x86: remove mach_reboot.h
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / x86 / kernel / setup_64.c
CommitLineData
1da177e4 1/*
1da177e4 2 * Copyright (C) 1995 Linus Torvalds
1da177e4
LT
3 */
4
5/*
6 * This file handles the architecture-dependent parts of initialization
7 */
8
9#include <linux/errno.h>
10#include <linux/sched.h>
11#include <linux/kernel.h>
12#include <linux/mm.h>
13#include <linux/stddef.h>
14#include <linux/unistd.h>
15#include <linux/ptrace.h>
16#include <linux/slab.h>
17#include <linux/user.h>
894673ee 18#include <linux/screen_info.h>
1da177e4
LT
19#include <linux/ioport.h>
20#include <linux/delay.h>
1da177e4
LT
21#include <linux/init.h>
22#include <linux/initrd.h>
23#include <linux/highmem.h>
24#include <linux/bootmem.h>
25#include <linux/module.h>
26#include <asm/processor.h>
27#include <linux/console.h>
28#include <linux/seq_file.h>
aac04b32 29#include <linux/crash_dump.h>
1da177e4
LT
30#include <linux/root_dev.h>
31#include <linux/pci.h>
5b83683f 32#include <linux/efi.h>
1da177e4
LT
33#include <linux/acpi.h>
34#include <linux/kallsyms.h>
35#include <linux/edd.h>
bbfceef4 36#include <linux/mmzone.h>
5f5609df 37#include <linux/kexec.h>
95235ca2 38#include <linux/cpufreq.h>
e9928674 39#include <linux/dmi.h>
17a941d8 40#include <linux/dma-mapping.h>
681558fd 41#include <linux/ctype.h>
746ef0cd 42#include <linux/uaccess.h>
f212ec4b 43#include <linux/init_ohci1394_dma.h>
bbfceef4 44
1da177e4
LT
45#include <asm/mtrr.h>
46#include <asm/uaccess.h>
47#include <asm/system.h>
e4026440 48#include <asm/vsyscall.h>
1da177e4
LT
49#include <asm/io.h>
50#include <asm/smp.h>
51#include <asm/msr.h>
52#include <asm/desc.h>
53#include <video/edid.h>
54#include <asm/e820.h>
55#include <asm/dma.h>
aaf23042 56#include <asm/gart.h>
1da177e4
LT
57#include <asm/mpspec.h>
58#include <asm/mmu_context.h>
1da177e4
LT
59#include <asm/proto.h>
60#include <asm/setup.h>
61#include <asm/mach_apic.h>
62#include <asm/numa.h>
2bc0414e 63#include <asm/sections.h>
f2d3efed 64#include <asm/dmi.h>
00bf4098 65#include <asm/cacheflush.h>
af7a78e9 66#include <asm/mce.h>
eee3af4a 67#include <asm/ds.h>
df3825c5 68#include <asm/topology.h>
1da177e4 69
746ef0cd
GOC
70#ifdef CONFIG_PARAVIRT
71#include <asm/paravirt.h>
72#else
73#define ARCH_SETUP
74#endif
75
1da177e4
LT
76/*
77 * Machine setup..
78 */
79
6c231b7b 80struct cpuinfo_x86 boot_cpu_data __read_mostly;
2ee60e17 81EXPORT_SYMBOL(boot_cpu_data);
1da177e4 82
7d851c8d
AK
83__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
84
1da177e4
LT
85unsigned long mmu_cr4_features;
86
1da177e4
LT
87/* Boot loader ID as an integer, for the benefit of proc_dointvec */
88int bootloader_type;
89
90unsigned long saved_video_mode;
91
f039b754
AK
92int force_mwait __cpuinitdata;
93
04e1ba85 94/*
f2d3efed
AK
95 * Early DMI memory
96 */
97int dmi_alloc_index;
98char dmi_alloc_data[DMI_MAX_DATA];
99
1da177e4
LT
100/*
101 * Setup options
102 */
1da177e4 103struct screen_info screen_info;
2ee60e17 104EXPORT_SYMBOL(screen_info);
1da177e4
LT
105struct sys_desc_table_struct {
106 unsigned short length;
107 unsigned char table[0];
108};
109
110struct edid_info edid_info;
ba70710e 111EXPORT_SYMBOL_GPL(edid_info);
1da177e4
LT
112
113extern int root_mountflags;
1da177e4 114
adf48856 115char __initdata command_line[COMMAND_LINE_SIZE];
1da177e4
LT
116
117struct resource standard_io_resources[] = {
118 { .name = "dma1", .start = 0x00, .end = 0x1f,
119 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
120 { .name = "pic1", .start = 0x20, .end = 0x21,
121 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
122 { .name = "timer0", .start = 0x40, .end = 0x43,
123 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
124 { .name = "timer1", .start = 0x50, .end = 0x53,
125 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
126 { .name = "keyboard", .start = 0x60, .end = 0x6f,
127 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
128 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
129 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
130 { .name = "pic2", .start = 0xa0, .end = 0xa1,
131 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
132 { .name = "dma2", .start = 0xc0, .end = 0xdf,
133 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
134 { .name = "fpu", .start = 0xf0, .end = 0xff,
135 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
136};
137
1da177e4
LT
138#define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
139
c9cce83d 140static struct resource data_resource = {
1da177e4
LT
141 .name = "Kernel data",
142 .start = 0,
143 .end = 0,
144 .flags = IORESOURCE_RAM,
145};
c9cce83d 146static struct resource code_resource = {
1da177e4
LT
147 .name = "Kernel code",
148 .start = 0,
149 .end = 0,
150 .flags = IORESOURCE_RAM,
151};
c9cce83d 152static struct resource bss_resource = {
00bf4098
BW
153 .name = "Kernel bss",
154 .start = 0,
155 .end = 0,
156 .flags = IORESOURCE_RAM,
157};
1da177e4 158
8c61b900
TG
159static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
160
2c8c0e6b
AK
161#ifdef CONFIG_PROC_VMCORE
162/* elfcorehdr= specifies the location of elf core header
163 * stored by the crashed kernel. This option will be passed
164 * by kexec loader to the capture kernel.
