arch/arm/mm/init.c

   1 /*
   2  *  linux/arch/arm/mm/init.c
   3  *
   4  *  Copyright (C) 1995-2002 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/config.h>
  11 #include <linux/kernel.h>
  12 #include <linux/errno.h>
  13 #include <linux/ptrace.h>
  14 #include <linux/swap.h>
  15 #include <linux/init.h>
  16 #include <linux/bootmem.h>
  17 #include <linux/mman.h>
  18 #include <linux/nodemask.h>
  19 #include <linux/initrd.h>
  20
  21 #include <asm/mach-types.h>
  22 #include <asm/hardware.h>
  23 #include <asm/setup.h>
  24 #include <asm/tlb.h>
  25
  26 #include <asm/mach/arch.h>
  27 #include <asm/mach/map.h>
  28
  29 #define TABLE_SIZE      (2 * PTRS_PER_PTE * sizeof(pte_t))
  30
  31 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  32
  33 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
  34 extern void _stext, _text, _etext, __data_start, _end, __init_begin, __init_end;
  35 extern unsigned long phys_initrd_start;
  36 extern unsigned long phys_initrd_size;
  37
  38 /*
  39  * The sole use of this is to pass memory configuration
  40  * data from paging_init to mem_init.
  41  */
  42 static struct meminfo meminfo __initdata = { 0, };
  43
  44 /*
  45  * empty_zero_page is a special page that is used for
  46  * zero-initialized data and COW.
  47  */
  48 struct page *empty_zero_page;
  49
  50 void show_mem(void)
  51 {
  52         int free = 0, total = 0, reserved = 0;
  53         int shared = 0, cached = 0, slab = 0, node;
  54
  55         printk("Mem-info:\n");
  56         show_free_areas();
  57         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
  58
  59         for_each_online_node(node) {
  60                 struct page *page, *end;
  61
  62                 page = NODE_MEM_MAP(node);
  63                 end  = page + NODE_DATA(node)->node_spanned_pages;
  64
  65                 do {
  66                         total++;
  67                         if (PageReserved(page))
  68                                 reserved++;
  69                         else if (PageSwapCache(page))
  70                                 cached++;
  71                         else if (PageSlab(page))
  72                                 slab++;
  73                         else if (!page_count(page))
  74                                 free++;
  75                         else
  76                                 shared += page_count(page) - 1;
  77                         page++;
  78                 } while (page < end);
  79         }
  80
  81         printk("%d pages of RAM\n", total);
  82         printk("%d free pages\n", free);
  83         printk("%d reserved pages\n", reserved);
  84         printk("%d slab pages\n", slab);
  85         printk("%d pages shared\n", shared);
  86         printk("%d pages swap cached\n", cached);
  87 }
  88
  89 struct node_info {
  90         unsigned int start;
  91         unsigned int end;
  92         int bootmap_pages;
  93 };
  94
  95 #define O_PFN_DOWN(x)   ((x) >> PAGE_SHIFT)
  96 #define O_PFN_UP(x)     (PAGE_ALIGN(x) >> PAGE_SHIFT)
  97
  98 /*
  99  * FIXME: We really want to avoid allocating the bootmap bitmap
 100  * over the top of the initrd.  Hopefully, this is located towards
 101  * the start of a bank, so if we allocate the bootmap bitmap at
 102  * the end, we won't clash.
