aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel/setup_64.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/kernel/setup_64.c')
-rw-r--r--arch/x86/kernel/setup_64.c1199
1 files changed, 0 insertions, 1199 deletions
diff --git a/arch/x86/kernel/setup_64.c b/arch/x86/kernel/setup_64.c
deleted file mode 100644
index 524b6850b2c0..000000000000
--- a/arch/x86/kernel/setup_64.c
+++ /dev/null
@@ -1,1199 +0,0 @@
1/*
2 * Copyright (C) 1995 Linus Torvalds
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>
18#include <linux/screen_info.h>
19#include <linux/ioport.h>
20#include <linux/delay.h>
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>
29#include <linux/crash_dump.h>
30#include <linux/root_dev.h>
31#include <linux/pci.h>
32#include <asm/pci-direct.h>
33#include <linux/efi.h>
34#include <linux/acpi.h>
35#include <linux/kallsyms.h>
36#include <linux/edd.h>
37#include <linux/iscsi_ibft.h>
38#include <linux/mmzone.h>
39#include <linux/kexec.h>
40#include <linux/cpufreq.h>
41#include <linux/dmi.h>
42#include <linux/dma-mapping.h>
43#include <linux/ctype.h>
44#include <linux/sort.h>
45#include <linux/uaccess.h>
46#include <linux/init_ohci1394_dma.h>
47#include <linux/kvm_para.h>
48
49#include <asm/mtrr.h>
50#include <asm/uaccess.h>
51#include <asm/system.h>
52#include <asm/vsyscall.h>
53#include <asm/io.h>
54#include <asm/smp.h>
55#include <asm/msr.h>
56#include <asm/desc.h>
57#include <video/edid.h>
58#include <asm/e820.h>
59#include <asm/dma.h>
60#include <asm/gart.h>
61#include <asm/mpspec.h>
62#include <asm/mmu_context.h>
63#include <asm/proto.h>
64#include <asm/setup.h>
65#include <asm/numa.h>
66#include <asm/sections.h>
67#include <asm/dmi.h>
68#include <asm/cacheflush.h>
69#include <asm/mce.h>
70#include <asm/ds.h>
71#include <asm/topology.h>
72#include <asm/trampoline.h>
73#include <asm/pat.h>
74
75#include <mach_apic.h>
76#ifdef CONFIG_PARAVIRT
77#include <asm/paravirt.h>
78#else
79#define ARCH_SETUP
80#endif
81
82/*
83 * Machine setup..
84 */
85
86struct cpuinfo_x86 boot_cpu_data __read_mostly;
87EXPORT_SYMBOL(boot_cpu_data);
88
89__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
90
91unsigned long mmu_cr4_features;
92
93/* Boot loader ID as an integer, for the benefit of proc_dointvec */
94int bootloader_type;
95
96unsigned long saved_video_mode;
97
98int force_mwait __cpuinitdata;
99
100/*
101 * Early DMI memory
102 */
103int dmi_alloc_index;
104char dmi_alloc_data[DMI_MAX_DATA];
105
106/*
107 * Setup options
108 */
109struct screen_info screen_info;
110EXPORT_SYMBOL(screen_info);
111struct sys_desc_table_struct {
112 unsigned short length;
113 unsigned char table[0];
114};
115
116struct edid_info edid_info;
117EXPORT_SYMBOL_GPL(edid_info);
118
119extern int root_mountflags;
120
121char __initdata command_line[COMMAND_LINE_SIZE];
122
123static struct resource standard_io_resources[] = {
124 { .name = "dma1", .start = 0x00, .end = 0x1f,
125 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
126 { .name = "pic1", .start = 0x20, .end = 0x21,
127 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
128 { .name = "timer0", .start = 0x40, .end = 0x43,
129 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
130 { .name = "timer1", .start = 0x50, .end = 0x53,
131 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
132 { .name = "keyboard", .start = 0x60, .end = 0x60,
133 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
134 { .name = "keyboard", .start = 0x64, .end = 0x64,
135 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
136 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
137 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
138 { .name = "pic2", .start = 0xa0, .end = 0xa1,
139 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
140 { .name = "dma2", .start = 0xc0, .end = 0xdf,
141 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
142 { .name = "fpu", .start = 0xf0, .end = 0xff,
143 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
144};
145
146#define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
147
148static struct resource data_resource = {
149 .name = "Kernel data",
150 .start = 0,
151 .end = 0,
152 .flags = IORESOURCE_RAM,
153};
154static struct resource code_resource = {
155 .name = "Kernel code",
156 .start = 0,
157 .end = 0,
158 .flags = IORESOURCE_RAM,
159};
160static struct resource bss_resource = {
161 .name = "Kernel bss",
162 .start = 0,
163 .end = 0,
164 .flags = IORESOURCE_RAM,
165};
166
167static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
168
169#ifdef CONFIG_PROC_VMCORE
170/* elfcorehdr= specifies the location of elf core header
171 * stored by the crashed kernel. This option will be passed
172 * by kexec loader to the capture kernel.
