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/tlbflush.h>
  21
  22
  23 DEFINE_RWLOCK(vmlist_lock);
  24 struct vm_struct *vmlist;
  25
  26 static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
  27 {
  28         pte_t *pte;
  29
  30         pte = pte_offset_kernel(pmd, addr);
  31         do {
  32                 pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
  33                 WARN_ON(!pte_none(ptent) && !pte_present(ptent));
  34         } while (pte++, addr += PAGE_SIZE, addr != end);
  35 }
  36
  37 static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
  38                                                 unsigned long end)
  39 {
  40         pmd_t *pmd;
  41         unsigned long next;
  42
  43         pmd = pmd_offset(pud, addr);
  44         do {
  45                 next = pmd_addr_end(addr, end);
  46                 if (pmd_none_or_clear_bad(pmd))
  47                         continue;
  48                 vunmap_pte_range(pmd, addr, next);
  49         } while (pmd++, addr = next, addr != end);
  50 }
  51
  52 static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
  53                                                 unsigned long end)
  54 {
  55         pud_t *pud;
  56         unsigned long next;
  57
  58         pud = pud_offset(pgd, addr);
  59         do {
  60                 next = pud_addr_end(addr, end);
  61                 if (pud_none_or_clear_bad(pud))
  62                         continue;
  63                 vunmap_pmd_range(pud, addr, next);
  64         } while (pud++, addr = next, addr != end);
  65 }
  66
  67 void unmap_vm_area(struct vm_struct *area)
  68 {
  69         pgd_t *pgd;
  70         unsigned long next;
  71         unsigned long addr = (unsigned long) area->addr;
  72         unsigned long end = addr + area->size;
  73
  74         BUG_ON(addr >= end);
  75         pgd = pgd_offset_k(addr);
  76         flush_cache_vunmap(addr, end);
  77         do {
  78                 next = pgd_addr_end(addr, end);
  79                 if (pgd_none_or_clear_bad(pgd))
  80                         continue;
  81                 vunmap_pud_range(pgd, addr, next);
  82         } while (pgd++, addr = next, addr != end);
  83         flush_tlb_kernel_range((unsigned long) area->addr, end);
  84 }
  85
  86 static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
  87                         unsigned long end, pgprot_t prot, struct page ***pages)
  88 {
  89         pte_t *pte;
  90
  91         pte = pte_alloc_kernel(&init_mm, pmd, addr);
  92         if (!pte)
  93                 return -ENOMEM;
  94         do {
  95                 struct page *page = **pages;
  96                 WARN_ON(!pte_none(*pte));
  97                 if (!page)
  98                         return -ENOMEM;
  99                 set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
 100                 (*pages)++;
 101         } while (pte++, addr += PAGE_SIZE, addr != end);
 102         return 0;
 103 }
 104
 105 static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
 106                         unsigned long end, pgprot_t prot, struct page ***pages)
 107 {
 108         pmd_t *pmd;
 109         unsigned long next;
 110
 111         pmd = pmd_alloc(&init_mm, pud, addr);
 112         if (!pmd)
 113                 return -ENOMEM;
 114         do {
 115                 next = pmd_addr_end(addr, end);
 116                 if (vmap_pte_range(pmd, addr, next, prot, pages))
 117                         return -ENOMEM;
 118         } while (pmd++, addr = next, addr != end);
 119         return 0;
 120 }
 121
 122 static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
 123                         unsigned long end, pgprot_t prot, struct page ***pages)
 124 {
 125         pud_t *pud;
 126         unsigned long next;
 127
 128         pud = pud_alloc(&init_mm, pgd, addr);
 129         if (!pud)
 130                 return -ENOMEM;
 131         do {
 132                 next = pud_addr_end(addr, end);
 133                 if (vmap_pmd_range(pud, addr, next, prot, pages))
 134                         return -ENOMEM;
 135         } while (pud++, addr = next, addr != end);
 136         return 0;
 137 }
 138
 139 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 140 {
 141         pgd_t *pgd;
 142         unsigned long next;
 143         unsigned long addr = (unsigned long) area->addr;
 144         unsigned long end = addr + area->size - PAGE_SIZE;
 145         int err;
 146
 147         BUG_ON(addr >= end);
 148         pgd = pgd_offset_k(addr);
 149         spin_lock(&init_mm.page_table_lock);
 150         do {
 151                 next = pgd_addr_end(addr, end);
 152                 err = vmap_pud_range(pgd, addr, next, prot, pages);
 153                 if (err)
 154                         break;
 155         } while (pgd++, addr = next, addr != end);
 156         spin_unlock(&init_mm.page_table_lock);
 157         flush_cache_vmap((unsigned long) area->addr, end);
 158         return err;
 159 }
 160
 161 struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
 162                                 unsigned long start, unsigned long end)
 163 {
 164         struct vm_struct **p, *tmp, *area;
 165         unsigned long align = 1;
 166         unsigned long addr;
 167
 168         if (flags & VM_IOREMAP) {
 169                 int bit = fls(size);
 170
 171                 if (bit > IOREMAP_MAX_ORDER)
 172                         bit = IOREMAP_MAX_ORDER;
 173                 else if (bit < PAGE_SHIFT)
 174                         bit = PAGE_SHIFT;
 175
 176                 align = 1ul << bit;
 177         }
 178         addr = ALIGN(start, align);
 179         size = PAGE_ALIGN(size);
 180
 181         area = kmalloc(sizeof(*area), GFP_KERNEL);
 182         if (unlikely(!area))
 183                 return NULL;
 184
 185         if (unlikely(!size)) {
 186                 kfree (area);
 187                 return NULL;
 188         }
 189
 190         /*
 191          * We always allocate a guard page.