 103  */
 104 static unsigned int __init
 105 find_bootmap_pfn(int node, struct meminfo *mi, unsigned int bootmap_pages)
 106 {
 107         unsigned int start_pfn, bank, bootmap_pfn;
 108
 109         start_pfn   = O_PFN_UP(__pa(&_end));
 110         bootmap_pfn = 0;
 111
 112         for (bank = 0; bank < mi->nr_banks; bank ++) {
 113                 unsigned int start, end;
 114
 115                 if (mi->bank[bank].node != node)
 116                         continue;
 117
 118                 start = mi->bank[bank].start >> PAGE_SHIFT;
 119                 end   = (mi->bank[bank].size +
 120                          mi->bank[bank].start) >> PAGE_SHIFT;
 121
 122                 if (end < start_pfn)
 123                         continue;
 124
 125                 if (start < start_pfn)
 126                         start = start_pfn;
 127
 128                 if (end <= start)
 129                         continue;
 130
 131                 if (end - start >= bootmap_pages) {
 132                         bootmap_pfn = start;
 133                         break;
 134                 }
 135         }
 136
 137         if (bootmap_pfn == 0)
 138                 BUG();
 139
 140         return bootmap_pfn;
 141 }
 142
 143 /*
 144  * Scan the memory info structure and pull out:
 145  *  - the end of memory
 146  *  - the number of nodes
 147  *  - the pfn range of each node
 148  *  - the number of bootmem bitmap pages
 149  */
 150 static unsigned int __init
 151 find_memend_and_nodes(struct meminfo *mi, struct node_info *np)
 152 {
 153         unsigned int i, bootmem_pages = 0, memend_pfn = 0;
 154
 155         for (i = 0; i < MAX_NUMNODES; i++) {
 156                 np[i].start = -1U;
 157                 np[i].end = 0;
 158                 np[i].bootmap_pages = 0;
 159         }
 160
 161         for (i = 0; i < mi->nr_banks; i++) {
 162                 unsigned long start, end;
 163                 int node;
 164
 165                 if (mi->bank[i].size == 0) {
 166                         /*
 167                          * Mark this bank with an invalid node number
 168                          */
 169                         mi->bank[i].node = -1;
 170                         continue;
 171                 }
 172
 173                 node = mi->bank[i].node;
 174
 175                 /*
 176                  * Make sure we haven't exceeded the maximum number of nodes
 177                  * that we have in this configuration.  If we have, we're in
 178                  * trouble.  (maybe we ought to limit, instead of bugging?)
 179                  */
 180                 if (node >= MAX_NUMNODES)
 181                         BUG();
 182                 node_set_online(node);
 183
 184                 /*
 185                  * Get the start and end pfns for this bank
 186                  */
 187                 start = mi->bank[i].start >> PAGE_SHIFT;
 188                 end   = (mi->bank[i].start + mi->bank[i].size) >> PAGE_SHIFT;
 189
 190                 if (np[node].start > start)
 191                         np[node].start = start;
 192
 193                 if (np[node].end < end)
 194                         np[node].end = end;
 195
 196                 if (memend_pfn < end)
 197                         memend_pfn = end;
 198         }
 199
 200         /*
 201          * Calculate the number of pages we require to
 202          * store the bootmem bitmaps.
 203          */
 204         for_each_online_node(i) {
 205                 if (np[i].end == 0)
 206                         continue;
 207
 208                 np[i].bootmap_pages = bootmem_bootmap_pages(np[i].end -
 209                                                             np[i].start);
 210                 bootmem_pages += np[i].bootmap_pages;
 211         }
 212
 213         high_memory = __va(memend_pfn << PAGE_SHIFT);
 214
 215         /*
 216          * This doesn't seem to be used by the Linux memory
 217          * manager any more.  If we can get rid of it, we
 218          * also get rid of some of the stuff above as well.
 219          *
 220          * Note: max_low_pfn and max_pfn reflect the number
 221          * of _pages_ in the system, not the maximum PFN.
 222          */
 223         max_low_pfn = memend_pfn - O_PFN_DOWN(PHYS_OFFSET);
 224         max_pfn = memend_pfn - O_PFN_DOWN(PHYS_OFFSET);
 225
 226         return bootmem_pages;
 227 }
 228
 229 static int __init check_initrd(struct meminfo *mi)
 230 {
 231         int initrd_node = -2;
 232 #ifdef CONFIG_BLK_DEV_INITRD
 233         unsigned long end = phys_initrd_start + phys_initrd_size;
 234
 235         /*
 236          * Make sure that the initrd is within a valid area of
 237          * memory.
