linux/arch/ia64/mm/discontig.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2000, 2003 Silicon Graphics, Inc.  All rights reserved.
   3 * Copyright (c) 2001 Intel Corp.
   4 * Copyright (c) 2001 Tony Luck <tony.luck@intel.com>
   5 * Copyright (c) 2002 NEC Corp.
   6 * Copyright (c) 2002 Kimio Suganuma <k-suganuma@da.jp.nec.com>
   7 * Copyright (c) 2004 Silicon Graphics, Inc
   8 *      Russ Anderson <rja@sgi.com>
   9 *      Jesse Barnes <jbarnes@sgi.com>
  10 *      Jack Steiner <steiner@sgi.com>
  11 */
  12
  13/*
  14 * Platform initialization for Discontig Memory
  15 */
  16
  17#include <linux/kernel.h>
  18#include <linux/mm.h>
  19#include <linux/nmi.h>
  20#include <linux/swap.h>
  21#include <linux/bootmem.h>
  22#include <linux/acpi.h>
  23#include <linux/efi.h>
  24#include <linux/nodemask.h>
  25#include <asm/pgalloc.h>
  26#include <asm/tlb.h>
  27#include <asm/meminit.h>
  28#include <asm/numa.h>
  29#include <asm/sections.h>
  30
  31/*
  32 * Track per-node information needed to setup the boot memory allocator, the
  33 * per-node areas, and the real VM.
  34 */
  35struct early_node_data {
  36        struct ia64_node_data *node_data;
  37        unsigned long pernode_addr;
  38        unsigned long pernode_size;
  39        unsigned long num_physpages;
  40#ifdef CONFIG_ZONE_DMA
  41        unsigned long num_dma_physpages;
  42#endif
  43        unsigned long min_pfn;
  44        unsigned long max_pfn;
  45};
  46
  47static struct early_node_data mem_data[MAX_NUMNODES] __initdata;
  48static nodemask_t memory_less_mask __initdata;
  49
  50pg_data_t *pgdat_list[MAX_NUMNODES];
  51
  52/*
  53 * To prevent cache aliasing effects, align per-node structures so that they
  54 * start at addresses that are strided by node number.
  55 */
  56#define MAX_NODE_ALIGN_OFFSET   (32 * 1024 * 1024)
  57#define NODEDATA_ALIGN(addr, node)                                              \
  58        ((((addr) + 1024*1024-1) & ~(1024*1024-1)) +                            \
  59             (((node)*PERCPU_PAGE_SIZE) & (MAX_NODE_ALIGN_OFFSET - 1)))
  60
  61/**
  62 * build_node_maps - callback to setup bootmem structs for each node
  63 * @start: physical start of range
  64 * @len: length of range
  65 * @node: node where this range resides
  66 *
  67 * We allocate a struct bootmem_data for each piece of memory that we wish to
  68 * treat as a virtually contiguous block (i.e. each node). Each such block
  69 * must start on an %IA64_GRANULE_SIZE boundary, so we round the address down
  70 * if necessary.  Any non-existent pages will simply be part of the virtual
  71 * memmap.  We also update min_low_pfn and max_low_pfn here as we receive
  72 * memory ranges from the caller.
  73 */
  74static int __init build_node_maps(unsigned long start, unsigned long len,
  75                                  int node)
  76{
  77        unsigned long spfn, epfn, end = start + len;
  78        struct bootmem_data *bdp = &bootmem_node_data[node];
  79
  80        epfn = GRANULEROUNDUP(end) >> PAGE_SHIFT;
  81        spfn = GRANULEROUNDDOWN(start) >> PAGE_SHIFT;
  82
  83        if (!bdp->node_low_pfn) {
  84                bdp->node_min_pfn = spfn;
  85                bdp->node_low_pfn = epfn;
  86        } else {
  87                bdp->node_min_pfn = min(spfn, bdp->node_min_pfn);
  88                bdp->node_low_pfn = max(epfn, bdp->node_low_pfn);
  89        }
  90
  91        return 0;
  92}
  93
  94/**
  95 * early_nr_cpus_node - return number of cpus on a given node
  96 * @node: node to check
  97 *
  98 * Count the number of cpus on @node.  We can't use nr_cpus_node() yet because
  99 * acpi_boot_init() (which builds the node_to_cpu_mask array) hasn't been
 100 * called yet.  Note that node 0 will also count all non-existent cpus.
