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 #include <linux/syscalls.h>
  23
  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 = OVERCOMMIT_GUESS; /* heuristic overcommit */
  35 int sysctl_overcommit_ratio = 50; /* default is 50% */
  36 int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT;
  37
  38 EXPORT_SYMBOL(sysctl_max_map_count);
  39 EXPORT_SYMBOL(mem_map);
  40
  41 /*
  42  * Handle all mappings that got truncated by a "truncate()"
  43  * system call.
  44  *
  45  * NOTE! We have to be ready to update the memory sharing
  46  * between the file and the memory map for a potential last
  47  * incomplete page.  Ugly, but necessary.
  48  */
  49 int vmtruncate(struct inode *inode, loff_t offset)
  50 {
  51         struct address_space *mapping = inode->i_mapping;
  52         unsigned long limit;
  53
  54         if (inode->i_size < offset)
  55                 goto do_expand;
  56         i_size_write(inode, offset);
  57
  58         truncate_inode_pages(mapping, offset);
  59         goto out_truncate;
  60
  61 do_expand:
  62         limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
  63         if (limit != RLIM_INFINITY && offset > limit)
  64                 goto out_sig;
  65         if (offset > inode->i_sb->s_maxbytes)
  66                 goto out;
  67         i_size_write(inode, offset);
  68
  69 out_truncate:
  70         if (inode->i_op && inode->i_op->truncate)
  71                 inode->i_op->truncate(inode);
  72         return 0;
  73 out_sig:
  74         send_sig(SIGXFSZ, current, 0);
  75 out:
  76         return -EFBIG;
  77 }
  78
  79 EXPORT_SYMBOL(vmtruncate);
  80
  81 /*
  82  * Return the total memory allocated for this pointer, not
  83  * just what the caller asked for.
  84  *
  85  * Doesn't have to be accurate, i.e. may have races.
  86  */
  87 unsigned int kobjsize(const void *objp)
  88 {
  89         struct page *page;
  90
  91         if (!objp || !((page = virt_to_page(objp))))
  92                 return 0;
  93
  94         if (PageSlab(page))
  95                 return ksize(objp);
  96
  97         BUG_ON(page->index < 0);
  98         BUG_ON(page->index >= MAX_ORDER);
  99
 100         return (PAGE_SIZE << page->index);
 101 }
 102
 103 /*
 104  * The nommu dodgy version :-)
 105  */
 106 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
 107         unsigned long start, int len, int write, int force,
 108         struct page **pages, struct vm_area_struct **vmas)
 109 {
 110         int i;
 111         static struct vm_area_struct dummy_vma;
 112
 113         for (i = 0; i < len; i++) {
 114                 if (pages) {
 115                         pages[i] = virt_to_page(start);
 116                         if (pages[i])
 117                                 page_cache_get(pages[i]);
 118                 }
 119                 if (vmas)
 120                         vmas[i] = &dummy_vma;
 121                 start += PAGE_SIZE;
 122         }
 123         return(i);
 124 }
 125
 126 rwlock_t vmlist_lock = RW_LOCK_UNLOCKED;
 127 struct vm_struct *vmlist;
 128
 129 void vfree(void *addr)
 130 {
 131         kfree(addr);
 132 }
 133
 134 void *__vmalloc(unsigned long size, int gfp_mask, pgprot_t prot)
 135 {
 136         /*
 137          * kmalloc doesn't like __GFP_HIGHMEM for some reason
 138          */
 139         return kmalloc(size, gfp_mask & ~__GFP_HIGHMEM);
 140 }
 141
 142 struct page * vmalloc_to_page(void *addr)
 143 {
 144         return virt_to_page(addr);
 145 }
 146
 147 unsigned long vmalloc_to_pfn(void *addr)
 148 {
 149         return page_to_pfn(virt_to_page(addr));
 150 }
 151
 152
 153 long vread(char *buf, char *addr, unsigned long count)
 154 {
 155         memcpy(buf, addr, count);
 156         return count;
 157 }
 158
 159 long vwrite(char *buf, char *addr, unsigned long count)
 160 {
 161         /* Don't allow overflow */
 162         if ((unsigned long) addr + count < count)
 163                 count = -(unsigned long) addr;
 164
 165         memcpy(addr, buf, count);
 166         return(count);
 167 }
 168
 169 /*
 170  *      vmalloc  -  allocate virtually continguos memory
 171  *
 172  *      @size:          allocation size
 173  *
 174  *      Allocate enough pages to cover @size from the page level
 175  *      allocator and map them into continguos kernel virtual space.