 238          */
 239         if (phys_initrd_size) {
 240                 unsigned int i;
 241
 242                 initrd_node = -1;
 243
 244                 for (i = 0; i < mi->nr_banks; i++) {
 245                         unsigned long bank_end;
 246
 247                         bank_end = mi->bank[i].start + mi->bank[i].size;
 248
 249                         if (mi->bank[i].start <= phys_initrd_start &&
 250                             end <= bank_end)
 251                                 initrd_node = mi->bank[i].node;
 252                 }
 253         }
 254
 255         if (initrd_node == -1) {
 256                 printk(KERN_ERR "initrd (0x%08lx - 0x%08lx) extends beyond "
 257                        "physical memory - disabling initrd\n",
 258                        phys_initrd_start, end);
 259                 phys_initrd_start = phys_initrd_size = 0;
 260         }
 261 #endif
 262
 263         return initrd_node;
 264 }
 265
 266 /*
 267  * Reserve the various regions of node 0
 268  */
 269 static __init void reserve_node_zero(unsigned int bootmap_pfn, unsigned int bootmap_pages)
 270 {
 271         pg_data_t *pgdat = NODE_DATA(0);
 272         unsigned long res_size = 0;
 273
 274         /*
 275          * Register the kernel text and data with bootmem.
 276          * Note that this can only be in node 0.
 277          */
 278 #ifdef CONFIG_XIP_KERNEL
 279         reserve_bootmem_node(pgdat, __pa(&__data_start), &_end - &__data_start);
 280 #else
 281         reserve_bootmem_node(pgdat, __pa(&_stext), &_end - &_stext);
 282 #endif
 283
 284         /*
 285          * Reserve the page tables.  These are already in use,
 286          * and can only be in node 0.
 287          */
 288         reserve_bootmem_node(pgdat, __pa(swapper_pg_dir),
 289                              PTRS_PER_PGD * sizeof(pgd_t));
 290
 291         /*
 292          * And don't forget to reserve the allocator bitmap,
 293          * which will be freed later.
 294          */
 295         reserve_bootmem_node(pgdat, bootmap_pfn << PAGE_SHIFT,
 296                              bootmap_pages << PAGE_SHIFT);
 297
 298         /*
 299          * Hmm... This should go elsewhere, but we really really need to
 300          * stop things allocating the low memory; ideally we need a better
 301          * implementation of GFP_DMA which does not assume that DMA-able
 302          * memory starts at zero.
 303          */
 304         if (machine_is_integrator() || machine_is_cintegrator())
 305                 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
 306
 307         /*
 308          * These should likewise go elsewhere.  They pre-reserve the
 309          * screen memory region at the start of main system memory.
 310          */
 311         if (machine_is_edb7211())
 312                 res_size = 0x00020000;
 313         if (machine_is_p720t())
 314                 res_size = 0x00014000;
 315
 316 #ifdef CONFIG_SA1111
 317         /*
 318          * Because of the SA1111 DMA bug, we want to preserve our
 319          * precious DMA-able memory...
 320          */
 321         res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
 322 #endif
 323         if (res_size)
 324                 reserve_bootmem_node(pgdat, PHYS_OFFSET, res_size);
 325 }
 326
 327 /*
 328  * Register all available RAM in this node with the bootmem allocator.
 329  */
 330 static inline void free_bootmem_node_bank(int node, struct meminfo *mi)
 331 {
 332         pg_data_t *pgdat = NODE_DATA(node);
 333         int bank;
 334
 335         for (bank = 0; bank < mi->nr_banks; bank++)
 336                 if (mi->bank[bank].node == node)
 337                         free_bootmem_node(pgdat, mi->bank[bank].start,
 338                                           mi->bank[bank].size);
 339 }
 340
 341 /*
 342  * Initialise the bootmem allocator for all nodes.  This is called
 343  * early during the architecture specific initialisation.
