arch/mips/loongson64/loongson-3/numa.c

   1 /*
   2  * Copyright (C) 2010 Loongson Inc. & Lemote Inc. &
   3  *                    Institute of Computing Technology
   4  * Author:  Xiang Gao, gaoxiang@ict.ac.cn
   5  *          Huacai Chen, chenhc@lemote.com
   6  *          Xiaofu Meng, Shuangshuang Zhang
   7  *
   8  * This program is free software; you can redistribute  it and/or modify it
   9  * under  the terms of  the GNU General  Public License as published by the
  10  * Free Software Foundation;  either version 2 of the  License, or (at your
  11  * option) any later version.
  12  */
  13 #include <linux/init.h>
  14 #include <linux/kernel.h>
  15 #include <linux/mm.h>
  16 #include <linux/mmzone.h>
  17 #include <linux/module.h>
  18 #include <linux/nodemask.h>
  19 #include <linux/swap.h>
  20 #include <linux/memblock.h>
  21 #include <linux/bootmem.h>
  22 #include <linux/pfn.h>
  23 #include <linux/highmem.h>
  24 #include <asm/page.h>
  25 #include <asm/pgalloc.h>
  26 #include <asm/sections.h>
  27 #include <linux/irq.h>
  28 #include <asm/bootinfo.h>
  29 #include <asm/mc146818-time.h>
  30 #include <asm/time.h>
  31 #include <asm/wbflush.h>
  32 #include <boot_param.h>
  33
  34 static struct node_data prealloc__node_data[MAX_NUMNODES];
  35 unsigned char __node_distances[MAX_NUMNODES][MAX_NUMNODES];
  36 EXPORT_SYMBOL(__node_distances);
  37 struct node_data *__node_data[MAX_NUMNODES];
  38 EXPORT_SYMBOL(__node_data);
  39
  40 static void enable_lpa(void)
  41 {
  42         unsigned long value;
  43
  44         value = __read_32bit_c0_register($16, 3);
  45         value |= 0x00000080;
  46         __write_32bit_c0_register($16, 3, value);
  47         value = __read_32bit_c0_register($16, 3);
  48         pr_info("CP0_Config3: CP0 16.3 (0x%lx)\n", value);
  49
  50         value = __read_32bit_c0_register($5, 1);
  51         value |= 0x20000000;
  52         __write_32bit_c0_register($5, 1, value);
  53         value = __read_32bit_c0_register($5, 1);
  54         pr_info("CP0_PageGrain: CP0 5.1 (0x%lx)\n", value);
  55 }
  56
  57 static void cpu_node_probe(void)
  58 {
  59         int i;
  60
  61         nodes_clear(node_possible_map);
  62         nodes_clear(node_online_map);
  63         for (i = 0; i < loongson_sysconf.nr_nodes; i++) {
  64                 node_set_state(num_online_nodes(), N_POSSIBLE);
  65                 node_set_online(num_online_nodes());
  66         }
  67
  68         pr_info("NUMA: Discovered %d cpus on %d nodes\n",
  69                 loongson_sysconf.nr_cpus, num_online_nodes());
  70 }
  71
  72 static int __init compute_node_distance(int row, int col)
  73 {
  74         int package_row = row * loongson_sysconf.cores_per_node /
  75                                 loongson_sysconf.cores_per_package;
  76         int package_col = col * loongson_sysconf.cores_per_node /
  77                                 loongson_sysconf.cores_per_package;
  78
  79         if (col == row)
  80                 return 0;
  81         else if (package_row == package_col)
  82                 return 40;
  83         else
  84                 return 100;
  85 }
  86
  87 static void __init init_topology_matrix(void)
  88 {
  89         int row, col;
  90
  91         for (row = 0; row < MAX_NUMNODES; row++)
  92                 for (col = 0; col < MAX_NUMNODES; col++)
  93                         __node_distances[row][col] = -1;
  94
  95         for_each_online_node(row) {
  96                 for_each_online_node(col) {
  97                         __node_distances[row][col] =
  98                                 compute_node_distance(row, col);
  99                 }
 100         }
 101 }
 102
 103 static unsigned long nid_to_addroffset(unsigned int nid)
 104 {
 105         unsigned long result;
 106         switch (nid) {
 107         case 0:
 108         default:
 109                 result = NODE0_ADDRSPACE_OFFSET;
 110                 break;
 111         case 1:
 112                 result = NODE1_ADDRSPACE_OFFSET;
