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 <linux/kexec.h>
   7 #include <asm/smp.h>
   8 #include <asm/percpu.h>
   9 #include <asm/sections.h>
  10 #include <asm/processor.h>
  11 #include <asm/setup.h>
  12 #include <asm/topology.h>
  13 #include <asm/mpspec.h>
  14 #include <asm/apicdef.h>
  15 #include <asm/highmem.h>
  16
  17 #ifdef CONFIG_X86_LOCAL_APIC
  18 unsigned int num_processors;
  19 unsigned disabled_cpus __cpuinitdata;
  20 /* Processor that is doing the boot up */
  21 unsigned int boot_cpu_physical_apicid = -1U;
  22 unsigned int max_physical_apicid;
  23 EXPORT_SYMBOL(boot_cpu_physical_apicid);
  24
  25 /* Bitmask of physically existing CPUs */
  26 physid_mask_t phys_cpu_present_map;
  27 #endif
  28
  29 /* map cpu index to physical APIC ID */
  30 DEFINE_EARLY_PER_CPU(u16, x86_cpu_to_apicid, BAD_APICID);
  31 DEFINE_EARLY_PER_CPU(u16, x86_bios_cpu_apicid, BAD_APICID);
  32 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
  33 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
  34
  35 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
  36 #define X86_64_NUMA     1
  37
  38 /* map cpu index to node index */
  39 DEFINE_EARLY_PER_CPU(int, x86_cpu_to_node_map, NUMA_NO_NODE);
  40 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_node_map);
  41
  42 /* which logical CPUs are on which nodes */
  43 cpumask_t *node_to_cpumask_map;
  44 EXPORT_SYMBOL(node_to_cpumask_map);
  45
  46 /* setup node_to_cpumask_map */
  47 static void __init setup_node_to_cpumask_map(void);
  48
  49 #else
  50 static inline void setup_node_to_cpumask_map(void) { }
  51 #endif
  52
  53 #if defined(CONFIG_HAVE_SETUP_PER_CPU_AREA) && defined(CONFIG_X86_SMP)
  54 /*
  55  * Copy data used in early init routines from the initial arrays to the
  56  * per cpu data areas.  These arrays then become expendable and the
  57  * *_early_ptr's are zeroed indicating that the static arrays are gone.
  58  */
  59 static void __init setup_per_cpu_maps(void)
  60 {
  61         int cpu;
  62
  63         for_each_possible_cpu(cpu) {
  64                 per_cpu(x86_cpu_to_apicid, cpu) =
  65                                 early_per_cpu_map(x86_cpu_to_apicid, cpu);
  66                 per_cpu(x86_bios_cpu_apicid, cpu) =
  67                                 early_per_cpu_map(x86_bios_cpu_apicid, cpu);
  68 #ifdef X86_64_NUMA
  69                 per_cpu(x86_cpu_to_node_map, cpu) =
  70                                 early_per_cpu_map(x86_cpu_to_node_map, cpu);
  71 #endif
  72         }
  73
  74         /* indicate the early static arrays will soon be gone */
  75         early_per_cpu_ptr(x86_cpu_to_apicid) = NULL;
  76         early_per_cpu_ptr(x86_bios_cpu_apicid) = NULL;
  77 #ifdef X86_64_NUMA
  78         early_per_cpu_ptr(x86_cpu_to_node_map) = NULL;
  79 #endif
  80 }
  81
  82 #ifdef CONFIG_HAVE_CPUMASK_OF_CPU_MAP
  83 cpumask_t *cpumask_of_cpu_map __read_mostly;
  84 EXPORT_SYMBOL(cpumask_of_cpu_map);
  85
  86 /* requires nr_cpu_ids to be initialized */
  87 static void __init setup_cpumask_of_cpu(void)
  88 {
  89         int i;
  90
  91         /* alloc_bootmem zeroes memory */
  92         cpumask_of_cpu_map = alloc_bootmem_low(sizeof(cpumask_t) * nr_cpu_ids);
  93         for (i = 0; i < nr_cpu_ids; i++)
  94                 cpu_set(i, cpumask_of_cpu_map[i]);
  95 }
  96 #else
  97 static inline void setup_cpumask_of_cpu(void) { }
  98 #endif
  99
 100 #ifdef CONFIG_X86_32
 101 /*
 102  * Great future not-so-futuristic plan: make i386 and x86_64 do it
 103  * the same way
 104  */
 105 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
 106 EXPORT_SYMBOL(__per_cpu_offset);
 107 static inline void setup_cpu_pda_map(void) { }
 108
 109 #elif !defined(CONFIG_SMP)
 110 static inline void setup_cpu_pda_map(void) { }
 111
 112 #else /* CONFIG_SMP && CONFIG_X86_64 */
 113
 114 /*
 115  * Allocate cpu_pda pointer table and array via alloc_bootmem.