 176  *
 177  *      For tight cotrol over page level allocator and protection flags
 178  *      use __vmalloc() instead.
 179  */
 180 void *vmalloc(unsigned long size)
 181 {
 182        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL);
 183 }
 184
 185 /*
 186  *      vmalloc_32  -  allocate virtually continguos memory (32bit addressable)
 187  *
 188  *      @size:          allocation size
 189  *
 190  *      Allocate enough 32bit PA addressable pages to cover @size from the
 191  *      page level allocator and map them into continguos kernel virtual space.
 192  */
 193 void *vmalloc_32(unsigned long size)
 194 {
 195         return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL);
 196 }
 197
 198 void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot)
 199 {
 200         BUG();
 201         return NULL;
 202 }
 203
 204 void vunmap(void *addr)
 205 {
 206         BUG();
 207 }
 208
 209 /*
 210  *  sys_brk() for the most part doesn't need the global kernel
 211  *  lock, except when an application is doing something nasty
 212  *  like trying to un-brk an area that has already been mapped
 213  *  to a regular file.  in this case, the unmapping will need
 214  *  to invoke file system routines that need the global lock.
 215  */
 216 asmlinkage unsigned long sys_brk(unsigned long brk)
 217 {
 218         struct mm_struct *mm = current->mm;
 219
 220         if (brk < mm->end_code || brk < mm->start_brk || brk > mm->context.end_brk)
 221                 return mm->brk;
 222
 223         if (mm->brk == brk)
 224                 return mm->brk;
 225
 226         /*
 227          * Always allow shrinking brk
 228          */
 229         if (brk <= mm->brk) {
 230                 mm->brk = brk;
 231                 return brk;
 232         }
 233
 234         /*
 235          * Ok, looks good - let it rip.
 236          */
 237         return mm->brk = brk;
 238 }
 239
 240 /*
 241  * Combine the mmap "prot" and "flags" argument into one "vm_flags" used
 242  * internally. Essentially, translate the "PROT_xxx" and "MAP_xxx" bits
 243  * into "VM_xxx".
 244  */
 245 static inline unsigned long calc_vm_flags(unsigned long prot, unsigned long flags)
 246 {
 247 #define _trans(x,bit1,bit2) \
 248 ((bit1==bit2)?(x&bit1):(x&bit1)?bit2:0)
 249
 250         unsigned long prot_bits, flag_bits;
 251         prot_bits =
 252                 _trans(prot, PROT_READ, VM_READ) |
 253                 _trans(prot, PROT_WRITE, VM_WRITE) |
 254                 _trans(prot, PROT_EXEC, VM_EXEC);
 255         flag_bits =
 256                 _trans(flags, MAP_GROWSDOWN, VM_GROWSDOWN) |
 257                 _trans(flags, MAP_DENYWRITE, VM_DENYWRITE) |
 258                 _trans(flags, MAP_EXECUTABLE, VM_EXECUTABLE);
 259         return prot_bits | flag_bits;
 260 #undef _trans
 261 }
 262
 263 #ifdef DEBUG
 264 static void show_process_blocks(void)
 265 {
 266         struct mm_tblock_struct *tblock;
 267
 268         printk("Process blocks %d:", current->pid);
 269
 270         for (tblock = &current->mm->context.tblock; tblock; tblock = tblock->next) {
 271                 printk(" %p: %p", tblock, tblock->rblock);
 272                 if (tblock->rblock)
 273                         printk(" (%d @%p #%d)", kobjsize(tblock->rblock->kblock), tblock->rblock->kblock, tblock->rblock->refcount);
 274                 printk(tblock->next ? " ->" : ".\n");
 275         }
 276 }
 277 #endif /* DEBUG */
 278
 279 unsigned long do_mmap_pgoff(
 280         struct file * file,
