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  *
 245  *      @size:          size of the area
 246  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
 247  *
 248  *      Search an area of @size in the kernel virtual mapping area,
 249  *      and reserved it for out purposes.  Returns the area descriptor
 250  *      on success or %NULL on failure.
 251  */
 252 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 253 {
 254         return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
 255 }
 256
 257 struct vm_struct *get_vm_area_node(unsigned long size, unsigned long flags, int node)
 258 {
 259         return __get_vm_area_node(size, flags, VMALLOC_START, VMALLOC_END, node);
 260 }
 261
 262 /* Caller must hold vmlist_lock */
 263 static struct vm_struct *__find_vm_area(void *addr)
 264 {
 265         struct vm_struct *tmp;
 266
 267         for (tmp = vmlist; tmp != NULL; tmp = tmp->next) {
 268                  if (tmp->addr == addr)
 269                         break;
 270         }
 271
 272         return tmp;
 273 }
 274
 275 /* Caller must hold vmlist_lock */
 276 struct vm_struct *__remove_vm_area(void *addr)
 277 {
 278         struct vm_struct **p, *tmp;
 279
 280         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 281                  if (tmp->addr == addr)
 282                          goto found;
 283         }
 284         return NULL;
 285
 286 found:
 287         unmap_vm_area(tmp);
 288         *p = tmp->next;
 289
 290         /*
 291          * Remove the guard page.
 292          */
 293         tmp->size -= PAGE_SIZE;
 294         return tmp;
 295 }
 296
 297 /**
 298  *      remove_vm_area  -  find and remove a contingous kernel virtual area
 299  *
 300  *      @addr:          base address
 301  *
 302  *      Search for the kernel VM area starting at @addr, and remove it.
 303  *      This function returns the found VM area, but using it is NOT safe
 304  *      on SMP machines, except for its size or flags.
 305  */
 306 struct vm_struct *remove_vm_area(void *addr)
 307 {
 308         struct vm_struct *v;
 309         write_lock(&vmlist_lock);
 310         v = __remove_vm_area(addr);
 311         write_unlock(&vmlist_lock);
 312         return v;
 313 }
 314
 315 void __vunmap(void *addr, int deallocate_pages)
 316 {
 317         struct vm_struct *area;
 318
 319         if (!addr)
 320                 return;
 321
 322         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 323                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 324                 WARN_ON(1);
 325                 return;
 326         }
 327
 328         area = remove_vm_area(addr);
 329         if (unlikely(!area)) {
 330                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 331                                 addr);
 332                 WARN_ON(1);
 333                 return;
 334         }
 335
 336         debug_check_no_locks_freed(addr, area->size);
 337
 338         if (deallocate_pages) {
 339                 int i;
 340
 341                 for (i = 0; i < area->nr_pages; i++) {
 342                         BUG_ON(!area->pages[i]);
 343                         __free_page(area->pages[i]);
 344                 }
 345
 346                 if (area->flags & VM_VPAGES)
 347                         vfree(area->pages);
 348                 else
 349                         kfree(area->pages);
 350         }
 351
 352         kfree(area);
 353         return;
 354 }
 355
 356 /**
 357  *      vfree  -  release memory allocated by vmalloc()
 358  *
 359  *      @addr:          memory base address
 360  *
 361  *      Free the virtually contiguous memory area starting at @addr, as
 362  *      obtained from vmalloc(), vmalloc_32() or __vmalloc(). If @addr is
 363  *      NULL, no operation is performed.
 364  *
 365  *      Must not be called in interrupt context.
 366  */
 367 void vfree(void *addr)
 368 {
 369         BUG_ON(in_interrupt());
 370         __vunmap(addr, 1);
 371 }
 372 EXPORT_SYMBOL(vfree);
 373
 374 /**
 375  *      vunmap  -  release virtual mapping obtained by vmap()
 376  *
 377  *      @addr:          memory base address
 378  *
 379  *      Free the virtually contiguous memory area starting at @addr,
 380  *      which was created from the page array passed to vmap().
