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 V_PFN_DOWN(x)   O_PFN_DOWN(__pa(x))
  97
  98 #define O_PFN_UP(x)     (PAGE_ALIGN(x) >> PAGE_SHIFT)
  99 #define V_PFN_UP(x)     O_PFN_UP(__pa(x))
 100
 101 #define PFN_SIZE(x)     ((x) >> PAGE_SHIFT)
 102 #define PFN_RANGE(s,e)  PFN_SIZE(PAGE_ALIGN((unsigned long)(e)) - \
 103                                 (((unsigned long)(s)) & PAGE_MASK))
 104
 105 /*
 106  * FIXME: We really want to avoid allocating the bootmap bitmap
 107  * over the top of the initrd.  Hopefully, this is located towards
 108  * the start of a bank, so if we allocate the bootmap bitmap at
 109  * the end, we won't clash.
 110  */
 111 static unsigned int __init
 112 find_bootmap_pfn(int node, struct meminfo *mi, unsigned int bootmap_pages)
 113 {
 114         unsigned int start_pfn, bank, bootmap_pfn;
 115
 116         start_pfn   = V_PFN_UP(&_end);
 117         bootmap_pfn = 0;
 118
 119         for (bank = 0; bank < mi->nr_banks; bank ++) {
 120                 unsigned int start, end;
 121
 122                 if (mi->bank[bank].node != node)
 123                         continue;
 124
 125                 start = O_PFN_UP(mi->bank[bank].start);
 126                 end   = O_PFN_DOWN(mi->bank[bank].size +
 127                                    mi->bank[bank].start);
 128
 129                 if (end < start_pfn)
 130                         continue;
 131
 132                 if (start < start_pfn)
 133                         start = start_pfn;
 134
 135                 if (end <= start)
 136                         continue;
 137
 138                 if (end - start >= bootmap_pages) {
 139                         bootmap_pfn = start;
 140                         break;
 141                 }
 142         }
 143
 144         if (bootmap_pfn == 0)
 145                 BUG();
 146
 147         return bootmap_pfn;
 148 }
 149
 150 /*
 151  * Scan the memory info structure and pull out:
 152  *  - the end of memory
 153  *  - the number of nodes
 154  *  - the pfn range of each node
 155  *  - the number of bootmem bitmap pages
 156  */
 157 static unsigned int __init
 158 find_memend_and_nodes(struct meminfo *mi, struct node_info *np)
 159 {
 160         unsigned int i, bootmem_pages = 0, memend_pfn = 0;
 161
 162         for (i = 0; i < MAX_NUMNODES; i++) {
 163                 np[i].start = -1U;
 164                 np[i].end = 0;
 165                 np[i].bootmap_pages = 0;
 166         }
 167
 168         for (i = 0; i < mi->nr_banks; i++) {
 169                 unsigned long start, end;
 170                 int node;
 171
 172                 if (mi->bank[i].size == 0) {
 173                         /*
 174                          * Mark this bank with an invalid node number
 175                          */
 176                         mi->bank[i].node = -1;
 177                         continue;
 178                 }
 179
 180                 node = mi->bank[i].node;
 181
 182                 /*
 183                  * Make sure we haven't exceeded the maximum number of nodes
 184                  * that we have in this configuration.  If we have, we're in
 185                  * trouble.  (maybe we ought to limit, instead of bugging?)
 186                  */
 187                 if (node >= MAX_NUMNODES)
 188                         BUG();
 189                 node_set_online(node);
 190
 191                 /*
 192                  * Get the start and end pfns for this bank
 193                  */
 194                 start = O_PFN_UP(mi->bank[i].start);
 195                 end   = O_PFN_DOWN(mi->bank[i].start + mi->bank[i].size);
 196
 197                 if (np[node].start > start)
 198                         np[node].start = start;
 199
 200                 if (np[node].end < end)
 201                         np[node].end = end;
 202
 203                 if (memend_pfn < end)
 204                         memend_pfn = end;
 205         }
 206
 207         /*
 208          * Calculate the number of pages we require to
 209          * store the bootmem bitmaps.
 210          */
 211         for_each_online_node(i) {
 212                 if (np[i].end == 0)
 213                         continue;
 214
 215                 np[i].bootmap_pages = bootmem_bootmap_pages(np[i].end -
 216                                                             np[i].start);
 217                 bootmem_pages += np[i].bootmap_pages;
 218         }
 219
 220         high_memory = __va(memend_pfn << PAGE_SHIFT);
 221
 222         /*
 223          * This doesn't seem to be used by the Linux memory
 224          * manager any more.  If we can get rid of it, we
 225          * also get rid of some of the stuff above as well.