 116  */
 117 static void __init setup_cpu_pda_map(void)
 118 {
 119         char *pda;
 120         struct x8664_pda **new_cpu_pda;
 121         unsigned long size;
 122         int cpu;
 123
 124         size = roundup(sizeof(struct x8664_pda), cache_line_size());
 125
 126         /* allocate cpu_pda array and pointer table */
 127         {
 128                 unsigned long tsize = nr_cpu_ids * sizeof(void *);
 129                 unsigned long asize = size * (nr_cpu_ids - 1);
 130
 131                 tsize = roundup(tsize, cache_line_size());
 132                 new_cpu_pda = alloc_bootmem(tsize + asize);
 133                 pda = (char *)new_cpu_pda + tsize;
 134         }
 135
 136         /* initialize pointer table to static pda's */
 137         for_each_possible_cpu(cpu) {
 138                 if (cpu == 0) {
 139                         /* leave boot cpu pda in place */
 140                         new_cpu_pda[0] = cpu_pda(0);
 141                         continue;
 142                 }
 143                 new_cpu_pda[cpu] = (struct x8664_pda *)pda;
 144                 new_cpu_pda[cpu]->in_bootmem = 1;
 145                 pda += size;
 146         }
 147
 148         /* point to new pointer table */
 149         _cpu_pda = new_cpu_pda;
 150 }
 151 #endif
 152
 153 /*
 154  * Great future plan:
 155  * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
 156  * Always point %gs to its beginning
 157  */
 158 void __init setup_per_cpu_areas(void)
 159 {
 160         ssize_t size = PERCPU_ENOUGH_ROOM;
 161         char *ptr;
 162         int cpu;
 163
 164         /* no processor from mptable or madt */
 165         if (!num_processors)
 166                 num_processors = 1;
 167
 168 #ifdef CONFIG_HOTPLUG_CPU
 169         prefill_possible_map();
 170 #else
 171         nr_cpu_ids = num_processors;
 172 #endif
 173
 174         /* Setup cpu_pda map */
 175         setup_cpu_pda_map();
 176
 177         /* Copy section for each CPU (we discard the original) */
 178         size = PERCPU_ENOUGH_ROOM;
 179         printk(KERN_INFO "PERCPU: Allocating %zd bytes of per cpu data\n",
 180                           size);
 181
 182         for_each_possible_cpu(cpu) {
 183 #ifndef CONFIG_NEED_MULTIPLE_NODES
 184                 ptr = alloc_bootmem_pages(size);
 185 #else
 186                 int node = early_cpu_to_node(cpu);
 187                 if (!node_online(node) || !NODE_DATA(node)) {
 188                         ptr = alloc_bootmem_pages(size);
 189                         printk(KERN_INFO
 190                                "cpu %d has no node %d or node-local memory\n",
 191                                 cpu, node);
 192                 }
 193                 else
 194                         ptr = alloc_bootmem_pages_node(NODE_DATA(node), size);
 195 #endif
 196                 per_cpu_offset(cpu) = ptr - __per_cpu_start;
 197                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 198
 199         }
 200
 201         printk(KERN_DEBUG "NR_CPUS: %d, nr_cpu_ids: %d, nr_node_ids %d\n",
 202                 NR_CPUS, nr_cpu_ids, nr_node_ids);
 203
 204         /* Setup percpu data maps */
 205         setup_per_cpu_maps();
 206
 207         /* Setup node to cpumask map */
 208         setup_node_to_cpumask_map();
 209
 210         /* Setup cpumask_of_cpu map */
 211         setup_cpumask_of_cpu();
 212 }
 213
 214 #endif
 215
 216 void __init parse_setup_data(void)
 217 {
 218         struct setup_data *data;
 219         u64 pa_data;
 220
 221         if (boot_params.hdr.version < 0x0209)
 222                 return;
 223         pa_data = boot_params.hdr.setup_data;
 224         while (pa_data) {
 225                 data = early_ioremap(pa_data, PAGE_SIZE);
 226                 switch (data->type) {
 227                 case SETUP_E820_EXT:
 228                         parse_e820_ext(data, pa_data);
 229                         break;
 230                 default:
 231                         break;
 232                 }
 233 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 234                 free_early(pa_data, pa_data+sizeof(*data)+data->len);
 235 #endif
 236                 pa_data = data->next;
 237                 early_iounmap(data, PAGE_SIZE);
 238         }
 239 }
 240
 241 #ifdef X86_64_NUMA
 242
 243 /*
 244  * Allocate node_to_cpumask_map based on number of available nodes
 245  * Requires node_possible_map to be valid.
