arch/ia64/mm/contig.c

   1 /*
   2  * This file is subject to the terms and conditions of the GNU General Public
   3  * License.  See the file "COPYING" in the main directory of this archive
   4  * for more details.
   5  *
   6  * Copyright (C) 1998-2003 Hewlett-Packard Co
   7  *      David Mosberger-Tang <davidm@hpl.hp.com>
   8  *      Stephane Eranian <eranian@hpl.hp.com>
   9  * Copyright (C) 2000, Rohit Seth <rohit.seth@intel.com>
  10  * Copyright (C) 1999 VA Linux Systems
  11  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
  12  * Copyright (C) 2003 Silicon Graphics, Inc. All rights reserved.
  13  *
  14  * Routines used by ia64 machines with contiguous (or virtually contiguous)
  15  * memory.
  16  */
  17 #include <linux/bootmem.h>
  18 #include <linux/efi.h>
  19 #include <linux/memblock.h>
  20 #include <linux/mm.h>
  21 #include <linux/nmi.h>
  22 #include <linux/swap.h>
  23
  24 #include <asm/meminit.h>
  25 #include <asm/pgalloc.h>
  26 #include <asm/pgtable.h>
  27 #include <asm/sections.h>
  28 #include <asm/mca.h>
  29
  30 #ifdef CONFIG_VIRTUAL_MEM_MAP
  31 static unsigned long max_gap;
  32 #endif
  33
  34 /**
  35  * show_mem - give short summary of memory stats
  36  *
  37  * Shows a simple page count of reserved and used pages in the system.
  38  * For discontig machines, it does this on a per-pgdat basis.
  39  */
  40 void show_mem(unsigned int filter)
  41 {
  42         int i, total_reserved = 0;
  43         int total_shared = 0, total_cached = 0;
  44         unsigned long total_present = 0;
  45         pg_data_t *pgdat;
  46
  47         printk(KERN_INFO "Mem-info:\n");
  48         show_free_areas(filter);
  49         printk(KERN_INFO "Node memory in pages:\n");
  50         if (filter & SHOW_MEM_FILTER_PAGE_COUNT)
  51                 return;
  52         for_each_online_pgdat(pgdat) {
  53                 unsigned long present;
  54                 unsigned long flags;
  55                 int shared = 0, cached = 0, reserved = 0;
  56                 int nid = pgdat->node_id;
  57
  58                 if (skip_free_areas_node(filter, nid))
  59                         continue;
  60                 pgdat_resize_lock(pgdat, &flags);
  61                 present = pgdat->node_present_pages;
  62                 for(i = 0; i < pgdat->node_spanned_pages; i++) {
  63                         struct page *page;
  64                         if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
  65                                 touch_nmi_watchdog();
  66                         if (pfn_valid(pgdat->node_start_pfn + i))
  67                                 page = pfn_to_page(pgdat->node_start_pfn + i);
  68                         else {
  69 #ifdef CONFIG_VIRTUAL_MEM_MAP
  70                                 if (max_gap < LARGE_GAP)
  71                                         continue;
  72 #endif
  73                                 i = vmemmap_find_next_valid_pfn(nid, i) - 1;
  74                                 continue;
  75                         }
  76                         if (PageReserved(page))
  77                                 reserved++;
  78                         else if (PageSwapCache(page))
  79                                 cached++;
  80                         else if (page_count(page))
  81                                 shared += page_count(page)-1;
  82                 }
  83                 pgdat_resize_unlock(pgdat, &flags);
  84                 total_present += present;
  85                 total_reserved += reserved;
  86                 total_cached += cached;
  87                 total_shared += shared;
  88                 printk(KERN_INFO "Node %4d:  RAM: %11ld, rsvd: %8d, "
  89                        "shrd: %10d, swpd: %10d\n", nid,
  90                        present, reserved, shared, cached);
  91         }
  92         printk(KERN_INFO "%ld pages of RAM\n", total_present);
  93         printk(KERN_INFO "%d reserved pages\n", total_reserved);
  94         printk(KERN_INFO "%d pages shared\n", total_shared);
  95         printk(KERN_INFO "%d pages swap cached\n", total_cached);
  96         printk(KERN_INFO "Total of %ld pages in page table cache\n",
  97                quicklist_total_size());
  98         printk(KERN_INFO "%ld free buffer pages\n", nr_free_buffer_pages());
  99 }
 100
 101
 102 /* physical address where the bootmem map is located */
 103 unsigned long bootmap_start;
 104
 105 /**
 106  * find_bootmap_location - callback to find a memory area for the bootmap
 107  * @start: start of region
 108  * @end: end of region
 109  * @arg: unused callback data
 110  *
 111  * Find a place to put the bootmap and return its starting address in
 112  * bootmap_start.  This address must be page-aligned.
