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 <linux/crash_dump.h>
9 #include <asm/percpu.h>
10 #include <asm/sections.h>
11 #include <asm/processor.h>
12 #include <asm/setup.h>
13 #include <asm/topology.h>
14 #include <asm/mpspec.h>
15 #include <asm/apicdef.h>
16 #include <asm/highmem.h>
18 #ifdef CONFIG_X86_LOCAL_APIC
19 unsigned int num_processors;
20 unsigned disabled_cpus __cpuinitdata;
21 /* Processor that is doing the boot up */
22 unsigned int boot_cpu_physical_apicid = -1U;
23 unsigned int max_physical_apicid;
24 EXPORT_SYMBOL(boot_cpu_physical_apicid);
26 /* Bitmask of physically existing CPUs */
27 physid_mask_t phys_cpu_present_map;
30 /* map cpu index to physical APIC ID */
31 DEFINE_EARLY_PER_CPU(u16, x86_cpu_to_apicid, BAD_APICID);
32 DEFINE_EARLY_PER_CPU(u16, x86_bios_cpu_apicid, BAD_APICID);
33 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
34 EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
36 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
39 /* map cpu index to node index */
40 DEFINE_EARLY_PER_CPU(int, x86_cpu_to_node_map, NUMA_NO_NODE);
41 EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_node_map);
43 /* which logical CPUs are on which nodes */
44 cpumask_t *node_to_cpumask_map;
45 EXPORT_SYMBOL(node_to_cpumask_map);
47 /* setup node_to_cpumask_map */
48 static void __init setup_node_to_cpumask_map(void);
51 static inline void setup_node_to_cpumask_map(void) { }
54 #if defined(CONFIG_HAVE_SETUP_PER_CPU_AREA) && defined(CONFIG_X86_SMP)
56 * Copy data used in early init routines from the initial arrays to the
57 * per cpu data areas. These arrays then become expendable and the
58 * *_early_ptr's are zeroed indicating that the static arrays are gone.
60 static void __init setup_per_cpu_maps(void)
64 for_each_possible_cpu(cpu) {
65 per_cpu(x86_cpu_to_apicid, cpu) =
66 early_per_cpu_map(x86_cpu_to_apicid, cpu);
67 per_cpu(x86_bios_cpu_apicid, cpu) =
68 early_per_cpu_map(x86_bios_cpu_apicid, cpu);
70 per_cpu(x86_cpu_to_node_map, cpu) =
71 early_per_cpu_map(x86_cpu_to_node_map, cpu);
75 /* indicate the early static arrays will soon be gone */
76 early_per_cpu_ptr(x86_cpu_to_apicid) = NULL;
77 early_per_cpu_ptr(x86_bios_cpu_apicid) = NULL;
79 early_per_cpu_ptr(x86_cpu_to_node_map) = NULL;
83 #ifdef CONFIG_HAVE_CPUMASK_OF_CPU_MAP
84 cpumask_t *cpumask_of_cpu_map __read_mostly;
85 EXPORT_SYMBOL(cpumask_of_cpu_map);
87 /* requires nr_cpu_ids to be initialized */
88 static void __init setup_cpumask_of_cpu(void)
92 /* alloc_bootmem zeroes memory */
93 cpumask_of_cpu_map = alloc_bootmem_low(sizeof(cpumask_t) * nr_cpu_ids);
94 for (i = 0; i < nr_cpu_ids; i++)
95 cpu_set(i, cpumask_of_cpu_map[i]);
98 static inline void setup_cpumask_of_cpu(void) { }
103 * Great future not-so-futuristic plan: make i386 and x86_64 do it
106 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
107 EXPORT_SYMBOL(__per_cpu_offset);
108 static inline void setup_cpu_pda_map(void) { }
110 #elif !defined(CONFIG_SMP)
111 static inline void setup_cpu_pda_map(void) { }
113 #else /* CONFIG_SMP && CONFIG_X86_64 */
116 * Allocate cpu_pda pointer table and array via alloc_bootmem.