165 */
166static int __init setup_elfcorehdr(char *arg)
681558fd 167{
2c8c0e6b
AK
168 char *end;
169 if (!arg)
170 return -EINVAL;
171 elfcorehdr_addr = memparse(arg, &end);
172 return end > arg ? 0 : -EINVAL;
681558fd 173}
2c8c0e6b 174early_param("elfcorehdr", setup_elfcorehdr);
e2c03888
AK
175#endif
176
2b97690f 177#ifndef CONFIG_NUMA
bbfceef4
MT
178static void __init
179contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
1da177e4 180{
bbfceef4
MT
181 unsigned long bootmap_size, bootmap;
182
bbfceef4 183 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
24a5da73
YL
184 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
185 PAGE_SIZE);
bbfceef4 186 if (bootmap == -1L)
04e1ba85 187 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
bbfceef4 188 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
5cb248ab
MG
189 e820_register_active_regions(0, start_pfn, end_pfn);
190 free_bootmem_with_active_regions(0, end_pfn);
72a7fe39 191 reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
04e1ba85 192}
1da177e4
LT
193#endif
194
1da177e4
LT
195#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
196struct edd edd;
197#ifdef CONFIG_EDD_MODULE
198EXPORT_SYMBOL(edd);
199#endif
200/**
201 * copy_edd() - Copy the BIOS EDD information
202 * from boot_params into a safe place.
203 *
204 */
205static inline void copy_edd(void)
206{
30c82645
PA
207 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
208 sizeof(edd.mbr_signature));
209 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
210 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
211 edd.edd_info_nr = boot_params.eddbuf_entries;
1da177e4
LT
212}
213#else
214static inline void copy_edd(void)
215{
216}
217#endif
218
5c3391f9
BW
219#ifdef CONFIG_KEXEC
220static void __init reserve_crashkernel(void)
221{
18a01a3b 222 unsigned long long total_mem;
5c3391f9
BW
223 unsigned long long crash_size, crash_base;
224 int ret;
225
18a01a3b 226 total_mem = ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
5c3391f9 227
18a01a3b 228 ret = parse_crashkernel(boot_command_line, total_mem,
5c3391f9
BW
229 &crash_size, &crash_base);
230 if (ret == 0 && crash_size) {
18a01a3b 231 if (crash_base <= 0) {
5c3391f9
BW
232 printk(KERN_INFO "crashkernel reservation failed - "
233 "you have to specify a base address\n");
18a01a3b
BW
234 return;
235 }
236
237 if (reserve_bootmem(crash_base, crash_size,
238 BOOTMEM_EXCLUSIVE) < 0) {
239 printk(KERN_INFO "crashkernel reservation failed - "
240 "memory is in use\n");
241 return;
242 }
243
244 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
245 "for crashkernel (System RAM: %ldMB)\n",
246 (unsigned long)(crash_size >> 20),
247 (unsigned long)(crash_base >> 20),
248 (unsigned long)(total_mem >> 20));
249 crashk_res.start = crash_base;
250 crashk_res.end = crash_base + crash_size - 1;
3def3d6d 251 insert_resource(&iomem_resource, &crashk_res);
5c3391f9
BW
252 }
253}
254#else
255static inline void __init reserve_crashkernel(void)
256{}
257#endif
258
746ef0cd 259/* Overridden in paravirt.c if CONFIG_PARAVIRT */
e3cfac84 260void __attribute__((weak)) __init memory_setup(void)
746ef0cd
GOC
261{
262 machine_specific_memory_setup();
263}
264
f212ec4b
BK
265/*
266 * setup_arch - architecture-specific boot-time initializations
267 *
268 * Note: On x86_64, fixmaps are ready for use even before this is called.
