arch/x86/mm/amdtopology_64.c

   1 /*
   2  * AMD NUMA support.
   3  * Discover the memory map and associated nodes.
   4  *
   5  * This version reads it directly from the AMD northbridge.
   6  *
   7  * Copyright 2002,2003 Andi Kleen, SuSE Labs.
   8  */
   9 #include <linux/kernel.h>
  10 #include <linux/init.h>
  11 #include <linux/string.h>
  12 #include <linux/module.h>
  13 #include <linux/nodemask.h>
  14 #include <linux/memblock.h>
  15
  16 #include <asm/io.h>
  17 #include <linux/pci_ids.h>
  18 #include <linux/acpi.h>
  19 #include <asm/types.h>
  20 #include <asm/mmzone.h>
  21 #include <asm/proto.h>
  22 #include <asm/e820.h>
  23 #include <asm/pci-direct.h>
  24 #include <asm/numa.h>
  25 #include <asm/mpspec.h>
  26 #include <asm/apic.h>
  27 #include <asm/amd_nb.h>
  28
  29 static unsigned char __initdata nodeids[8];
  30
  31 static __init int find_northbridge(void)
  32 {
  33         int num;
  34
  35         for (num = 0; num < 32; num++) {
  36                 u32 header;
  37
  38                 header = read_pci_config(0, num, 0, 0x00);
  39                 if (header != (PCI_VENDOR_ID_AMD | (0x1100<<16)) &&
  40                         header != (PCI_VENDOR_ID_AMD | (0x1200<<16)) &&
  41                         header != (PCI_VENDOR_ID_AMD | (0x1300<<16)))
  42                         continue;
  43
  44                 header = read_pci_config(0, num, 1, 0x00);
  45                 if (header != (PCI_VENDOR_ID_AMD | (0x1101<<16)) &&
  46                         header != (PCI_VENDOR_ID_AMD | (0x1201<<16)) &&
  47                         header != (PCI_VENDOR_ID_AMD | (0x1301<<16)))
  48                         continue;
  49                 return num;
  50         }
  51
  52         return -ENOENT;
  53 }
  54
  55 static __init void early_get_boot_cpu_id(void)
  56 {
  57         /*
  58          * need to get the APIC ID of the BSP so can use that to
  59          * create apicid_to_node in amd_scan_nodes()
  60          */
  61 #ifdef CONFIG_X86_MPPARSE
  62         /*
  63          * get boot-time SMP configuration:
  64          */
  65         if (smp_found_config)
  66                 early_get_smp_config();
  67 #endif
  68 }
  69
  70 int __init amd_numa_init(void)
  71 {
  72         unsigned long start = PFN_PHYS(0);
  73         unsigned long end = PFN_PHYS(max_pfn);
  74         unsigned numnodes;
  75         unsigned long prevbase;
  76         int i, j, nb;
  77         u32 nodeid, reg;
  78         unsigned int bits, cores, apicid_base;
  79
  80         if (!early_pci_allowed())
  81                 return -EINVAL;
  82
  83         nb = find_northbridge();
  84         if (nb < 0)
  85                 return nb;
  86
  87         pr_info("Scanning NUMA topology in Northbridge %d\n", nb);
  88
  89         reg = read_pci_config(0, nb, 0, 0x60);
  90         numnodes = ((reg >> 4) & 0xF) + 1;
  91         if (numnodes <= 1)
  92                 return -ENOENT;
  93
  94         pr_info("Number of physical nodes %d\n", numnodes);
  95
  96         prevbase = 0;
  97         for (i = 0; i < 8; i++) {
  98                 unsigned long base, limit;
  99
 100                 base = read_pci_config(0, nb, 1, 0x40 + i*8);
 101                 limit = read_pci_config(0, nb, 1, 0x44 + i*8);
 102
 103                 nodeids[i] = nodeid = limit & 7;
 104                 if ((base & 3) == 0) {
 105                         if (i < numnodes)
 106                                 pr_info("Skipping disabled node %d\n", i);
 107                         continue;
 108                 }
 109                 if (nodeid >= numnodes) {
 110                         pr_info("Ignoring excess node %d (%lx:%lx)\n", nodeid,
 111                                 base, limit);
 112                         continue;
 113                 }
 114
 115                 if (!limit) {
 116                         pr_info("Skipping node entry %d (base %lx)\n",
 117                                 i, base);
 118                         continue;
 119                 }
 120                 if ((base >> 8) & 3 || (limit >> 8) & 3) {
 121                         pr_err("Node %d using interleaving mode %lx/%lx\n",
 122                                nodeid, (base >> 8) & 3, (limit >> 8) & 3);
 123                         return -EINVAL;
 124                 }
 125                 if (node_isset(nodeid, mem_nodes_parsed)) {
 126                         pr_info("Node %d already present, skipping\n",
 127                                 nodeid);
 128                         continue;
 129                 }
 130
 131                 limit >>= 16;
 132                 limit <<= 24;
 133                 limit |= (1<<24)-1;
 134                 limit++;
 135
 136                 if (limit > end)
 137                         limit = end;
 138                 if (limit <= base)
 139                         continue;
 140
 141                 base >>= 16;
 142                 base <<= 24;
 143
 144                 if (base < start)
 145                         base = start;
 146                 if (limit > end)
 147                         limit = end;
 148                 if (limit == base) {
 149                         pr_err("Empty node %d\n", nodeid);
 150                         continue;
 151                 }
 152                 if (limit < base) {
 153                         pr_err("Node %d bogus settings %lx-%lx.\n",
 154                                nodeid, base, limit);
 155                         continue;
 156                 }
 157
 158                 /* Could sort here, but pun for now. Should not happen anyroads. */
 159                 if (prevbase > base) {
 160                         pr_err("Node map not sorted %lx,%lx\n",
 161                                prevbase, base);
 162                         return -EINVAL;
 163                 }
 164
 165                 pr_info("Node %d MemBase %016lx Limit %016lx\n",
 166                         nodeid, base, limit);
 167
 168                 numa_nodes[nodeid].start = base;
 169                 numa_nodes[nodeid].end = limit;
 170
 171                 prevbase = base;
 172
 173                 node_set(nodeid, mem_nodes_parsed);
 174                 node_set(nodeid, cpu_nodes_parsed);
 175         }
 176
 177         if (!nodes_weight(mem_nodes_parsed))
 178                 return -ENOENT;
 179
 180         /*
 181          * We seem to have valid NUMA configuration.  Map apicids to nodes
 182          * using the coreid bits from early_identify_cpu.