173 */
174static int __init setup_elfcorehdr(char *arg)
175{
176 char *end;
177 if (!arg)
178 return -EINVAL;
179 elfcorehdr_addr = memparse(arg, &end);
180 return end > arg ? 0 : -EINVAL;
181}
182early_param("elfcorehdr", setup_elfcorehdr);
183#endif
184
185#ifndef CONFIG_NUMA
186static void __init
187contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
188{
189 unsigned long bootmap_size, bootmap;
190
191 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
192 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
193 PAGE_SIZE);
194 if (bootmap == -1L)
195 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
196 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
197 e820_register_active_regions(0, start_pfn, end_pfn);
198 free_bootmem_with_active_regions(0, end_pfn);
199 early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
200 reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
201}
202#endif
203
204#if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
205struct edd edd;
206#ifdef CONFIG_EDD_MODULE
207EXPORT_SYMBOL(edd);
208#endif
209/**
210 * copy_edd() - Copy the BIOS EDD information
211 * from boot_params into a safe place.
212 *
213 */
214static inline void copy_edd(void)
215{
216 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
217 sizeof(edd.mbr_signature));
218 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
219 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
220 edd.edd_info_nr = boot_params.eddbuf_entries;
221}
222#else
223static inline void copy_edd(void)
224{
225}
226#endif
227
228#ifdef CONFIG_KEXEC
229static void __init reserve_crashkernel(void)
230{
231 unsigned long long total_mem;
232 unsigned long long crash_size, crash_base;
233 int ret;
234
235 total_mem = ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
236
237 ret = parse_crashkernel(boot_command_line, total_mem,
238 &crash_size, &crash_base);
239 if (ret == 0 && crash_size) {
240 if (crash_base <= 0) {
241 printk(KERN_INFO "crashkernel reservation failed - "
242 "you have to specify a base address\n");
243 return;
244 }
245
246 if (reserve_bootmem(crash_base, crash_size,
247 BOOTMEM_EXCLUSIVE) < 0) {
248 printk(KERN_INFO "crashkernel reservation failed - "
249 "memory is in use\n");
250 return;
251 }
252
253 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
254 "for crashkernel (System RAM: %ldMB)\n",
255 (unsigned long)(crash_size >> 20),
256 (unsigned long)(crash_base >> 20),
257 (unsigned long)(total_mem >> 20));
258 crashk_res.start = crash_base;
259 crashk_res.end = crash_base + crash_size - 1;
260 insert_resource(&iomem_resource, &crashk_res);
261 }
262}
263#else
264static inline void __init reserve_crashkernel(void)
265{}
266#endif
267
268/* Overridden in paravirt.c if CONFIG_PARAVIRT */
269void __attribute__((weak)) __init memory_setup(void)
270{
271 machine_specific_memory_setup();
272}
273
274static void __init parse_setup_data(void)
275{
276 struct setup_data *data;
277 unsigned long pa_data;
278
279 if (boot_params.hdr.version < 0x0209)
280 return;
281 pa_data = boot_params.hdr.setup_data;
282 while (pa_data) {
283 data = early_ioremap(pa_data, PAGE_SIZE);
284 switch (data->type) {
285 default:
286 break;
287 }
288#ifndef CONFIG_DEBUG_BOOT_PARAMS
289 free_early(pa_data, pa_data+sizeof(*data)+data->len);
290#endif
291 pa_data = data->next;
292 early_iounmap(data, PAGE_SIZE);
293 }
294}
295
296#ifdef CONFIG_PCI_MMCONFIG
297extern void __cpuinit fam10h_check_enable_mmcfg(void);
298extern void __init check_enable_amd_mmconf_dmi(void);
