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
  21 static struct acpi_table_slit *acpi_slit;
  22
  23 static nodemask_t nodes_parsed __initdata;
  24 static nodemask_t nodes_found __initdata;
  25 static struct node nodes[MAX_NUMNODES] __initdata;
  26 static __u8  pxm2node[256] = { [0 ... 255] = 0xff };
  27
  28 static __init int setup_node(int pxm)
  29 {
  30         unsigned node = pxm2node[pxm];
  31         if (node == 0xff) {
  32                 if (nodes_weight(nodes_found) >= MAX_NUMNODES)
  33                         return -1;
  34                 node = first_unset_node(nodes_found);
  35                 node_set(node, nodes_found);
  36                 pxm2node[pxm] = node;
  37         }
  38         return pxm2node[pxm];
  39 }
  40
  41 static __init int conflicting_nodes(unsigned long start, unsigned long end)
  42 {
  43         int i;
  44         for_each_online_node(i) {
  45                 struct node *nd = &nodes[i];
  46                 if (nd->start == nd->end)
  47                         continue;
  48                 if (nd->end > start && nd->start < end)
  49                         return 1;
  50                 if (nd->end == end && nd->start == start)
  51                         return 1;
  52         }
  53         return -1;
  54 }
  55
  56 static __init void cutoff_node(int i, unsigned long start, unsigned long end)
  57 {
  58         struct node *nd = &nodes[i];
  59         if (nd->start < start) {
  60                 nd->start = start;
  61                 if (nd->end < nd->start)
  62                         nd->start = nd->end;
  63         }
  64         if (nd->end > end) {
  65                 if (!(end & 0xfff))
  66                         end--;
  67                 nd->end = end;
  68                 if (nd->start > nd->end)
  69                         nd->start = nd->end;
  70         }
  71 }
  72
  73 static __init void bad_srat(void)
  74 {
  75         printk(KERN_ERR "SRAT: SRAT not used.\n");
  76         acpi_numa = -1;
  77 }
  78
  79 static __init inline int srat_disabled(void)
  80 {
  81         return numa_off || acpi_numa < 0;
  82 }
  83
  84 /* Callback for SLIT parsing */
  85 void __init acpi_numa_slit_init(struct acpi_table_slit *slit)
  86 {
  87         acpi_slit = slit;
  88 }
  89
  90 /* Callback for Proximity Domain -> LAPIC mapping */
  91 void __init
  92 acpi_numa_processor_affinity_init(struct acpi_table_processor_affinity *pa)
  93 {
  94         int pxm, node;
  95         if (srat_disabled() || pa->flags.enabled == 0)
  96                 return;
  97         pxm = pa->proximity_domain;
  98         node = setup_node(pxm);
  99         if (node < 0) {
 100                 printk(KERN_ERR "SRAT: Too many proximity domains %x\n", pxm);
 101                 bad_srat();
 102                 return;
 103         }
 104         if (pa->apic_id >= NR_CPUS) {
 105                 printk(KERN_ERR "SRAT: lapic %u too large.\n",
 106                        pa->apic_id);
 107                 bad_srat();
 108                 return;
 109         }
 110         cpu_to_node[pa->apic_id] = node;
 111         acpi_numa = 1;
 112         printk(KERN_INFO "SRAT: PXM %u -> APIC %u -> Node %u\n",
 113                pxm, pa->apic_id, node);
 114 }
 115
 116 /* Callback for parsing of the Proximity Domain <-> Memory Area mappings */
 117 void __init
 118 acpi_numa_memory_affinity_init(struct acpi_table_memory_affinity *ma)
 119 {
 120         struct node *nd;
 121         unsigned long start, end;
 122         int node, pxm;
 123         int i;
 124
 125         if (srat_disabled() || ma->flags.enabled == 0)
 126                 return;
 127         pxm = ma->proximity_domain;
 128         node = setup_node(pxm);
 129         if (node < 0) {
 130                 printk(KERN_ERR "SRAT: Too many proximity domains.\n");
 131                 bad_srat();
 132                 return;
 133         }
 134         start = ma->base_addr_lo | ((u64)ma->base_addr_hi << 32);
 135         end = start + (ma->length_lo | ((u64)ma->length_hi << 32));
 136         /* It is fine to add this area to the nodes data it will be used later*/
 137         if (ma->flags.hot_pluggable == 1)
 138                 printk(KERN_INFO "SRAT: hot plug zone found %lx - %lx \n",
 139                                 start, end);
 140         i = conflicting_nodes(start, end);
 141         if (i >= 0) {
 142                 printk(KERN_ERR
 143                        "SRAT: pxm %d overlap %lx-%lx with node %d(%Lx-%Lx)\n",
 144                        pxm, start, end, i, nodes[i].start, nodes[i].end);
 145                 bad_srat();
 146                 return;
 147         }
 148         nd = &nodes[node];
 149         if (!node_test_and_set(node, nodes_parsed)) {
 150                 nd->start = start;
 151                 nd->end = end;
 152         } else {
 153                 if (start < nd->start)
 154                         nd->start = start;
 155                 if (nd->end < end)
 156                         nd->end = end;
 157         }
 158         if (!(nd->end & 0xfff))
 159                 nd->end--;
 160         printk(KERN_INFO "SRAT: Node %u PXM %u %Lx-%Lx\n", node, pxm,
 161                nd->start, nd->end);
 162 }
 163
 164 void __init acpi_numa_arch_fixup(void) {}
 165
 166 /* Use the information discovered above to actually set up the nodes. */
 167 int __init acpi_scan_nodes(unsigned long start, unsigned long end)
 168 {
 169         int i;
 170         if (acpi_numa <= 0)
 171                 return -1;
 172         memnode_shift = compute_hash_shift(nodes, nodes_weight(nodes_parsed));
 173         if (memnode_shift < 0) {
 174                 printk(KERN_ERR
 175                      "SRAT: No NUMA node hash function found. Contact maintainer\n");
 176                 bad_srat();
 177                 return -1;
 178         }
 179         for (i = 0; i < MAX_NUMNODES; i++) {
 180                 if (!node_isset(i, nodes_parsed))
 181                         continue;
 182                 cutoff_node(i, start, end);
 183                 if (nodes[i].start == nodes[i].end) {
 184                         node_clear(i, nodes_parsed);
 185                         continue;
 186                 }
 187                 setup_node_bootmem(i, nodes[i].start, nodes[i].end);
 188         }
 189         for (i = 0; i < NR_CPUS; i++) {
 190                 if (cpu_to_node[i] == NUMA_NO_NODE)
 191                         continue;
 192                 if (!node_isset(cpu_to_node[i], nodes_parsed))
 193                         cpu_to_node[i] = NUMA_NO_NODE;
 194         }
 195         numa_init_array();
 196         return 0;
 197 }
 198
 199 int node_to_pxm(int n)
 200 {
 201        int i;
 202        if (pxm2node[n] == n)
 203                return n;
 204        for (i = 0; i < 256; i++)
 205                if (pxm2node[i] == n)
 206                        return i;
 207        return 0;
 208 }
 209
 210 int __node_distance(int a, int b)
 211 {
 212         int index;
 213
 214         if (!acpi_slit)
 215                 return a == b ? 10 : 20;
 216         index = acpi_slit->localities * node_to_pxm(a);
 217         return acpi_slit->entry[index + node_to_pxm(b)];
 218 }
 219
 220 EXPORT_SYMBOL(__node_distance);