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