arch/arm/mm/init.c

   1 /*
   2  *  linux/arch/arm/mm/init.c
   3  *
   4  *  Copyright (C) 1995-2005 Russell King
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10 #include <linux/kernel.h>
  11 #include <linux/errno.h>
  12 #include <linux/swap.h>
  13 #include <linux/init.h>
  14 #include <linux/bootmem.h>
  15 #include <linux/mman.h>
  16 #include <linux/nodemask.h>
  17 #include <linux/initrd.h>
  18 #include <linux/sort.h>
  19 #include <linux/highmem.h>
  20
  21 #include <asm/mach-types.h>
  22 #include <asm/sections.h>
  23 #include <asm/setup.h>
  24 #include <asm/sizes.h>
  25 #include <asm/tlb.h>
  26
  27 #include <asm/mach/arch.h>
  28 #include <asm/mach/map.h>
  29
  30 #include "mm.h"
  31
  32 static unsigned long phys_initrd_start __initdata = 0;
  33 static unsigned long phys_initrd_size __initdata = 0;
  34
  35 static void __init early_initrd(char **p)
  36 {
  37         unsigned long start, size;
  38
  39         start = memparse(*p, p);
  40         if (**p == ',') {
  41                 size = memparse((*p) + 1, p);
  42
  43                 phys_initrd_start = start;
  44                 phys_initrd_size = size;
  45         }
  46 }
  47 __early_param("initrd=", early_initrd);
  48
  49 static int __init parse_tag_initrd(const struct tag *tag)
  50 {
  51         printk(KERN_WARNING "ATAG_INITRD is deprecated; "
  52                 "please update your bootloader.\n");
  53         phys_initrd_start = __virt_to_phys(tag->u.initrd.start);
  54         phys_initrd_size = tag->u.initrd.size;
  55         return 0;
  56 }
  57
  58 __tagtable(ATAG_INITRD, parse_tag_initrd);
  59
  60 static int __init parse_tag_initrd2(const struct tag *tag)
  61 {
  62         phys_initrd_start = tag->u.initrd.start;
  63         phys_initrd_size = tag->u.initrd.size;
  64         return 0;
  65 }
  66
  67 __tagtable(ATAG_INITRD2, parse_tag_initrd2);
  68
  69 /*
  70  * This keeps memory configuration data used by a couple memory
  71  * initialization functions, as well as show_mem() for the skipping
  72  * of holes in the memory map.  It is populated by arm_add_memory().
  73  */
  74 struct meminfo meminfo;
  75
  76 void show_mem(void)
  77 {
  78         int free = 0, total = 0, reserved = 0;
  79         int shared = 0, cached = 0, slab = 0, node, i;
  80         struct meminfo * mi = &meminfo;
  81
  82         printk("Mem-info:\n");
  83         show_free_areas();
  84         for_each_online_node(node) {
  85                 pg_data_t *n = NODE_DATA(node);
  86                 struct page *map = pgdat_page_nr(n, 0) - n->node_start_pfn;
  87
  88                 for_each_nodebank (i,mi,node) {
  89                         struct membank *bank = &mi->bank[i];
  90                         unsigned int pfn1, pfn2;
  91                         struct page *page, *end;
  92
  93                         pfn1 = bank_pfn_start(bank);
  94                         pfn2 = bank_pfn_end(bank);
  95
  96                         page = map + pfn1;
  97                         end  = map + pfn2;
  98
  99                         do {
 100                                 total++;
 101                                 if (PageReserved(page))
 102                                         reserved++;
 103                                 else if (PageSwapCache(page))
 104                                         cached++;
 105                                 else if (PageSlab(page))
 106                                         slab++;
 107                                 else if (!page_count(page))
 108                                         free++;
 109                                 else
 110                                         shared += page_count(page) - 1;
 111                                 page++;
 112                         } while (page < end);
 113                 }
 114         }
 115
 116         printk("%d pages of RAM\n", total);
 117         printk("%d free pages\n", free);
 118         printk("%d reserved pages\n", reserved);
 119         printk("%d slab pages\n", slab);
 120         printk("%d pages shared\n", shared);
 121         printk("%d pages swap cached\n", cached);
 122 }