269 */
1da177e4
LT
270void __init setup_arch(char **cmdline_p)
271{
04e1ba85
TG
272 unsigned i;
273
adf48856 274 printk(KERN_INFO "Command line: %s\n", boot_command_line);
43c85c9c 275
30c82645
PA
276 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
277 screen_info = boot_params.screen_info;
278 edid_info = boot_params.edid_info;
279 saved_video_mode = boot_params.hdr.vid_mode;
280 bootloader_type = boot_params.hdr.type_of_loader;
1da177e4
LT
281
282#ifdef CONFIG_BLK_DEV_RAM
30c82645
PA
283 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
284 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
285 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
1da177e4 286#endif
5b83683f
HY
287#ifdef CONFIG_EFI
288 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
289 "EL64", 4))
290 efi_enabled = 1;
291#endif
746ef0cd
GOC
292
293 ARCH_SETUP
294
295 memory_setup();
1da177e4
LT
296 copy_edd();
297
30c82645 298 if (!boot_params.hdr.root_flags)
1da177e4
LT
299 root_mountflags &= ~MS_RDONLY;
300 init_mm.start_code = (unsigned long) &_text;
301 init_mm.end_code = (unsigned long) &_etext;
302 init_mm.end_data = (unsigned long) &_edata;
303 init_mm.brk = (unsigned long) &_end;
304
e3ebadd9
LT
305 code_resource.start = virt_to_phys(&_text);
306 code_resource.end = virt_to_phys(&_etext)-1;
307 data_resource.start = virt_to_phys(&_etext);
308 data_resource.end = virt_to_phys(&_edata)-1;
00bf4098
BW
309 bss_resource.start = virt_to_phys(&__bss_start);
310 bss_resource.end = virt_to_phys(&__bss_stop)-1;
1da177e4 311
1da177e4
LT
312 early_identify_cpu(&boot_cpu_data);
313
adf48856 314 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
2c8c0e6b
AK
315 *cmdline_p = command_line;
316
317 parse_early_param();
318
f212ec4b
BK
319#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
320 if (init_ohci1394_dma_early)
321 init_ohci1394_dma_on_all_controllers();
322#endif
323
2c8c0e6b 324 finish_e820_parsing();
9ca33eb6 325
3def3d6d
YL
326 /* after parse_early_param, so could debug it */
327 insert_resource(&iomem_resource, &code_resource);
328 insert_resource(&iomem_resource, &data_resource);
329 insert_resource(&iomem_resource, &bss_resource);
330
aaf23042
YL
331 early_gart_iommu_check();
332
5cb248ab 333 e820_register_active_regions(0, 0, -1UL);
1da177e4
LT
334 /*
335 * partially used pages are not usable - thus
336 * we are rounding upwards:
337 */
338 end_pfn = e820_end_of_ram();
99fc8d42
JB
339 /* update e820 for memory not covered by WB MTRRs */
340 mtrr_bp_init();
341 if (mtrr_trim_uncached_memory(end_pfn)) {
342 e820_register_active_regions(0, 0, -1UL);
343 end_pfn = e820_end_of_ram();
344 }
345
caff0710 346 num_physpages = end_pfn;
1da177e4
LT
347
348 check_efer();
349
350 init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
5b83683f
HY
351 if (efi_enabled)
352 efi_init();
1da177e4 353
2785c8d0
GC
354#ifdef CONFIG_PARAVIRT
355 vsmp_init();
356#endif
357
f2d3efed
AK
358 dmi_scan_machine();
359
b02aae9c
RH
360 io_delay_init();
361
71fff5e6 362#ifdef CONFIG_SMP
df3825c5 363 /* setup to use the early static init tables during kernel startup */
3effef1f
YL
364 x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
365 x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
e8c10ef9 366#ifdef CONFIG_NUMA
3effef1f 367 x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
71fff5e6 368#endif
e8c10ef9 369#endif
71fff5e6 370
888ba6c6 371#ifdef CONFIG_ACPI
1da177e4
LT
372 /*
373 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
374 * Call this early for SRAT node setup.
375 */
376 acpi_boot_table_init();
377#endif
378
caff0710
JB
379 /* How many end-of-memory variables you have, grandma! */
380 max_low_pfn = end_pfn;
381 max_pfn = end_pfn;
382 high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
383
5cb248ab
MG
384 /* Remove active ranges so rediscovery with NUMA-awareness happens */
385 remove_all_active_ranges();
386
1da177e4
LT
387#ifdef CONFIG_ACPI_NUMA
388 /*
389 * Parse SRAT to discover nodes.
390 */
391 acpi_numa_init();
392#endif
393
2b97690f 394#ifdef CONFIG_NUMA
04e1ba85 395 numa_initmem_init(0, end_pfn);
1da177e4 396#else
bbfceef4 397 contig_initmem_init(0, end_pfn);
1da177e4
LT
398#endif
399
75175278 400 early_res_to_bootmem();
1da177e4 401
673d5b43 402#ifdef CONFIG_ACPI_SLEEP
1da177e4 403 /*
04e1ba85 404 * Reserve low memory region for sleep support.
