mm/nommu.c

   1 /*
   2  *  linux/mm/nommu.c
   3  *
   4  *  Replacement code for mm functions to support CPU's that don't
   5  *  have any form of memory management unit (thus no virtual memory).
   6  *
   7  *  Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com>
   8  *  Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org>
   9  *  Copyright (c) 2002      Greg Ungerer <gerg@snapgear.com>
  10  */
  11
  12 #include <linux/mm.h>
  13 #include <linux/mman.h>
  14 #include <linux/swap.h>
  15 #include <linux/smp_lock.h>
  16 #include <linux/highmem.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/slab.h>
  19 #include <linux/vmalloc.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/backing-dev.h>
  22
  23 #include <asm/pgalloc.h>
  24 #include <asm/uaccess.h>
  25 #include <asm/tlb.h>
  26 #include <asm/tlbflush.h>
  27
  28 void *high_memory;
  29 struct page *mem_map;
  30 unsigned long max_mapnr;
  31 unsigned long num_physpages;
  32 unsigned long askedalloc, realalloc;
  33 atomic_t vm_committed_space = ATOMIC_INIT(0);
  34 int sysctl_overcommit_memory; /* default is heuristic overcommit */
  35 int sysctl_overcommit_ratio = 50; /* default is 50% */
  36
  37 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
  38 EXPORT_SYMBOL(sysctl_max_map_count);
  39
  40 /*
  41  * Handle all mappings that got truncated by a "truncate()"
  42  * system call.
  43  *
  44  * NOTE! We have to be ready to update the memory sharing
  45  * between the file and the memory map for a potential last
  46  * incomplete page.  Ugly, but necessary.
  47  */
  48 int vmtruncate(struct inode *inode, loff_t offset)
  49 {
  50         struct address_space *mapping = inode->i_mapping;
  51         unsigned long limit;
  52
  53         if (inode->i_size < offset)
  54                 goto do_expand;
  55         i_size_write(inode, offset);
  56
  57         truncate_inode_pages(mapping, offset);
  58         goto out_truncate;
  59
  60 do_expand:
  61         limit = current->rlim[RLIMIT_FSIZE].rlim_cur;
  62         if (limit != RLIM_INFINITY && offset > limit)
  63                 goto out_sig;
  64         if (offset > inode->i_sb->s_maxbytes)
  65                 goto out;
  66         i_size_write(inode, offset);
  67
  68 out_truncate:
  69         if (inode->i_op && inode->i_op->truncate)
  70                 inode->i_op->truncate(inode);
  71         return 0;
  72 out_sig:
  73         send_sig(SIGXFSZ, current, 0);
  74 out:
  75         return -EFBIG;
  76 }
  77
  78 /*
  79  * Return the total memory allocated for this pointer, not
  80  * just what the caller asked for.
  81  *
  82  * Doesn't have to be accurate, i.e. may have races.