 192          */
 193         size += PAGE_SIZE;
 194
 195         write_lock(&vmlist_lock);
 196         for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
 197                 if ((unsigned long)tmp->addr < addr) {
 198                         if((unsigned long)tmp->addr + tmp->size >= addr)
 199                                 addr = ALIGN(tmp->size +
 200                                              (unsigned long)tmp->addr, align);
 201                         continue;
 202                 }
 203                 if ((size + addr) < addr)
 204                         goto out;
 205                 if (size + addr <= (unsigned long)tmp->addr)
 206                         goto found;
 207                 addr = ALIGN(tmp->size + (unsigned long)tmp->addr, align);
 208                 if (addr > end - size)
 209                         goto out;
 210         }
 211
 212 found:
 213         area->next = *p;
 214         *p = area;
 215
 216         area->flags = flags;
 217         area->addr = (void *)addr;
 218         area->size = size;
 219         area->pages = NULL;
 220         area->nr_pages = 0;
 221         area->phys_addr = 0;
 222         write_unlock(&vmlist_lock);
 223
 224         return area;
 225
 226 out:
 227         write_unlock(&vmlist_lock);
 228         kfree(area);
 229         if (printk_ratelimit())
 230                 printk(KERN_WARNING "allocation failed: out of vmalloc space - use vmalloc=<size> to increase size.\n");
 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 /* Caller must hold vmlist_lock */
 250 struct vm_struct *__remove_vm_area(void *addr)
 251 {
 252         struct vm_struct **p, *tmp;
 253
 254         for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
 255                  if (tmp->addr == addr)
 256                          goto found;
 257         }
 258         return NULL;
 259
 260 found:
 261         unmap_vm_area(tmp);
 262         *p = tmp->next;
 263
 264         /*
 265          * Remove the guard page.
 266          */
 267         tmp->size -= PAGE_SIZE;
 268         return tmp;
 269 }
 270
 271 /**
 272  *      remove_vm_area  -  find and remove a contingous kernel virtual area
 273  *
 274  *      @addr:          base address
 275  *
 276  *      Search for the kernel VM area starting at @addr, and remove it.
 277  *      This function returns the found VM area, but using it is NOT safe
 278  *      on SMP machines, except for its size or flags.
 279  */
 280 struct vm_struct *remove_vm_area(void *addr)
 281 {
 282         struct vm_struct *v;
 283         write_lock(&vmlist_lock);
 284         v = __remove_vm_area(addr);
 285         write_unlock(&vmlist_lock);
 286         return v;
 287 }
 288
 289 void __vunmap(void *addr, int deallocate_pages)
 290 {
 291         struct vm_struct *area;
 292
 293         if (!addr)
 294                 return;
 295
 296         if ((PAGE_SIZE-1) & (unsigned long)addr) {
 297                 printk(KERN_ERR "Trying to vfree() bad address (%p)\n", addr);
 298                 WARN_ON(1);
 299                 return;
 300         }
 301
 302         area = remove_vm_area(addr);
 303         if (unlikely(!area)) {
 304                 printk(KERN_ERR "Trying to vfree() nonexistent vm area (%p)\n",
 305                                 addr);
 306                 WARN_ON(1);
 307                 return;
 308         }
 309
 310         if (deallocate_pages) {
 311                 int i;
 312
 313                 for (i = 0; i < area->nr_pages; i++) {
 314                         if (unlikely(!area->pages[i]))
 315                                 BUG();
 316                         __free_page(area->pages[i]);
 317                 }
 318
 319                 if (area->nr_pages > PAGE_SIZE/sizeof(struct page *))
 320                         vfree(area->pages);
 321                 else
 322                         kfree(area->pages);
 323         }
 324
 325         kfree(area);
 326         return;
 327 }
 328
 329 /**
 330  *      vfree  -  release memory allocated by vmalloc()
 331  *
 332  *      @addr:          memory base address
 333  *
 334  *      Free the virtually contiguous memory area starting at @addr, as
 335  *      obtained from vmalloc(), vmalloc_32() or __vmalloc().