 281         unsigned long addr,
 282         unsigned long len,
 283         unsigned long prot,
 284         unsigned long flags,
 285         unsigned long pgoff)
 286 {
 287         void * result;
 288         struct mm_tblock_struct * tblock;
 289         unsigned int vm_flags;
 290
 291         /*
 292          * Get the !CONFIG_MMU specific checks done first
 293          */
 294         if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && (file)) {
 295                 printk("MAP_SHARED not supported (cannot write mappings to disk)\n");
 296                 return -EINVAL;
 297         }
 298
 299         if ((prot & PROT_WRITE) && (flags & MAP_PRIVATE)) {
 300                 printk("Private writable mappings not supported\n");
 301                 return -EINVAL;
 302         }
 303
 304         /*
 305          *      now all the standard checks
 306          */
 307         if (file && (!file->f_op || !file->f_op->mmap))
 308                 return -ENODEV;
 309
 310         if (PAGE_ALIGN(len) == 0)
 311                 return addr;
 312
 313         if (len > TASK_SIZE)
 314                 return -EINVAL;
 315
 316         /* offset overflow? */
 317         if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
 318                 return -EINVAL;
 319
 320         /* Do simple checking here so the lower-level routines won't have
 321          * to. we assume access permissions have been handled by the open
 322          * of the memory object, so we don't do any here.
 323          */
 324         vm_flags = calc_vm_flags(prot,flags) /* | mm->def_flags */ | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
 325
 326         /*
 327          * determine the object being mapped and call the appropriate
 328          * specific mapper.
 329          */
 330         if (file) {
 331                 struct vm_area_struct vma;
 332                 int error;
 333
 334                 if (!file->f_op)
 335                         return -ENODEV;
 336
 337                 vma.vm_start = addr;
 338                 vma.vm_end = addr + len;
 339                 vma.vm_flags = vm_flags;
 340                 vma.vm_pgoff = pgoff;
 341
 342 #ifdef MAGIC_ROM_PTR
 343                 /* First, try simpler routine designed to give us a ROM pointer. */
 344
 345                 if (file->f_op->romptr && !(prot & PROT_WRITE)) {
 346                         error = file->f_op->romptr(file, &vma);
 347 #ifdef DEBUG
 348                         printk("romptr mmap returned %d, start 0x%.8x\n", error,
 349                                         vma.vm_start);
 350 #endif
 351                         if (!error)
 352                                 return vma.vm_start;
 353                         else if (error != -ENOSYS)
 354                                 return error;
 355                 } else
 356 #endif /* MAGIC_ROM_PTR */
 357                 /* Then try full mmap routine, which might return a RAM pointer,
 358                    or do something truly complicated. */
 359
 360                 if (file->f_op->mmap) {
 361                         error = file->f_op->mmap(file, &vma);
 362
 363 #ifdef DEBUG
 364                         printk("f_op->mmap() returned %d/%lx\n", error, vma.vm_start);
 365 #endif
 366                         if (!error)
 367                                 return vma.vm_start;
 368                         else if (error != -ENOSYS)
 369                                 return error;
 370                 } else
 371                         return -ENODEV; /* No mapping operations defined */