1da177e4 405 */
04e1ba85
TG
406 acpi_reserve_bootmem();
407#endif
5b83683f 408
a3828064 409 if (efi_enabled)
5b83683f 410 efi_reserve_bootmem();
5b83683f 411
04e1ba85
TG
412 /*
413 * Find and reserve possible boot-time SMP configuration:
414 */
1da177e4 415 find_smp_config();
1da177e4 416#ifdef CONFIG_BLK_DEV_INITRD
30c82645
PA
417 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
418 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
419 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
420 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
421 unsigned long end_of_mem = end_pfn << PAGE_SHIFT;
422
423 if (ramdisk_end <= end_of_mem) {
424 reserve_bootmem_generic(ramdisk_image, ramdisk_size);
425 initrd_start = ramdisk_image + PAGE_OFFSET;
426 initrd_end = initrd_start+ramdisk_size;
427 } else {
75175278
AK
428 /* Assumes everything on node 0 */
429 free_bootmem(ramdisk_image, ramdisk_size);
1da177e4 430 printk(KERN_ERR "initrd extends beyond end of memory "
30c82645
PA
431 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
432 ramdisk_end, end_of_mem);
1da177e4
LT
433 initrd_start = 0;
434 }
435 }
436#endif
5c3391f9 437 reserve_crashkernel();
1da177e4 438 paging_init();
e4026440 439 map_vsyscall();
1da177e4 440
dfa4698c 441 early_quirks();
1da177e4 442
888ba6c6 443#ifdef CONFIG_ACPI
1da177e4
LT
444 /*
445 * Read APIC and some other early information from ACPI tables.
446 */
447 acpi_boot_init();
448#endif
449
05b3cbd8
RT
450 init_cpu_to_node();
451
1da177e4
LT
452 /*
453 * get boot-time SMP configuration:
454 */
455 if (smp_found_config)
456 get_smp_config();
457 init_apic_mappings();
3e35a0e5 458 ioapic_init_mappings();
1da177e4
LT
459
460 /*
fc986db4 461 * We trust e820 completely. No explicit ROM probing in memory.
04e1ba85 462 */
3def3d6d 463 e820_reserve_resources();
e8eff5ac 464 e820_mark_nosave_regions();
1da177e4 465
1da177e4 466 /* request I/O space for devices used on all i[345]86 PCs */
9d0ef4fd 467 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
1da177e4 468 request_resource(&ioport_resource, &standard_io_resources[i]);
1da177e4 469
a1e97782 470 e820_setup_gap();
1da177e4 471
1da177e4
LT
472#ifdef CONFIG_VT
473#if defined(CONFIG_VGA_CONSOLE)
5b83683f
HY
474 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
475 conswitchp = &vga_con;
1da177e4
LT
476#elif defined(CONFIG_DUMMY_CONSOLE)
477 conswitchp = &dummy_con;
478#endif
479#endif
480}
481
e6982c67 482static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
1da177e4
LT
483{
484 unsigned int *v;
485
ebfcaa96 486 if (c->extended_cpuid_level < 0x80000004)
1da177e4
LT
487 return 0;
488
489 v = (unsigned int *) c->x86_model_id;
490 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
491 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
492 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
493 c->x86_model_id[48] = 0;
494 return 1;
495}
496
497
e6982c67 498static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
1da177e4
LT
499{
500 unsigned int n, dummy, eax, ebx, ecx, edx;
501
ebfcaa96 502 n = c->extended_cpuid_level;
1da177e4
LT
503
504 if (n >= 0x80000005) {
505 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
04e1ba85
TG
506 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
507 "D cache %dK (%d bytes/line)\n",
508 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
509 c->x86_cache_size = (ecx>>24) + (edx>>24);
1da177e4
LT
510 /* On K8 L1 TLB is inclusive, so don't count it */
511 c->x86_tlbsize = 0;
512 }
513
514 if (n >= 0x80000006) {
515 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
516 ecx = cpuid_ecx(0x80000006);
517 c->x86_cache_size = ecx >> 16;
518 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
519
520 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
521 c->x86_cache_size, ecx & 0xFF);
522 }
1da177e4 523 if (n >= 0x80000008) {
04e1ba85 524 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
1da177e4
LT
525 c->x86_virt_bits = (eax >> 8) & 0xff;
526 c->x86_phys_bits = eax & 0xff;
527 }
528}
529
3f098c26 530#ifdef CONFIG_NUMA
08acb672 531static int __cpuinit nearby_node(int apicid)
3f098c26 532{
04e1ba85
TG
533 int i, node;
534
3f098c26 535 for (i = apicid - 1; i >= 0; i--) {
04e1ba85 536 node = apicid_to_node[i];
3f098c26
AK
537 if (node != NUMA_NO_NODE && node_online(node))
538 return node;
539 }
540 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
04e1ba85 541 node = apicid_to_node[i];
3f098c26
AK
542 if (node != NUMA_NO_NODE && node_online(node))
543 return node;
544 }
545 return first_node(node_online_map); /* Shouldn't happen */
546}
547#endif
548
63518644
AK
549/*
550 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
551 * Assumes number of cores is a power of two.