 113                 break;
 114         case 2:
 115                 result = NODE2_ADDRSPACE_OFFSET;
 116                 break;
 117         case 3:
 118                 result = NODE3_ADDRSPACE_OFFSET;
 119                 break;
 120         }
 121         return result;
 122 }
 123
 124 static void __init szmem(unsigned int node)
 125 {
 126         u32 i, mem_type;
 127         static unsigned long num_physpages = 0;
 128         u64 node_id, node_psize, start_pfn, end_pfn, mem_start, mem_size;
 129
 130         /* Parse memory information and activate */
 131         for (i = 0; i < loongson_memmap->nr_map; i++) {
 132                 node_id = loongson_memmap->map[i].node_id;
 133                 if (node_id != node)
 134                         continue;
 135
 136                 mem_type = loongson_memmap->map[i].mem_type;
 137                 mem_size = loongson_memmap->map[i].mem_size;
 138                 mem_start = loongson_memmap->map[i].mem_start;
 139
 140                 switch (mem_type) {
 141                 case SYSTEM_RAM_LOW:
 142                         start_pfn = ((node_id << 44) + mem_start) >> PAGE_SHIFT;
 143                         node_psize = (mem_size << 20) >> PAGE_SHIFT;
 144                         end_pfn  = start_pfn + node_psize;
 145                         num_physpages += node_psize;
 146                         pr_info("Node%d: mem_type:%d, mem_start:0x%llx, mem_size:0x%llx MB\n",
 147                                 (u32)node_id, mem_type, mem_start, mem_size);
 148                         pr_info("       start_pfn:0x%llx, end_pfn:0x%llx, num_physpages:0x%lx\n",
 149                                 start_pfn, end_pfn, num_physpages);
 150                         add_memory_region((node_id << 44) + mem_start,
 151                                 (u64)mem_size << 20, BOOT_MEM_RAM);
 152                         memblock_add_node(PFN_PHYS(start_pfn),
 153                                 PFN_PHYS(end_pfn - start_pfn), node);
 154                         break;
 155                 case SYSTEM_RAM_HIGH:
 156                         start_pfn = ((node_id << 44) + mem_start) >> PAGE_SHIFT;
 157                         node_psize = (mem_size << 20) >> PAGE_SHIFT;
 158                         end_pfn  = start_pfn + node_psize;
 159                         num_physpages += node_psize;
 160                         pr_info("Node%d: mem_type:%d, mem_start:0x%llx, mem_size:0x%llx MB\n",
 161                                 (u32)node_id, mem_type, mem_start, mem_size);
 162                         pr_info("       start_pfn:0x%llx, end_pfn:0x%llx, num_physpages:0x%lx\n",
 163                                 start_pfn, end_pfn, num_physpages);
 164                         add_memory_region((node_id << 44) + mem_start,
 165                                 (u64)mem_size << 20, BOOT_MEM_RAM);
 166                         memblock_add_node(PFN_PHYS(start_pfn),
 167                                 PFN_PHYS(end_pfn - start_pfn), node);
 168                         break;
 169                 case MEM_RESERVED:
 170                         pr_info("Node%d: mem_type:%d, mem_start:0x%llx, mem_size:0x%llx MB\n",
 171                                 (u32)node_id, mem_type, mem_start, mem_size);
 172                         add_memory_region((node_id << 44) + mem_start,
 173                                 (u64)mem_size << 20, BOOT_MEM_RESERVED);
 174                         memblock_reserve(((node_id << 44) + mem_start),
 175                                 mem_size << 20);
 176                         break;
 177                 }
 178         }
 179 }
 180
 181 static void __init node_mem_init(unsigned int node)
 182 {
 183         unsigned long bootmap_size;
 184         unsigned long node_addrspace_offset;
 185         unsigned long start_pfn, end_pfn, freepfn;
 186
 187         node_addrspace_offset = nid_to_addroffset(node);
 188         pr_info("Node%d's addrspace_offset is 0x%lx\n",
 189                         node, node_addrspace_offset);