 246  *
 247  * Note: node_to_cpumask() is not valid until after this is done.
 248  */
 249 static void __init setup_node_to_cpumask_map(void)
 250 {
 251         unsigned int node, num = 0;
 252         cpumask_t *map;
 253
 254         /* setup nr_node_ids if not done yet */
 255         if (nr_node_ids == MAX_NUMNODES) {
 256                 for_each_node_mask(node, node_possible_map)
 257                         num = node;
 258                 nr_node_ids = num + 1;
 259         }
 260
 261         /* allocate the map */
 262         map = alloc_bootmem_low(nr_node_ids * sizeof(cpumask_t));
 263
 264         Dprintk(KERN_DEBUG "Node to cpumask map at %p for %d nodes\n",
 265                 map, nr_node_ids);
 266
 267         /* node_to_cpumask() will now work */
 268         node_to_cpumask_map = map;
 269 }
 270
 271 void __cpuinit numa_set_node(int cpu, int node)
 272 {
 273         int *cpu_to_node_map = early_per_cpu_ptr(x86_cpu_to_node_map);
 274
 275         if (cpu_pda(cpu) && node != NUMA_NO_NODE)
 276                 cpu_pda(cpu)->nodenumber = node;
 277
 278         if (cpu_to_node_map)
 279                 cpu_to_node_map[cpu] = node;
 280
 281         else if (per_cpu_offset(cpu))
 282                 per_cpu(x86_cpu_to_node_map, cpu) = node;
 283
 284         else
 285                 Dprintk(KERN_INFO "Setting node for non-present cpu %d\n", cpu);
 286 }
 287
 288 void __cpuinit numa_clear_node(int cpu)
 289 {
 290         numa_set_node(cpu, NUMA_NO_NODE);
 291 }
 292
 293 #ifndef CONFIG_DEBUG_PER_CPU_MAPS
 294
 295 void __cpuinit numa_add_cpu(int cpu)
 296 {
 297         cpu_set(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
 298 }
 299
 300 void __cpuinit numa_remove_cpu(int cpu)
 301 {
 302         cpu_clear(cpu, node_to_cpumask_map[cpu_to_node(cpu)]);
 303 }
 304
 305 #else /* CONFIG_DEBUG_PER_CPU_MAPS */
 306
 307 /*
 308  * --------- debug versions of the numa functions ---------
 309  */
 310 static void __cpuinit numa_set_cpumask(int cpu, int enable)
 311 {
 312         int node = cpu_to_node(cpu);
 313         cpumask_t *mask;
 314         char buf[64];
 315
 316         if (node_to_cpumask_map == NULL) {
 317                 printk(KERN_ERR "node_to_cpumask_map NULL\n");
 318                 dump_stack();
 319                 return;
 320         }
 321
 322         mask = &node_to_cpumask_map[node];
 323         if (enable)
 324                 cpu_set(cpu, *mask);
 325         else
 326                 cpu_clear(cpu, *mask);
 327
 328         cpulist_scnprintf(buf, sizeof(buf), *mask);
 329         printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
 330                 enable? "numa_add_cpu":"numa_remove_cpu", cpu, node, buf);
 331  }
 332
 333 void __cpuinit numa_add_cpu(int cpu)
 334 {
 335         numa_set_cpumask(cpu, 1);
 336 }
 337
 338 void __cpuinit numa_remove_cpu(int cpu)
 339 {
 340         numa_set_cpumask(cpu, 0);
 341 }
 342
 343 int cpu_to_node(int cpu)
 344 {
 345         if (early_per_cpu_ptr(x86_cpu_to_node_map)) {
 346                 printk(KERN_WARNING
 347                         "cpu_to_node(%d): usage too early!\n", cpu);
 348                 dump_stack();
 349                 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
 350         }
 351         return per_cpu(x86_cpu_to_node_map, cpu);
 352 }
 353 EXPORT_SYMBOL(cpu_to_node);
 354
 355 /*
 356  * Same function as cpu_to_node() but used if called before the
 357  * per_cpu areas are setup.