118 static void __init setup_cpu_pda_map(void)
121 struct x8664_pda **new_cpu_pda;
125 size = roundup(sizeof(struct x8664_pda), cache_line_size());
127 /* allocate cpu_pda array and pointer table */
129 unsigned long tsize = nr_cpu_ids * sizeof(void *);
130 unsigned long asize = size * (nr_cpu_ids - 1);
132 tsize = roundup(tsize, cache_line_size());
133 new_cpu_pda = alloc_bootmem(tsize + asize);
134 pda = (char *)new_cpu_pda + tsize;
137 /* initialize pointer table to static pda's */
138 for_each_possible_cpu(cpu) {
140 /* leave boot cpu pda in place */
141 new_cpu_pda[0] = cpu_pda(0);
144 new_cpu_pda[cpu] = (struct x8664_pda *)pda;
145 new_cpu_pda[cpu]->in_bootmem = 1;
149 /* point to new pointer table */
150 _cpu_pda = new_cpu_pda;
156 * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
157 * Always point %gs to its beginning
159 void __init setup_per_cpu_areas(void)
161 ssize_t size = PERCPU_ENOUGH_ROOM;
165 /* no processor from mptable or madt */
169 #ifdef CONFIG_HOTPLUG_CPU
170 prefill_possible_map();
172 nr_cpu_ids = num_processors;
175 /* Setup cpu_pda map */
178 /* Copy section for each CPU (we discard the original) */
179 size = PERCPU_ENOUGH_ROOM;
180 printk(KERN_INFO "PERCPU: Allocating %zd bytes of per cpu data\n",
183 for_each_possible_cpu(cpu) {
184 #ifndef CONFIG_NEED_MULTIPLE_NODES
185 ptr = alloc_bootmem_pages(size);
187 int node = early_cpu_to_node(cpu);
188 if (!node_online(node) || !NODE_DATA(node)) {
189 ptr = alloc_bootmem_pages(size);
191 "cpu %d has no node %d or node-local memory\n",
195 ptr = alloc_bootmem_pages_node(NODE_DATA(node), size);
197 per_cpu_offset(cpu) = ptr - __per_cpu_start;
198 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
202 printk(KERN_DEBUG "NR_CPUS: %d, nr_cpu_ids: %d, nr_node_ids %d\n",
203 NR_CPUS, nr_cpu_ids, nr_node_ids);
205 /* Setup percpu data maps */
206 setup_per_cpu_maps();
208 /* Setup node to cpumask map */
209 setup_node_to_cpumask_map();
211 /* Setup cpumask_of_cpu map */
212 setup_cpumask_of_cpu();
217 void __init parse_setup_data(void)
219 struct setup_data *data;
222 if (boot_params.hdr.version < 0x0209)
224 pa_data = boot_params.hdr.setup_data;
226 data = early_ioremap(pa_data, PAGE_SIZE);
227 switch (data->type) {
229 parse_e820_ext(data, pa_data);
234 #ifndef CONFIG_DEBUG_BOOT_PARAMS
235 free_early(pa_data, pa_data+sizeof(*data)+data->len);
237 pa_data = data->next;
238 early_iounmap(data, PAGE_SIZE);
245 * Allocate node_to_cpumask_map based on number of available nodes
246 * Requires node_possible_map to be valid.
248 * Note: node_to_cpumask() is not valid until after this is done.
250 static void __init setup_node_to_cpumask_map(void)
252 unsigned int node, num = 0;
255 /* setup nr_node_ids if not done yet */
256 if (nr_node_ids == MAX_NUMNODES) {
257 for_each_node_mask(node, node_possible_map)
259 nr_node_ids = num + 1;
262 /* allocate the map */
263 map = alloc_bootmem_low(nr_node_ids * sizeof(cpumask_t));
265 Dprintk(KERN_DEBUG "Node to cpumask map at %p for %d nodes\n",
268 /* node_to_cpumask() will now work */
269 node_to_cpumask_map = map;
272 void __cpuinit numa_set_node(int cpu, int node)
274 int *cpu_to_node_map = early_per_cpu_ptr(x86_cpu_to_node_map);
276 if (cpu_pda(cpu) && node != NUMA_NO_NODE)
277 cpu_pda(cpu)->nodenumber = node;
280 cpu_to_node_map[cpu] = node;
282 else if (per_cpu_offset(cpu))
283 per_cpu(x86_cpu_to_node_map, cpu) = node;
286 Dprintk(KERN_INFO "Setting node for non-present cpu %d\n", cpu);
289 void __cpuinit numa_clear_node(int cpu)
291 numa_set_node(cpu, NUMA_NO_NODE);
294 #ifndef CONFIG_DEBUG_PER_CPU_MAPS
296 void __cpuinit numa_add_cpu(int cpu)
298 cpu_set(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
301 void __cpuinit numa_remove_cpu(int cpu)
303 cpu_clear(cpu, node_to_cpumask_map[cpu_to_node(cpu)]);
306 #else /* CONFIG_DEBUG_PER_CPU_MAPS */
309 * --------- debug versions of the numa functions ---------
311 static void __cpuinit numa_set_cpumask(int cpu, int enable)
313 int node = cpu_to_node(cpu);
317 if (node_to_cpumask_map == NULL) {
318 printk(KERN_ERR "node_to_cpumask_map NULL\n");
323 mask = &node_to_cpumask_map[node];
327 cpu_clear(cpu, *mask);
329 cpulist_scnprintf(buf, sizeof(buf), *mask);
330 printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
331 enable? "numa_add_cpu":"numa_remove_cpu", cpu, node, buf);
334 void __cpuinit numa_add_cpu(int cpu)
336 numa_set_cpumask(cpu, 1);
339 void __cpuinit numa_remove_cpu(int cpu)
341 numa_set_cpumask(cpu, 0);
344 int cpu_to_node(int cpu)
346 if (early_per_cpu_ptr(x86_cpu_to_node_map)) {
348 "cpu_to_node(%d): usage too early!\n", cpu);
350 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
352 return per_cpu(x86_cpu_to_node_map, cpu);
354 EXPORT_SYMBOL(cpu_to_node);
357 * Same function as cpu_to_node() but used if called before the
358 * per_cpu areas are setup.
