arch/x86/kernel/setup.c

   1 #include <linux/kernel.h>
   2 #include <linux/module.h>
   3 #include <linux/init.h>
   4 #include <linux/bootmem.h>
   5 #include <linux/percpu.h>
   6 #include <asm/smp.h>
   7 #include <asm/percpu.h>
   8 #include <asm/sections.h>
   9 #include <asm/processor.h>
  10 #include <asm/setup.h>
  11 #include <asm/topology.h>
  12 #include <asm/mpspec.h>
  13 #include <asm/apicdef.h>
  14
  15 #ifdef CONFIG_X86_LOCAL_APIC
  16 unsigned int num_processors;
  17 unsigned disabled_cpus __cpuinitdata;
  18 /* Processor that is doing the boot up */
  19 unsigned int boot_cpu_physical_apicid = -1U;
  20 EXPORT_SYMBOL(boot_cpu_physical_apicid);
  21
  22 /* Bitmask of physically existing CPUs */
  23 physid_mask_t phys_cpu_present_map;
  24 #endif
  25
  26 /* map cpu index to physical APIC ID */
  27 DEFINE_EARLY_PER_CPU(u16, x86_cpu_to_apicid, BAD_APICID);
  28 DEFINE_EARLY_PER_CPU(u16, x86_bios_cpu_apicid, BAD_APICID);
  29 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
  30 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
  31
  32 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
  33 #define X86_64_NUMA     1
  34
  35 /* map cpu index to node index */
  36 DEFINE_EARLY_PER_CPU(int, x86_cpu_to_node_map, NUMA_NO_NODE);
  37 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_node_map);
  38
  39 /* which logical CPUs are on which nodes */
  40 cpumask_t *node_to_cpumask_map;
  41 EXPORT_SYMBOL(node_to_cpumask_map);
  42
  43 /* setup node_to_cpumask_map */
  44 static void __init setup_node_to_cpumask_map(void);
  45
  46 #else
  47 static inline void setup_node_to_cpumask_map(void) { }
  48 #endif
  49
  50 #if defined(CONFIG_HAVE_SETUP_PER_CPU_AREA) && defined(CONFIG_X86_SMP)
  51 /*
  52  * Copy data used in early init routines from the initial arrays to the
  53  * per cpu data areas.  These arrays then become expendable and the
  54  * *_early_ptr's are zeroed indicating that the static arrays are gone.
  55  */
  56 static void __init setup_per_cpu_maps(void)
  57 {
  58         int cpu;
  59
  60         for_each_possible_cpu(cpu) {
  61                 per_cpu(x86_cpu_to_apicid, cpu) =
  62                                 early_per_cpu_map(x86_cpu_to_apicid, cpu);
  63                 per_cpu(x86_bios_cpu_apicid, cpu) =
  64                                 early_per_cpu_map(x86_bios_cpu_apicid, cpu);
  65 #ifdef X86_64_NUMA
  66                 per_cpu(x86_cpu_to_node_map, cpu) =
  67                                 early_per_cpu_map(x86_cpu_to_node_map, cpu);
  68 #endif
  69         }
  70
  71         /* indicate the early static arrays will soon be gone */
  72         early_per_cpu_ptr(x86_cpu_to_apicid) = NULL;
  73         early_per_cpu_ptr(x86_bios_cpu_apicid) = NULL;
  74 #ifdef X86_64_NUMA
  75         early_per_cpu_ptr(x86_cpu_to_node_map) = NULL;
  76 #endif
  77 }
  78
  79 #ifdef CONFIG_HAVE_CPUMASK_OF_CPU_MAP
  80 cpumask_t *cpumask_of_cpu_map __read_mostly;
  81 EXPORT_SYMBOL(cpumask_of_cpu_map);
  82
  83 /* requires nr_cpu_ids to be initialized */
  84 static void __init setup_cpumask_of_cpu(void)
  85 {
  86         int i;
  87
  88         /* alloc_bootmem zeroes memory */
  89         cpumask_of_cpu_map = alloc_bootmem_low(sizeof(cpumask_t) * nr_cpu_ids);
  90         for (i = 0; i < nr_cpu_ids; i++)
  91                 cpu_set(i, cpumask_of_cpu_map[i]);
  92 }
  93 #else
  94 static inline void setup_cpumask_of_cpu(void) { }
  95 #endif
  96
  97 #ifdef CONFIG_X86_32
  98 /*
  99  * Great future not-so-futuristic plan: make i386 and x86_64 do it
 100  * the same way
 101  */
 102 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
 103 EXPORT_SYMBOL(__per_cpu_offset);
 104 static inline void setup_cpu_pda_map(void) { }
 105
 106 #elif !defined(CONFIG_SMP)
 107 static inline void setup_cpu_pda_map(void) { }
 108
 109 #else /* CONFIG_SMP && CONFIG_X86_64 */
 110
 111 /*
 112  * Allocate cpu_pda pointer table and array via alloc_bootmem.