 372
 373                 /* An ENOSYS error indicates that mmap isn't possible (as opposed to
 374                    tried but failed) so we'll fall through to the copy. */
 375         }
 376
 377         tblock = (struct mm_tblock_struct *)
 378                         kmalloc(sizeof(struct mm_tblock_struct), GFP_KERNEL);
 379         if (!tblock) {
 380                 printk("Allocation of tblock for %lu byte allocation from process %d failed\n", len, current->pid);
 381                 show_free_areas();
 382                 return -ENOMEM;
 383         }
 384
 385         tblock->rblock = (struct mm_rblock_struct *)
 386                         kmalloc(sizeof(struct mm_rblock_struct), GFP_KERNEL);
 387
 388         if (!tblock->rblock) {
 389                 printk("Allocation of rblock for %lu byte allocation from process %d failed\n", len, current->pid);
 390                 show_free_areas();
 391                 kfree(tblock);
 392                 return -ENOMEM;
 393         }
 394
 395         result = kmalloc(len, GFP_KERNEL);
 396         if (!result) {
 397                 printk("Allocation of length %lu from process %d failed\n", len,
 398                                 current->pid);
 399                 show_free_areas();
 400                 kfree(tblock->rblock);
 401                 kfree(tblock);
 402                 return -ENOMEM;
 403         }
 404
 405         tblock->rblock->refcount = 1;
 406         tblock->rblock->kblock = result;
 407         tblock->rblock->size = len;
 408
 409         realalloc += kobjsize(result);
 410         askedalloc += len;
 411
 412 #ifdef WARN_ON_SLACK
 413         if ((len+WARN_ON_SLACK) <= kobjsize(result))
 414                 printk("Allocation of %lu bytes from process %d has %lu bytes of slack\n", len, current->pid, kobjsize(result)-len);
 415 #endif
 416
 417         if (file) {
 418                 int error;
 419                 mm_segment_t old_fs = get_fs();
 420                 set_fs(KERNEL_DS);
 421                 error = file->f_op->read(file, (char *) result, len, &file->f_pos);
 422                 set_fs(old_fs);
 423                 if (error < 0) {
 424                         kfree(result);
 425                         kfree(tblock->rblock);
 426                         kfree(tblock);
 427                         return error;
 428                 }
 429                 if (error < len)
 430                         memset(result+error, '\0', len-error);
 431         } else {
 432                 memset(result, '\0', len);
 433         }
 434
 435         realalloc += kobjsize(tblock);
 436         askedalloc += sizeof(struct mm_tblock_struct);
 437
 438         realalloc += kobjsize(tblock->rblock);
 439         askedalloc += sizeof(struct mm_rblock_struct);
 440
 441         tblock->next = current->mm->context.tblock.next;
 442         current->mm->context.tblock.next = tblock;
 443         // current->mm->total_vm += len >> PAGE_SHIFT;
 444         vx_vmpages_add(current->mm, len >> PAGE_SHIFT);
 445
 446 #ifdef DEBUG
 447         printk("do_mmap:\n");
 448         show_process_blocks();
 449 #endif
 450
 451         return (unsigned long)result;
 452 }
 453
 454 int do_munmap(struct mm_struct * mm, unsigned long addr, size_t len)
 455 {
 456         struct mm_tblock_struct * tblock, *tmp;
 457
 458 #ifdef MAGIC_ROM_PTR
 459         /*
 460          * For efficiency's sake, if the pointer is obviously in ROM,
 461          * don't bother walking the lists to free it.