 101 */
 102static int __meminit early_nr_cpus_node(int node)
 103{
 104        int cpu, n = 0;
 105
 106        for_each_possible_early_cpu(cpu)
 107                if (node == node_cpuid[cpu].nid)
 108                        n++;
 109
 110        return n;
 111}
 112
 113/**
 114 * compute_pernodesize - compute size of pernode data
 115 * @node: the node id.
 116 */
 117static unsigned long __meminit compute_pernodesize(int node)
 118{
 119        unsigned long pernodesize = 0, cpus;
 120
 121        cpus = early_nr_cpus_node(node);
 122        pernodesize += PERCPU_PAGE_SIZE * cpus;
 123        pernodesize += node * L1_CACHE_BYTES;
 124        pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t));
 125        pernodesize += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
 126        pernodesize += L1_CACHE_ALIGN(sizeof(pg_data_t));
 127        pernodesize = PAGE_ALIGN(pernodesize);
 128        return pernodesize;
 129}
 130
 131/**
 132 * per_cpu_node_setup - setup per-cpu areas on each node
 133 * @cpu_data: per-cpu area on this node
 134 * @node: node to setup
 135 *
 136 * Copy the static per-cpu data into the region we just set aside and then
 137 * setup __per_cpu_offset for each CPU on this node.  Return a pointer to
 138 * the end of the area.
 139 */
 140static void *per_cpu_node_setup(void *cpu_data, int node)
 141{
 142#ifdef CONFIG_SMP
 143        int cpu;
 144
 145        for_each_possible_early_cpu(cpu) {
 146                if (cpu == 0) {
 147                        void *cpu0_data = __cpu0_per_cpu;
 148                        __per_cpu_offset[cpu] = (char*)cpu0_data -
 149                                __per_cpu_start;
 150                } else if (node == node_cpuid[cpu].nid) {
 151                        memcpy(__va(cpu_data), __phys_per_cpu_start,
 152                               __per_cpu_end - __per_cpu_start);
 153                        __per_cpu_offset[cpu] = (char*)__va(cpu_data) -
 154                                __per_cpu_start;
 155                        cpu_data += PERCPU_PAGE_SIZE;
 156                }
 157        }
 158#endif
 159        return cpu_data;
 160}
 161
 162/**
 163 * fill_pernode - initialize pernode data.
 164 * @node: the node id.
 165 * @pernode: physical address of pernode data
 166 * @pernodesize: size of the pernode data
 167 */
 168static void __init fill_pernode(int node, unsigned long pernode,
 169        unsigned long pernodesize)
 170{
 171        void *cpu_data;
 172        int cpus = early_nr_cpus_node(node);
 173        struct bootmem_data *bdp = &bootmem_node_data[node];
 174
 175        mem_data[node].pernode_addr = pernode;
 176        mem_data[node].pernode_size = pernodesize;
 177        memset(__va(pernode), 0, pernodesize);
 178
 179        cpu_data = (void *)pernode;
 180        pernode += PERCPU_PAGE_SIZE * cpus;
 181        pernode += node * L1_CACHE_BYTES;
 182
 183        pgdat_list[node] = __va(pernode);
 184        pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
 185
 186        mem_data[node].node_data = __va(pernode);
 187        pernode += L1_CACHE_ALIGN(sizeof(struct ia64_node_data));
 188
 189        pgdat_list[node]->bdata = bdp;
 190        pernode += L1_CACHE_ALIGN(sizeof(pg_data_t));
 191
 192        cpu_data = per_cpu_node_setup(cpu_data, node);
 193
 194        return;
 195}
 196
 197/**
 198 * find_pernode_space - allocate memory for memory map and per-node structures
 199 * @start: physical start of range
 200 * @len: length of range
 201 * @node: node where this range resides
 202 *
 203 * This routine reserves space for the per-cpu data struct, the list of
 204 * pg_data_ts and the per-node data struct.  Each node will have something like
 205 * the following in the first chunk of addr. space large enough to hold it.
