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  */
   9
  10 #include <linux/mm.h>
  11 #include <linux/module.h>
  12 #include <linux/highmem.h>
  13 #include <linux/slab.h>
  14 #include <linux/spinlock.h>
  15 #include <linux/interrupt.h>
  16
  17 #include <linux/vmalloc.h>
  18
  19 #include <asm/uaccess.h>
  20 #include <asm/pgalloc.h>
  21 #include <asm/tlbflush.h>
  22
  23
  24 rwlock_t vmlist_lock = RW_LOCK_UNLOCKED;
  25 struct vm_struct *vmlist;
  26
  27 static void unmap_area_pte(pmd_t *pmd, unsigned long address,
  28                                   unsigned long size)
  29 {
  30         unsigned long end;
  31         pte_t *pte;
  32
  33         if (pmd_none(*pmd))
  34                 return;
  35         if (pmd_bad(*pmd)) {
  36                 pmd_ERROR(*pmd);
  37                 pmd_clear(pmd);
  38                 return;
  39         }
  40
  41         pte = pte_offset_kernel(pmd, address);
  42         address &= ~PMD_MASK;
  43         end = address + size;
  44         if (end > PMD_SIZE)
  45                 end = PMD_SIZE;
  46
  47         do {
  48                 pte_t page;
  49                 page = ptep_get_and_clear(pte);
  50                 address += PAGE_SIZE;
  51                 pte++;
  52                 if (pte_none(page))
  53                         continue;
  54                 if (pte_present(page))
  55                         continue;
  56                 printk(KERN_CRIT "Whee.. Swapped out page in kernel page table\n");
  57         } while (address < end);
  58 }
  59
  60 static void unmap_area_pmd(pgd_t *dir, unsigned long address,
  61                                   unsigned long size)
  62 {
  63         unsigned long end;
  64         pmd_t *pmd;
  65
  66         if (pgd_none(*dir))
  67                 return;
  68         if (pgd_bad(*dir)) {
  69                 pgd_ERROR(*dir);
  70                 pgd_clear(dir);
  71                 return;
  72         }
  73
  74         pmd = pmd_offset(dir, address);
  75         address &= ~PGDIR_MASK;
  76         end = address + size;
  77         if (end > PGDIR_SIZE)
  78                 end = PGDIR_SIZE;
  79
  80         do {
  81                 unmap_area_pte(pmd, address, end - address);
  82                 address = (address + PMD_SIZE) & PMD_MASK;
  83                 pmd++;
  84         } while (address < end);
  85 }
  86
  87 static int map_area_pte(pte_t *pte, unsigned long address,
  88                                unsigned long size, pgprot_t prot,
  89                                struct page ***pages)
  90 {
  91         unsigned long end;
  92
  93         address &= ~PMD_MASK;
  94         end = address + size;
  95         if (end > PMD_SIZE)
  96                 end = PMD_SIZE;
  97
  98         do {
  99                 struct page *page = **pages;
 100
 101                 WARN_ON(!pte_none(*pte));
 102                 if (!page)
 103                         return -ENOMEM;
 104
 105                 set_pte(pte, mk_pte(page, prot));
 106                 address += PAGE_SIZE;
 107                 pte++;
 108                 (*pages)++;
 109         } while (address < end);
 110         return 0;
 111 }
 112
 113 static int map_area_pmd(pmd_t *pmd, unsigned long address,
 114                                unsigned long size, pgprot_t prot,
 115                                struct page ***pages)
 116 {
 117         unsigned long end;
 118
 119         address &= ~PGDIR_MASK;
 120         end = address + size;
 121         if (end > PGDIR_SIZE)
 122                 end = PGDIR_SIZE;
 123
 124         do {
 125                 pte_t * pte = pte_alloc_kernel(&init_mm, pmd, address);
 126                 if (!pte)
 127                         return -ENOMEM;
 128                 if (map_area_pte(pte, address, end - address, prot, pages))
 129                         return -ENOMEM;