 344  */
 345 static void __init bootmem_init(struct meminfo *mi)
 346 {
 347         struct node_info node_info[MAX_NUMNODES], *np = node_info;
 348         unsigned int bootmap_pages, bootmap_pfn, map_pg;
 349         int node, initrd_node;
 350
 351         bootmap_pages = find_memend_and_nodes(mi, np);
 352         bootmap_pfn   = find_bootmap_pfn(0, mi, bootmap_pages);
 353         initrd_node   = check_initrd(mi);
 354
 355         map_pg = bootmap_pfn;
 356
 357         /*
 358          * Initialise the bootmem nodes.
 359          *
 360          * What we really want to do is:
 361          *
 362          *   unmap_all_regions_except_kernel();
 363          *   for_each_node_in_reverse_order(node) {
 364          *     map_node(node);
 365          *     allocate_bootmem_map(node);
 366          *     init_bootmem_node(node);
 367          *     free_bootmem_node(node);
 368          *   }
 369          *
 370          * but this is a 2.5-type change.  For now, we just set
 371          * the nodes up in reverse order.
 372          *
 373          * (we could also do with rolling bootmem_init and paging_init
 374          * into one generic "memory_init" type function).
 375          */
 376         np += num_online_nodes() - 1;
 377         for (node = num_online_nodes() - 1; node >= 0; node--, np--) {
 378                 /*
 379                  * If there are no pages in this node, ignore it.
 380                  * Note that node 0 must always have some pages.
 381                  */
 382                 if (np->end == 0 || !node_online(node)) {
 383                         if (node == 0)
 384                                 BUG();
 385                         continue;
 386                 }
 387
 388                 /*
 389                  * Initialise the bootmem allocator.
 390                  */
 391                 init_bootmem_node(NODE_DATA(node), map_pg, np->start, np->end);
 392                 free_bootmem_node_bank(node, mi);
 393                 map_pg += np->bootmap_pages;
 394
 395                 /*
 396                  * If this is node 0, we need to reserve some areas ASAP -
 397                  * we may use bootmem on node 0 to setup the other nodes.
 398                  */
 399                 if (node == 0)
 400                         reserve_node_zero(bootmap_pfn, bootmap_pages);
 401         }
 402
 403
 404 #ifdef CONFIG_BLK_DEV_INITRD
 405         if (phys_initrd_size && initrd_node >= 0) {
 406                 reserve_bootmem_node(NODE_DATA(initrd_node), phys_initrd_start,
 407                                      phys_initrd_size);
 408                 initrd_start = __phys_to_virt(phys_initrd_start);
 409                 initrd_end = initrd_start + phys_initrd_size;
 410         }
 411 #endif
 412
 413         BUG_ON(map_pg != bootmap_pfn + bootmap_pages);
 414 }
 415
 416 /*
 417  * paging_init() sets up the page tables, initialises the zone memory
 418  * maps, and sets up the zero page, bad page and bad page tables.
 419  */
 420 void __init paging_init(struct meminfo *mi, struct machine_desc *mdesc)
 421 {
 422         void *zero_page;
 423         int node;
 424
 425         bootmem_init(mi);
 426
 427         memcpy(&meminfo, mi, sizeof(meminfo));
 428
 429         /*
 430          * allocate the zero page.  Note that we count on this going ok.
 431          */
 432         zero_page = alloc_bootmem_low_pages(PAGE_SIZE);
 433
 434         /*
 435          * initialise the page tables.
 436          */
 437         memtable_init(mi);
 438         if (mdesc->map_io)
 439                 mdesc->map_io();
 440         local_flush_tlb_all();
 441
 442         /*
 443          * initialise the zones within each node
 444          */
 445         for_each_online_node(node) {
 446                 unsigned long zone_size[MAX_NR_ZONES];
 447                 unsigned long zhole_size[MAX_NR_ZONES];
 448                 struct bootmem_data *bdata;
 449                 pg_data_t *pgdat;
 450                 int i;
 451
 452                 /*
 453                  * Initialise the zone size information.