 462          */
 463         if (is_in_rom(addr))
 464                 return 0;
 465 #endif
 466
 467 #ifdef DEBUG
 468         printk("do_munmap:\n");
 469 #endif
 470
 471         tmp = &mm->context.tblock; /* dummy head */
 472         while ((tblock=tmp->next) && tblock->rblock &&
 473                         tblock->rblock->kblock != (void*)addr)
 474                 tmp = tblock;
 475
 476         if (!tblock) {
 477                 printk("munmap of non-mmaped memory by process %d (%s): %p\n",
 478                                 current->pid, current->comm, (void*)addr);
 479                 return -EINVAL;
 480         }
 481         if (tblock->rblock) {
 482                 if (!--tblock->rblock->refcount) {
 483                         if (tblock->rblock->kblock) {
 484                                 realalloc -= kobjsize(tblock->rblock->kblock);
 485                                 askedalloc -= tblock->rblock->size;
 486                                 kfree(tblock->rblock->kblock);
 487                         }
 488
 489                         realalloc -= kobjsize(tblock->rblock);
 490                         askedalloc -= sizeof(struct mm_rblock_struct);
 491                         kfree(tblock->rblock);
 492                 }
 493         }
 494         tmp->next = tblock->next;
 495         realalloc -= kobjsize(tblock);
 496         askedalloc -= sizeof(struct mm_tblock_struct);
 497         kfree(tblock);
 498         // mm->total_vm -= len >> PAGE_SHIFT;
 499         vx_vmpages_sub(mm, len >> PAGE_SHIFT);
 500
 501 #ifdef DEBUG
 502         show_process_blocks();
 503 #endif
 504
 505         return 0;
 506 }
 507
 508 /* Release all mmaps. */
 509 void exit_mmap(struct mm_struct * mm)
 510 {
 511         struct mm_tblock_struct *tmp;
 512         // mm->total_vm = 0;
 513         vx_vmpages_sub(mm, mm->total_vm);
 514
 515         if (!mm)
 516                 return;
 517
 518 #ifdef DEBUG
 519         printk("Exit_mmap:\n");
 520 #endif
 521
 522         while((tmp = mm->context.tblock.next)) {
 523                 if (tmp->rblock) {
 524                         if (!--tmp->rblock->refcount) {
 525                                 if (tmp->rblock->kblock) {
 526                                         realalloc -= kobjsize(tmp->rblock->kblock);
 527                                         askedalloc -= tmp->rblock->size;
 528                                         kfree(tmp->rblock->kblock);
 529                                 }
 530                                 realalloc -= kobjsize(tmp->rblock);
 531                                 askedalloc -= sizeof(struct mm_rblock_struct);
 532                                 kfree(tmp->rblock);
 533                         }
 534                         tmp->rblock = 0;
 535                 }
 536                 mm->context.tblock.next = tmp->next;
 537                 realalloc -= kobjsize(tmp);
 538                 askedalloc -= sizeof(struct mm_tblock_struct);
 539                 kfree(tmp);
 540         }
 541
 542 #ifdef DEBUG
 543         show_process_blocks();
 544 #endif
 545 }
 546
 547 asmlinkage long sys_munmap(unsigned long addr, size_t len)
 548 {
 549         int ret;
 550         struct mm_struct *mm = current->mm;
 551
 552         down_write(&mm->mmap_sem);
 553         ret = do_munmap(mm, addr, len);
 554         up_write(&mm->mmap_sem);
 555         return ret;
 556 }
 557
 558 unsigned long do_brk(unsigned long addr, unsigned long len)
 559 {
 560         return -ENOMEM;
 561 }
 562
 563 struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr)
 564 {
 565         return NULL;
 566 }
 567
 568 struct page * follow_page(struct mm_struct *mm, unsigned long addr, int write)
 569 {
 570         return NULL;
 571 }
 572
 573 struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr)
 574 {
 575         return NULL;
 576 }
 577
 578 int remap_pfn_range(struct vm_area_struct *vma, unsigned long from,
 579                 unsigned long to, unsigned long size, pgprot_t prot)
 580 {
 581         return -EPERM;
 582 }
 583
 584 unsigned long get_unmapped_area(struct file *file, unsigned long addr,
 585         unsigned long len, unsigned long pgoff, unsigned long flags)
 586 {
 587         return -ENOMEM;
 588 }
 589
 590 void swap_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 591 {
 592 }
 593
 594 unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr,
 595         unsigned long len, unsigned long pgoff, unsigned long flags)
 596 {
 597         return -ENOMEM;
 598 }
 599
 600 void arch_unmap_area(struct vm_area_struct *area)
 601 {
 602 }
 603