 206 *
 207 *    ________________________
 208 *   |                        |
 209 *   |~~~~~~~~~~~~~~~~~~~~~~~~| <-- NODEDATA_ALIGN(start, node) for the first
 210 *   |    PERCPU_PAGE_SIZE *  |     start and length big enough
 211 *   |    cpus_on_this_node   | Node 0 will also have entries for all non-existent cpus.
 212 *   |------------------------|
 213 *   |   local pg_data_t *    |
 214 *   |------------------------|
 215 *   |  local ia64_node_data  |
 216 *   |------------------------|
 217 *   |          ???           |
 218 *   |________________________|
 219 *
 220 * Once this space has been set aside, the bootmem maps are initialized.  We
 221 * could probably move the allocation of the per-cpu and ia64_node_data space
 222 * outside of this function and use alloc_bootmem_node(), but doing it here
 223 * is straightforward and we get the alignments we want so...
 224 */
 225static int __init find_pernode_space(unsigned long start, unsigned long len,
 226                                     int node)
 227{
 228        unsigned long spfn, epfn;
 229        unsigned long pernodesize = 0, pernode, pages, mapsize;
 230        struct bootmem_data *bdp = &bootmem_node_data[node];
 231
 232        spfn = start >> PAGE_SHIFT;
 233        epfn = (start + len) >> PAGE_SHIFT;
 234
 235        pages = bdp->node_low_pfn - bdp->node_min_pfn;
 236        mapsize = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
 237
 238        /*
 239         * Make sure this memory falls within this node's usable memory
 240         * since we may have thrown some away in build_maps().
 241         */
 242        if (spfn < bdp->node_min_pfn || epfn > bdp->node_low_pfn)
 243                return 0;
 244
 245        /* Don't setup this node's local space twice... */
 246        if (mem_data[node].pernode_addr)
 247                return 0;
 248
 249        /*
 250         * Calculate total size needed, incl. what's necessary
 251         * for good alignment and alias prevention.
 252         */
 253        pernodesize = compute_pernodesize(node);
 254        pernode = NODEDATA_ALIGN(start, node);
 255
 256        /* Is this range big enough for what we want to store here? */
 257        if (start + len > (pernode + pernodesize + mapsize))
 258                fill_pernode(node, pernode, pernodesize);
 259
 260        return 0;
 261}
 262
 263/**
 264 * free_node_bootmem - free bootmem allocator memory for use
 265 * @start: physical start of range
 266 * @len: length of range
 267 * @node: node where this range resides
 268 *
 269 * Simply calls the bootmem allocator to free the specified ranged from
 270 * the given pg_data_t's bdata struct.  After this function has been called
 271 * for all the entries in the EFI memory map, the bootmem allocator will
 272 * be ready to service allocation requests.
 273 */
 274static int __init free_node_bootmem(unsigned long start, unsigned long len,
 275                                    int node)
 276{
 277        free_bootmem_node(pgdat_list[node], start, len);
 278
 279        return 0;
 280}
 281
 282/**
 283 * reserve_pernode_space - reserve memory for per-node space
 284 *
 285 * Reserve the space used by the bootmem maps & per-node space in the boot
 286 * allocator so that when we actually create the real mem maps we don't
 287 * use their memory.