 123
 124 static void __init find_node_limits(int node, struct meminfo *mi,
 125         unsigned long *min, unsigned long *max_low, unsigned long *max_high)
 126 {
 127         int i;
 128
 129         *min = -1UL;
 130         *max_low = *max_high = 0;
 131
 132         for_each_nodebank(i, mi, node) {
 133                 struct membank *bank = &mi->bank[i];
 134                 unsigned long start, end;
 135
 136                 start = bank_pfn_start(bank);
 137                 end = bank_pfn_end(bank);
 138
 139                 if (*min > start)
 140                         *min = start;
 141                 if (*max_high < end)
 142                         *max_high = end;
 143                 if (bank->highmem)
 144                         continue;
 145                 if (*max_low < end)
 146                         *max_low = end;
 147         }
 148 }
 149
 150 /*
 151  * FIXME: We really want to avoid allocating the bootmap bitmap
 152  * over the top of the initrd.  Hopefully, this is located towards
 153  * the start of a bank, so if we allocate the bootmap bitmap at
 154  * the end, we won't clash.
 155  */
 156 static unsigned int __init
 157 find_bootmap_pfn(int node, struct meminfo *mi, unsigned int bootmap_pages)
 158 {
 159         unsigned int start_pfn, i, bootmap_pfn;
 160
 161         start_pfn   = PAGE_ALIGN(__pa(_end)) >> PAGE_SHIFT;
 162         bootmap_pfn = 0;
 163
 164         for_each_nodebank(i, mi, node) {
 165                 struct membank *bank = &mi->bank[i];
 166                 unsigned int start, end;
 167
 168                 start = bank_pfn_start(bank);
 169                 end   = bank_pfn_end(bank);
 170
 171                 if (end < start_pfn)
 172                         continue;
 173
 174                 if (start < start_pfn)
 175                         start = start_pfn;
 176
 177                 if (end <= start)
 178                         continue;
 179
 180                 if (end - start >= bootmap_pages) {
 181                         bootmap_pfn = start;
 182                         break;
 183                 }
 184         }
 185
 186         if (bootmap_pfn == 0)
 187                 BUG();
 188
 189         return bootmap_pfn;
 190 }
 191
 192 static int __init check_initrd(struct meminfo *mi)
 193 {
 194         int initrd_node = -2;
 195 #ifdef CONFIG_BLK_DEV_INITRD
 196         unsigned long end = phys_initrd_start + phys_initrd_size;
 197
 198         /*
 199          * Make sure that the initrd is within a valid area of
 200          * memory.
 201          */
 202         if (phys_initrd_size) {
 203                 unsigned int i;
 204
 205                 initrd_node = -1;
 206
 207                 for (i = 0; i < mi->nr_banks; i++) {
 208                         struct membank *bank = &mi->bank[i];
 209                         if (bank_phys_start(bank) <= phys_initrd_start &&
 210                             end <= bank_phys_end(bank))
 211                                 initrd_node = bank->node;
 212                 }
 213         }
 214
 215         if (initrd_node == -1) {
 216                 printk(KERN_ERR "INITRD: 0x%08lx+0x%08lx extends beyond "
 217                        "physical memory - disabling initrd\n",
 218                        phys_initrd_start, phys_initrd_size);
 219                 phys_initrd_start = phys_initrd_size = 0;
 220         }
 221 #endif
 222
 223         return initrd_node;
 224 }
 225
 226 static inline void map_memory_bank(struct membank *bank)
 227 {
 228 #ifdef CONFIG_MMU
 229         struct map_desc map;
 230
 231         map.pfn = bank_pfn_start(bank);
 232         map.virtual = __phys_to_virt(bank_phys_start(bank));
 233         map.length = bank_phys_size(bank);
 234         map.type = MT_MEMORY;
 235
 236         create_mapping(&map);
 237 #endif
 238 }
 239
 240 static void __init bootmem_init_node(int node, struct meminfo *mi,
 241         unsigned long start_pfn, unsigned long end_pfn)
 242 {
 243         unsigned long boot_pfn;
 244         unsigned int boot_pages;
 245         pg_data_t *pgdat;
 246         int i;
 247
 248         /*
 249          * Map the memory banks for this node.