 226          */
 227         max_low_pfn = memend_pfn - O_PFN_DOWN(PHYS_OFFSET);
 228         max_pfn = memend_pfn - O_PFN_DOWN(PHYS_OFFSET);
 229
 230         return bootmem_pages;
 231 }
 232
 233 static int __init check_initrd(struct meminfo *mi)
 234 {
 235         int initrd_node = -2;
 236 #ifdef CONFIG_BLK_DEV_INITRD
 237         unsigned long end = phys_initrd_start + phys_initrd_size;
 238
 239         /*
 240          * Make sure that the initrd is within a valid area of
 241          * memory.
 242          */
 243         if (phys_initrd_size) {
 244                 unsigned int i;
 245
 246                 initrd_node = -1;
 247
 248                 for (i = 0; i < mi->nr_banks; i++) {
 249                         unsigned long bank_end;
 250
 251                         bank_end = mi->bank[i].start + mi->bank[i].size;
 252
 253                         if (mi->bank[i].start <= phys_initrd_start &&
 254                             end <= bank_end)
 255                                 initrd_node = mi->bank[i].node;
 256                 }
 257         }
 258
 259         if (initrd_node == -1) {
 260                 printk(KERN_ERR "initrd (0x%08lx - 0x%08lx) extends beyond "
 261                        "physical memory - disabling initrd\n",
 262                        phys_initrd_start, end);
 263                 phys_initrd_start = phys_initrd_size = 0;
 264         }
 265 #endif
 266
 267         return initrd_node;
 268 }
 269
 270 /*
 271  * Reserve the various regions of node 0
 272  */
 273 static __init void reserve_node_zero(unsigned int bootmap_pfn, unsigned int bootmap_pages)
 274 {
 275         pg_data_t *pgdat = NODE_DATA(0);
 276         unsigned long res_size = 0;
 277
 278         /*
 279          * Register the kernel text and data with bootmem.
 280          * Note that this can only be in node 0.
 281          */
 282 #ifdef CONFIG_XIP_KERNEL
 283         reserve_bootmem_node(pgdat, __pa(&__data_start), &_end - &__data_start);
 284 #else
 285         reserve_bootmem_node(pgdat, __pa(&_stext), &_end - &_stext);
 286 #endif
 287
 288         /*
 289          * Reserve the page tables.  These are already in use,
 290          * and can only be in node 0.
 291          */
 292         reserve_bootmem_node(pgdat, __pa(swapper_pg_dir),
 293                              PTRS_PER_PGD * sizeof(pgd_t));
 294
 295         /*
 296          * And don't forget to reserve the allocator bitmap,
 297          * which will be freed later.
 298          */
 299         reserve_bootmem_node(pgdat, bootmap_pfn << PAGE_SHIFT,
 300                              bootmap_pages << PAGE_SHIFT);
 301
 302         /*
 303          * Hmm... This should go elsewhere, but we really really need to
 304          * stop things allocating the low memory; ideally we need a better
 305          * implementation of GFP_DMA which does not assume that DMA-able
 306          * memory starts at zero.
 307          */
 308         if (machine_is_integrator() || machine_is_cintegrator())
 309                 res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
 310
 311         /*
 312          * These should likewise go elsewhere.  They pre-reserve the
 313          * screen memory region at the start of main system memory.
 314          */
 315         if (machine_is_edb7211())
 316                 res_size = 0x00020000;
 317         if (machine_is_p720t())
 318                 res_size = 0x00014000;
 319
 320 #ifdef CONFIG_SA1111
 321         /*
 322          * Because of the SA1111 DMA bug, we want to preserve our
 323          * precious DMA-able memory...
 324          */
 325         res_size = __pa(swapper_pg_dir) - PHYS_OFFSET;
 326 #endif
 327         if (res_size)
 328                 reserve_bootmem_node(pgdat, PHYS_OFFSET, res_size);
 329 }
 330
 331 /*
 332  * Register all available RAM in this node with the bootmem allocator.
 333  */
 334 static inline void free_bootmem_node_bank(int node, struct meminfo *mi)
 335 {
 336         pg_data_t *pgdat = NODE_DATA(node);
 337         int bank;
 338
 339         for (bank = 0; bank < mi->nr_banks; bank++)
 340                 if (mi->bank[bank].node == node)
 341                         free_bootmem_node(pgdat, mi->bank[bank].start,
 342                                           mi->bank[bank].size);
 343 }
 344
 345 /*
 346  * Initialise the bootmem allocator for all nodes.  This is called
 347  * early during the architecture specific initialisation.
 348  */
 349 static void __init bootmem_init(struct meminfo *mi)
 350 {
 351         struct node_info node_info[MAX_NUMNODES], *np = node_info;
 352         unsigned int bootmap_pages, bootmap_pfn, map_pg;
 353         int node, initrd_node;
 354
 355         bootmap_pages = find_memend_and_nodes(mi, np);
 356         bootmap_pfn   = find_bootmap_pfn(0, mi, bootmap_pages);
 357         initrd_node   = check_initrd(mi);
 358
 359         map_pg = bootmap_pfn;
 360
 361         /*
 362          * Initialise the bootmem nodes.