 381  *
 382  *      Must not be called in interrupt context.
 383  */
 384 void vunmap(void *addr)
 385 {
 386         BUG_ON(in_interrupt());
 387         __vunmap(addr, 0);
 388 }
 389 EXPORT_SYMBOL(vunmap);
 390
 391 /**
 392  *      vmap  -  map an array of pages into virtually contiguous space
 393  *
 394  *      @pages:         array of page pointers
 395  *      @count:         number of pages to map
 396  *      @flags:         vm_area->flags
 397  *      @prot:          page protection for the mapping
 398  *
 399  *      Maps @count pages from @pages into contiguous kernel virtual
 400  *      space.
 401  */
 402 void *vmap(struct page **pages, unsigned int count,
 403                 unsigned long flags, pgprot_t prot)
 404 {
 405         struct vm_struct *area;
 406
 407         if (count > num_physpages)
 408                 return NULL;
 409
 410         area = get_vm_area((count << PAGE_SHIFT), flags);
 411         if (!area)
 412                 return NULL;
 413         if (map_vm_area(area, prot, &pages)) {
 414                 vunmap(area->addr);
 415                 return NULL;
 416         }
 417
 418         return area->addr;
 419 }
 420 EXPORT_SYMBOL(vmap);
 421
 422 void *__vmalloc_area_node(struct vm_struct *area, gfp_t gfp_mask,
 423                                 pgprot_t prot, int node)
 424 {
 425         struct page **pages;
 426         unsigned int nr_pages, array_size, i;
 427
 428         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 429         array_size = (nr_pages * sizeof(struct page *));
 430
 431         area->nr_pages = nr_pages;
 432         /* Please note that the recursion is strictly bounded. */
 433         if (array_size > PAGE_SIZE) {
 434                 pages = __vmalloc_node(array_size, gfp_mask, PAGE_KERNEL, node);
 435                 area->flags |= VM_VPAGES;
 436         } else
 437                 pages = kmalloc_node(array_size, (gfp_mask & ~__GFP_HIGHMEM), node);
 438         area->pages = pages;
 439         if (!area->pages) {
 440                 remove_vm_area(area->addr);
 441                 kfree(area);
 442                 return NULL;
 443         }
 444         memset(area->pages, 0, array_size);
 445
 446         for (i = 0; i < area->nr_pages; i++) {
 447                 if (node < 0)
 448                         area->pages[i] = alloc_page(gfp_mask);
 449                 else
 450                         area->pages[i] = alloc_pages_node(node, gfp_mask, 0);
 451                 if (unlikely(!area->pages[i])) {
 452                         /* Successfully allocated i pages, free them in __vunmap() */
 453                         area->nr_pages = i;
 454                         goto fail;
 455                 }
 456         }
 457
 458         if (map_vm_area(area, prot, &pages))
 459                 goto fail;
 460         return area->addr;
 461
 462 fail:
 463         vfree(area->addr);
 464         return NULL;
 465 }
 466
 467 void *__vmalloc_area(struct vm_struct *area, gfp_t gfp_mask, pgprot_t prot)
 468 {
 469         return __vmalloc_area_node(area, gfp_mask, prot, -1);
 470 }
 471
 472 /**
 473  *      __vmalloc_node  -  allocate virtually contiguous memory
 474  *
 475  *      @size:          allocation size
 476  *      @gfp_mask:      flags for the page level allocator
 477  *      @prot:          protection mask for the allocated pages
 478  *      @node:          node to use for allocation or -1
 479  *
 480  *      Allocate enough pages to cover @size from the page level
 481  *      allocator with @gfp_mask flags.  Map them into contiguous
 482  *      kernel virtual space, using a pagetable protection of @prot.