 130                 address = (address + PMD_SIZE) & PMD_MASK;
 131                 pmd++;
 132         } while (address < end);
 133
 134         return 0;
 135 }
 136
 137 void unmap_vm_area(struct vm_struct *area)
 138 {
 139         unsigned long address = (unsigned long) area->addr;
 140         unsigned long end = (address + area->size);
 141         pgd_t *dir;
 142
 143         dir = pgd_offset_k(address);
 144         flush_cache_vunmap(address, end);
 145         do {
 146                 unmap_area_pmd(dir, address, end - address);
 147                 address = (address + PGDIR_SIZE) & PGDIR_MASK;
 148                 dir++;
 149         } while (address && (address < end));
 150         flush_tlb_kernel_range((unsigned long) area->addr, end);
 151 }
 152
 153 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 154 {
 155         unsigned long address = (unsigned long) area->addr;
 156         unsigned long end = address + (area->size-PAGE_SIZE);
 157         pgd_t *dir;
 158         int err = 0;
 159
 160         dir = pgd_offset_k(address);
 161         spin_lock(&init_mm.page_table_lock);
 162         do {
 163                 pmd_t *pmd = pmd_alloc(&init_mm, dir, address);
 164                 if (!pmd) {
 165                         err = -ENOMEM;
 166                         break;
 167                 }
 168                 if (map_area_pmd(pmd, address, end - address, prot, pages)) {
 169                         err = -ENOMEM;
 170                         break;
 171                 }
 172
 173                 address = (address + PGDIR_SIZE) & PGDIR_MASK;
 174                 dir++;
 175         } while (address && (address < end));
 176
 177         spin_unlock(&init_mm.page_table_lock);
 178         flush_cache_vmap((unsigned long) area->addr, end);
 179         return err;
 180 }
 181
 182 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 183                                 unsigned long start, unsigned long end)
 184 {
 185         struct vm_struct **p, *tmp, *area;
 186         unsigned long addr = start;
 187
 188         area = kmalloc(sizeof(*area), GFP_KERNEL);
 189         if (unlikely(!area))
 190                 return NULL;
 191
 192         /*
 193          * We always allocate a guard page.
 194          */
 195         size += PAGE_SIZE;
 196         if (unlikely(!size)) {
 197                 kfree (area);
 198                 return NULL;
 199         }
 200
 201         write_lock(&vmlist_lock);
 202         for (p = &vmlist; (tmp = *p) ;p = &tmp->next) {
 203                 if ((unsigned long)tmp->addr < addr)
 204                         continue;
 205                 if ((size + addr) < addr)
 206                         goto out;
 207                 if (size + addr <= (unsigned long)tmp->addr)
 208                         goto found;
 209                 addr = tmp->size + (unsigned long)tmp->addr;
 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         return NULL;
 232 }
 233
 234 /**
 235  *      get_vm_area  -  reserve a contingous kernel virtual area
 236  *
 237  *      @size:          size of the area
 238  *      @flags:         %VM_IOREMAP for I/O mappings or VM_ALLOC
 239  *
 240  *      Search an area of @size in the kernel virtual mapping area,
 241  *      and reserved it for out purposes.  Returns the area descriptor
 242  *      on success or %NULL on failure.
 243  */
 244 struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
 245 {
 246         return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
 247 }
 248
 249 /**
 250  *      remove_vm_area  -  find and remove a contingous kernel virtual area
 251  *
 252  *      @addr:          base address
 253  *
 254  *      Search for the kernel VM area starting at @addr, and remove it.
 255  *      This function returns the found VM area, but using it is NOT safe
 256  *      on SMP machines.