 113  */
 114 static int __init
 115 find_bootmap_location (u64 start, u64 end, void *arg)
 116 {
 117         u64 needed = *(unsigned long *)arg;
 118         u64 range_start, range_end, free_start;
 119         int i;
 120
 121 #if IGNORE_PFN0
 122         if (start == PAGE_OFFSET) {
 123                 start += PAGE_SIZE;
 124                 if (start >= end)
 125                         return 0;
 126         }
 127 #endif
 128
 129         free_start = PAGE_OFFSET;
 130
 131         for (i = 0; i < num_rsvd_regions; i++) {
 132                 range_start = max(start, free_start);
 133                 range_end   = min(end, rsvd_region[i].start & PAGE_MASK);
 134
 135                 free_start = PAGE_ALIGN(rsvd_region[i].end);
 136
 137                 if (range_end <= range_start)
 138                         continue; /* skip over empty range */
 139
 140                 if (range_end - range_start >= needed) {
 141                         bootmap_start = __pa(range_start);
 142                         return -1;      /* done */
 143                 }
 144
 145                 /* nothing more available in this segment */
 146                 if (range_end == end)
 147                         return 0;
 148         }
 149         return 0;
 150 }
 151
 152 #ifdef CONFIG_SMP
 153 static void *cpu_data;
 154 /**
 155  * per_cpu_init - setup per-cpu variables
 156  *
 157  * Allocate and setup per-cpu data areas.
 158  */
 159 void * __cpuinit
 160 per_cpu_init (void)
 161 {
 162         static bool first_time = true;
 163         void *cpu0_data = __cpu0_per_cpu;
 164         unsigned int cpu;
 165
 166         if (!first_time)
 167                 goto skip;
 168         first_time = false;
 169
 170         /*
 171          * get_free_pages() cannot be used before cpu_init() done.
 172          * BSP allocates PERCPU_PAGE_SIZE bytes for all possible CPUs
 173          * to avoid that AP calls get_zeroed_page().
 174          */
 175         for_each_possible_cpu(cpu) {
 176                 void *src = cpu == 0 ? cpu0_data : __phys_per_cpu_start;
 177
 178                 memcpy(cpu_data, src, __per_cpu_end - __per_cpu_start);
 179                 __per_cpu_offset[cpu] = (char *)cpu_data - __per_cpu_start;
 180                 per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu];
 181
 182                 /*
 183                  * percpu area for cpu0 is moved from the __init area
 184                  * which is setup by head.S and used till this point.
 185                  * Update ar.k3.  This move is ensures that percpu
 186                  * area for cpu0 is on the correct node and its
 187                  * virtual address isn't insanely far from other
 188                  * percpu areas which is important for congruent
 189                  * percpu allocator.
 190                  */
 191                 if (cpu == 0)
 192                         ia64_set_kr(IA64_KR_PER_CPU_DATA, __pa(cpu_data) -
 193                                     (unsigned long)__per_cpu_start);
 194
 195                 cpu_data += PERCPU_PAGE_SIZE;
 196         }
 197 skip:
 198         return __per_cpu_start + __per_cpu_offset[smp_processor_id()];
 199 }
 200
 201 static inline void
 202 alloc_per_cpu_data(void)
 203 {
 204         cpu_data = __alloc_bootmem(PERCPU_PAGE_SIZE * num_possible_cpus(),
 205                                    PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
 206 }
 207
 208 /**
 209  * setup_per_cpu_areas - setup percpu areas
 210  *
 211  * Arch code has already allocated and initialized percpu areas.  All
 212  * this function has to do is to teach the determined layout to the
 213  * dynamic percpu allocator, which happens to be more complex than
 214  * creating whole new ones using helpers.
 215  */
 216 void __init
 217 setup_per_cpu_areas(void)
 218 {
 219         struct pcpu_alloc_info *ai;
 220         struct pcpu_group_info *gi;
 221         unsigned int cpu;
 222         ssize_t static_size, reserved_size, dyn_size;
 223         int rc;
 224
 225         ai = pcpu_alloc_alloc_info(1, num_possible_cpus());
 226         if (!ai)
 227                 panic("failed to allocate pcpu_alloc_info");
 228         gi = &ai->groups[0];
 229
 230         /* units are assigned consecutively to possible cpus */
 231         for_each_possible_cpu(cpu)
 232                 gi->cpu_map[gi->nr_units++] = cpu;
 233
 234         /* set parameters */
 235         static_size = __per_cpu_end - __per_cpu_start;
 236         reserved_size = PERCPU_MODULE_RESERVE;
 237         dyn_size = PERCPU_PAGE_SIZE - static_size - reserved_size;
 238         if (dyn_size < 0)
 239                 panic("percpu area overflow static=%zd reserved=%zd\n",
 240                       static_size, reserved_size);
 241
 242         ai->static_size         = static_size;
 243         ai->reserved_size       = reserved_size;
 244         ai->dyn_size            = dyn_size;
 245         ai->unit_size           = PERCPU_PAGE_SIZE;
 246         ai->atom_size           = PAGE_SIZE;
 247         ai->alloc_size          = PERCPU_PAGE_SIZE;
 248
 249         rc = pcpu_setup_first_chunk(ai, __per_cpu_start + __per_cpu_offset[0]);
 250         if (rc)
 251                 panic("failed to setup percpu area (err=%d)", rc);
 252
 253         pcpu_free_alloc_info(ai);
 254 }
 255 #else
 256 #define alloc_per_cpu_data() do { } while (0)
 257 #endif /* CONFIG_SMP */
 258
 259 /**
 260  * find_memory - setup memory map
 261  *
 262  * Walk the EFI memory map and find usable memory for the system, taking
 263  * into account reserved areas.
