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_kernel_range(unsigned long addr, unsigned long size)
  72 {
  73         pgd_t *pgd;
  74         unsigned long next;
  75         unsigned long start = addr;
  76         unsigned long end = addr + 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(start, end);
  88 }
  89
  90 static void unmap_vm_area(struct vm_struct *area)
  91 {
  92         unmap_kernel_range((unsigned long)area->addr, area->size);
  93 }
  94
  95 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  96                         unsigned long end, pgprot_t prot, struct page ***pages)
  97 {
  98         pte_t *pte;
  99
 100         pte = pte_alloc_kernel(pmd, addr);
 101         if (!pte)
 102                 return -ENOMEM;
 103         do {
 104                 struct page *page = **pages;
 105                 WARN_ON(!pte_none(*pte));
 106                 if (!page)
 107                         return -ENOMEM;
 108                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
 109                 (*pages)++;
 110         } while (pte++, addr += PAGE_SIZE, addr != end);
 111         return 0;
 112 }
 113
 114 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
 115                         unsigned long end, pgprot_t prot, struct page ***pages)
 116 {
 117         pmd_t *pmd;
 118         unsigned long next;
 119
 120         pmd = pmd_alloc(&init_mm, pud, addr);
 121         if (!pmd)
 122                 return -ENOMEM;
 123         do {
 124                 next = pmd_addr_end(addr, end);
 125                 if (vmap_pte_range(pmd, addr, next, prot, pages))
 126                         return -ENOMEM;
 127         } while (pmd++, addr = next, addr != end);
 128         return 0;
 129 }
 130
 131 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
 132                         unsigned long end, pgprot_t prot, struct page ***pages)
 133 {
 134         pud_t *pud;
 135         unsigned long next;
 136
 137         pud = pud_alloc(&init_mm, pgd, addr);
 138         if (!pud)
 139                 return -ENOMEM;
 140         do {
 141                 next = pud_addr_end(addr, end);
 142                 if (vmap_pmd_range(pud, addr, next, prot, pages))
 143                         return -ENOMEM;
 144         } while (pud++, addr = next, addr != end);
 145         return 0;
 146 }
 147
 148 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 149 {
 150         pgd_t *pgd;
 151         unsigned long next;
 152         unsigned long addr = (unsigned long) area->addr;
 153         unsigned long end = addr + area->size - PAGE_SIZE;
 154         int err;
 155
 156         BUG_ON(addr >= end);
 157         pgd = pgd_offset_k(addr);
 158         do {
 159                 next = pgd_addr_end(addr, end);
 160                 err = vmap_pud_range(pgd, addr, next, prot, pages);
 161                 if (err)
 162                         break;
 163         } while (pgd++, addr = next, addr != end);
 164         flush_cache_vmap((unsigned long) area->addr, end);
 165         return err;
 166 }
 167
 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)
 171 {
 172         struct vm_struct **p, *tmp, *area;
 173         unsigned long align = 1;
 174         unsigned long addr;
 175
 176         BUG_ON(in_interrupt());
 177         if (flags & VM_IOREMAP) {
 178                 int bit = fls(size);
 179
 180                 if (bit > IOREMAP_MAX_ORDER)
 181                         bit = IOREMAP_MAX_ORDER;
 182                 else if (bit < PAGE_SHIFT)
 183                         bit = PAGE_SHIFT;
 184
 185                 align = 1ul << bit;
 186         }
 187         addr = ALIGN(start, align);
 188         size = PAGE_ALIGN(size);
 189         if (unlikely(!size))
 190                 return NULL;
 191
 192         area = kmalloc_node(sizeof(*area), gfp_mask & GFP_LEVEL_MASK, node);
 193         if (unlikely(!area))
 194                 return NULL;
 195
 196         /*
 197          * We always allocate a guard page.