 257  */
 258 struct vm_struct *remove_vm_area(void *addr)
 259 {
 260         struct vm_struct **p, *tmp;
 261
 262         write_lock(&vmlist_lock);
 263         for (p = &vmlist ; (tmp = *p) ;p = &tmp->next) {
 264                  if (tmp->addr == addr)
 265                          goto found;
 266         }
 267         write_unlock(&vmlist_lock);
 268         return NULL;
 269
 270 found:
 271         unmap_vm_area(tmp);
 272         *p = tmp->next;
 273         write_unlock(&vmlist_lock);
 274         return tmp;
 275 }
 276
 277 void __vunmap(void *addr, int deallocate_pages)
 278 {
 279         struct vm_struct *area;
 280
 281         if (!addr)
 282                 return;
 283
 284         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 285                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 286                 return;
 287         }
 288
 289         area = remove_vm_area(addr);
 290         if (unlikely(!area)) {
 291                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 292                                 addr);
 293                 return;
 294         }
 295
 296         if (deallocate_pages) {
 297                 int i;
 298
 299                 for (i = 0; i < area->nr_pages; i++) {
 300                         if (unlikely(!area->pages[i]))
 301                                 BUG();
 302                         __free_page(area->pages[i]);
 303                 }
 304
 305                 kfree(area->pages);
 306         }
 307
 308         kfree(area);
 309         return;
 310 }
 311
 312 /**
 313  *      vfree  -  release memory allocated by vmalloc()
 314  *
 315  *      @addr:          memory base address
 316  *
 317  *      Free the virtually contiguous memory area starting at @addr, as
 318  *      obtained from vmalloc(), vmalloc_32() or __vmalloc().
 319  *
 320  *      May not be called in interrupt context.
 321  */
 322 void vfree(void *addr)
 323 {
 324         BUG_ON(in_interrupt());
 325         __vunmap(addr, 1);
 326 }
 327
 328 EXPORT_SYMBOL(vfree);
 329
 330 /**
 331  *      vunmap  -  release virtual mapping obtained by vmap()
 332  *
 333  *      @addr:          memory base address
 334  *
 335  *      Free the virtually contiguous memory area starting at @addr,
 336  *      which was created from the page array passed to vmap().
 337  *
 338  *      May not be called in interrupt context.
 339  */
 340 void vunmap(void *addr)
 341 {
 342         BUG_ON(in_interrupt());
 343         __vunmap(addr, 0);
 344 }
 345
 346 EXPORT_SYMBOL(vunmap);
 347
 348 /**
 349  *      vmap  -  map an array of pages into virtually contiguous space
 350  *
 351  *      @pages:         array of page pointers
 352  *      @count:         number of pages to map
 353  *      @flags:         vm_area->flags
 354  *      @prot:          page protection for the mapping
 355  *
 356  *      Maps @count pages from @pages into contiguous kernel virtual
 357  *      space.
 358  */
 359 void *vmap(struct page **pages, unsigned int count,
 360                 unsigned long flags, pgprot_t prot)
 361 {
 362         struct vm_struct *area;
 363
 364         if (count > num_physpages)
 365                 return NULL;
 366
 367         area = get_vm_area((count << PAGE_SHIFT), flags);
 368         if (!area)
 369                 return NULL;
 370         if (map_vm_area(area, prot, &pages)) {
 371                 vunmap(area->addr);
 372                 return NULL;
 373         }
 374
 375         return area->addr;
 376 }
 377
 378 EXPORT_SYMBOL(vmap);
 379
 380 /**
 381  *      __vmalloc  -  allocate virtually contiguous memory
 382  *
 383  *      @size:          allocation size
 384  *      @gfp_mask:      flags for the page level allocator
 385  *      @prot:          protection mask for the allocated pages
 386  *
 387  *      Allocate enough pages to cover @size from the page level
 388  *      allocator with @gfp_mask flags.  Map them into contiguous
 389  *      kernel virtual space, using a pagetable protection of @prot.