 113  */
 114 static void __init setup_cpu_pda_map(void)
 115 {
 116         char *pda;
 117         struct x8664_pda **new_cpu_pda;
 118         unsigned long size;
 119         int cpu;
 120
 121         size = roundup(sizeof(struct x8664_pda), cache_line_size());
 122
 123         /* allocate cpu_pda array and pointer table */
 124         {
 125                 unsigned long tsize = nr_cpu_ids * sizeof(void *);
 126                 unsigned long asize = size * (nr_cpu_ids - 1);
 127
 128                 tsize = roundup(tsize, cache_line_size());
 129                 new_cpu_pda = alloc_bootmem(tsize + asize);
 130                 pda = (char *)new_cpu_pda + tsize;
 131         }
 132
 133         /* initialize pointer table to static pda's */
 134         for_each_possible_cpu(cpu) {
 135                 if (cpu == 0) {
 136                         /* leave boot cpu pda in place */
 137                         new_cpu_pda[0] = cpu_pda(0);
 138                         continue;
 139                 }
 140                 new_cpu_pda[cpu] = (struct x8664_pda *)pda;
 141                 new_cpu_pda[cpu]->in_bootmem = 1;
 142                 pda += size;
 143         }
 144
 145         /* point to new pointer table */
 146         _cpu_pda = new_cpu_pda;
 147 }
 148 #endif
 149
 150 /*
 151  * Great future plan:
 152  * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
 153  * Always point %gs to its beginning
 154  */
 155 void __init setup_per_cpu_areas(void)
 156 {
 157         ssize_t size = PERCPU_ENOUGH_ROOM;
 158         char *ptr;
 159         int cpu;
 160
 161 #ifdef CONFIG_HOTPLUG_CPU
 162         prefill_possible_map();
 163 #else
 164         nr_cpu_ids = num_processors;
 165 #endif
 166
 167         /* Setup cpu_pda map */
 168         setup_cpu_pda_map();
 169
 170         /* Copy section for each CPU (we discard the original) */
 171         size = PERCPU_ENOUGH_ROOM;
 172         printk(KERN_INFO "PERCPU: Allocating %lu bytes of per cpu data\n",
 173                           size);
 174
 175         for_each_possible_cpu(cpu) {
 176 #ifndef CONFIG_NEED_MULTIPLE_NODES
 177                 ptr = alloc_bootmem_pages(size);
 178 #else
 179                 int node = early_cpu_to_node(cpu);
 180                 if (!node_online(node) || !NODE_DATA(node)) {
 181                         ptr = alloc_bootmem_pages(size);
 182                         printk(KERN_INFO
 183                                "cpu %d has no node %d or node-local memory\n",
 184                                 cpu, node);
 185                 }
 186                 else
 187                         ptr = alloc_bootmem_pages_node(NODE_DATA(node), size);
 188 #endif
 189                 per_cpu_offset(cpu) = ptr - __per_cpu_start;
 190                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 191
 192         }
 193
 194         printk(KERN_DEBUG "NR_CPUS: %d, nr_cpu_ids: %d, nr_node_ids %d\n",
 195                 NR_CPUS, nr_cpu_ids, nr_node_ids);
 196
 197         /* Setup percpu data maps */
 198         setup_per_cpu_maps();
 199
 200         /* Setup node to cpumask map */
 201         setup_node_to_cpumask_map();
 202
 203         /* Setup cpumask_of_cpu map */
 204         setup_cpumask_of_cpu();
 205 }
 206
 207 #endif
 208
 209 #ifdef X86_64_NUMA
 210
 211 /*
 212  * Allocate node_to_cpumask_map based on number of available nodes
 213  * Requires node_possible_map to be valid.
 214  *
 215  * Note: node_to_cpumask() is not valid until after this is done.