 336  *
 337  *      May not be called in interrupt context.
 338  */
 339 void vfree(void *addr)
 340 {
 341         BUG_ON(in_interrupt());
 342         __vunmap(addr, 1);
 343 }
 344
 345 EXPORT_SYMBOL(vfree);
 346
 347 /**
 348  *      vunmap  -  release virtual mapping obtained by vmap()
 349  *
 350  *      @addr:          memory base address
 351  *
 352  *      Free the virtually contiguous memory area starting at @addr,
 353  *      which was created from the page array passed to vmap().
 354  *
 355  *      May not be called in interrupt context.
 356  */
 357 void vunmap(void *addr)
 358 {
 359         BUG_ON(in_interrupt());
 360         __vunmap(addr, 0);
 361 }
 362
 363 EXPORT_SYMBOL(vunmap);
 364
 365 /**
 366  *      vmap  -  map an array of pages into virtually contiguous space
 367  *
 368  *      @pages:         array of page pointers
 369  *      @count:         number of pages to map
 370  *      @flags:         vm_area->flags
 371  *      @prot:          page protection for the mapping
 372  *
 373  *      Maps @count pages from @pages into contiguous kernel virtual
 374  *      space.
 375  */
 376 void *vmap(struct page **pages, unsigned int count,
 377                 unsigned long flags, pgprot_t prot)
 378 {
 379         struct vm_struct *area;
 380
 381         if (count > num_physpages)
 382                 return NULL;
 383
 384         area = get_vm_area((count << PAGE_SHIFT), flags);
 385         if (!area)
 386                 return NULL;
 387         if (map_vm_area(area, prot, &pages)) {
 388                 vunmap(area->addr);
 389                 return NULL;
 390         }
 391
 392         return area->addr;
 393 }
 394
 395 EXPORT_SYMBOL(vmap);
 396
 397 void *__vmalloc_area(struct vm_struct *area, unsigned int __nocast gfp_mask, pgprot_t prot)
 398 {
 399         struct page **pages;
 400         unsigned int nr_pages, array_size, i;
 401
 402         nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
 403         array_size = (nr_pages * sizeof(struct page *));
 404
 405         area->nr_pages = nr_pages;
 406         /* Please note that the recursion is strictly bounded. */
 407         if (array_size > PAGE_SIZE)
 408                 pages = __vmalloc(array_size, gfp_mask, PAGE_KERNEL);
 409         else
 410                 pages = kmalloc(array_size, (gfp_mask & ~__GFP_HIGHMEM));
 411         area->pages = pages;
 412         if (!area->pages) {
 413                 remove_vm_area(area->addr);
 414                 kfree(area);
 415                 return NULL;
 416         }
 417         memset(area->pages, 0, array_size);
 418
 419         for (i = 0; i < area->nr_pages; i++) {
 420                 area->pages[i] = alloc_page(gfp_mask);
 421                 if (unlikely(!area->pages[i])) {
 422                         /* Successfully allocated i pages, free them in __vunmap() */
 423                         area->nr_pages = i;
 424                         goto fail;
 425                 }
 426         }
 427
 428         if (map_vm_area(area, prot, &pages))
 429                 goto fail;
 430         return area->addr;
 431
 432 fail:
 433         vfree(area->addr);
 434         return NULL;
 435 }
 436
 437 /**
 438  *      __vmalloc  -  allocate virtually contiguous memory
 439  *
 440  *      @size:          allocation size
 441  *      @gfp_mask:      flags for the page level allocator
 442  *      @prot:          protection mask for the allocated pages
 443  *
 444  *      Allocate enough pages to cover @size from the page level
 445  *      allocator with @gfp_mask flags.  Map them into contiguous
 446  *      kernel virtual space, using a pagetable protection of @prot.
