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