299#else
300void __cpuinit fam10h_check_enable_mmcfg(void)
301{
302}
303void __init check_enable_amd_mmconf_dmi(void)
304{
305}
306#endif
307
308/*
309 * setup_arch - architecture-specific boot-time initializations
310 *
311 * Note: On x86_64, fixmaps are ready for use even before this is called.
312 */
313void __init setup_arch(char **cmdline_p)
314{
315 unsigned i;
316
317 printk(KERN_INFO "Command line: %s\n", boot_command_line);
318
319 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
320 screen_info = boot_params.screen_info;
321 edid_info = boot_params.edid_info;
322 saved_video_mode = boot_params.hdr.vid_mode;
323 bootloader_type = boot_params.hdr.type_of_loader;
324
325#ifdef CONFIG_BLK_DEV_RAM
326 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
327 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
328 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
329#endif
330#ifdef CONFIG_EFI
331 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
332 "EL64", 4))
333 efi_enabled = 1;
334#endif
335
336 ARCH_SETUP
337
338 memory_setup();
339 copy_edd();
340
341 if (!boot_params.hdr.root_flags)
342 root_mountflags &= ~MS_RDONLY;
343 init_mm.start_code = (unsigned long) &_text;
344 init_mm.end_code = (unsigned long) &_etext;
345 init_mm.end_data = (unsigned long) &_edata;
346 init_mm.brk = (unsigned long) &_end;
347
348 code_resource.start = virt_to_phys(&_text);
349 code_resource.end = virt_to_phys(&_etext)-1;
350 data_resource.start = virt_to_phys(&_etext);
351 data_resource.end = virt_to_phys(&_edata)-1;
352 bss_resource.start = virt_to_phys(&__bss_start);
353 bss_resource.end = virt_to_phys(&__bss_stop)-1;
354
355 early_identify_cpu(&boot_cpu_data);
356
357 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
358 *cmdline_p = command_line;
359
360 parse_setup_data();
361
362 parse_early_param();
363
364#ifdef CONFIG_PCI
365 if (pci_early_dump_regs)
366 early_dump_pci_devices();
367#endif
368
369#ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
370 if (init_ohci1394_dma_early)
371 init_ohci1394_dma_on_all_controllers();
372#endif
373
374 finish_e820_parsing();
375
376 /* after parse_early_param, so could debug it */
377 insert_resource(&iomem_resource, &code_resource);
378 insert_resource(&iomem_resource, &data_resource);
379 insert_resource(&iomem_resource, &bss_resource);
380
381 early_gart_iommu_check();
382
383 e820_register_active_regions(0, 0, -1UL);
384 /*
385 * partially used pages are not usable - thus
386 * we are rounding upwards:
387 */
388 end_pfn = e820_end_of_ram();
389 /* update e820 for memory not covered by WB MTRRs */
390 mtrr_bp_init();
391 if (mtrr_trim_uncached_memory(end_pfn)) {
392 e820_register_active_regions(0, 0, -1UL);
393 end_pfn = e820_end_of_ram();
394 }
395
396 num_physpages = end_pfn;
397
398 check_efer();
399
400 max_pfn_mapped = init_memory_mapping(0, (max_pfn_mapped << PAGE_SHIFT));
401 if (efi_enabled)
402 efi_init();
403
404 vsmp_init();
405
406 dmi_scan_machine();
407
408 io_delay_init();
409
410#ifdef CONFIG_KVM_CLOCK
411 kvmclock_init();
412#endif
413
414#ifdef CONFIG_SMP
415 /* setup to use the early static init tables during kernel startup */
416 x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
417 x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
418#ifdef CONFIG_NUMA
419 x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
420#endif
421#endif
422
423#ifdef CONFIG_ACPI
424 /*
425 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
426 * Call this early for SRAT node setup.
