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