 216  */
 217 static void __init setup_node_to_cpumask_map(void)
 218 {
 219         unsigned int node, num = 0;
 220         cpumask_t *map;
 221
 222         /* setup nr_node_ids if not done yet */
 223         if (nr_node_ids == MAX_NUMNODES) {
 224                 for_each_node_mask(node, node_possible_map)
 225                         num = node;
 226                 nr_node_ids = num + 1;
 227         }
 228
 229         /* allocate the map */
 230         map = alloc_bootmem_low(nr_node_ids * sizeof(cpumask_t));
 231
 232         Dprintk(KERN_DEBUG "Node to cpumask map at %p for %d nodes\n",
 233                 map, nr_node_ids);
 234
 235         /* node_to_cpumask() will now work */
 236         node_to_cpumask_map = map;
 237 }
 238
 239 void __cpuinit numa_set_node(int cpu, int node)
 240 {
 241         int *cpu_to_node_map = early_per_cpu_ptr(x86_cpu_to_node_map);
 242
 243         if (cpu_pda(cpu) && node != NUMA_NO_NODE)
 244                 cpu_pda(cpu)->nodenumber = node;
 245
 246         if (cpu_to_node_map)
 247                 cpu_to_node_map[cpu] = node;
 248
 249         else if (per_cpu_offset(cpu))
 250                 per_cpu(x86_cpu_to_node_map, cpu) = node;
 251
 252         else
 253                 Dprintk(KERN_INFO "Setting node for non-present cpu %d\n", cpu);
 254 }
 255
 256 void __cpuinit numa_clear_node(int cpu)
 257 {
 258         numa_set_node(cpu, NUMA_NO_NODE);
 259 }
 260
 261 #ifndef CONFIG_DEBUG_PER_CPU_MAPS
 262
 263 void __cpuinit numa_add_cpu(int cpu)
 264 {
 265         cpu_set(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
 266 }
 267
 268 void __cpuinit numa_remove_cpu(int cpu)
 269 {
 270         cpu_clear(cpu, node_to_cpumask_map[cpu_to_node(cpu)]);
 271 }
 272
 273 #else /* CONFIG_DEBUG_PER_CPU_MAPS */
 274
 275 /*
 276  * --------- debug versions of the numa functions ---------
 277  */
 278 static void __cpuinit numa_set_cpumask(int cpu, int enable)
 279 {
 280         int node = cpu_to_node(cpu);
 281         cpumask_t *mask;
 282         char buf[64];
 283
 284         if (node_to_cpumask_map == NULL) {
 285                 printk(KERN_ERR "node_to_cpumask_map NULL\n");
 286                 dump_stack();
 287                 return;
 288         }
 289
 290         mask = &node_to_cpumask_map[node];
 291         if (enable)
 292                 cpu_set(cpu, *mask);
 293         else
 294                 cpu_clear(cpu, *mask);
 295
 296         cpulist_scnprintf(buf, sizeof(buf), *mask);
 297         printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
 298                 enable? "numa_add_cpu":"numa_remove_cpu", cpu, node, buf);
 299  }
 300
 301 void __cpuinit numa_add_cpu(int cpu)
 302 {
 303         numa_set_cpumask(cpu, 1);
 304 }
 305
 306 void __cpuinit numa_remove_cpu(int cpu)
 307 {
 308         numa_set_cpumask(cpu, 0);
 309 }
 310
 311 int cpu_to_node(int cpu)
 312 {
 313         if (early_per_cpu_ptr(x86_cpu_to_node_map)) {
 314                 printk(KERN_WARNING
 315                         "cpu_to_node(%d): usage too early!\n", cpu);
 316                 dump_stack();
 317                 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
 318         }
 319         return per_cpu(x86_cpu_to_node_map, cpu);
 320 }
 321 EXPORT_SYMBOL(cpu_to_node);
 322
 323 /*
 324  * Same function as cpu_to_node() but used if called before the
 325  * per_cpu areas are setup.
 326  */
 327 int early_cpu_to_node(int cpu)
 328 {
 329         if (early_per_cpu_ptr(x86_cpu_to_node_map))
 330                 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
 331
 332         if (!per_cpu_offset(cpu)) {
 333                 printk(KERN_WARNING
 334                         "early_cpu_to_node(%d): no per_cpu area!\n", cpu);
 335                 dump_stack();
 336                 return NUMA_NO_NODE;
 337         }
 338         return per_cpu(x86_cpu_to_node_map, cpu);
 339 }
 340
 341 /*
 342  * Returns a pointer to the bitmask of CPUs on Node 'node'.
 343  */
 344 cpumask_t *_node_to_cpumask_ptr(int node)
 345 {
 346         if (node_to_cpumask_map == NULL) {
 347                 printk(KERN_WARNING
 348                         "_node_to_cpumask_ptr(%d): no node_to_cpumask_map!\n",
 349                         node);
 350                 dump_stack();
 351                 return &cpu_online_map;
 352         }
 353         BUG_ON(node >= nr_node_ids);
 354         return &node_to_cpumask_map[node];
 355 }
 356 EXPORT_SYMBOL(_node_to_cpumask_ptr);
 357
 358 /*
 359  * Returns a bitmask of CPUs on Node 'node'.
 360  */
 361 cpumask_t node_to_cpumask(int node)
 362 {
 363         if (node_to_cpumask_map == NULL) {
 364                 printk(KERN_WARNING
 365                         "node_to_cpumask(%d): no node_to_cpumask_map!\n", node);
 366                 dump_stack();
 367                 return cpu_online_map;
 368         }
 369         BUG_ON(node >= nr_node_ids);
 370         return node_to_cpumask_map[node];
 371 }
 372 EXPORT_SYMBOL(node_to_cpumask);
 373
 374 /*
 375  * --------- end of debug versions of the numa functions ---------
 376  */
 377
 378 #endif /* CONFIG_DEBUG_PER_CPU_MAPS */
 379
 380 #endif /* X86_64_NUMA */