 483  */
 484 static void *__vmalloc_node(unsigned long size, gfp_t gfp_mask, pgprot_t prot,
 485                             int node)
 486 {
 487         struct vm_struct *area;
 488
 489         size = PAGE_ALIGN(size);
 490         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 491                 return NULL;
 492
 493         area = get_vm_area_node(size, VM_ALLOC, node);
 494         if (!area)
 495                 return NULL;
 496
 497         return __vmalloc_area_node(area, gfp_mask, prot, node);
 498 }
 499
 500 void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot)
 501 {
 502         return __vmalloc_node(size, gfp_mask, prot, -1);
 503 }
 504 EXPORT_SYMBOL(__vmalloc);
 505
 506 /**
 507  *      vmalloc  -  allocate virtually contiguous memory
 508  *
 509  *      @size:          allocation size
 510  *
 511  *      Allocate enough pages to cover @size from the page level
 512  *      allocator and map them into contiguous kernel virtual space.
 513  *
 514  *      For tight cotrol over page level allocator and protection flags
 515  *      use __vmalloc() instead.
 516  */
 517 void *vmalloc(unsigned long size)
 518 {
 519         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 520 }
 521 EXPORT_SYMBOL(vmalloc);
 522
 523 /**
 524  *      vmalloc_user  -  allocate virtually contiguous memory which has
 525  *                         been zeroed so it can be mapped to userspace without
 526  *                         leaking data.
 527  *
 528  *      @size:          allocation size
 529  */
 530 void *vmalloc_user(unsigned long size)
 531 {
 532         struct vm_struct *area;
 533         void *ret;
 534
 535         ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
 536         write_lock(&vmlist_lock);
 537         area = __find_vm_area(ret);
 538         area->flags |= VM_USERMAP;
 539         write_unlock(&vmlist_lock);
 540
 541         return ret;
 542 }
 543 EXPORT_SYMBOL(vmalloc_user);
 544
 545 /**
 546  *      vmalloc_node  -  allocate memory on a specific node
 547  *
 548  *      @size:          allocation size
 549  *      @node:          numa node
 550  *
 551  *      Allocate enough pages to cover @size from the page level
 552  *      allocator and map them into contiguous kernel virtual space.
 553  *
 554  *      For tight cotrol over page level allocator and protection flags
 555  *      use __vmalloc() instead.
 556  */
 557 void *vmalloc_node(unsigned long size, int node)
 558 {
 559         return __vmalloc_node(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL, node);
 560 }
 561 EXPORT_SYMBOL(vmalloc_node);
 562
 563 #ifndef PAGE_KERNEL_EXEC
 564 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 565 #endif
 566
 567 /**
 568  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 569  *
 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 cotrol 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 /**
 586  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 587  *
 588  *      @size:          allocation size
 589  *
 590  *      Allocate enough 32bit PA addressable pages to cover @size from the
 591  *      page level allocator and map them into contiguous kernel virtual space.
 592  */
 593 void *vmalloc_32(unsigned long size)
 594 {
 595         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 596 }
 597 EXPORT_SYMBOL(vmalloc_32);
 598
 599 /**
 600  *      vmalloc_32_user  -  allocate virtually contiguous memory (32bit
 601  *                            addressable) which is zeroed so it can be
 602  *                            mapped to userspace without leaking data.