  83  */
  84 unsigned int kobjsize(const void *objp)
  85 {
  86         struct page *page;
  87
  88         if (!objp || !((page = virt_to_page(objp))))
  89                 return 0;
  90
  91         if (PageSlab(page))
  92                 return ksize(objp);
  93
  94         BUG_ON(page->index < 0);
  95         BUG_ON(page->index >= MAX_ORDER);
  96
  97         return (PAGE_SIZE << page->index);
  98 }
  99
 100 /*
 101  * The nommu dodgy version :-)
 102  */
 103 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 104         unsigned long start, int len, int write, int force,
 105         struct page **pages, struct vm_area_struct **vmas)
 106 {
 107         int i;
 108         static struct vm_area_struct dummy_vma;
 109
 110         for (i = 0; i < len; i++) {
 111                 if (pages) {
 112                         pages[i] = virt_to_page(start);
 113                         if (pages[i])
 114                                 page_cache_get(pages[i]);
 115                 }
 116                 if (vmas)
 117                         vmas[i] = &dummy_vma;
 118                 start += PAGE_SIZE;
 119         }
 120         return(i);
 121 }
 122
 123 rwlock_t vmlist_lock = RW_LOCK_UNLOCKED;
 124 struct vm_struct *vmlist;
 125
 126 void vfree(void *addr)
 127 {
 128         kfree(addr);
 129 }
 130
 131 void *__vmalloc(unsigned long size, int gfp_mask, pgprot_t prot)
 132 {
 133         /*
 134          * kmalloc doesn't like __GFP_HIGHMEM for some reason
 135          */
 136         return kmalloc(size, gfp_mask & ~__GFP_HIGHMEM);
 137 }
 138
 139 struct page * vmalloc_to_page(void *addr)
 140 {
 141         return virt_to_page(addr);
 142 }
 143
 144 long vread(char *buf, char *addr, unsigned long count)
 145 {
 146         memcpy(buf, addr, count);
 147         return count;
 148 }
 149
 150 long vwrite(char *buf, char *addr, unsigned long count)
 151 {
 152         /* Don't allow overflow */
 153         if ((unsigned long) addr + count < count)
 154                 count = -(unsigned long) addr;
 155
 156         memcpy(addr, buf, count);
 157         return(count);
 158 }
 159
 160 /*
 161  *      vmalloc  -  allocate virtually continguos memory
 162  *
 163  *      @size:          allocation size
 164  *
 165  *      Allocate enough pages to cover @size from the page level
 166  *      allocator and map them into continguos kernel virtual space.
 167  *
 168  *      For tight cotrol over page level allocator and protection flags
 169  *      use __vmalloc() instead.
 170  */
 171 void *vmalloc(unsigned long size)
 172 {
 173        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 174 }
 175
 176 /*
 177  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
 178  *
 179  *      @size:          allocation size
 180  *
 181  *      Allocate enough 32bit PA addressable pages to cover @size from the
 182  *      page level allocator and map them into continguos kernel virtual space.
 183  */
 184 void *vmalloc_32(unsigned long size)
 185 {
 186         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 187 }
 188
 189 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 190 {
 191         BUG();
 192         return NULL;
 193 }
 194
 195 void vunmap(void *addr)
 196 {
 197         BUG();
 198 }
 199
 200 /*
 201  *  sys_brk() for the most part doesn't need the global kernel
 202  *  lock, except when an application is doing something nasty
 203  *  like trying to un-brk an area that has already been mapped
 204  *  to a regular file.  in this case, the unmapping will need
 205  *  to invoke file system routines that need the global lock.
 206  */
 207 asmlinkage unsigned long sys_brk(unsigned long brk)
 208 {
 209         struct mm_struct *mm = current->mm;
 210
 211         if (brk < mm->end_code || brk < mm->start_brk || brk > mm->context.end_brk)
 212                 return mm->brk;
 213
 214         if (mm->brk == brk)
 215                 return mm->brk;
 216
 217         /*
 218          * Always allow shrinking brk
 219          */
 220         if (brk <= mm->brk) {
 221                 mm->brk = brk;
 222                 return brk;
 223         }
 224
 225         /*
 226          * Ok, looks good - let it rip.
 227          */
 228         return mm->brk = brk;
 229 }
 230
 231 /*
 232  * Combine the mmap "prot" and "flags" argument into one "vm_flags" used
 233  * internally. Essentially, translate the "PROT_xxx" and "MAP_xxx" bits
 234  * into "VM_xxx".