 288 */
 289static void __init reserve_pernode_space(void)
 290{
 291        unsigned long base, size, pages;
 292        struct bootmem_data *bdp;
 293        int node;
 294
 295        for_each_online_node(node) {
 296                pg_data_t *pdp = pgdat_list[node];
 297
 298                if (node_isset(node, memory_less_mask))
 299                        continue;
 300
 301                bdp = pdp->bdata;
 302
 303                /* First the bootmem_map itself */
 304                pages = bdp->node_low_pfn - bdp->node_min_pfn;
 305                size = bootmem_bootmap_pages(pages) << PAGE_SHIFT;
 306                base = __pa(bdp->node_bootmem_map);
 307                reserve_bootmem_node(pdp, base, size, BOOTMEM_DEFAULT);
 308
 309                /* Now the per-node space */
 310                size = mem_data[node].pernode_size;
 311                base = __pa(mem_data[node].pernode_addr);
 312                reserve_bootmem_node(pdp, base, size, BOOTMEM_DEFAULT);
 313        }
 314}
 315
 316static void __meminit scatter_node_data(void)
 317{
 318        pg_data_t **dst;
 319        int node;
 320
 321        /*
 322         * for_each_online_node() can't be used at here.
 323         * node_online_map is not set for hot-added nodes at this time,
 324         * because we are halfway through initialization of the new node's
 325         * structures.  If for_each_online_node() is used, a new node's
 326         * pg_data_ptrs will be not initialized. Instead of using it,
 327         * pgdat_list[] is checked.
 328         */
 329        for_each_node(node) {
 330                if (pgdat_list[node]) {
 331                        dst = LOCAL_DATA_ADDR(pgdat_list[node])->pg_data_ptrs;
 332                        memcpy(dst, pgdat_list, sizeof(pgdat_list));
 333                }
 334        }
 335}
 336
 337/**
 338 * initialize_pernode_data - fixup per-cpu & per-node pointers
 339 *
 340 * Each node's per-node area has a copy of the global pg_data_t list, so
 341 * we copy that to each node here, as well as setting the per-cpu pointer
 342 * to the local node data structure.  The active_cpus field of the per-node
 343 * structure gets setup by the platform_cpu_init() function later.
 344 */
 345static void __init initialize_pernode_data(void)
 346{
 347        int cpu, node;
 348
 349        scatter_node_data();
 350
 351#ifdef CONFIG_SMP
 352        /* Set the node_data pointer for each per-cpu struct */
 353        for_each_possible_early_cpu(cpu) {
 354                node = node_cpuid[cpu].nid;
 355                per_cpu(cpu_info, cpu).node_data = mem_data[node].node_data;
 356        }
 357#else
 358        {
 359                struct cpuinfo_ia64 *cpu0_cpu_info;
 360                cpu = 0;
 361                node = node_cpuid[cpu].nid;
 362                cpu0_cpu_info = (struct cpuinfo_ia64 *)(__phys_per_cpu_start +
 363                        ((char *)&per_cpu__cpu_info - __per_cpu_start));
 364                cpu0_cpu_info->node_data = mem_data[node].node_data;
 365        }
 366#endif /* CONFIG_SMP */
 367}
 368
 369/**
 370 * memory_less_node_alloc - * attempt to allocate memory on the best NUMA slit
 371 *      node but fall back to any other node when __alloc_bootmem_node fails
 372 *      for best.
 373 * @nid: node id
 374 * @pernodesize: size of this node's pernode data
 375 */
 376static void __init *memory_less_node_alloc(int nid, unsigned long pernodesize)
 377{
 378        void *ptr = NULL;
 379        u8 best = 0xff;
 380        int bestnode = -1, node, anynode = 0;
 381
 382        for_each_online_node(node) {
 383                if (node_isset(node, memory_less_mask))
 384                        continue;
 385                else if (node_distance(nid, node) < best) {
 386                        best = node_distance(nid, node);
 387                        bestnode = node;
 388                }
 389                anynode = node;
 390        }
 391
 392        if (bestnode == -1)
 393                bestnode = anynode;
 394
 395        ptr = __alloc_bootmem_node(pgdat_list[bestnode], pernodesize,
 396                PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
 397
 398        return ptr;
 399}
 400
 401/**
 402 * memory_less_nodes - allocate and initialize CPU only nodes pernode
 403 *      information.