 390  */
 391 void *__vmalloc(unsigned long size, int gfp_mask, pgprot_t prot)
 392 {
 393         struct vm_struct *area;
 394         struct page **pages;
 395         unsigned int nr_pages, array_size, i;
 396
 397         size = PAGE_ALIGN(size);
 398         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 399                 return NULL;
 400
 401         area = get_vm_area(size, VM_ALLOC);
 402         if (!area)
 403                 return NULL;
 404
 405         nr_pages = size >> PAGE_SHIFT;
 406         array_size = (nr_pages * sizeof(struct page *));
 407
 408         area->nr_pages = nr_pages;
 409         area->pages = pages = kmalloc(array_size, (gfp_mask & ~__GFP_HIGHMEM));
 410         if (!area->pages) {
 411                 remove_vm_area(area->addr);
 412                 kfree(area);
 413                 return NULL;
 414         }
 415         memset(area->pages, 0, array_size);
 416
 417         for (i = 0; i < area->nr_pages; i++) {
 418                 area->pages[i] = alloc_page(gfp_mask);
 419                 if (unlikely(!area->pages[i])) {
 420                         /* Successfully allocated i pages, free them in __vunmap() */
 421                         area->nr_pages = i;
 422                         goto fail;
 423                 }
 424         }
 425
 426         if (map_vm_area(area, prot, &pages))
 427                 goto fail;
 428         return area->addr;
 429
 430 fail:
 431         vfree(area->addr);
 432         return NULL;
 433 }
 434
 435 EXPORT_SYMBOL(__vmalloc);
 436
 437 /**
 438  *      vmalloc  -  allocate virtually contiguous memory
 439  *
 440  *      @size:          allocation size
 441  *
 442  *      Allocate enough pages to cover @size from the page level
 443  *      allocator and map them into contiguous kernel virtual space.
 444  *
 445  *      For tight cotrol over page level allocator and protection flags
 446  *      use __vmalloc() instead.
 447  */
 448 void *vmalloc(unsigned long size)
 449 {
 450        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 451 }
 452
 453 EXPORT_SYMBOL(vmalloc);
 454
 455 /**
 456  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 457  *
 458  *      @size:          allocation size
 459  *
 460  *      Allocate enough 32bit PA addressable pages to cover @size from the
 461  *      page level allocator and map them into contiguous kernel virtual space.
 462  */
 463 void *vmalloc_32(unsigned long size)
 464 {
 465         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 466 }
 467
 468 EXPORT_SYMBOL(vmalloc_32);
 469
 470 long vread(char *buf, char *addr, unsigned long count)
 471 {
 472         struct vm_struct *tmp;
 473         char *vaddr, *buf_start = buf;
 474         unsigned long n;
 475
 476         /* Don't allow overflow */
 477         if ((unsigned long) addr + count < count)
 478                 count = -(unsigned long) addr;
 479
 480         read_lock(&vmlist_lock);
 481         for (tmp = vmlist; tmp; tmp = tmp->next) {
 482                 vaddr = (char *) tmp->addr;
 483                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 484                         continue;
 485                 while (addr < vaddr) {
 486                         if (count == 0)
 487                                 goto finished;
 488                         *buf = '\0';
 489                         buf++;
 490                         addr++;
 491                         count--;
 492                 }
 493                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 494                 do {
 495                         if (count == 0)
 496                                 goto finished;
 497                         *buf = *addr;
 498                         buf++;
 499                         addr++;
 500                         count--;
 501                 } while (--n > 0);
 502         }
 503 finished:
 504         read_unlock(&vmlist_lock);
 505         return buf - buf_start;
 506 }
 507
 508 long vwrite(char *buf, char *addr, unsigned long count)
 509 {
 510         struct vm_struct *tmp;
 511         char *vaddr, *buf_start = buf;
 512         unsigned long n;
 513
 514         /* Don't allow overflow */
 515         if ((unsigned long) addr + count < count)
 516                 count = -(unsigned long) addr;
 517
 518         read_lock(&vmlist_lock);
 519         for (tmp = vmlist; tmp; tmp = tmp->next) {
 520                 vaddr = (char *) tmp->addr;
 521                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 522                         continue;
 523                 while (addr < vaddr) {
 524                         if (count == 0)
 525                                 goto finished;
 526                         buf++;
 527                         addr++;
 528                         count--;
 529                 }
 530                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 531                 do {
 532                         if (count == 0)
 533                                 goto finished;
 534                         *addr = *buf;
 535                         buf++;
 536                         addr++;
 537                         count--;
 538                 } while (--n > 0);
 539         }
 540 finished:
 541         read_unlock(&vmlist_lock);
 542         return buf - buf_start;
 543 }