 603  *
 604  *      @size:          allocation size
 605  */
 606 void *vmalloc_32_user(unsigned long size)
 607 {
 608         struct vm_struct *area;
 609         void *ret;
 610
 611         ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL);
 612         write_lock(&vmlist_lock);
 613         area = __find_vm_area(ret);
 614         area->flags |= VM_USERMAP;
 615         write_unlock(&vmlist_lock);
 616
 617         return ret;
 618 }
 619 EXPORT_SYMBOL(vmalloc_32_user);
 620
 621 long vread(char *buf, char *addr, unsigned long count)
 622 {
 623         struct vm_struct *tmp;
 624         char *vaddr, *buf_start = buf;
 625         unsigned long n;
 626
 627         /* Don't allow overflow */
 628         if ((unsigned long) addr + count < count)
 629                 count = -(unsigned long) addr;
 630
 631         read_lock(&vmlist_lock);
 632         for (tmp = vmlist; tmp; tmp = tmp->next) {
 633                 vaddr = (char *) tmp->addr;
 634                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 635                         continue;
 636                 while (addr < vaddr) {
 637                         if (count == 0)
 638                                 goto finished;
 639                         *buf = '\0';
 640                         buf++;
 641                         addr++;
 642                         count--;
 643                 }
 644                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 645                 do {
 646                         if (count == 0)
 647                                 goto finished;
 648                         *buf = *addr;
 649                         buf++;
 650                         addr++;
 651                         count--;
 652                 } while (--n > 0);
 653         }
 654 finished:
 655         read_unlock(&vmlist_lock);
 656         return buf - buf_start;
 657 }
 658
 659 long vwrite(char *buf, char *addr, unsigned long count)
 660 {
 661         struct vm_struct *tmp;
 662         char *vaddr, *buf_start = buf;
 663         unsigned long n;
 664
 665         /* Don't allow overflow */
 666         if ((unsigned long) addr + count < count)
 667                 count = -(unsigned long) addr;
 668
 669         read_lock(&vmlist_lock);
 670         for (tmp = vmlist; tmp; tmp = tmp->next) {
 671                 vaddr = (char *) tmp->addr;
 672                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 673                         continue;
 674                 while (addr < vaddr) {
 675                         if (count == 0)
 676                                 goto finished;
 677                         buf++;
 678                         addr++;
 679                         count--;
 680                 }
 681                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 682                 do {
 683                         if (count == 0)
 684                                 goto finished;
 685                         *addr = *buf;
 686                         buf++;
 687                         addr++;
 688                         count--;
 689                 } while (--n > 0);
 690         }
 691 finished:
 692         read_unlock(&vmlist_lock);
 693         return buf - buf_start;
 694 }
 695
 696 /**
 697  *      remap_vmalloc_range  -  map vmalloc pages to userspace
 698  *
 699  *      @vma:           vma to cover (map full range of vma)
 700  *      @addr:          vmalloc memory
 701  *      @pgoff:         number of pages into addr before first page to map
 702  *      @returns:       0 for success, -Exxx on failure
 703  *
 704  *      This function checks that addr is a valid vmalloc'ed area, and
 705  *      that it is big enough to cover the vma. Will return failure if
 706  *      that criteria isn't met.
 707  *
 708  *      Similar to remap_pfn_range (see mm/memory.c)
 709  */
 710 int remap_vmalloc_range(struct vm_area_struct *vma, void *addr,
 711                                                 unsigned long pgoff)
 712 {
 713         struct vm_struct *area;
 714         unsigned long uaddr = vma->vm_start;
 715         unsigned long usize = vma->vm_end - vma->vm_start;
 716         int ret;
 717
 718         if ((PAGE_SIZE-1) & (unsigned long)addr)
 719                 return -EINVAL;
 720
 721         read_lock(&vmlist_lock);
 722         area = __find_vm_area(addr);
 723         if (!area)
 724                 goto out_einval_locked;
 725
 726         if (!(area->flags & VM_USERMAP))
 727                 goto out_einval_locked;
 728
 729         if (usize + (pgoff << PAGE_SHIFT) > area->size - PAGE_SIZE)
 730                 goto out_einval_locked;
 731         read_unlock(&vmlist_lock);
 732
 733         addr += pgoff << PAGE_SHIFT;
 734         do {
 735                 struct page *page = vmalloc_to_page(addr);
 736                 ret = vm_insert_page(vma, uaddr, page);
 737                 if (ret)
 738                         return ret;
 739
 740                 uaddr += PAGE_SIZE;
 741                 addr += PAGE_SIZE;
 742                 usize -= PAGE_SIZE;
 743         } while (usize > 0);
 744
 745         /* Prevent "things" like memory migration? VM_flags need a cleanup... */
 746         vma->vm_flags |= VM_RESERVED;
 747
 748         return ret;
 749
 750 out_einval_locked:
 751         read_unlock(&vmlist_lock);
 752         return -EINVAL;
 753 }
 754 EXPORT_SYMBOL(remap_vmalloc_range);
 755