 404 */
 405static void __init memory_less_nodes(void)
 406{
 407        unsigned long pernodesize;
 408        void *pernode;
 409        int node;
 410
 411        for_each_node_mask(node, memory_less_mask) {
 412                pernodesize = compute_pernodesize(node);
 413                pernode = memory_less_node_alloc(node, pernodesize);
 414                fill_pernode(node, __pa(pernode), pernodesize);
 415        }
 416
 417        return;
 418}
 419
 420/**
 421 * find_memory - walk the EFI memory map and setup the bootmem allocator
 422 *
 423 * Called early in boot to setup the bootmem allocator, and to
 424 * allocate the per-cpu and per-node structures.
 425 */
 426void __init find_memory(void)
 427{
 428        int node;
 429
 430        reserve_memory();
 431
 432        if (num_online_nodes() == 0) {
 433                printk(KERN_ERR "node info missing!\n");
 434                node_set_online(0);
 435        }
 436
 437        nodes_or(memory_less_mask, memory_less_mask, node_online_map);
 438        min_low_pfn = -1;
 439        max_low_pfn = 0;
 440
 441        /* These actually end up getting called by call_pernode_memory() */
 442        efi_memmap_walk(filter_rsvd_memory, build_node_maps);
 443        efi_memmap_walk(filter_rsvd_memory, find_pernode_space);
 444        efi_memmap_walk(find_max_min_low_pfn, NULL);
 445
 446        for_each_online_node(node)
 447                if (bootmem_node_data[node].node_low_pfn) {
 448                        node_clear(node, memory_less_mask);
 449                        mem_data[node].min_pfn = ~0UL;
 450                }
 451
 452        efi_memmap_walk(filter_memory, register_active_ranges);
 453
 454        /*
 455         * Initialize the boot memory maps in reverse order since that's
 456         * what the bootmem allocator expects
 457         */
 458        for (node = MAX_NUMNODES - 1; node >= 0; node--) {
 459                unsigned long pernode, pernodesize, map;
 460                struct bootmem_data *bdp;
 461
 462                if (!node_online(node))
 463                        continue;
 464                else if (node_isset(node, memory_less_mask))
 465                        continue;
 466
 467                bdp = &bootmem_node_data[node];
 468                pernode = mem_data[node].pernode_addr;
 469                pernodesize = mem_data[node].pernode_size;
 470                map = pernode + pernodesize;
 471
 472                init_bootmem_node(pgdat_list[node],
 473                                  map>>PAGE_SHIFT,
 474                                  bdp->node_min_pfn,
 475                                  bdp->node_low_pfn);
 476        }
 477
 478        efi_memmap_walk(filter_rsvd_memory, free_node_bootmem);
 479
 480        reserve_pernode_space();
 481        memory_less_nodes();
 482        initialize_pernode_data();
 483
 484        max_pfn = max_low_pfn;
 485
 486        find_initrd();
 487}
 488
 489#ifdef CONFIG_SMP
 490/**
 491 * per_cpu_init - setup per-cpu variables
 492 *
 493 * find_pernode_space() does most of this already, we just need to set
 494 * local_per_cpu_offset
 495 */
 496void __cpuinit *per_cpu_init(void)
 497{
 498        int cpu;
 499        static int first_time = 1;
 500
 501        if (first_time) {
 502                first_time = 0;
 503                for_each_possible_early_cpu(cpu)
 504                        per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu];
 505        }
 506
 507        return __per_cpu_start + __per_cpu_offset[smp_processor_id()];
 508}
 509#endif /* CONFIG_SMP */
 510
 511/**
 512 * show_mem - give short summary of memory stats
 513 *
 514 * Shows a simple page count of reserved and used pages in the system.
 515 * For discontig machines, it does this on a per-pgdat basis.