 198          */
 199         size += PAGE_SIZE;
 200
 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);
 207                         continue;
 208                 }
 209                 if ((size + addr) < addr)
 210                         goto out;
 211                 if (size + addr <= (unsigned long)tmp->addr)
 212                         goto found;
 213                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 214                 if (addr > end - size)
 215                         goto out;
 216         }
 217
 218 found:
 219         area->next = *p;
 220         *p = area;
 221
 222         area->flags = flags;
 223         area->addr = (void *)addr;
 224         area->size = size;
 225         area->pages = NULL;
 226         area->nr_pages = 0;
 227         area->phys_addr = 0;
 228         write_unlock(&vmlist_lock);
 229
 230         return area;
 231
 232 out:
 233         write_unlock(&vmlist_lock);
 234         kfree(area);
 235         if (printk_ratelimit())
 236                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 237         return NULL;
 238 }
 239
 240 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 241                                 unsigned long start, unsigned long end)
 242 {
 243         return __get_vm_area_node(size, flags, start, end, -1, GFP_KERNEL);
 244 }
 245
 246 /**
 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
 250  *
 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.
 254  */
 255 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 256 {
 257         return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
 258 }
 259
 260 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags,
 261                                    int node, gfp_t gfp_mask)
 262 {
 263         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node,
 264                                   gfp_mask);
 265 }
 266
 267 /* Caller must hold vmlist_lock */
 268 static struct vm_struct *__find_vm_area(void *addr)
 269 {
 270         struct vm_struct *tmp;
 271
 272         for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
 273                  if (tmp->addr == addr)
 274                         break;
 275         }
 276
 277         return tmp;
 278 }
 279
 280 /* Caller must hold vmlist_lock */
 281 static struct vm_struct *__remove_vm_area(void *addr)
 282 {
 283         struct vm_struct **p, *tmp;
 284
 285         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 286                  if (tmp->addr == addr)
 287                          goto found;
 288         }
 289         return NULL;
 290
 291 found:
 292         unmap_vm_area(tmp);
 293         *p = tmp->next;
 294
 295         /*
 296          * Remove the guard page.
 297          */
 298         tmp->size -= PAGE_SIZE;
 299         return tmp;
 300 }
 301
 302 /**
 303  *      remove_vm_area  -  find and remove a contingous kernel virtual area
 304  *      @addr:          base address
 305  *
 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.
 309  */
 310 struct vm_struct *remove_vm_area(void *addr)
 311 {
 312         struct vm_struct *v;
 313         write_lock(&vmlist_lock);
 314         v = __remove_vm_area(addr);
 315         write_unlock(&vmlist_lock);
 316         return v;
 317 }
 318
 319 static void __vunmap(void *addr, int deallocate_pages)
 320 {
 321         struct vm_struct *area;
 322
 323         if (!addr)
 324                 return;
 325
 326         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 327                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 328                 WARN_ON(1);
 329                 return;
 330         }
 331
 332         area = remove_vm_area(addr);
 333         if (unlikely(!area)) {
 334                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 335                                 addr);
 336                 WARN_ON(1);
 337                 return;
 338         }
 339
 340         debug_check_no_locks_freed(addr, area->size);
 341
 342         if (deallocate_pages) {
 343                 int i;
 344
 345                 for (i = 0; i < area->nr_pages; i++) {
 346                         BUG_ON(!area->pages[i]);
 347                         __free_page(area->pages[i]);
 348                 }
 349
 350                 if (area->flags & VM_VPAGES)
 351                         vfree(area->pages);
 352                 else
 353                         kfree(area->pages);
 354         }
 355
 356         kfree(area);
 357         return;
 358 }
 359
 360 /**
 361  *      vfree  -  release memory allocated by vmalloc()
 362  *      @addr:          memory base address
 363  *
 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.
 367  *
 368  *      Must not be called in interrupt context.
 369  */
 370 void vfree(void *addr)
 371 {
 372         BUG_ON(in_interrupt());
 373         __vunmap(addr, 1);
 374 }
 375 EXPORT_SYMBOL(vfree);
 376
 377 /**
 378  *      vunmap  -  release virtual mapping obtained by vmap()
 379  *      @addr:          memory base address
 380  *
 381  *      Free the virtually contiguous memory area starting at @addr,
 382  *      which was created from the page array passed to vmap().