 363          *
 364          * What we really want to do is:
 365          *
 366          *   unmap_all_regions_except_kernel();
 367          *   for_each_node_in_reverse_order(node) {
 368          *     map_node(node);
 369          *     allocate_bootmem_map(node);
 370          *     init_bootmem_node(node);
 371          *     free_bootmem_node(node);
 372          *   }
 373          *
 374          * but this is a 2.5-type change.  For now, we just set
 375          * the nodes up in reverse order.
 376          *
 377          * (we could also do with rolling bootmem_init and paging_init
 378          * into one generic "memory_init" type function).
 379          */
 380         np += num_online_nodes() - 1;
 381         for (node = num_online_nodes() - 1; node >= 0; node--, np--) {
 382                 /*
 383                  * If there are no pages in this node, ignore it.
 384                  * Note that node 0 must always have some pages.
 385                  */
 386                 if (np->end == 0 || !node_online(node)) {
 387                         if (node == 0)
 388                                 BUG();
 389                         continue;
 390                 }
 391
 392                 /*
 393                  * Initialise the bootmem allocator.
 394                  */
 395                 init_bootmem_node(NODE_DATA(node), map_pg, np->start, np->end);
 396                 free_bootmem_node_bank(node, mi);
 397                 map_pg += np->bootmap_pages;
 398
 399                 /*
 400                  * If this is node 0, we need to reserve some areas ASAP -
 401                  * we may use bootmem on node 0 to setup the other nodes.
 402                  */
 403                 if (node == 0)
 404                         reserve_node_zero(bootmap_pfn, bootmap_pages);
 405         }
 406
 407
 408 #ifdef CONFIG_BLK_DEV_INITRD
 409         if (phys_initrd_size && initrd_node >= 0) {
 410                 reserve_bootmem_node(NODE_DATA(initrd_node), phys_initrd_start,
 411                                      phys_initrd_size);
 412                 initrd_start = __phys_to_virt(phys_initrd_start);
 413                 initrd_end = initrd_start + phys_initrd_size;
 414         }
 415 #endif
 416
 417         BUG_ON(map_pg != bootmap_pfn + bootmap_pages);
 418 }
 419
 420 /*
 421  * paging_init() sets up the page tables, initialises the zone memory
 422  * maps, and sets up the zero page, bad page and bad page tables.
 423  */
 424 void __init paging_init(struct meminfo *mi, struct machine_desc *mdesc)
 425 {
 426         void *zero_page;
 427         int node;
 428
 429         bootmem_init(mi);
 430
 431         memcpy(&meminfo, mi, sizeof(meminfo));
 432
 433         /*
 434          * allocate the zero page.  Note that we count on this going ok.
 435          */
 436         zero_page = alloc_bootmem_low_pages(PAGE_SIZE);
 437
 438         /*
 439          * initialise the page tables.
 440          */
 441         memtable_init(mi);
 442         if (mdesc->map_io)
 443                 mdesc->map_io();
 444         flush_tlb_all();
 445
 446         /*
 447          * initialise the zones within each node
 448          */
 449         for_each_online_node(node) {
 450                 unsigned long zone_size[MAX_NR_ZONES];
 451                 unsigned long zhole_size[MAX_NR_ZONES];
 452                 struct bootmem_data *bdata;
 453                 pg_data_t *pgdat;
 454                 int i;
 455
 456                 /*
 457                  * Initialise the zone size information.
 458                  */
 459                 for (i = 0; i < MAX_NR_ZONES; i++) {
 460                         zone_size[i]  = 0;
 461                         zhole_size[i] = 0;
 462                 }
 463
 464                 pgdat = NODE_DATA(node);
 465                 bdata = pgdat->bdata;
 466
 467                 /*
 468                  * The size of this node has already been determined.
 469                  * If we need to do anything fancy with the allocation
 470                  * of this memory to the zones, now is the time to do
 471                  * it.
 472                  */
 473                 zone_size[0] = bdata->node_low_pfn -
 474                                 (bdata->node_boot_start >> PAGE_SHIFT);
 475
 476                 /*
 477                  * If this zone has zero size, skip it.
 478                  */
 479                 if (!zone_size[0])
 480                         continue;
 481
 482                 /*
 483                  * For each bank in this node, calculate the size of the
 484                  * holes.  holes = node_size - sum(bank_sizes_in_node)
 485                  */
 486                 zhole_size[0] = zone_size[0];
 487                 for (i = 0; i < mi->nr_banks; i++) {
 488                         if (mi->bank[i].node != node)
 489                                 continue;
 490
 491                         zhole_size[0] -= mi->bank[i].size >> PAGE_SHIFT;
 492                 }
 493
 494                 /*
 495                  * Adjust the sizes according to any special
 496                  * requirements for this machine type.