 516 */
 517void show_mem(void)
 518{
 519        int i, total_reserved = 0;
 520        int total_shared = 0, total_cached = 0;
 521        unsigned long total_present = 0;
 522        pg_data_t *pgdat;
 523
 524        printk(KERN_INFO "Mem-info:\n");
 525        show_free_areas();
 526        printk(KERN_INFO "Node memory in pages:\n");
 527        for_each_online_pgdat(pgdat) {
 528                unsigned long present;
 529                unsigned long flags;
 530                int shared = 0, cached = 0, reserved = 0;
 531
 532                pgdat_resize_lock(pgdat, &flags);
 533                present = pgdat->node_present_pages;
 534                for(i = 0; i < pgdat->node_spanned_pages; i++) {
 535                        struct page *page;
 536                        if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
 537                                touch_nmi_watchdog();
 538                        if (pfn_valid(pgdat->node_start_pfn + i))
 539                                page = pfn_to_page(pgdat->node_start_pfn + i);
 540                        else {
 541                                i = vmemmap_find_next_valid_pfn(pgdat->node_id,
 542                                         i) - 1;
 543                                continue;
 544                        }
 545                        if (PageReserved(page))
 546                                reserved++;
 547                        else if (PageSwapCache(page))
 548                                cached++;
 549                        else if (page_count(page))
 550                                shared += page_count(page)-1;
 551                }
 552                pgdat_resize_unlock(pgdat, &flags);
 553                total_present += present;
 554                total_reserved += reserved;
 555                total_cached += cached;
 556                total_shared += shared;
 557                printk(KERN_INFO "Node %4d:  RAM: %11ld, rsvd: %8d, "
 558                       "shrd: %10d, swpd: %10d\n", pgdat->node_id,
 559                       present, reserved, shared, cached);
 560        }
 561        printk(KERN_INFO "%ld pages of RAM\n", total_present);
 562        printk(KERN_INFO "%d reserved pages\n", total_reserved);
 563        printk(KERN_INFO "%d pages shared\n", total_shared);
 564        printk(KERN_INFO "%d pages swap cached\n", total_cached);
 565        printk(KERN_INFO "Total of %ld pages in page table cache\n",
 566               quicklist_total_size());
 567        printk(KERN_INFO "%d free buffer pages\n", nr_free_buffer_pages());
 568}
 569
 570/**
 571 * call_pernode_memory - use SRAT to call callback functions with node info
 572 * @start: physical start of range
 573 * @len: length of range
 574 * @arg: function to call for each range
 575 *
 576 * efi_memmap_walk() knows nothing about layout of memory across nodes. Find
 577 * out to which node a block of memory belongs.  Ignore memory that we cannot
 578 * identify, and split blocks that run across multiple nodes.
 579 *
 580 * Take this opportunity to round the start address up and the end address
 581 * down to page boundaries.
 582 */
 583void call_pernode_memory(unsigned long start, unsigned long len, void *arg)
 584{
 585        unsigned long rs, re, end = start + len;
 586        void (*func)(unsigned long, unsigned long, int);
 587        int i;
 588
 589        start = PAGE_ALIGN(start);
 590        end &= PAGE_MASK;
 591        if (start >= end)
 592                return;
 593
 594        func = arg;
 595
 596        if (!num_node_memblks) {
 597                /* No SRAT table, so assume one node (node 0) */
 598                if (start < end)
 599                        (*func)(start, end - start, 0);
 600                return;
 601        }
 602
 603        for (i = 0; i < num_node_memblks; i++) {
 604                rs = max(start, node_memblk[i].start_paddr);
 605                re = min(end, node_memblk[i].start_paddr +
 606                         node_memblk[i].size);
 607
 608                if (rs < re)
 609                        (*func)(rs, re - rs, node_memblk[i].nid);
 610
 611                if (re == end)
 612                        break;
 613        }
 614}
 615
 616/**
 617 * count_node_pages - callback to build per-node memory info structures
 618 * @start: physical start of range
 619 * @len: length of range
 620 * @node: node where this range resides
 621 *
 622 * Each node has it's own number of physical pages, DMAable pages, start, and
 623 * end page frame number.  This routine will be called by call_pernode_memory()
 624 * for each piece of usable memory and will setup these values for each node.