 383  *
 384  *      Must not be called in interrupt context.
 385  */
 386 void vunmap(void *addr)
 387 {
 388         BUG_ON(in_interrupt());
 389         __vunmap(addr, 0);
 390 }
 391 EXPORT_SYMBOL(vunmap);
 392
 393 /**
 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
 399  *
 400  *      Maps @count pages from @pages into contiguous kernel virtual
 401  *      space.
 402  */
 403 void *vmap(struct page **pages, unsigned int count,
 404                 unsigned long flags, pgprot_t prot)
 405 {
 406         struct vm_struct *area;
 407
 408         if (count > num_physpages)
 409                 return NULL;
 410
 411         area = get_vm_area((count << PAGE_SHIFT), flags);
 412         if (!area)
 413                 return NULL;
 414         if (map_vm_area(area, prot, &pages)) {
 415                 vunmap(area->addr);
 416                 return NULL;
 417         }
 418
 419         return area->addr;
 420 }
 421 EXPORT_SYMBOL(vmap);
 422
 423 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
 424                                 pgprot_t prot, int node)
 425 {
 426         struct page **pages;
 427         unsigned int nr_pages, array_size, i;
 428
 429         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 430         array_size = (nr_pages * sizeof(struct page *));
 431
 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,
 436                                         PAGE_KERNEL, node);
 437                 area->flags |= VM_VPAGES;
 438         } else {
 439                 pages = kmalloc_node(array_size,
 440                                 (gfp_mask & GFP_LEVEL_MASK) | __GFP_ZERO,
 441                                 node);
 442         }
 443         area->pages = pages;
 444         if (!area->pages) {
 445                 remove_vm_area(area->addr);
 446                 kfree(area);
 447                 return NULL;
 448         }
 449
 450         for (i = 0; i < area->nr_pages; i++) {
 451                 if (node < 0)
 452                         area->pages[i] = alloc_page(gfp_mask);
 453                 else
 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() */
 457                         area->nr_pages = i;
 458                         goto fail;
 459                 }
 460         }
 461
 462         if (map_vm_area(area, prot, &pages))
 463                 goto fail;
 464         return area->addr;
 465
 466 fail:
 467         vfree(area->addr);
 468         return NULL;
 469 }
 470
 471 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
 472 {
 473         return __vmalloc_area_node(area, gfp_mask, prot, -1);
 474 }
 475
 476 /**
 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
 482  *
 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.
 486  */
 487 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
 488                             int node)
 489 {
 490         struct vm_struct *area;
 491
 492         size = PAGE_ALIGN(size);
 493         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 494                 return NULL;
 495
 496         area = get_vm_area_node(size, VM_ALLOC, node, gfp_mask);
 497         if (!area)
 498                 return NULL;
 499
 500         return __vmalloc_area_node(area, gfp_mask, prot, node);
 501 }
 502
 503 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 504 {
 505         return __vmalloc_node(size, gfp_mask, prot, -1);
 506 }
 507 EXPORT_SYMBOL(__vmalloc);
 508
 509 /**
 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.
 514  *
 515  *      For tight control over page level allocator and protection flags
 516  *      use __vmalloc() instead.
 517  */
 518 void *vmalloc(unsigned long size)
 519 {
 520         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 521 }
 522 EXPORT_SYMBOL(vmalloc);
 523
 524 /**
 525  * vmalloc_user - allocate zeroed virtually contiguous memory for userspace
 526  * @size: allocation size
 527  *
 528  * The resulting memory area is zeroed so it can be mapped to userspace
 529  * without leaking data.