 235  */
 236 static inline unsigned long calc_vm_flags(unsigned long prot, unsigned long flags)
 237 {
 238 #define _trans(x,bit1,bit2) \
 239 ((bit1==bit2)?(x&bit1):(x&bit1)?bit2:0)
 240
 241         unsigned long prot_bits, flag_bits;
 242         prot_bits =
 243                 _trans(prot, PROT_READ, VM_READ) |
 244                 _trans(prot, PROT_WRITE, VM_WRITE) |
 245                 _trans(prot, PROT_EXEC, VM_EXEC);
 246         flag_bits =
 247                 _trans(flags, MAP_GROWSDOWN, VM_GROWSDOWN) |
 248                 _trans(flags, MAP_DENYWRITE, VM_DENYWRITE) |
 249                 _trans(flags, MAP_EXECUTABLE, VM_EXECUTABLE);
 250         return prot_bits | flag_bits;
 251 #undef _trans
 252 }
 253
 254 #ifdef DEBUG
 255 static void show_process_blocks(void)
 256 {
 257         struct mm_tblock_struct *tblock;
 258
 259         printk("Process blocks %d:", current->pid);
 260
 261         for (tblock = &current->mm->context.tblock; tblock; tblock = tblock->next) {
 262                 printk(" %p: %p", tblock, tblock->rblock);
 263                 if (tblock->rblock)
 264                         printk(" (%d @%p #%d)", kobjsize(tblock->rblock->kblock), tblock->rblock->kblock, tblock->rblock->refcount);
 265                 printk(tblock->next ? " ->" : ".\n");
 266         }
 267 }
 268 #endif /* DEBUG */
 269
 270 unsigned long do_mmap_pgoff(
 271         struct file * file,
 272         unsigned long addr,
 273         unsigned long len,
 274         unsigned long prot,
 275         unsigned long flags,
 276         unsigned long pgoff)
 277 {
 278         void * result;
 279         struct mm_tblock_struct * tblock;
 280         unsigned int vm_flags;
 281
 282         /*
 283          * Get the !CONFIG_MMU specific checks done first
 284          */
 285         if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && (file)) {
 286                 printk("MAP_SHARED not supported (cannot write mappings to disk)\n");
 287                 return -EINVAL;
 288         }
 289
 290         if ((prot & PROT_WRITE) && (flags & MAP_PRIVATE)) {
 291                 printk("Private writable mappings not supported\n");
 292                 return -EINVAL;
 293         }
 294
 295         /*
 296          *      now all the standard checks
 297          */
 298         if (file && (!file->f_op || !file->f_op->mmap))
 299                 return -ENODEV;
 300
 301         if (PAGE_ALIGN(len) == 0)
 302                 return addr;
 303
 304         if (len > TASK_SIZE)
 305                 return -EINVAL;
 306
 307         /* offset overflow? */
 308         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 309                 return -EINVAL;
 310
 311         /* Do simple checking here so the lower-level routines won't have
 312          * to. we assume access permissions have been handled by the open
 313          * of the memory object, so we don't do any here.
 314          */
 315         vm_flags = calc_vm_flags(prot,flags) /* | mm->def_flags */ | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 316
 317         /*
 318          * determine the object being mapped and call the appropriate
 319          * specific mapper.