 250          */
 251         for_each_nodebank(i, mi, node) {
 252                 struct membank *bank = &mi->bank[i];
 253
 254                 if (!bank->highmem)
 255                         map_memory_bank(bank);
 256         }
 257
 258         /*
 259          * Allocate the bootmem bitmap page.
 260          */
 261         boot_pages = bootmem_bootmap_pages(end_pfn - start_pfn);
 262         boot_pfn = find_bootmap_pfn(node, mi, boot_pages);
 263
 264         /*
 265          * Initialise the bootmem allocator for this node, handing the
 266          * memory banks over to bootmem.
 267          */
 268         node_set_online(node);
 269         pgdat = NODE_DATA(node);
 270         init_bootmem_node(pgdat, boot_pfn, start_pfn, end_pfn);
 271
 272         for_each_nodebank(i, mi, node) {
 273                 struct membank *bank = &mi->bank[i];
 274                 if (!bank->highmem)
 275                         free_bootmem_node(pgdat, bank_phys_start(bank), bank_phys_size(bank));
 276         }
 277
 278         /*
 279          * Reserve the bootmem bitmap for this node.
 280          */
 281         reserve_bootmem_node(pgdat, boot_pfn << PAGE_SHIFT,
 282                              boot_pages << PAGE_SHIFT, BOOTMEM_DEFAULT);
 283 }
 284
 285 static void __init bootmem_reserve_initrd(int node)
 286 {
 287 #ifdef CONFIG_BLK_DEV_INITRD
 288         pg_data_t *pgdat = NODE_DATA(node);
 289         int res;
 290
 291         res = reserve_bootmem_node(pgdat, phys_initrd_start,
 292                              phys_initrd_size, BOOTMEM_EXCLUSIVE);
 293
 294         if (res == 0) {
 295                 initrd_start = __phys_to_virt(phys_initrd_start);
 296                 initrd_end = initrd_start + phys_initrd_size;
 297         } else {
 298                 printk(KERN_ERR
 299                         "INITRD: 0x%08lx+0x%08lx overlaps in-use "
 300                         "memory region - disabling initrd\n",
 301                         phys_initrd_start, phys_initrd_size);
 302         }
 303 #endif
 304 }
 305
 306 static void __init bootmem_free_node(int node, struct meminfo *mi)
 307 {
 308         unsigned long zone_size[MAX_NR_ZONES], zhole_size[MAX_NR_ZONES];
 309         unsigned long min, max_low, max_high;
 310         int i;
 311
 312         find_node_limits(node, mi, &min, &max_low, &max_high);
 313
 314         /*
 315          * initialise the zones within this node.
 316          */
 317         memset(zone_size, 0, sizeof(zone_size));
 318
 319         /*
 320          * The size of this node has already been determined.  If we need
 321          * to do anything fancy with the allocation of this memory to the
 322          * zones, now is the time to do it.
 323          */
 324         zone_size[0] = max_low - min;
 325 #ifdef CONFIG_HIGHMEM
 326         zone_size[ZONE_HIGHMEM] = max_high - max_low;
 327 #endif
 328
 329         /*
 330          * For each bank in this node, calculate the size of the holes.
 331          *  holes = node_size - sum(bank_sizes_in_node)
 332          */
 333         memcpy(zhole_size, zone_size, sizeof(zhole_size));
 334         for_each_nodebank(i, mi, node) {
 335                 int idx = 0;
 336 #ifdef CONFIG_HIGHMEM
 337                 if (mi->bank[i].highmem)
 338                         idx = ZONE_HIGHMEM;
 339 #endif
 340                 zhole_size[idx] -= bank_pfn_size(&mi->bank[i]);
 341         }
 342
 343         /*
 344          * Adjust the sizes according to any special requirements for
 345          * this machine type.