552 */
adb8daed 553static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
63518644
AK
554{
555#ifdef CONFIG_SMP
b41e2939 556 unsigned bits;
3f098c26 557#ifdef CONFIG_NUMA
f3fa8ebc 558 int cpu = smp_processor_id();
3f098c26 559 int node = 0;
60c1bc82 560 unsigned apicid = hard_smp_processor_id();
3f098c26 561#endif
a860b63c 562 bits = c->x86_coreid_bits;
b41e2939
AK
563
564 /* Low order bits define the core id (index of core in socket) */
f3fa8ebc 565 c->cpu_core_id = c->phys_proc_id & ((1 << bits)-1);
b41e2939 566 /* Convert the APIC ID into the socket ID */
a7062211 567 c->phys_proc_id = (c->apicid - boot_cpu_id) >> bits;
63518644
AK
568
569#ifdef CONFIG_NUMA
04e1ba85
TG
570 node = c->phys_proc_id;
571 if (apicid_to_node[apicid] != NUMA_NO_NODE)
572 node = apicid_to_node[apicid];
573 if (!node_online(node)) {
574 /* Two possibilities here:
575 - The CPU is missing memory and no node was created.
576 In that case try picking one from a nearby CPU
577 - The APIC IDs differ from the HyperTransport node IDs
578 which the K8 northbridge parsing fills in.
579 Assume they are all increased by a constant offset,
580 but in the same order as the HT nodeids.
581 If that doesn't result in a usable node fall back to the
582 path for the previous case. */
583
a7062211 584 int ht_nodeid = apicid - boot_cpu_id;
04e1ba85
TG
585
586 if (ht_nodeid >= 0 &&
587 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
588 node = apicid_to_node[ht_nodeid];
589 /* Pick a nearby node */
590 if (!node_online(node))
591 node = nearby_node(apicid);
592 }
69d81fcd 593 numa_set_node(cpu, node);
3f098c26 594
e42f9437 595 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
63518644 596#endif
63518644
AK
597#endif
598}
1da177e4 599
2b16a235 600static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
a860b63c
YL
601{
602#ifdef CONFIG_SMP
603 unsigned bits, ecx;
604
605 /* Multi core CPU? */
606 if (c->extended_cpuid_level < 0x80000008)
607 return;
608
609 ecx = cpuid_ecx(0x80000008);
610
611 c->x86_max_cores = (ecx & 0xff) + 1;
612
613 /* CPU telling us the core id bits shift? */
614 bits = (ecx >> 12) & 0xF;
615
616 /* Otherwise recompute */
617 if (bits == 0) {
618 while ((1 << bits) < c->x86_max_cores)
619 bits++;
620 }
621
622 c->x86_coreid_bits = bits;
623
624#endif
625}
626
fb79d22e
TG
627#define ENABLE_C1E_MASK 0x18000000
628#define CPUID_PROCESSOR_SIGNATURE 1
629#define CPUID_XFAM 0x0ff00000
630#define CPUID_XFAM_K8 0x00000000
631#define CPUID_XFAM_10H 0x00100000
632#define CPUID_XFAM_11H 0x00200000
633#define CPUID_XMOD 0x000f0000
634#define CPUID_XMOD_REV_F 0x00040000
635
636/* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
637static __cpuinit int amd_apic_timer_broken(void)
638{
04e1ba85
TG
639 u32 lo, hi, eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
640
fb79d22e
TG
641 switch (eax & CPUID_XFAM) {
642 case CPUID_XFAM_K8:
643 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
644 break;
645 case CPUID_XFAM_10H:
646 case CPUID_XFAM_11H:
647 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
648 if (lo & ENABLE_C1E_MASK)
649 return 1;
650 break;
651 default:
652 /* err on the side of caution */
653 return 1;
654 }
655 return 0;
656}
657
2b16a235
AK
658static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
659{
660 early_init_amd_mc(c);
661
662 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
663 if (c->x86_power & (1<<8))
664 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
665}
666
ed77504b 667static void __cpuinit init_amd(struct cpuinfo_x86 *c)
1da177e4 668{
7bcd3f34 669 unsigned level;
1da177e4 670
bc5e8fdf
LT
671#ifdef CONFIG_SMP
672 unsigned long value;
673
7d318d77
AK
674 /*
675 * Disable TLB flush filter by setting HWCR.FFDIS on K8
676 * bit 6 of msr C001_0015
04e1ba85 677 *
7d318d77
AK
678 * Errata 63 for SH-B3 steppings
679 * Errata 122 for all steppings (F+ have it disabled by default)
680 */
681 if (c->x86 == 15) {
682 rdmsrl(MSR_K8_HWCR, value);
683 value |= 1 << 6;
684 wrmsrl(MSR_K8_HWCR, value);
685 }
bc5e8fdf
LT
686#endif
687
1da177e4
LT
688 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