 320          */
 321         if (file) {
 322                 struct vm_area_struct vma;
 323                 int error;
 324
 325                 if (!file->f_op)
 326                         return -ENODEV;
 327
 328                 vma.vm_start = addr;
 329                 vma.vm_end = addr + len;
 330                 vma.vm_flags = vm_flags;
 331                 vma.vm_pgoff = pgoff;
 332
 333 #ifdef MAGIC_ROM_PTR
 334                 /* First, try simpler routine designed to give us a ROM pointer. */
 335
 336                 if (file->f_op->romptr && !(prot & PROT_WRITE)) {
 337                         error = file->f_op->romptr(file, &vma);
 338 #ifdef DEBUG
 339                         printk("romptr mmap returned %d, start 0x%.8x\n", error,
 340                                         vma.vm_start);
 341 #endif
 342                         if (!error)
 343                                 return vma.vm_start;
 344                         else if (error != -ENOSYS)
 345                                 return error;
 346                 } else
 347 #endif /* MAGIC_ROM_PTR */
 348                 /* Then try full mmap routine, which might return a RAM pointer,
 349                    or do something truly complicated. */
 350
 351                 if (file->f_op->mmap) {
 352                         error = file->f_op->mmap(file, &vma);
 353
 354 #ifdef DEBUG
 355                         printk("f_op->mmap() returned %d/%lx\n", error, vma.vm_start);
 356 #endif
 357                         if (!error)
 358                                 return vma.vm_start;
 359                         else if (error != -ENOSYS)
 360                                 return error;
 361                 } else
 362                         return -ENODEV; /* No mapping operations defined */
 363
 364                 /* An ENOSYS error indicates that mmap isn't possible (as opposed to
 365                    tried but failed) so we'll fall through to the copy. */
 366         }
 367
 368         tblock = (struct mm_tblock_struct *)
 369                         kmalloc(sizeof(struct mm_tblock_struct), GFP_KERNEL);
 370         if (!tblock) {
 371                 printk("Allocation of tblock for %lu byte allocation from process %d failed\n", len, current->pid);
 372                 show_free_areas();
 373                 return -ENOMEM;
 374         }
 375
 376         tblock->rblock = (struct mm_rblock_struct *)
 377                         kmalloc(sizeof(struct mm_rblock_struct), GFP_KERNEL);
 378
 379         if (!tblock->rblock) {
 380                 printk("Allocation of rblock for %lu byte allocation from process %d failed\n", len, current->pid);
 381                 show_free_areas();
 382                 kfree(tblock);
 383                 return -ENOMEM;
 384         }
 385
 386         result = kmalloc(len, GFP_KERNEL);
 387         if (!result) {
 388                 printk("Allocation of length %lu from process %d failed\n", len,
 389                                 current->pid);
 390                 show_free_areas();
 391                 kfree(tblock->rblock);
 392                 kfree(tblock);
 393                 return -ENOMEM;
 394         }
 395
 396         tblock->rblock->refcount = 1;
 397         tblock->rblock->kblock = result;
 398         tblock->rblock->size = len;
 399
 400         realalloc += kobjsize(result);
 401         askedalloc += len;
 402
 403 #ifdef WARN_ON_SLACK
 404         if ((len+WARN_ON_SLACK) <= kobjsize(result))
 405                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n", len, current->pid, kobjsize(result)-len);
 406 #endif
 407
 408         if (file) {
 409                 int error;
 410                 mm_segment_t old_fs = get_fs();
 411                 set_fs(KERNEL_DS);
 412                 error = file->f_op->read(file, (char *) result, len, &file->f_pos);
 413                 set_fs(old_fs);
 414                 if (error < 0) {
 415                         kfree(result);
 416                         kfree(tblock->rblock);
 417                         kfree(tblock);
 418                         return error;
 419                 }
 420                 if (error < len)
 421                         memset(result+error, '\0', len-error);
 422         } else {
 423                 memset(result, '\0', len);
 424         }
 425
 426         realalloc += kobjsize(tblock);
 427         askedalloc += sizeof(struct mm_tblock_struct);
 428
 429         realalloc += kobjsize(tblock->rblock);
 430         askedalloc += sizeof(struct mm_rblock_struct);
 431
 432         tblock->next = current->mm->context.tblock.next;
 433         current->mm->context.tblock.next = tblock;
 434
 435 #ifdef DEBUG
 436         printk("do_mmap:\n");
 437         show_process_blocks();
 438 #endif
 439
 440         return (unsigned long)result;
 441 }
 442
 443 int do_munmap(struct mm_struct * mm, unsigned long addr, size_t len)
 444 {
 445         struct mm_tblock_struct * tblock, *tmp;
 446
 447 #ifdef MAGIC_ROM_PTR
 448         /*
 449          * For efficiency's sake, if the pointer is obviously in ROM,
 450          * don't bother walking the lists to free it.
