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