 346          */
 347         arch_adjust_zones(node, zone_size, zhole_size);
 348
 349         free_area_init_node(node, zone_size, min, zhole_size);
 350 }
 351
 352 #ifndef CONFIG_SPARSEMEM
 353 int pfn_valid(unsigned long pfn)
 354 {
 355         struct meminfo *mi = &meminfo;
 356         unsigned int left = 0, right = mi->nr_banks;
 357
 358         do {
 359                 unsigned int mid = (right + left) / 2;
 360                 struct membank *bank = &mi->bank[mid];
 361
 362                 if (pfn < bank_pfn_start(bank))
 363                         right = mid;
 364                 else if (pfn >= bank_pfn_end(bank))
 365                         left = mid + 1;
 366                 else
 367                         return 1;
 368         } while (left < right);
 369         return 0;
 370 }
 371 EXPORT_SYMBOL(pfn_valid);
 372
 373 static void arm_memory_present(struct meminfo *mi, int node)
 374 {
 375 }
 376 #else
 377 static void arm_memory_present(struct meminfo *mi, int node)
 378 {
 379         int i;
 380         for_each_nodebank(i, mi, node) {
 381                 struct membank *bank = &mi->bank[i];
 382                 memory_present(node, bank_pfn_start(bank), bank_pfn_end(bank));
 383         }
 384 }
 385 #endif
 386
 387 static int __init meminfo_cmp(const void *_a, const void *_b)
 388 {
 389         const struct membank *a = _a, *b = _b;
 390         long cmp = bank_pfn_start(a) - bank_pfn_start(b);
 391         return cmp < 0 ? -1 : cmp > 0 ? 1 : 0;
 392 }
 393
 394 void __init bootmem_init(void)
 395 {
 396         struct meminfo *mi = &meminfo;
 397         unsigned long min, max_low, max_high;
 398         int node, initrd_node;
 399
 400         sort(&mi->bank, mi->nr_banks, sizeof(mi->bank[0]), meminfo_cmp, NULL);
 401
 402         /*
 403          * Locate which node contains the ramdisk image, if any.
 404          */
 405         initrd_node = check_initrd(mi);
 406
 407         max_low = max_high = 0;
 408
 409         /*
 410          * Run through each node initialising the bootmem allocator.
 411          */
 412         for_each_node(node) {
 413                 unsigned long node_low, node_high;
 414
 415                 find_node_limits(node, mi, &min, &node_low, &node_high);
 416
 417                 if (node_low > max_low)
 418                         max_low = node_low;
 419                 if (node_high > max_high)
 420                         max_high = node_high;
 421
 422                 /*
 423                  * If there is no memory in this node, ignore it.
 424                  * (We can't have nodes which have no lowmem)
 425                  */
 426                 if (node_low == 0)
 427                         continue;
 428
 429                 bootmem_init_node(node, mi, min, node_low);
 430
 431                 /*
 432                  * Reserve any special node zero regions.
 433                  */
 434                 if (node == 0)
 435                         reserve_node_zero(NODE_DATA(node));
 436
 437                 /*
 438                  * If the initrd is in this node, reserve its memory.
 439                  */
 440                 if (node == initrd_node)
 441                         bootmem_reserve_initrd(node);
 442
 443                 /*
 444                  * Sparsemem tries to allocate bootmem in memory_present(),
 445                  * so must be done after the fixed reservations
 446                  */
 447                 arm_memory_present(mi, node);
 448         }
 449
 450         /*
 451          * sparse_init() needs the bootmem allocator up and running.
 452          */
 453         sparse_init();
 454
 455         /*
 456          * Now free memory in each node - free_area_init_node needs
 457          * the sparse mem_map arrays initialized by sparse_init()
 458          * for memmap_init_zone(), otherwise all PFNs are invalid.