689 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
9716951e 690 clear_cpu_cap(c, 0*32+31);
04e1ba85 691
7bcd3f34
AK
692 /* On C+ stepping K8 rep microcode works well for copy/memset */
693 level = cpuid_eax(1);
04e1ba85
TG
694 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) ||
695 level >= 0x0f58))
53756d37 696 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
99741faa 697 if (c->x86 == 0x10 || c->x86 == 0x11)
53756d37 698 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
7bcd3f34 699
18bd057b
AK
700 /* Enable workaround for FXSAVE leak */
701 if (c->x86 >= 6)
53756d37 702 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
18bd057b 703
e42f9437
RS
704 level = get_model_name(c);
705 if (!level) {
04e1ba85 706 switch (c->x86) {
1da177e4
LT
707 case 15:
708 /* Should distinguish Models here, but this is only
709 a fallback anyways. */
710 strcpy(c->x86_model_id, "Hammer");
04e1ba85
TG
711 break;
712 }
713 }
1da177e4
LT
714 display_cacheinfo(c);
715
faee9a5d
AK
716 /* Multi core CPU? */
717 if (c->extended_cpuid_level >= 0x80000008)
63518644 718 amd_detect_cmp(c);
1da177e4 719
67cddd94
AK
720 if (c->extended_cpuid_level >= 0x80000006 &&
721 (cpuid_edx(0x80000006) & 0xf000))
722 num_cache_leaves = 4;
723 else
724 num_cache_leaves = 3;
2049336f 725
0bd8acd1 726 if (c->x86 == 0xf || c->x86 == 0x10 || c->x86 == 0x11)
53756d37 727 set_cpu_cap(c, X86_FEATURE_K8);
0bd8acd1 728
de421863
AK
729 /* MFENCE stops RDTSC speculation */
730 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
f039b754 731
fb79d22e
TG
732 if (amd_apic_timer_broken())
733 disable_apic_timer = 1;
1da177e4
LT
734}
735
1a53905a 736void __cpuinit detect_ht(struct cpuinfo_x86 *c)
1da177e4
LT
737{
738#ifdef CONFIG_SMP
04e1ba85
TG
739 u32 eax, ebx, ecx, edx;
740 int index_msb, core_bits;
94605eff
SS
741
742 cpuid(1, &eax, &ebx, &ecx, &edx);
743
94605eff 744
e42f9437 745 if (!cpu_has(c, X86_FEATURE_HT))
1da177e4 746 return;
04e1ba85 747 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
e42f9437 748 goto out;
1da177e4 749
1da177e4 750 smp_num_siblings = (ebx & 0xff0000) >> 16;
94605eff 751
1da177e4
LT
752 if (smp_num_siblings == 1) {
753 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
04e1ba85 754 } else if (smp_num_siblings > 1) {
94605eff 755
1da177e4 756 if (smp_num_siblings > NR_CPUS) {
04e1ba85
TG
757 printk(KERN_WARNING "CPU: Unsupported number of "
758 "siblings %d", smp_num_siblings);
1da177e4
LT
759 smp_num_siblings = 1;
760 return;
761 }
94605eff
SS
762
763 index_msb = get_count_order(smp_num_siblings);
f3fa8ebc 764 c->phys_proc_id = phys_pkg_id(index_msb);
3dd9d514 765
94605eff 766 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
3dd9d514 767
04e1ba85 768 index_msb = get_count_order(smp_num_siblings);
94605eff
SS
769
770 core_bits = get_count_order(c->x86_max_cores);
3dd9d514 771
f3fa8ebc 772 c->cpu_core_id = phys_pkg_id(index_msb) &
94605eff 773 ((1 << core_bits) - 1);
1da177e4 774 }
e42f9437
RS
775out:
776 if ((c->x86_max_cores * smp_num_siblings) > 1) {
04e1ba85
TG
777 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
778 c->phys_proc_id);
779 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
780 c->cpu_core_id);
e42f9437
RS
781 }
782
1da177e4
LT
783#endif
784}
785
3dd9d514
AK
786/*
787 * find out the number of processor cores on the die
788 */
e6982c67 789static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
3dd9d514 790{
2bbc419f 791 unsigned int eax, t;
3dd9d514
AK
792
793 if (c->cpuid_level < 4)
794 return 1;
795
2bbc419f 796 cpuid_count(4, 0, &eax, &t, &t, &t);
3dd9d514
AK
797
798 if (eax & 0x1f)
799 return ((eax >> 26) + 1);
800 else
801 return 1;
802}
803
04d733bd 804static void __cpuinit srat_detect_node(void)
df0cc26b
AK
805{
806#ifdef CONFIG_NUMA
ddea7be0 807 unsigned node;
df0cc26b 808 int cpu = smp_processor_id();
e42f9437 809 int apicid = hard_smp_processor_id();
df0cc26b
AK
810
811 /* Don't do the funky fallback heuristics the AMD version employs
812 for now. */
e42f9437 813 node = apicid_to_node[apicid];
475613b9 814 if (node == NUMA_NO_NODE || !node_online(node))
0d015324 815 node = first_node(node_online_map);