 447  */
 448 void *__vmalloc(unsigned long size, unsigned int __nocast gfp_mask, pgprot_t prot)
 449 {
 450         struct vm_struct *area;
 451
 452         size = PAGE_ALIGN(size);
 453         if (!size || (size >> PAGE_SHIFT) > num_physpages)
 454                 return NULL;
 455
 456         area = get_vm_area(size, VM_ALLOC);
 457         if (!area)
 458                 return NULL;
 459
 460         return __vmalloc_area(area, gfp_mask, prot);
 461 }
 462
 463 EXPORT_SYMBOL(__vmalloc);
 464
 465 /**
 466  *      vmalloc  -  allocate virtually contiguous memory
 467  *
 468  *      @size:          allocation size
 469  *
 470  *      Allocate enough pages to cover @size from the page level
 471  *      allocator and map them into contiguous kernel virtual space.
 472  *
 473  *      For tight cotrol over page level allocator and protection flags
 474  *      use __vmalloc() instead.
 475  */
 476 void *vmalloc(unsigned long size)
 477 {
 478        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 479 }
 480
 481 EXPORT_SYMBOL(vmalloc);
 482
 483 #ifndef PAGE_KERNEL_EXEC
 484 # define PAGE_KERNEL_EXEC PAGE_KERNEL
 485 #endif
 486
 487 /**
 488  *      vmalloc_exec  -  allocate virtually contiguous, executable memory
 489  *
 490  *      @size:          allocation size
 491  *
 492  *      Kernel-internal function to allocate enough pages to cover @size
 493  *      the page level allocator and map them into contiguous and
 494  *      executable kernel virtual space.
 495  *
 496  *      For tight cotrol over page level allocator and protection flags
 497  *      use __vmalloc() instead.
 498  */
 499
 500 void *vmalloc_exec(unsigned long size)
 501 {
 502         return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);
 503 }
 504
 505 /**
 506  *      vmalloc_32  -  allocate virtually contiguous memory (32bit addressable)
 507  *
 508  *      @size:          allocation size
 509  *
 510  *      Allocate enough 32bit PA addressable pages to cover @size from the
 511  *      page level allocator and map them into contiguous kernel virtual space.
 512  */
 513 void *vmalloc_32(unsigned long size)
 514 {
 515         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 516 }
 517
 518 EXPORT_SYMBOL(vmalloc_32);
 519
 520 long vread(char *buf, char *addr, unsigned long count)
 521 {
 522         struct vm_struct *tmp;
 523         char *vaddr, *buf_start = buf;
 524         unsigned long n;
 525
 526         /* Don't allow overflow */
 527         if ((unsigned long) addr + count < count)
 528                 count = -(unsigned long) addr;
 529
 530         read_lock(&vmlist_lock);
 531         for (tmp = vmlist; tmp; tmp = tmp->next) {
 532                 vaddr = (char *) tmp->addr;
 533                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 534                         continue;
 535                 while (addr < vaddr) {
 536                         if (count == 0)
 537                                 goto finished;
 538                         *buf = '\0';
 539                         buf++;
 540                         addr++;
 541                         count--;
 542                 }
 543                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 544                 do {
 545                         if (count == 0)
 546                                 goto finished;
 547                         *buf = *addr;
 548                         buf++;
 549                         addr++;
 550                         count--;
 551                 } while (--n > 0);
 552         }
 553 finished:
 554         read_unlock(&vmlist_lock);
 555         return buf - buf_start;
 556 }
 557
 558 long vwrite(char *buf, char *addr, unsigned long count)
 559 {
 560         struct vm_struct *tmp;
 561         char *vaddr, *buf_start = buf;
 562         unsigned long n;
 563
 564         /* Don't allow overflow */
 565         if ((unsigned long) addr + count < count)
 566                 count = -(unsigned long) addr;
 567
 568         read_lock(&vmlist_lock);
 569         for (tmp = vmlist; tmp; tmp = tmp->next) {
 570                 vaddr = (char *) tmp->addr;
 571                 if (addr >= vaddr + tmp->size - PAGE_SIZE)
 572                         continue;
 573                 while (addr < vaddr) {
 574                         if (count == 0)
 575                                 goto finished;
 576                         buf++;
 577                         addr++;
 578                         count--;
 579                 }
 580                 n = vaddr + tmp->size - PAGE_SIZE - addr;
 581                 do {
 582                         if (count == 0)
 583                                 goto finished;
 584                         *addr = *buf;
 585                         buf++;
 586                         addr++;
 587                         count--;
 588                 } while (--n > 0);
 589         }
 590 finished:
 591         read_unlock(&vmlist_lock);
 592         return buf - buf_start;
 593 }