598 void *vmalloc_32_user(unsigned long size)
 599 {
 600         struct vm_struct *area;
 601         void *ret;
 602
 603         ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
 604         write_lock(&vmlist_lock);
 605         area = __find_vm_area(ret);
 606         area->flags |= VM_USERMAP;
 607         write_unlock(&vmlist_lock);
 608
 609         return ret;
 610 }
 611 EXPORT_SYMBOL(vmalloc_32_user);
 612
 613 long vread(char *buf, char *addr, unsigned long count)
 614 {
 615         struct vm_struct *tmp;
 616         char *vaddr, *buf_start = buf;
 617         unsigned long n;
 618
 619         /* Don't allow overflow */
 620         if ((unsigned long) addr + count < count)
 621                 count = -(unsigned long) addr;
 622
 623         read_lock(&vmlist_lock);
 624         for (tmp = vmlist; tmp; tmp = tmp->next) {
 625                 vaddr = (char *) tmp->addr;
 626                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 627                         continue;
 628                 while (addr < vaddr) {
 629                         if (count == 0)
 630                                 goto finished;
 631                         *buf = '\0';
 632                         buf++;
 633                         addr++;
 634                         count--;
 635                 }
 636                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 637                 do {
 638                         if (count == 0)
 639                                 goto finished;
 640                         *buf = *addr;
 641                         buf++;
 642                         addr++;
 643                         count--;
 644                 } while (--n > 0);
 645         }
 646 finished:
 647         read_unlock(&vmlist_lock);
 648         return buf - buf_start;
 649 }
 650
 651 long vwrite(char *buf, char *addr, unsigned long count)
 652 {
 653         struct vm_struct *tmp;
 654         char *vaddr, *buf_start = buf;
 655         unsigned long n;
 656
 657         /* Don't allow overflow */
 658         if ((unsigned long) addr + count < count)
 659                 count = -(unsigned long) addr;
 660
 661         read_lock(&vmlist_lock);
 662         for (tmp = vmlist; tmp; tmp = tmp->next) {
 663                 vaddr = (char *) tmp->addr;
 664                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 665                         continue;
 666                 while (addr < vaddr) {
 667                         if (count == 0)
 668                                 goto finished;
 669                         buf++;
 670                         addr++;
 671                         count--;
 672                 }
 673                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 674                 do {
 675                         if (count == 0)
 676                                 goto finished;
 677                         *addr = *buf;
 678                         buf++;
 679                         addr++;
 680                         count--;
 681                 } while (--n > 0);
 682         }
 683 finished:
 684         read_unlock(&vmlist_lock);
 685         return buf - buf_start;
 686 }
 687
 688 /**
 689  *      remap_vmalloc_range  -  map vmalloc pages to userspace
 690  *      @vma:           vma to cover (map full range of vma)
 691  *      @addr:          vmalloc memory
 692  *      @pgoff:         number of pages into addr before first page to map
 693  *      @returns:       0 for success, -Exxx on failure
 694  *
 695  *      This function checks that addr is a valid vmalloc'ed area, and
 696  *      that it is big enough to cover the vma. Will return failure if
 697  *      that criteria isn't met.
 698  *
 699  *      Similar to remap_pfn_range (see mm/memory.c)
 700  */
 701 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
 702                                                 unsigned long pgoff)
 703 {
 704         struct vm_struct *area;
 705         unsigned long uaddr = vma->vm_start;
 706         unsigned long usize = vma->vm_end - vma->vm_start;
 707         int ret;
 708
 709         if ((PAGE_SIZE-1) & (unsigned long)addr)
 710                 return -EINVAL;
 711
 712         read_lock(&vmlist_lock);
 713         area = __find_vm_area(addr);
 714         if (!area)
 715                 goto out_einval_locked;
 716
 717         if (!(area->flags & VM_USERMAP))
 718                 goto out_einval_locked;
 719
 720         if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
 721                 goto out_einval_locked;
 722         read_unlock(&vmlist_lock);
 723
 724         addr += pgoff << PAGE_SHIFT;
 725         do {
 726                 struct page *page = vmalloc_to_page(addr);
 727                 ret = vm_insert_page(vma, uaddr, page);
 728                 if (ret)
 729                         return ret;
 730
 731                 uaddr += PAGE_SIZE;
 732                 addr += PAGE_SIZE;
 733                 usize -= PAGE_SIZE;
 734         } while (usize > 0);
 735
 736         /* Prevent "things" like memory migration? VM_flags need a cleanup... */
 737         vma->vm_flags |= VM_RESERVED;
 738
 739         return ret;
 740
 741 out_einval_locked:
 742         read_unlock(&vmlist_lock);
 743         return -EINVAL;
 744 }
 745 EXPORT_SYMBOL(remap_vmalloc_range);
 746