 530  */
 531 void *vmalloc_user(unsigned long size)
 532 {
 533         struct vm_struct *area;
 534         void *ret;
 535
 536         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
 537         if (ret) {
 538                 write_lock(&vmlist_lock);
 539                 area = __find_vm_area(ret);
 540                 area->flags |= VM_USERMAP;
 541                 write_unlock(&vmlist_lock);
 542         }
 543         return ret;
 544 }
 545 EXPORT_SYMBOL(vmalloc_user);
 546
 547 /**
 548  *      vmalloc_node  -  allocate memory on a specific node
 549  *      @size:          allocation size
 550  *      @node:          numa node
 551  *
 552  *      Allocate enough pages to cover @size from the page level
 553  *      allocator and map them into contiguous kernel virtual space.
 554  *
 555  *      For tight control over page level allocator and protection flags
 556  *      use __vmalloc() instead.
 557  */
 558 void *vmalloc_node(unsigned long size, int node)
 559 {
 560         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
 561 }
 562 EXPORT_SYMBOL(vmalloc_node);
 563
 564 #ifndef PAGE_KERNEL_EXEC
 565 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 566 #endif
 567
 568 /**
 569  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 570  *      @size:          allocation size
 571  *
 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.
 575  *
 576  *      For tight control over page level allocator and protection flags
 577  *      use __vmalloc() instead.
 578  */
 579
 580 void *vmalloc_exec(unsigned long size)
 581 {
 582         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 583 }
 584
 585 #if defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA32)
 586 #define GFP_VMALLOC32 GFP_DMA32
 587 #elif defined(CONFIG_64BIT) && defined(CONFIG_ZONE_DMA)
 588 #define GFP_VMALLOC32 GFP_DMA
 589 #else
 590 #define GFP_VMALLOC32 GFP_KERNEL
 591 #endif
 592
 593 /**
 594  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 595  *      @size:          allocation size
 596  *
 597  *      Allocate enough 32bit PA addressable pages to cover @size from the
 598  *      page level allocator and map them into contiguous kernel virtual space.
 599  */
 600 void *vmalloc_32(unsigned long size)
 601 {
 602         return __vmalloc(size, GFP_VMALLOC32, PAGE_KERNEL);
 603 }
 604 EXPORT_SYMBOL(vmalloc_32);
 605
 606 /**
 607  * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory
 608  *      @size:          allocation size
 609  *
 610  * The resulting memory area is 32bit addressable and zeroed so it can be
 611  * mapped to userspace without leaking data.
 612  */
 613 void *vmalloc_32_user(unsigned long size)
 614 {
 615         struct vm_struct *area;
 616         void *ret;
 617
 618         ret = __vmalloc(size, GFP_VMALLOC32 | __GFP_ZERO, PAGE_KERNEL);
 619         if (ret) {
 620                 write_lock(&vmlist_lock);
 621                 area = __find_vm_area(ret);
 622                 area->flags |= VM_USERMAP;
 623                 write_unlock(&vmlist_lock);
 624         }
 625         return ret;
 626 }
 627 EXPORT_SYMBOL(vmalloc_32_user);
 628
 629 long vread(char *buf, char *addr, unsigned long count)
 630 {
 631         struct vm_struct *tmp;
 632         char *vaddr, *buf_start = buf;
 633         unsigned long n;
 634
 635         /* Don't allow overflow */
 636         if ((unsigned long) addr + count < count)
 637                 count = -(unsigned long) addr;
 638
 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)
 643                         continue;
 644                 while (addr < vaddr) {
 645                         if (count == 0)
 646                                 goto finished;
 647                         *buf = '\0';
 648                         buf++;
 649                         addr++;
 650                         count--;
 651                 }
 652                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 653                 do {
 654                         if (count == 0)
 655                                 goto finished;
 656                         *buf = *addr;
 657                         buf++;
 658                         addr++;
 659                         count--;
 660                 } while (--n > 0);
 661         }
 662 finished:
 663         read_unlock(&vmlist_lock);
 664         return buf - buf_start;
 665 }
 666
 667 long vwrite(char *buf, char *addr, unsigned long count)
 668 {
 669         struct vm_struct *tmp;
 670         char *vaddr, *buf_start = buf;
 671         unsigned long n;
 672
 673         /* Don't allow overflow */
 674         if ((unsigned long) addr + count < count)
 675                 count = -(unsigned long) addr;
 676
 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)
 681                         continue;
 682                 while (addr < vaddr) {
 683                         if (count == 0)
 684                                 goto finished;
 685                         buf++;
 686                         addr++;
 687                         count--;
 688                 }
 689                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 690                 do {
 691                         if (count == 0)
 692                                 goto finished;
 693                         *addr = *buf;
 694                         buf++;
 695                         addr++;
 696                         count--;
 697                 } while (--n > 0);
 698         }
 699 finished:
 700         read_unlock(&vmlist_lock);
 701         return buf - buf_start;
 702 }
 703
 704 /**
 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
 710  *
 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.