 459          */
 460         for_each_node(node)
 461                 bootmem_free_node(node, mi);
 462
 463         high_memory = __va((max_low << PAGE_SHIFT) - 1) + 1;
 464
 465         /*
 466          * This doesn't seem to be used by the Linux memory manager any
 467          * more, but is used by ll_rw_block.  If we can get rid of it, we
 468          * also get rid of some of the stuff above as well.
 469          *
 470          * Note: max_low_pfn and max_pfn reflect the number of _pages_ in
 471          * the system, not the maximum PFN.
 472          */
 473         max_low_pfn = max_low - PHYS_PFN_OFFSET;
 474         max_pfn = max_high - PHYS_PFN_OFFSET;
 475 }
 476
 477 static inline int free_area(unsigned long pfn, unsigned long end, char *s)
 478 {
 479         unsigned int pages = 0, size = (end - pfn) << (PAGE_SHIFT - 10);
 480
 481         for (; pfn < end; pfn++) {
 482                 struct page *page = pfn_to_page(pfn);
 483                 ClearPageReserved(page);
 484                 init_page_count(page);
 485                 __free_page(page);
 486                 pages++;
 487         }
 488
 489         if (size && s)
 490                 printk(KERN_INFO "Freeing %s memory: %dK\n", s, size);
 491
 492         return pages;
 493 }
 494
 495 static inline void
 496 free_memmap(int node, unsigned long start_pfn, unsigned long end_pfn)
 497 {
 498         struct page *start_pg, *end_pg;
 499         unsigned long pg, pgend;
 500
 501         /*
 502          * Convert start_pfn/end_pfn to a struct page pointer.
 503          */
 504         start_pg = pfn_to_page(start_pfn - 1) + 1;
 505         end_pg = pfn_to_page(end_pfn);
 506
 507         /*
 508          * Convert to physical addresses, and
 509          * round start upwards and end downwards.
 510          */
 511         pg = PAGE_ALIGN(__pa(start_pg));
 512         pgend = __pa(end_pg) & PAGE_MASK;
 513
 514         /*
 515          * If there are free pages between these,
 516          * free the section of the memmap array.
 517          */
 518         if (pg < pgend)
 519                 free_bootmem_node(NODE_DATA(node), pg, pgend - pg);
 520 }
 521
 522 /*
 523  * The mem_map array can get very big.  Free the unused area of the memory map.
 524  */
 525 static void __init free_unused_memmap_node(int node, struct meminfo *mi)
 526 {
 527         unsigned long bank_start, prev_bank_end = 0;
 528         unsigned int i;
 529
 530         /*
 531          * [FIXME] This relies on each bank being in address order.  This
 532          * may not be the case, especially if the user has provided the
 533          * information on the command line.
 534          */
 535         for_each_nodebank(i, mi, node) {
 536                 struct membank *bank = &mi->bank[i];
 537
 538                 bank_start = bank_pfn_start(bank);
 539                 if (bank_start < prev_bank_end) {
 540                         printk(KERN_ERR "MEM: unordered memory banks.  "
 541                                 "Not freeing memmap.\n");
 542                         break;
 543                 }
 544
 545                 /*
 546                  * If we had a previous bank, and there is a space
 547                  * between the current bank and the previous, free it.
 548                  */
 549                 if (prev_bank_end && prev_bank_end != bank_start)
 550                         free_memmap(node, prev_bank_end, bank_start);
 551
 552                 prev_bank_end = bank_pfn_end(bank);
 553         }
 554 }
 555
 556 /*
 557  * mem_init() marks the free areas in the mem_map and tells us how much
 558  * memory is free.  This is done after various parts of the system have
 559  * claimed their memory after the kernel image.