 190
 191         get_pfn_range_for_nid(node, &start_pfn, &end_pfn);
 192         freepfn = start_pfn;
 193         if (node == 0)
 194                 freepfn = PFN_UP(__pa_symbol(&_end)); /* kernel end address */
 195         pr_info("Node%d: start_pfn=0x%lx, end_pfn=0x%lx, freepfn=0x%lx\n",
 196                 node, start_pfn, end_pfn, freepfn);
 197
 198         __node_data[node] = prealloc__node_data + node;
 199
 200         NODE_DATA(node)->bdata = &bootmem_node_data[node];
 201         NODE_DATA(node)->node_start_pfn = start_pfn;
 202         NODE_DATA(node)->node_spanned_pages = end_pfn - start_pfn;
 203
 204         bootmap_size = init_bootmem_node(NODE_DATA(node), freepfn,
 205                                         start_pfn, end_pfn);
 206         free_bootmem_with_active_regions(node, end_pfn);
 207         if (node == 0) /* used by finalize_initrd() */
 208                 max_low_pfn = end_pfn;
 209
 210         /* This is reserved for the kernel and bdata->node_bootmem_map */
 211         reserve_bootmem_node(NODE_DATA(node), start_pfn << PAGE_SHIFT,
 212                 ((freepfn - start_pfn) << PAGE_SHIFT) + bootmap_size,
 213                 BOOTMEM_DEFAULT);
 214
 215         if (node == 0 && node_end_pfn(0) >= (0xffffffff >> PAGE_SHIFT)) {
 216                 /* Reserve 0xfe000000~0xffffffff for RS780E integrated GPU */
 217                 reserve_bootmem_node(NODE_DATA(node),
 218                                 (node_addrspace_offset | 0xfe000000),
 219                                 32 << 20, BOOTMEM_DEFAULT);
 220         }
 221
 222         sparse_memory_present_with_active_regions(node);
 223 }
 224
 225 static __init void prom_meminit(void)
 226 {
 227         unsigned int node, cpu, active_cpu = 0;
 228
 229         cpu_node_probe();
 230         init_topology_matrix();
 231
 232         for (node = 0; node < loongson_sysconf.nr_nodes; node++) {
 233                 if (node_online(node)) {
 234                         szmem(node);
 235                         node_mem_init(node);
 236                         cpumask_clear(&__node_data[(node)]->cpumask);
 237                 }
 238         }
 239         for (cpu = 0; cpu < loongson_sysconf.nr_cpus; cpu++) {
 240                 node = cpu / loongson_sysconf.cores_per_node;
 241                 if (node >= num_online_nodes())
 242                         node = 0;
 243
 244                 if (loongson_sysconf.reserved_cpus_mask & (1<<cpu))
 245                         continue;
 246
 247                 cpumask_set_cpu(active_cpu, &__node_data[(node)]->cpumask);
 248                 pr_info("NUMA: set cpumask cpu %d on node %d\n", active_cpu, node);
 249
 250                 active_cpu++;
 251         }
 252 }
 253
 254 void __init paging_init(void)
 255 {
 256         unsigned node;
 257         unsigned long zones_size[MAX_NR_ZONES] = {0, };
 258
 259         pagetable_init();
 260
 261         for_each_online_node(node) {
 262                 unsigned long  start_pfn, end_pfn;
 263
 264                 get_pfn_range_for_nid(node, &start_pfn, &end_pfn);
 265
 266                 if (end_pfn > max_low_pfn)
 267                         max_low_pfn = end_pfn;
 268         }
 269 #ifdef CONFIG_ZONE_DMA32
 270         zones_size[ZONE_DMA32] = MAX_DMA32_PFN;
 271 #endif
 272         zones_size[ZONE_NORMAL] = max_low_pfn;
 273         free_area_init_nodes(zones_size);
 274 }
 275
 276 void __init mem_init(void)
 277 {
 278         high_memory = (void *) __va(get_num_physpages() << PAGE_SHIFT);
 279         free_all_bootmem();
 280         setup_zero_pages();     /* This comes from node 0 */
 281         mem_init_print_info(NULL);
 282 }
 283
 284 /* All PCI device belongs to logical Node-0 */
 285 int pcibus_to_node(struct pci_bus *bus)
 286 {
 287         return 0;
 288 }
 289 EXPORT_SYMBOL(pcibus_to_node);
 290
 291 void __init prom_init_numa_memory(void)
 292 {
 293         enable_lpa();
 294         prom_meminit();
 295 }
 296 EXPORT_SYMBOL(prom_init_numa_memory);