 714  *
 715  *      Similar to remap_pfn_range() (see mm/memory.c)
 716  */
 717 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
 718                                                 unsigned long pgoff)
 719 {
 720         struct vm_struct *area;
 721         unsigned long uaddr = vma->vm_start;
 722         unsigned long usize = vma->vm_end - vma->vm_start;
 723         int ret;
 724
 725         if ((PAGE_SIZE-1) & (unsigned long)addr)
 726                 return -EINVAL;
 727
 728         read_lock(&vmlist_lock);
 729         area = __find_vm_area(addr);
 730         if (!area)
 731                 goto out_einval_locked;
 732
 733         if (!(area->flags & VM_USERMAP))
 734                 goto out_einval_locked;
 735
 736         if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
 737                 goto out_einval_locked;
 738         read_unlock(&vmlist_lock);
 739
 740         addr += pgoff << PAGE_SHIFT;
 741         do {
 742                 struct page *page = vmalloc_to_page(addr);
 743                 ret = vm_insert_page(vma, uaddr, page);
 744                 if (ret)
 745                         return ret;
 746
 747                 uaddr += PAGE_SIZE;
 748                 addr += PAGE_SIZE;
 749                 usize -= PAGE_SIZE;
 750         } while (usize > 0);
 751
 752         /* Prevent "things" like memory migration? VM_flags need a cleanup... */
 753         vma->vm_flags |= VM_RESERVED;
 754
 755         return ret;
 756
 757 out_einval_locked:
 758         read_unlock(&vmlist_lock);
 759         return -EINVAL;
 760 }
 761 EXPORT_SYMBOL(remap_vmalloc_range);
 762
 763 /*
 764  * Implement a stub for vmalloc_sync_all() if the architecture chose not to
 765  * have one.
 766  */
 767 void  __attribute__((weak)) vmalloc_sync_all(void)
 768 {
 769 }
 770
 771
 772 static int f(pte_t *pte, struct page *pmd_page, unsigned long addr, void *data)
 773 {
 774         /* apply_to_page_range() does all the hard work. */
 775         return 0;
 776 }
 777
 778 /**
 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
 782  *
 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
 787  *      processes.
 788  */
 789 struct vm_struct *alloc_vm_area(size_t size)
 790 {
 791         struct vm_struct *area;
 792
 793         area = get_vm_area(size, VM_IOREMAP);
 794         if (area == NULL)
 795                 return NULL;
 796
 797         /*
 798          * This ensures that page tables are constructed for this region
 799          * of kernel virtual address space and mapped into init_mm.
 800          */
 801         if (apply_to_page_range(&init_mm, (unsigned long)area->addr,
 802                                 area->size, f, NULL)) {
 803                 free_vm_area(area);
 804                 return NULL;
 805         }
 806
 807         /* Make sure the pagetables are constructed in process kernel
 808            mappings */
 809         vmalloc_sync_all();
 810
 811         return area;
 812 }
 813 EXPORT_SYMBOL_GPL(alloc_vm_area);
 814
 815 void free_vm_area(struct vm_struct *area)
 816 {
 817         struct vm_struct *ret;
 818         ret = remove_vm_area(area->addr);
 819         BUG_ON(ret != area);
 820         kfree(area);
 821 }
 822 EXPORT_SYMBOL_GPL(free_vm_area);