arch/x86_64/mm/srat.c

   1 /*
   2  * ACPI 3.0 based NUMA setup
   3  * Copyright 2004 Andi Kleen, SuSE Labs.
   4  *
   5  * Reads the ACPI SRAT table to figure out what memory belongs to which CPUs.
   6  *
   7  * Called from acpi_numa_init while reading the SRAT and SLIT tables.
   8  * Assumes all memory regions belonging to a single proximity domain
   9  * are in one chunk. Holes between them will be included in the node.
  10  */
  11
  12 #include <linux/kernel.h>
  13 #include <linux/acpi.h>
  14 #include <linux/mmzone.h>
  15 #include <linux/bitmap.h>
  16 #include <linux/module.h>
  17 #include <linux/topology.h>
  18 #include <asm/proto.h>
  19 #include <asm/numa.h>
  20 #include <asm/e820.h>
  21
  22 static struct acpi_table_slit *acpi_slit;
  23
  24 static nodemask_t nodes_parsed __initdata;
  25 static nodemask_t nodes_found __initdata;
  26 static struct node nodes[MAX_NUMNODES] __initdata;
  27 static u8 pxm2node[256] = { [0 ... 255] = 0xff };
  28
  29 static int node_to_pxm(int n);
  30
  31 int pxm_to_node(int pxm)
  32 {
  33         if ((unsigned)pxm >= 256)
  34                 return -1;
  35         /* Extend 0xff to (int)-1 */
  36         return (signed char)pxm2node[pxm];
  37 }
  38
  39 static __init int setup_node(int pxm)
  40 {
  41         unsigned node = pxm2node[pxm];
  42         if (node == 0xff) {
  43                 if (nodes_weight(nodes_found) >= MAX_NUMNODES)
  44                         return -1;
  45                 node = first_unset_node(nodes_found);
  46                 node_set(node, nodes_found);
  47                 pxm2node[pxm] = node;
  48         }
  49         return pxm2node[pxm];
  50 }
  51
  52 static __init int conflicting_nodes(unsigned long start, unsigned long end)
  53 {
  54         int i;
  55         for_each_node_mask(i, nodes_parsed) {
  56                 struct node *nd = &nodes[i];
  57                 if (nd->start == nd->end)
  58                         continue;
  59                 if (nd->end > start && nd->start < end)
  60                         return i;
  61                 if (nd->end == end && nd->start == start)
  62                         return i;
  63         }
  64         return -1;
  65 }
  66
  67 static __init void cutoff_node(int i, unsigned long start, unsigned long end)
  68 {
  69         struct node *nd = &nodes[i];
  70         if (nd->start < start) {
  71                 nd->start = start;
  72                 if (nd->end < nd->start)
  73                         nd->start = nd->end;
  74         }
  75         if (nd->end > end) {
  76                 nd->end = end;
  77                 if (nd->start > nd->end)
  78                         nd->start = nd->end;
  79         }
  80 }
  81
  82 static __init void bad_srat(void)
  83 {
  84         int i;
  85         printk(KERN_ERR "SRAT: SRAT not used.\n");
  86         acpi_numa = -1;
  87         for (i = 0; i < MAX_LOCAL_APIC; i++)
  88                 apicid_to_node[i] = NUMA_NO_NODE;
  89 }
  90
  91 static __init inline int srat_disabled(void)
  92 {
  93         return numa_off || acpi_numa < 0;
  94 }
  95
  96 /*
  97  * A lot of BIOS fill in 10 (= no distance) everywhere. This messes
  98  * up the NUMA heuristics which wants the local node to have a smaller
  99  * distance than the others.
 100  * Do some quick checks here and only use the SLIT if it passes.
 101  */
 102 static __init int slit_valid(struct acpi_table_slit *slit)
 103 {
 104         int i, j;
 105         int d = slit->localities;
 106         for (i = 0; i < d; i++) {
 107                 for (j = 0; j < d; j++)  {
 108                         u8 val = slit->entry[d*i + j];
 109                         if (i == j) {
 110                                 if (val != 10)
 111                                         return 0;
 112                         } else if (val <= 10)
 113                                 return 0;
 114                 }
 115         }
 116         return 1;
 117 }
 118
 119 /* Callback for SLIT parsing */
 120 void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
 121 {
 122         if (!slit_valid(slit)) {
 123                 printk(KERN_INFO "ACPI: SLIT table looks invalid. Not used.\n");
 124                 return;
 125         }
 126         acpi_slit = slit;
 127 }
 128
 129 /* Callback for Proximity Domain -> LAPIC mapping */
 130 void __init
 131 acpi_numa_processor_affinity_init(struct acpi_table_processor_affinity *pa)
 132 {
 133         int pxm, node;
 134         if (srat_disabled() || pa->flags.enabled == 0)
 135                 return;
 136         pxm = pa->proximity_domain;
 137         node = setup_node(pxm);
 138         if (node < 0) {
 139                 printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm);
 140                 bad_srat();
 141                 return;
 142         }
 143         apicid_to_node[pa->apic_id] = node;
 144         acpi_numa = 1;
 145         printk(KERN_INFO "SRAT: PXM %u -> APIC %u -> Node %u\n",
 146                pxm, pa->apic_id, node);
 147 }
 148
 149 /* Callback for parsing of the Proximity Domain <-> Memory Area mappings */
 150 void __init
 151 acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma)