69d81fcd 816 numa_set_node(cpu, node);
df0cc26b 817
c31fbb1a 818 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
df0cc26b
AK
819#endif
820}
821
2b16a235
AK
822static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
823{
824 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
825 (c->x86 == 0x6 && c->x86_model >= 0x0e))
9716951e 826 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
2b16a235
AK
827}
828
e6982c67 829static void __cpuinit init_intel(struct cpuinfo_x86 *c)
1da177e4
LT
830{
831 /* Cache sizes */
832 unsigned n;
833
834 init_intel_cacheinfo(c);
04e1ba85 835 if (c->cpuid_level > 9) {
0080e667
VP
836 unsigned eax = cpuid_eax(10);
837 /* Check for version and the number of counters */
838 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
53756d37 839 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
0080e667
VP
840 }
841
36b2a8d5
SE
842 if (cpu_has_ds) {
843 unsigned int l1, l2;
844 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
ee58fad5 845 if (!(l1 & (1<<11)))
53756d37 846 set_cpu_cap(c, X86_FEATURE_BTS);
36b2a8d5 847 if (!(l1 & (1<<12)))
53756d37 848 set_cpu_cap(c, X86_FEATURE_PEBS);
36b2a8d5
SE
849 }
850
eee3af4a
MM
851
852 if (cpu_has_bts)
853 ds_init_intel(c);
854
ebfcaa96 855 n = c->extended_cpuid_level;
1da177e4
LT
856 if (n >= 0x80000008) {
857 unsigned eax = cpuid_eax(0x80000008);
858 c->x86_virt_bits = (eax >> 8) & 0xff;
859 c->x86_phys_bits = eax & 0xff;
af9c142d
SL
860 /* CPUID workaround for Intel 0F34 CPU */
861 if (c->x86_vendor == X86_VENDOR_INTEL &&
862 c->x86 == 0xF && c->x86_model == 0x3 &&
863 c->x86_mask == 0x4)
864 c->x86_phys_bits = 36;
1da177e4
LT
865 }
866
867 if (c->x86 == 15)
868 c->x86_cache_alignment = c->x86_clflush_size * 2;
27fbe5b2 869 if (c->x86 == 6)
53756d37 870 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
707fa8ed 871 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
04e1ba85 872 c->x86_max_cores = intel_num_cpu_cores(c);
df0cc26b
AK
873
874 srat_detect_node();
1da177e4
LT
875}
876
672289e9 877static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
1da177e4
LT
878{
879 char *v = c->x86_vendor_id;
880
881 if (!strcmp(v, "AuthenticAMD"))
882 c->x86_vendor = X86_VENDOR_AMD;
883 else if (!strcmp(v, "GenuineIntel"))
884 c->x86_vendor = X86_VENDOR_INTEL;
885 else
886 c->x86_vendor = X86_VENDOR_UNKNOWN;
887}
888
1da177e4
LT
889/* Do some early cpuid on the boot CPU to get some parameter that are
890 needed before check_bugs. Everything advanced is in identify_cpu
891 below. */
8c61b900 892static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
1da177e4 893{
a860b63c 894 u32 tfms, xlvl;
1da177e4
LT
895
896 c->loops_per_jiffy = loops_per_jiffy;
897 c->x86_cache_size = -1;
898 c->x86_vendor = X86_VENDOR_UNKNOWN;
899 c->x86_model = c->x86_mask = 0; /* So far unknown... */
900 c->x86_vendor_id[0] = '\0'; /* Unset */
901 c->x86_model_id[0] = '\0'; /* Unset */
902 c->x86_clflush_size = 64;
903 c->x86_cache_alignment = c->x86_clflush_size;
94605eff 904 c->x86_max_cores = 1;
a860b63c 905 c->x86_coreid_bits = 0;
ebfcaa96 906 c->extended_cpuid_level = 0;
1da177e4
LT
907 memset(&c->x86_capability, 0, sizeof c->x86_capability);
908
909 /* Get vendor name */
910 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
911 (unsigned int *)&c->x86_vendor_id[0],
912 (unsigned int *)&c->x86_vendor_id[8],
913 (unsigned int *)&c->x86_vendor_id[4]);
04e1ba85 914
1da177e4
LT
915 get_cpu_vendor(c);
916
917 /* Initialize the standard set of capabilities */
918 /* Note that the vendor-specific code below might override */
919
920 /* Intel-defined flags: level 0x00000001 */
921 if (c->cpuid_level >= 0x00000001) {
922 __u32 misc;
923 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
924 &c->x86_capability[0]);
925 c->x86 = (tfms >> 8) & 0xf;
926 c->x86_model = (tfms >> 4) & 0xf;
927 c->x86_mask = tfms & 0xf;
f5f786d0 928 if (c->x86 == 0xf)
1da177e4 929 c->x86 += (tfms >> 20) & 0xff;
f5f786d0 930 if (c->x86 >= 0x6)
1da177e4 931 c->x86_model += ((tfms >> 16) & 0xF) << 4;
9716951e 932 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
1da177e4 933 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1da177e4