 625 * Very similar to build_maps().
 626 */
 627static __init int count_node_pages(unsigned long start, unsigned long len, int node)
 628{
 629        unsigned long end = start + len;
 630
 631        mem_data[node].num_physpages += len >> PAGE_SHIFT;
 632#ifdef CONFIG_ZONE_DMA
 633        if (start <= __pa(MAX_DMA_ADDRESS))
 634                mem_data[node].num_dma_physpages +=
 635                        (min(end, __pa(MAX_DMA_ADDRESS)) - start) >>PAGE_SHIFT;
 636#endif
 637        start = GRANULEROUNDDOWN(start);
 638        start = ORDERROUNDDOWN(start);
 639        end = GRANULEROUNDUP(end);
 640        mem_data[node].max_pfn = max(mem_data[node].max_pfn,
 641                                     end >> PAGE_SHIFT);
 642        mem_data[node].min_pfn = min(mem_data[node].min_pfn,
 643                                     start >> PAGE_SHIFT);
 644
 645        return 0;
 646}
 647
 648/**
 649 * paging_init - setup page tables
 650 *
 651 * paging_init() sets up the page tables for each node of the system and frees
 652 * the bootmem allocator memory for general use.
 653 */
 654void __init paging_init(void)
 655{
 656        unsigned long max_dma;
 657        unsigned long pfn_offset = 0;
 658        unsigned long max_pfn = 0;
 659        int node;
 660        unsigned long max_zone_pfns[MAX_NR_ZONES];
 661
 662        max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 663
 664        efi_memmap_walk(filter_rsvd_memory, count_node_pages);
 665
 666        sparse_memory_present_with_active_regions(MAX_NUMNODES);
 667        sparse_init();
 668
 669#ifdef CONFIG_VIRTUAL_MEM_MAP
 670        vmalloc_end -= PAGE_ALIGN(ALIGN(max_low_pfn, MAX_ORDER_NR_PAGES) *
 671                sizeof(struct page));
 672        vmem_map = (struct page *) vmalloc_end;
 673        efi_memmap_walk(create_mem_map_page_table, NULL);
 674        printk("Virtual mem_map starts at 0x%p\n", vmem_map);
 675#endif
 676
 677        for_each_online_node(node) {
 678                num_physpages += mem_data[node].num_physpages;
 679                pfn_offset = mem_data[node].min_pfn;
 680
 681#ifdef CONFIG_VIRTUAL_MEM_MAP
 682                NODE_DATA(node)->node_mem_map = vmem_map + pfn_offset;
 683#endif
 684                if (mem_data[node].max_pfn > max_pfn)
 685                        max_pfn = mem_data[node].max_pfn;
 686        }
 687
 688        memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 689#ifdef CONFIG_ZONE_DMA
 690        max_zone_pfns[ZONE_DMA] = max_dma;
 691#endif
 692        max_zone_pfns[ZONE_NORMAL] = max_pfn;
 693        free_area_init_nodes(max_zone_pfns);
 694
 695        zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
 696}
 697
 698#ifdef CONFIG_MEMORY_HOTPLUG
 699pg_data_t *arch_alloc_nodedata(int nid)
 700{
 701        unsigned long size = compute_pernodesize(nid);
 702
 703        return kzalloc(size, GFP_KERNEL);
 704}
 705
 706void arch_free_nodedata(pg_data_t *pgdat)
 707{
 708        kfree(pgdat);
 709}
 710
 711void arch_refresh_nodedata(int update_node, pg_data_t *update_pgdat)
 712{
 713        pgdat_list[update_node] = update_pgdat;
 714        scatter_node_data();
 715}
 716#endif
 717
 718#ifdef CONFIG_SPARSEMEM_VMEMMAP
 719int __meminit vmemmap_populate(struct page *start_page,
 720                                                unsigned long size, int node)
 721{
 722        return vmemmap_populate_basepages(start_page, size, node);
 723}
 724#endif
 725