 454                  */
 455                 for (i = 0; i < MAX_NR_ZONES; i++) {
 456                         zone_size[i]  = 0;
 457                         zhole_size[i] = 0;
 458                 }
 459
 460                 pgdat = NODE_DATA(node);
 461                 bdata = pgdat->bdata;
 462
 463                 /*
 464                  * The size of this node has already been determined.
 465                  * If we need to do anything fancy with the allocation
 466                  * of this memory to the zones, now is the time to do
 467                  * it.
 468                  */
 469                 zone_size[0] = bdata->node_low_pfn -
 470                                 (bdata->node_boot_start >> PAGE_SHIFT);
 471
 472                 /*
 473                  * If this zone has zero size, skip it.
 474                  */
 475                 if (!zone_size[0])
 476                         continue;
 477
 478                 /*
 479                  * For each bank in this node, calculate the size of the
 480                  * holes.  holes = node_size - sum(bank_sizes_in_node)
 481                  */
 482                 zhole_size[0] = zone_size[0];
 483                 for (i = 0; i < mi->nr_banks; i++) {
 484                         if (mi->bank[i].node != node)
 485                                 continue;
 486
 487                         zhole_size[0] -= mi->bank[i].size >> PAGE_SHIFT;
 488                 }
 489
 490                 /*
 491                  * Adjust the sizes according to any special
 492                  * requirements for this machine type.
 493                  */
 494                 arch_adjust_zones(node, zone_size, zhole_size);
 495
 496                 free_area_init_node(node, pgdat, zone_size,
 497                                 bdata->node_boot_start >> PAGE_SHIFT, zhole_size);
 498         }
 499
 500         /*
 501          * finish off the bad pages once
 502          * the mem_map is initialised
 503          */
 504         memzero(zero_page, PAGE_SIZE);
 505         empty_zero_page = virt_to_page(zero_page);
 506         flush_dcache_page(empty_zero_page);
 507 }
 508
 509 static inline void free_area(unsigned long addr, unsigned long end, char *s)
 510 {
 511         unsigned int size = (end - addr) >> 10;
 512
 513         for (; addr < end; addr += PAGE_SIZE) {
 514                 struct page *page = virt_to_page(addr);
 515                 ClearPageReserved(page);
 516                 set_page_count(page, 1);
 517                 free_page(addr);
 518                 totalram_pages++;
 519         }
 520
 521         if (size && s)
 522                 printk(KERN_INFO "Freeing %s memory: %dK\n", s, size);
 523 }
 524
 525 static inline void
 526 free_memmap(int node, unsigned long start_pfn, unsigned long end_pfn)
 527 {
 528         struct page *start_pg, *end_pg;
 529         unsigned long pg, pgend;
 530
 531         /*
 532          * Convert start_pfn/end_pfn to a struct page pointer.
 533          */
 534         start_pg = pfn_to_page(start_pfn);
 535         end_pg = pfn_to_page(end_pfn);
 536
 537         /*
 538          * Convert to physical addresses, and
 539          * round start upwards and end downwards.
 540          */
 541         pg = PAGE_ALIGN(__pa(start_pg));
 542         pgend = __pa(end_pg) & PAGE_MASK;
 543
 544         /*
 545          * If there are free pages between these,
 546          * free the section of the memmap array.
 547          */
 548         if (pg < pgend)
 549                 free_bootmem_node(NODE_DATA(node), pg, pgend - pg);
 550 }
 551
 552 /*
 553  * The mem_map array can get very big.  Free the unused area of the memory map.
 554  */
 555 static void __init free_unused_memmap_node(int node, struct meminfo *mi)
 556 {
 557         unsigned long bank_start, prev_bank_end = 0;
 558         unsigned int i;
 559
 560         /*
 561          * [FIXME] This relies on each bank being in address order.  This
 562          * may not be the case, especially if the user has provided the
 563          * information on the command line.