 451          */
 452         if (is_in_rom(addr))
 453                 return 0;
 454 #endif
 455
 456 #ifdef DEBUG
 457         printk("do_munmap:\n");
 458 #endif
 459
 460         tmp = &mm->context.tblock; /* dummy head */
 461         while ((tblock=tmp->next) && tblock->rblock &&
 462                         tblock->rblock->kblock != (void*)addr)
 463                 tmp = tblock;
 464
 465         if (!tblock) {
 466                 printk("munmap of non-mmaped memory by process %d (%s): %p\n",
 467                                 current->pid, current->comm, (void*)addr);
 468                 return -EINVAL;
 469         }
 470         if (tblock->rblock) {
 471                 if (!--tblock->rblock->refcount) {
 472                         if (tblock->rblock->kblock) {
 473                                 realalloc -= kobjsize(tblock->rblock->kblock);
 474                                 askedalloc -= tblock->rblock->size;
 475                                 kfree(tblock->rblock->kblock);
 476                         }
 477
 478                         realalloc -= kobjsize(tblock->rblock);
 479                         askedalloc -= sizeof(struct mm_rblock_struct);
 480                         kfree(tblock->rblock);
 481                 }
 482         }
 483         tmp->next = tblock->next;
 484         realalloc -= kobjsize(tblock);
 485         askedalloc -= sizeof(struct mm_tblock_struct);
 486         kfree(tblock);
 487
 488 #ifdef DEBUG
 489         show_process_blocks();
 490 #endif
 491
 492         return 0;
 493 }
 494
 495 /* Release all mmaps. */
 496 void exit_mmap(struct mm_struct * mm)
 497 {
 498         struct mm_tblock_struct *tmp;
 499
 500         if (!mm)
 501                 return;
 502
 503 #ifdef DEBUG
 504         printk("Exit_mmap:\n");
 505 #endif
 506
 507         while((tmp = mm->context.tblock.next)) {
 508                 if (tmp->rblock) {
 509                         if (!--tmp->rblock->refcount) {
 510                                 if (tmp->rblock->kblock) {
 511                                         realalloc -= kobjsize(tmp->rblock->kblock);
 512                                         askedalloc -= tmp->rblock->size;
 513                                         kfree(tmp->rblock->kblock);
 514                                 }
 515                                 realalloc -= kobjsize(tmp->rblock);
 516                                 askedalloc -= sizeof(struct mm_rblock_struct);
 517                                 kfree(tmp->rblock);
 518                         }
 519                         tmp->rblock = 0;
 520                 }
 521                 mm->context.tblock.next = tmp->next;
 522                 realalloc -= kobjsize(tmp);
 523                 askedalloc -= sizeof(struct mm_tblock_struct);
 524                 kfree(tmp);
 525         }
 526
 527 #ifdef DEBUG
 528         show_process_blocks();
 529 #endif
 530 }
 531
 532 asmlinkage long sys_munmap(unsigned long addr, size_t len)
 533 {
 534         int ret;
 535         struct mm_struct *mm = current->mm;
 536
 537         down_write(&mm->mmap_sem);
 538         ret = do_munmap(mm, addr, len);
 539         up_write(&mm->mmap_sem);
 540         return ret;
 541 }
 542
 543 unsigned long do_brk(unsigned long addr, unsigned long len)
 544 {
 545         return -ENOMEM;
 546 }
 547
 548 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
 549 {
 550         return NULL;
 551 }
 552
 553 struct page * follow_page(struct mm_struct *mm, unsigned long addr, int write)
 554 {
 555         return NULL;
 556 }
 557
 558 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 559 {
 560         return NULL;
 561 }
 562
 563 int remap_page_range(struct vm_area_struct *vma, unsigned long from,
 564                 unsigned long to, unsigned long size, pgprot_t prot)
 565 {
 566         return -EPERM;
 567 }
 568
 569 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
 570         unsigned long len, unsigned long pgoff, unsigned long flags)
 571 {
 572         return -ENOMEM;
 573 }
 574
 575 void swap_unplug_io_fn(struct backing_dev_info *)
 576 {
 577 }