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