LT
934 } else {
935 /* Have CPUID level 0 only - unheard of */
936 c->x86 = 4;
937 }
a158608b
AK
938
939#ifdef CONFIG_SMP
f3fa8ebc 940 c->phys_proc_id = (cpuid_ebx(1) >> 24) & 0xff;
a158608b 941#endif
1da177e4
LT
942 /* AMD-defined flags: level 0x80000001 */
943 xlvl = cpuid_eax(0x80000000);
ebfcaa96 944 c->extended_cpuid_level = xlvl;
1da177e4
LT
945 if ((xlvl & 0xffff0000) == 0x80000000) {
946 if (xlvl >= 0x80000001) {
947 c->x86_capability[1] = cpuid_edx(0x80000001);
5b7abc6f 948 c->x86_capability[6] = cpuid_ecx(0x80000001);
1da177e4
LT
949 }
950 if (xlvl >= 0x80000004)
951 get_model_name(c); /* Default name */
952 }
953
954 /* Transmeta-defined flags: level 0x80860001 */
955 xlvl = cpuid_eax(0x80860000);
956 if ((xlvl & 0xffff0000) == 0x80860000) {
957 /* Don't set x86_cpuid_level here for now to not confuse. */
958 if (xlvl >= 0x80860001)
959 c->x86_capability[2] = cpuid_edx(0x80860001);
960 }
961
9566e91d
AH
962 c->extended_cpuid_level = cpuid_eax(0x80000000);
963 if (c->extended_cpuid_level >= 0x80000007)
964 c->x86_power = cpuid_edx(0x80000007);
965
a860b63c
YL
966 switch (c->x86_vendor) {
967 case X86_VENDOR_AMD:
968 early_init_amd(c);
969 break;
71617bf1
YL
970 case X86_VENDOR_INTEL:
971 early_init_intel(c);
972 break;
a860b63c
YL
973 }
974
975}
976
977/*
978 * This does the hard work of actually picking apart the CPU stuff...
979 */
980void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
981{
982 int i;
983
984 early_identify_cpu(c);
985
1d67953f
VP
986 init_scattered_cpuid_features(c);
987
1e9f28fa
SS
988 c->apicid = phys_pkg_id(0);
989
1da177e4
LT
990 /*
991 * Vendor-specific initialization. In this section we
992 * canonicalize the feature flags, meaning if there are
993 * features a certain CPU supports which CPUID doesn't
994 * tell us, CPUID claiming incorrect flags, or other bugs,
995 * we handle them here.
996 *
997 * At the end of this section, c->x86_capability better
998 * indicate the features this CPU genuinely supports!
999 */
1000 switch (c->x86_vendor) {
1001 case X86_VENDOR_AMD:
1002 init_amd(c);
1003 break;
1004
1005 case X86_VENDOR_INTEL:
1006 init_intel(c);
1007 break;
1008
1009 case X86_VENDOR_UNKNOWN:
1010 default:
1011 display_cacheinfo(c);
1012 break;
1013 }
1014
04e1ba85 1015 detect_ht(c);
1da177e4
LT
1016
1017 /*
1018 * On SMP, boot_cpu_data holds the common feature set between
1019 * all CPUs; so make sure that we indicate which features are
1020 * common between the CPUs. The first time this routine gets
1021 * executed, c == &boot_cpu_data.
1022 */
1023 if (c != &boot_cpu_data) {
1024 /* AND the already accumulated flags with these */
04e1ba85 1025 for (i = 0; i < NCAPINTS; i++)
1da177e4
LT
1026 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1027 }
1028
7d851c8d
AK
1029 /* Clear all flags overriden by options */
1030 for (i = 0; i < NCAPINTS; i++)
12c247a6 1031 c->x86_capability[i] &= ~cleared_cpu_caps[i];
7d851c8d 1032
1da177e4
LT
1033#ifdef CONFIG_X86_MCE
1034 mcheck_init(c);
1035#endif
74ff305b
HS
1036 select_idle_routine(c);
1037
8bd99481 1038 if (c != &boot_cpu_data)
3b520b23 1039 mtrr_ap_init();
1da177e4 1040#ifdef CONFIG_NUMA
3019e8eb 1041 numa_add_cpu(smp_processor_id());
1da177e4 1042#endif
2b16a235 1043
1da177e4 1044}
1da177e4 1045
191679fd
AK
1046static __init int setup_noclflush(char *arg)
1047{
1048 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
1049 return 1;
1050}
1051__setup("noclflush", setup_noclflush);
1052
e6982c67 1053void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1da177e4
LT
1054{
1055 if (c->x86_model_id[0])
d8ff0bbf 1056 printk(KERN_CONT "%s", c->x86_model_id);
1da177e4 1057
04e1ba85
TG
1058 if (c->x86_mask || c->cpuid_level >= 0)
1059 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
1da177e4 1060 else
04e1ba85 1061 printk(KERN_CONT "\n");
1da177e4
LT
1062}
1063
ac72e788
AK
1064static __init int setup_disablecpuid(char *arg)
1065{
1066 int bit;
1067 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1068 setup_clear_cpu_cap(bit);
1069 else
1070 return 0;
1071 return 1;
1072}
1073__setup("clearcpuid=", setup_disablecpuid);