 358  */
 359 int early_cpu_to_node(int cpu)
 360 {
 361         if (early_per_cpu_ptr(x86_cpu_to_node_map))
 362                 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
 363
 364         if (!per_cpu_offset(cpu)) {
 365                 printk(KERN_WARNING
 366                         "early_cpu_to_node(%d): no per_cpu area!\n", cpu);
 367                 dump_stack();
 368                 return NUMA_NO_NODE;
 369         }
 370         return per_cpu(x86_cpu_to_node_map, cpu);
 371 }
 372
 373 /*
 374  * Returns a pointer to the bitmask of CPUs on Node 'node'.
 375  */
 376 cpumask_t *_node_to_cpumask_ptr(int node)
 377 {
 378         if (node_to_cpumask_map == NULL) {
 379                 printk(KERN_WARNING
 380                         "_node_to_cpumask_ptr(%d): no node_to_cpumask_map!\n",
 381                         node);
 382                 dump_stack();
 383                 return &cpu_online_map;
 384         }
 385         BUG_ON(node >= nr_node_ids);
 386         return &node_to_cpumask_map[node];
 387 }
 388 EXPORT_SYMBOL(_node_to_cpumask_ptr);
 389
 390 /*
 391  * Returns a bitmask of CPUs on Node 'node'.
 392  */
 393 cpumask_t node_to_cpumask(int node)
 394 {
 395         if (node_to_cpumask_map == NULL) {
 396                 printk(KERN_WARNING
 397                         "node_to_cpumask(%d): no node_to_cpumask_map!\n", node);
 398                 dump_stack();
 399                 return cpu_online_map;
 400         }
 401         BUG_ON(node >= nr_node_ids);
 402         return node_to_cpumask_map[node];
 403 }
 404 EXPORT_SYMBOL(node_to_cpumask);
 405
 406 /*
 407  * --------- end of debug versions of the numa functions ---------
 408  */
 409
 410 #endif /* CONFIG_DEBUG_PER_CPU_MAPS */
 411
 412 #endif /* X86_64_NUMA */
 413
 414
 415 /*
 416  * --------- Crashkernel reservation ------------------------------
 417  */
 418
 419 static inline unsigned long long get_total_mem(void)
 420 {
 421         unsigned long long total;
 422
 423         total = max_low_pfn - min_low_pfn;
 424 #ifdef CONFIG_HIGHMEM
 425         total += highend_pfn - highstart_pfn;
 426 #endif
 427
 428         return total << PAGE_SHIFT;
 429 }
 430
 431 #ifdef CONFIG_KEXEC
 432 void __init reserve_crashkernel(void)
 433 {
 434         unsigned long long total_mem;
 435         unsigned long long crash_size, crash_base;
 436         int ret;
 437
 438         total_mem = get_total_mem();
 439
 440         ret = parse_crashkernel(boot_command_line, total_mem,
 441                         &crash_size, &crash_base);
 442         if (ret == 0 && crash_size > 0) {
 443                 if (crash_base <= 0) {
 444                         printk(KERN_INFO "crashkernel reservation failed - "
 445                                         "you have to specify a base address\n");
 446                         return;
 447                 }
 448
 449                 if (reserve_bootmem_generic(crash_base, crash_size,
 450                                         BOOTMEM_EXCLUSIVE) < 0) {
 451                         printk(KERN_INFO "crashkernel reservation failed - "
 452                                         "memory is in use\n");
 453                         return;
 454                 }
 455
 456                 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 457                                 "for crashkernel (System RAM: %ldMB)\n",
 458                                 (unsigned long)(crash_size >> 20),
 459                                 (unsigned long)(crash_base >> 20),
 460                                 (unsigned long)(total_mem >> 20));
 461
 462                 crashk_res.start = crash_base;
 463                 crashk_res.end   = crash_base + crash_size - 1;
 464                 insert_resource(&iomem_resource, &crashk_res);
 465         }
 466 }
 467 #else
 468 void __init reserve_crashkernel(void)
 469 {}
 470 #endif
 471