 183          */
 184         bits = boot_cpu_data.x86_coreid_bits;
 185         cores = 1 << bits;
 186         apicid_base = 0;
 187
 188         /* get the APIC ID of the BSP early for systems with apicid lifting */
 189         early_get_boot_cpu_id();
 190         if (boot_cpu_physical_apicid > 0) {
 191                 pr_info("BSP APIC ID: %02x\n", boot_cpu_physical_apicid);
 192                 apicid_base = boot_cpu_physical_apicid;
 193         }
 194
 195         for_each_node_mask(i, cpu_nodes_parsed)
 196                 for (j = apicid_base; j < cores + apicid_base; j++)
 197                         set_apicid_to_node((i << bits) + j, i);
 198
 199         return 0;
 200 }
 201
 202 #ifdef CONFIG_NUMA_EMU
 203 static s16 fake_apicid_to_node[MAX_LOCAL_APIC] __initdata = {
 204         [0 ... MAX_LOCAL_APIC-1] = NUMA_NO_NODE
 205 };
 206
 207 static int __init find_node_by_addr(unsigned long addr)
 208 {
 209         int ret = NUMA_NO_NODE;
 210         int i;
 211
 212         for (i = 0; i < 8; i++)
 213                 if (addr >= numa_nodes[i].start && addr < numa_nodes[i].end) {
 214                         ret = i;
 215                         break;
 216                 }
 217         return ret;
 218 }
 219
 220 /*
 221  * For NUMA emulation, fake proximity domain (_PXM) to node id mappings must be
 222  * setup to represent the physical topology but reflect the emulated
 223  * environment.  For each emulated node, the real node which it appears on is
 224  * found and a fake pxm to nid mapping is created which mirrors the actual
 225  * locality.  node_distance() then represents the correct distances between
 226  * emulated nodes by using the fake acpi mappings to pxms.
 227  */
 228 void __init amd_fake_nodes(const struct bootnode *nodes, int nr_nodes)
 229 {
 230         unsigned int bits;
 231         unsigned int cores;
 232         unsigned int apicid_base = 0;
 233         int i;
 234
 235         bits = boot_cpu_data.x86_coreid_bits;
 236         cores = 1 << bits;
 237         early_get_boot_cpu_id();
 238         if (boot_cpu_physical_apicid > 0)
 239                 apicid_base = boot_cpu_physical_apicid;
 240
 241         for (i = 0; i < nr_nodes; i++) {
 242                 int index;
 243                 int nid;
 244                 int j;
 245
 246                 nid = find_node_by_addr(nodes[i].start);
 247                 if (nid == NUMA_NO_NODE)
 248                         continue;
 249
 250                 index = nodeids[nid] << bits;
 251                 if (fake_apicid_to_node[index + apicid_base] == NUMA_NO_NODE)
 252                         for (j = apicid_base; j < cores + apicid_base; j++)
 253                                 fake_apicid_to_node[index + j] = i;
 254 #ifdef CONFIG_ACPI_NUMA
 255                 __acpi_map_pxm_to_node(nid, i);
 256 #endif
 257         }
 258         memcpy(__apicid_to_node, fake_apicid_to_node, sizeof(__apicid_to_node));
 259 }
 260 #endif /* CONFIG_NUMA_EMU */
 261
 262 int __init amd_scan_nodes(void)
 263 {
 264         int i;
 265
 266         memnode_shift = compute_hash_shift(numa_nodes, 8, NULL);
 267         if (memnode_shift < 0) {
 268                 pr_err("No NUMA node hash function found. Contact maintainer\n");
 269                 return -1;
 270         }
 271         pr_info("Using node hash shift of %d\n", memnode_shift);
 272
 273         /* use the coreid bits from early_identify_cpu */
 274         for_each_node_mask(i, node_possible_map)
 275                 memblock_x86_register_active_regions(i,
 276                                 numa_nodes[i].start >> PAGE_SHIFT,
 277                                 numa_nodes[i].end >> PAGE_SHIFT);
 278         init_memory_mapping_high();
 279         for_each_node_mask(i, node_possible_map)
 280                 setup_node_bootmem(i, numa_nodes[i].start, numa_nodes[i].end);
 281
 282         numa_init_array();
 283         return 0;
 284 }