 564          */
 565         for (i = 0; i < mi->nr_banks; i++) {
 566                 if (mi->bank[i].size == 0 || mi->bank[i].node != node)
 567                         continue;
 568
 569                 bank_start = mi->bank[i].start >> PAGE_SHIFT;
 570                 if (bank_start < prev_bank_end) {
 571                         printk(KERN_ERR "MEM: unordered memory banks.  "
 572                                 "Not freeing memmap.\n");
 573                         break;
 574                 }
 575
 576                 /*
 577                  * If we had a previous bank, and there is a space
 578                  * between the current bank and the previous, free it.
 579                  */
 580                 if (prev_bank_end && prev_bank_end != bank_start)
 581                         free_memmap(node, prev_bank_end, bank_start);
 582
 583                 prev_bank_end = (mi->bank[i].start +
 584                                  mi->bank[i].size) >> PAGE_SHIFT;
 585         }
 586 }
 587
 588 /*
 589  * mem_init() marks the free areas in the mem_map and tells us how much
 590  * memory is free.  This is done after various parts of the system have
 591  * claimed their memory after the kernel image.
 592  */
 593 void __init mem_init(void)
 594 {
 595         unsigned int codepages, datapages, initpages;
 596         int i, node;
 597
 598         codepages = &_etext - &_text;
 599         datapages = &_end - &__data_start;
 600         initpages = &__init_end - &__init_begin;
 601
 602 #ifndef CONFIG_DISCONTIGMEM
 603         max_mapnr   = virt_to_page(high_memory) - mem_map;
 604 #endif
 605
 606         /* this will put all unused low memory onto the freelists */
 607         for_each_online_node(node) {
 608                 pg_data_t *pgdat = NODE_DATA(node);
 609
 610                 free_unused_memmap_node(node, &meminfo);
 611
 612                 if (pgdat->node_spanned_pages != 0)
 613                         totalram_pages += free_all_bootmem_node(pgdat);
 614         }
 615
 616 #ifdef CONFIG_SA1111
 617         /* now that our DMA memory is actually so designated, we can free it */
 618         free_area(PAGE_OFFSET, (unsigned long)swapper_pg_dir, NULL);
 619 #endif
 620
 621         /*
 622          * Since our memory may not be contiguous, calculate the
 623          * real number of pages we have in this system
 624          */
 625         printk(KERN_INFO "Memory:");
 626
 627         num_physpages = 0;
 628         for (i = 0; i < meminfo.nr_banks; i++) {
 629                 num_physpages += meminfo.bank[i].size >> PAGE_SHIFT;
 630                 printk(" %ldMB", meminfo.bank[i].size >> 20);
 631         }
 632
 633         printk(" = %luMB total\n", num_physpages >> (20 - PAGE_SHIFT));
 634         printk(KERN_NOTICE "Memory: %luKB available (%dK code, "
 635                 "%dK data, %dK init)\n",
 636                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 637                 codepages >> 10, datapages >> 10, initpages >> 10);
 638
 639         if (PAGE_SIZE >= 16384 && num_physpages <= 128) {
 640                 extern int sysctl_overcommit_memory;
 641                 /*
 642                  * On a machine this small we won't get
 643                  * anywhere without overcommit, so turn
 644                  * it on by default.
 645                  */
 646                 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
 647         }
 648 }
 649
 650 void free_initmem(void)
 651 {
 652         if (!machine_is_integrator() && !machine_is_cintegrator()) {
 653                 free_area((unsigned long)(&__init_begin),
 654                           (unsigned long)(&__init_end),
 655                           "init");
 656         }
 657 }
 658
 659 #ifdef CONFIG_BLK_DEV_INITRD
 660
 661 static int keep_initrd;
 662
 663 void free_initrd_mem(unsigned long start, unsigned long end)
 664 {
 665         if (!keep_initrd)
 666                 free_area(start, end, "initrd");
 667 }
 668
 669 static int __init keepinitrd_setup(char *__unused)
 670 {
 671         keep_initrd = 1;
 672         return 1;
 673 }
 674
 675 __setup("keepinitrd", keepinitrd_setup);
 676 #endif