 560  */
 561 void __init mem_init(void)
 562 {
 563         unsigned int codesize, datasize, initsize;
 564         int i, node;
 565
 566 #ifndef CONFIG_DISCONTIGMEM
 567         max_mapnr   = pfn_to_page(max_pfn + PHYS_PFN_OFFSET) - mem_map;
 568 #endif
 569
 570         /* this will put all unused low memory onto the freelists */
 571         for_each_online_node(node) {
 572                 pg_data_t *pgdat = NODE_DATA(node);
 573
 574                 free_unused_memmap_node(node, &meminfo);
 575
 576                 if (pgdat->node_spanned_pages != 0)
 577                         totalram_pages += free_all_bootmem_node(pgdat);
 578         }
 579
 580 #ifdef CONFIG_SA1111
 581         /* now that our DMA memory is actually so designated, we can free it */
 582         totalram_pages += free_area(PHYS_PFN_OFFSET,
 583                                     __phys_to_pfn(__pa(swapper_pg_dir)), NULL);
 584 #endif
 585
 586 #ifdef CONFIG_HIGHMEM
 587         /* set highmem page free */
 588         for_each_online_node(node) {
 589                 for_each_nodebank (i, &meminfo, node) {
 590                         unsigned long start = bank_pfn_start(&meminfo.bank[i]);
 591                         unsigned long end = bank_pfn_end(&meminfo.bank[i]);
 592                         if (start >= max_low_pfn + PHYS_PFN_OFFSET)
 593                                 totalhigh_pages += free_area(start, end, NULL);
 594                 }
 595         }
 596         totalram_pages += totalhigh_pages;
 597 #endif
 598
 599         /*
 600          * Since our memory may not be contiguous, calculate the
 601          * real number of pages we have in this system
 602          */
 603         printk(KERN_INFO "Memory:");
 604         num_physpages = 0;
 605         for (i = 0; i < meminfo.nr_banks; i++) {
 606                 num_physpages += bank_pfn_size(&meminfo.bank[i]);
 607                 printk(" %ldMB", bank_phys_size(&meminfo.bank[i]) >> 20);
 608         }
 609         printk(" = %luMB total\n", num_physpages >> (20 - PAGE_SHIFT));
 610
 611         codesize = _etext - _text;
 612         datasize = _end - _data;
 613         initsize = __init_end - __init_begin;
 614
 615         printk(KERN_NOTICE "Memory: %luKB available (%dK code, "
 616                 "%dK data, %dK init, %luK highmem)\n",
 617                 nr_free_pages() << (PAGE_SHIFT-10), codesize >> 10,
 618                 datasize >> 10, initsize >> 10,
 619                 totalhigh_pages << (PAGE_SHIFT-10));
 620
 621         if (PAGE_SIZE >= 16384 && num_physpages <= 128) {
 622                 extern int sysctl_overcommit_memory;
 623                 /*
 624                  * On a machine this small we won't get
 625                  * anywhere without overcommit, so turn
 626                  * it on by default.
 627                  */
 628                 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
 629         }
 630 }
 631
 632 void free_initmem(void)
 633 {
 634 #ifdef CONFIG_HAVE_TCM
 635         extern char *__tcm_start, *__tcm_end;
 636
 637         totalram_pages += free_area(__phys_to_pfn(__pa(__tcm_start)),
 638                                     __phys_to_pfn(__pa(__tcm_end)),
 639                                     "TCM link");
 640 #endif
 641
 642         if (!machine_is_integrator() && !machine_is_cintegrator())
 643                 totalram_pages += free_area(__phys_to_pfn(__pa(__init_begin)),
 644                                             __phys_to_pfn(__pa(__init_end)),
 645                                             "init");
 646 }
 647
 648 #ifdef CONFIG_BLK_DEV_INITRD
 649
 650 static int keep_initrd;
 651
 652 void free_initrd_mem(unsigned long start, unsigned long end)
 653 {
 654         if (!keep_initrd)
 655                 totalram_pages += free_area(__phys_to_pfn(__pa(start)),
 656                                             __phys_to_pfn(__pa(end)),
 657                                             "initrd");
 658 }
 659
 660 static int __init keepinitrd_setup(char *__unused)
 661 {
 662         keep_initrd = 1;
 663         return 1;
 664 }
 665
 666 __setup("keepinitrd", keepinitrd_setup);
 667 #endif