 497                  */
 498                 arch_adjust_zones(node, zone_size, zhole_size);
 499
 500                 free_area_init_node(node, pgdat, zone_size,
 501                                 bdata->node_boot_start >> PAGE_SHIFT, zhole_size);
 502         }
 503
 504         /*
 505          * finish off the bad pages once
 506          * the mem_map is initialised
 507          */
 508         memzero(zero_page, PAGE_SIZE);
 509         empty_zero_page = virt_to_page(zero_page);
 510         flush_dcache_page(empty_zero_page);
 511 }
 512
 513 static inline void free_area(unsigned long addr, unsigned long end, char *s)
 514 {
 515         unsigned int size = (end - addr) >> 10;
 516
 517         for (; addr < end; addr += PAGE_SIZE) {
 518                 struct page *page = virt_to_page(addr);
 519                 ClearPageReserved(page);
 520                 set_page_count(page, 1);
 521                 free_page(addr);
 522                 totalram_pages++;
 523         }
 524
 525         if (size && s)
 526                 printk(KERN_INFO "Freeing %s memory: %dK\n", s, size);
 527 }
 528
 529 /*
 530  * mem_init() marks the free areas in the mem_map and tells us how much
 531  * memory is free.  This is done after various parts of the system have
 532  * claimed their memory after the kernel image.
 533  */
 534 void __init mem_init(void)
 535 {
 536         unsigned int codepages, datapages, initpages;
 537         int i, node;
 538
 539         codepages = &_etext - &_text;
 540         datapages = &_end - &__data_start;
 541         initpages = &__init_end - &__init_begin;
 542
 543 #ifndef CONFIG_DISCONTIGMEM
 544         max_mapnr   = virt_to_page(high_memory) - mem_map;
 545 #endif
 546
 547         /*
 548          * We may have non-contiguous memory.
 549          */
 550         if (meminfo.nr_banks != 1)
 551                 create_memmap_holes(&meminfo);
 552
 553         /* this will put all unused low memory onto the freelists */
 554         for_each_online_node(node) {
 555                 pg_data_t *pgdat = NODE_DATA(node);
 556
 557                 if (pgdat->node_spanned_pages != 0)
 558                         totalram_pages += free_all_bootmem_node(pgdat);
 559         }
 560
 561 #ifdef CONFIG_SA1111
 562         /* now that our DMA memory is actually so designated, we can free it */
 563         free_area(PAGE_OFFSET, (unsigned long)swapper_pg_dir, NULL);
 564 #endif
 565
 566         /*
 567          * Since our memory may not be contiguous, calculate the
 568          * real number of pages we have in this system
 569          */
 570         printk(KERN_INFO "Memory:");
 571
 572         num_physpages = 0;
 573         for (i = 0; i < meminfo.nr_banks; i++) {
 574                 num_physpages += meminfo.bank[i].size >> PAGE_SHIFT;
 575                 printk(" %ldMB", meminfo.bank[i].size >> 20);
 576         }
 577
 578         printk(" = %luMB total\n", num_physpages >> (20 - PAGE_SHIFT));
 579         printk(KERN_NOTICE "Memory: %luKB available (%dK code, "
 580                 "%dK data, %dK init)\n",
 581                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 582                 codepages >> 10, datapages >> 10, initpages >> 10);
 583
 584         if (PAGE_SIZE >= 16384 && num_physpages <= 128) {
 585                 extern int sysctl_overcommit_memory;
 586                 /*
 587                  * On a machine this small we won't get
 588                  * anywhere without overcommit, so turn
 589                  * it on by default.
 590                  */
 591                 sysctl_overcommit_memory = OVERCOMMIT_ALWAYS;
 592         }
 593 }
 594
 595 void free_initmem(void)
 596 {
 597         if (!machine_is_integrator() && !machine_is_cintegrator()) {
 598                 free_area((unsigned long)(&__init_begin),
 599                           (unsigned long)(&__init_end),
 600                           "init");
 601         }
 602 }
 603
 604 #ifdef CONFIG_BLK_DEV_INITRD
 605
 606 static int keep_initrd;
 607
 608 void free_initrd_mem(unsigned long start, unsigned long end)
 609 {
 610         if (!keep_initrd)
 611                 free_area(start, end, "initrd");
 612 }
 613
 614 static int __init keepinitrd_setup(char *__unused)
 615 {
 616         keep_initrd = 1;
 617         return 1;
 618 }
 619
 620 __setup("keepinitrd", keepinitrd_setup);
 621 #endif