 152 {
 153         struct node *nd;
 154         unsigned long start, end;
 155         int node, pxm;
 156         int i;
 157
 158         if (srat_disabled() || ma->flags.enabled == 0)
 159                 return;
 160         pxm = ma->proximity_domain;
 161         node = setup_node(pxm);
 162         if (node < 0) {
 163                 printk(KERN_ERR "SRAT: Too many proximity domains.\n");
 164                 bad_srat();
 165                 return;
 166         }
 167         start = ma->base_addr_lo | ((u64)ma->base_addr_hi << 32);
 168         end = start + (ma->length_lo | ((u64)ma->length_hi << 32));
 169         /* It is fine to add this area to the nodes data it will be used later*/
 170         if (ma->flags.hot_pluggable == 1)
 171                 printk(KERN_INFO "SRAT: hot plug zone found %lx - %lx \n",
 172                                 start, end);
 173         i = conflicting_nodes(start, end);
 174         if (i == node) {
 175                 printk(KERN_WARNING
 176                 "SRAT: Warning: PXM %d (%lx-%lx) overlaps with itself (%Lx-%Lx)\n",
 177                         pxm, start, end, nodes[i].start, nodes[i].end);
 178         } else if (i >= 0) {
 179                 printk(KERN_ERR
 180                        "SRAT: PXM %d (%lx-%lx) overlaps with PXM %d (%Lx-%Lx)\n",
 181                        pxm, start, end, node_to_pxm(i),
 182                         nodes[i].start, nodes[i].end);
 183                 bad_srat();
 184                 return;
 185         }
 186         nd = &nodes[node];
 187         if (!node_test_and_set(node, nodes_parsed)) {
 188                 nd->start = start;
 189                 nd->end = end;
 190         } else {
 191                 if (start < nd->start)
 192                         nd->start = start;
 193                 if (nd->end < end)
 194                         nd->end = end;
 195         }
 196         printk(KERN_INFO "SRAT: Node %u PXM %u %Lx-%Lx\n", node, pxm,
 197                nd->start, nd->end);
 198 }
 199
 200 /* Sanity check to catch more bad SRATs (they are amazingly common).
 201    Make sure the PXMs cover all memory. */
 202 static int nodes_cover_memory(void)
 203 {
 204         int i;
 205         unsigned long pxmram, e820ram;
 206
 207         pxmram = 0;
 208         for_each_node_mask(i, nodes_parsed) {
 209                 unsigned long s = nodes[i].start >> PAGE_SHIFT;
 210                 unsigned long e = nodes[i].end >> PAGE_SHIFT;
 211                 pxmram += e - s;
 212                 pxmram -= e820_hole_size(s, e);
 213         }
 214
 215         e820ram = end_pfn - e820_hole_size(0, end_pfn);
 216         if (pxmram < e820ram) {
 217                 printk(KERN_ERR
 218         "SRAT: PXMs only cover %luMB of your %luMB e820 RAM. Not used.\n",
 219                         (pxmram << PAGE_SHIFT) >> 20,
 220                         (e820ram << PAGE_SHIFT) >> 20);
 221                 return 0;
 222         }
 223         return 1;
 224 }
 225
 226 void __init acpi_numa_arch_fixup(void) {}
 227
 228 /* Use the information discovered above to actually set up the nodes. */
 229 int __init acpi_scan_nodes(unsigned long start, unsigned long end)
 230 {
 231         int i;
 232
 233         if (acpi_numa <= 0)
 234                 return -1;
 235
 236         /* First clean up the node list */
 237         for_each_node_mask(i, nodes_parsed) {
 238                 cutoff_node(i, start, end);
 239                 if (nodes[i].start == nodes[i].end)
 240                         node_clear(i, nodes_parsed);
 241         }
 242
 243         if (!nodes_cover_memory()) {
 244                 bad_srat();
 245                 return -1;
 246         }
 247
 248         memnode_shift = compute_hash_shift(nodes, nodes_weight(nodes_parsed));
 249         if (memnode_shift < 0) {
 250                 printk(KERN_ERR
 251                      "SRAT: No NUMA node hash function found. Contact maintainer\n");
 252                 bad_srat();
 253                 return -1;
 254         }
 255
 256         /* Finally register nodes */
 257         for_each_node_mask(i, nodes_parsed)
 258                 setup_node_bootmem(i, nodes[i].start, nodes[i].end);
 259         for (i = 0; i < NR_CPUS; i++) {
 260                 if (cpu_to_node[i] == NUMA_NO_NODE)
 261                         continue;
 262                 if (!node_isset(cpu_to_node[i], nodes_parsed))
 263                         numa_set_node(i, NUMA_NO_NODE);
 264         }
 265         numa_init_array();
 266         return 0;
 267 }
 268
 269 static int node_to_pxm(int n)
 270 {
 271        int i;
 272        if (pxm2node[n] == n)
 273                return n;
 274        for (i = 0; i < 256; i++)
 275                if (pxm2node[i] == n)
 276                        return i;
 277        return 0;
 278 }
 279
 280 int __node_distance(int a, int b)
 281 {
 282         int index;
 283
 284         if (!acpi_slit)
 285                 return a == b ? 10 : 20;
 286         index = acpi_slit->localities * node_to_pxm(a);
 287         return acpi_slit->entry[index + node_to_pxm(b)];
 288 }
 289
 290 EXPORT_SYMBOL(__node_distance);