360 int early_cpu_to_node(int cpu)
362 if (early_per_cpu_ptr(x86_cpu_to_node_map))
363 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
365 if (!per_cpu_offset(cpu)) {
367 "early_cpu_to_node(%d): no per_cpu area!\n", cpu);
371 return per_cpu(x86_cpu_to_node_map, cpu);
375 * Returns a pointer to the bitmask of CPUs on Node 'node'.
377 cpumask_t *_node_to_cpumask_ptr(int node)
379 if (node_to_cpumask_map == NULL) {
381 "_node_to_cpumask_ptr(%d): no node_to_cpumask_map!\n",
384 return &cpu_online_map;
386 BUG_ON(node >= nr_node_ids);
387 return &node_to_cpumask_map[node];
389 EXPORT_SYMBOL(_node_to_cpumask_ptr);
392 * Returns a bitmask of CPUs on Node 'node'.
394 cpumask_t node_to_cpumask(int node)
396 if (node_to_cpumask_map == NULL) {
398 "node_to_cpumask(%d): no node_to_cpumask_map!\n", node);
400 return cpu_online_map;
402 BUG_ON(node >= nr_node_ids);
403 return node_to_cpumask_map[node];
405 EXPORT_SYMBOL(node_to_cpumask);
408 * --------- end of debug versions of the numa functions ---------
411 #endif /* CONFIG_DEBUG_PER_CPU_MAPS */
413 #endif /* X86_64_NUMA */
417 * --------- Crashkernel reservation ------------------------------
420 static inline unsigned long long get_total_mem(void)
422 unsigned long long total;
424 total = max_low_pfn - min_low_pfn;
425 #ifdef CONFIG_HIGHMEM
426 total += highend_pfn - highstart_pfn;
429 return total << PAGE_SHIFT;
433 void __init reserve_crashkernel(void)
435 unsigned long long total_mem;
436 unsigned long long crash_size, crash_base;
439 total_mem = get_total_mem();
441 ret = parse_crashkernel(boot_command_line, total_mem,
442 &crash_size, &crash_base);
443 if (ret == 0 && crash_size > 0) {
444 if (crash_base <= 0) {
445 printk(KERN_INFO "crashkernel reservation failed - "
446 "you have to specify a base address\n");
450 if (reserve_bootmem_generic(crash_base, crash_size,
451 BOOTMEM_EXCLUSIVE) < 0) {
452 printk(KERN_INFO "crashkernel reservation failed - "
453 "memory is in use\n");
457 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
458 "for crashkernel (System RAM: %ldMB)\n",
459 (unsigned long)(crash_size >> 20),
460 (unsigned long)(crash_base >> 20),
461 (unsigned long)(total_mem >> 20));
463 crashk_res.start = crash_base;
464 crashk_res.end = crash_base + crash_size - 1;
465 insert_resource(&iomem_resource, &crashk_res);
469 void __init reserve_crashkernel(void)
472 static struct resource standard_io_resources[] = {
473 { .name = "dma1", .start = 0x00, .end = 0x1f,
474 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
475 { .name = "pic1", .start = 0x20, .end = 0x21,
476 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
477 { .name = "timer0", .start = 0x40, .end = 0x43,
478 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
479 { .name = "timer1", .start = 0x50, .end = 0x53,
480 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
481 { .name = "keyboard", .start = 0x60, .end = 0x60,
482 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
483 { .name = "keyboard", .start = 0x64, .end = 0x64,
484 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
485 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
486 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
487 { .name = "pic2", .start = 0xa0, .end = 0xa1,
488 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
489 { .name = "dma2", .start = 0xc0, .end = 0xdf,
490 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
491 { .name = "fpu", .start = 0xf0, .end = 0xff,
492 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
495 void __init reserve_standard_io_resources(void)
499 /* request I/O space for devices used on all i[345]86 PCs */
500 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
501 request_resource(&ioport_resource, &standard_io_resources[i]);
505 #ifdef CONFIG_PROC_VMCORE
506 /* elfcorehdr= specifies the location of elf core header
507 * stored by the crashed kernel. This option will be passed
508 * by kexec loader to the capture kernel.
510 static int __init setup_elfcorehdr(char *arg)
515 elfcorehdr_addr = memparse(arg, &end);
516 return end > arg ? 0 : -EINVAL;
518 early_param("elfcorehdr", setup_elfcorehdr);