 264  */
 265 void __init
 266 find_memory (void)
 267 {
 268         unsigned long bootmap_size;
 269
 270         reserve_memory();
 271
 272         /* first find highest page frame number */
 273         min_low_pfn = ~0UL;
 274         max_low_pfn = 0;
 275         efi_memmap_walk(find_max_min_low_pfn, NULL);
 276         max_pfn = max_low_pfn;
 277         /* how many bytes to cover all the pages */
 278         bootmap_size = bootmem_bootmap_pages(max_pfn) << PAGE_SHIFT;
 279
 280         /* look for a location to hold the bootmap */
 281         bootmap_start = ~0UL;
 282         efi_memmap_walk(find_bootmap_location, &bootmap_size);
 283         if (bootmap_start == ~0UL)
 284                 panic("Cannot find %ld bytes for bootmap\n", bootmap_size);
 285
 286         bootmap_size = init_bootmem_node(NODE_DATA(0),
 287                         (bootmap_start >> PAGE_SHIFT), 0, max_pfn);
 288
 289         /* Free all available memory, then mark bootmem-map as being in use. */
 290         efi_memmap_walk(filter_rsvd_memory, free_bootmem);
 291         reserve_bootmem(bootmap_start, bootmap_size, BOOTMEM_DEFAULT);
 292
 293         find_initrd();
 294
 295         alloc_per_cpu_data();
 296 }
 297
 298 static int count_pages(u64 start, u64 end, void *arg)
 299 {
 300         unsigned long *count = arg;
 301
 302         *count += (end - start) >> PAGE_SHIFT;
 303         return 0;
 304 }
 305
 306 /*
 307  * Set up the page tables.
 308  */
 309
 310 void __init
 311 paging_init (void)
 312 {
 313         unsigned long max_dma;
 314         unsigned long max_zone_pfns[MAX_NR_ZONES];
 315
 316         num_physpages = 0;
 317         efi_memmap_walk(count_pages, &num_physpages);
 318
 319         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 320 #ifdef CONFIG_ZONE_DMA
 321         max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 322         max_zone_pfns[ZONE_DMA] = max_dma;
 323 #endif
 324         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 325
 326 #ifdef CONFIG_VIRTUAL_MEM_MAP
 327         efi_memmap_walk(filter_memory, register_active_ranges);
 328         efi_memmap_walk(find_largest_hole, (u64 *)&max_gap);
 329         if (max_gap < LARGE_GAP) {
 330                 vmem_map = (struct page *) 0;
 331                 free_area_init_nodes(max_zone_pfns);
 332         } else {
 333                 unsigned long map_size;
 334
 335                 /* allocate virtual_mem_map */
 336
 337                 map_size = PAGE_ALIGN(ALIGN(max_low_pfn, MAX_ORDER_NR_PAGES) *
 338                         sizeof(struct page));
 339                 VMALLOC_END -= map_size;
 340                 vmem_map = (struct page *) VMALLOC_END;
 341                 efi_memmap_walk(create_mem_map_page_table, NULL);
 342
 343                 /*
 344                  * alloc_node_mem_map makes an adjustment for mem_map
 345                  * which isn't compatible with vmem_map.
 346                  */
 347                 NODE_DATA(0)->node_mem_map = vmem_map +
 348                         find_min_pfn_with_active_regions();
 349                 free_area_init_nodes(max_zone_pfns);
 350
 351                 printk("Virtual mem_map starts at 0x%p\n", mem_map);
 352         }
 353 #else /* !CONFIG_VIRTUAL_MEM_MAP */
 354         memblock_add_node(0, PFN_PHYS(max_low_pfn), 0);
 355         free_area_init_nodes(max_zone_pfns);
 356 #endif /* !CONFIG_VIRTUAL_MEM_MAP */
 357         zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
 358 }