427 */
428 acpi_boot_table_init();
429#endif
430
431 /* How many end-of-memory variables you have, grandma! */
432 max_low_pfn = end_pfn;
433 max_pfn = end_pfn;
434 high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
435
436 /* Remove active ranges so rediscovery with NUMA-awareness happens */
437 remove_all_active_ranges();
438
439#ifdef CONFIG_ACPI_NUMA
440 /*
441 * Parse SRAT to discover nodes.
442 */
443 acpi_numa_init();
444#endif
445
446#ifdef CONFIG_NUMA
447 numa_initmem_init(0, end_pfn);
448#else
449 contig_initmem_init(0, end_pfn);
450#endif
451
452 dma32_reserve_bootmem();
453
454#ifdef CONFIG_ACPI_SLEEP
455 /*
456 * Reserve low memory region for sleep support.
457 */
458 acpi_reserve_bootmem();
459#endif
460
461 if (efi_enabled)
462 efi_reserve_bootmem();
463
464 /*
465 * Find and reserve possible boot-time SMP configuration:
466 */
467 find_smp_config();
468#ifdef CONFIG_BLK_DEV_INITRD
469 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
470 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
471 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
472 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
473 unsigned long end_of_mem = end_pfn << PAGE_SHIFT;
474
475 if (ramdisk_end <= end_of_mem) {
476 /*
477 * don't need to reserve again, already reserved early
478 * in x86_64_start_kernel, and early_res_to_bootmem
479 * convert that to reserved in bootmem
480 */
481 initrd_start = ramdisk_image + PAGE_OFFSET;
482 initrd_end = initrd_start+ramdisk_size;
483 } else {
484 free_bootmem(ramdisk_image, ramdisk_size);
485 printk(KERN_ERR "initrd extends beyond end of memory "
486 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
487 ramdisk_end, end_of_mem);
488 initrd_start = 0;
489 }
490 }
491#endif
492 reserve_crashkernel();
493
494 reserve_ibft_region();
495
496 paging_init();
497 map_vsyscall();
498
499 early_quirks();
500
501#ifdef CONFIG_ACPI
502 /*
503 * Read APIC and some other early information from ACPI tables.
504 */
505 acpi_boot_init();
506#endif
507
508 init_cpu_to_node();
509
510 /*
511 * get boot-time SMP configuration:
512 */
513 if (smp_found_config)
514 get_smp_config();
515 init_apic_mappings();
516 ioapic_init_mappings();
517
518 kvm_guest_init();
519
520 /*
521 * We trust e820 completely. No explicit ROM probing in memory.
522 */
523 e820_reserve_resources();
524 e820_mark_nosave_regions();
525
526 /* request I/O space for devices used on all i[345]86 PCs */
527 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
528 request_resource(&ioport_resource, &standard_io_resources[i]);
529
530 e820_setup_gap();
531
532#ifdef CONFIG_VT
533#if defined(CONFIG_VGA_CONSOLE)
534 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
535 conswitchp = &vga_con;
536#elif defined(CONFIG_DUMMY_CONSOLE)
537 conswitchp = &dummy_con;
538#endif
539#endif
540
541 /* do this before identify_cpu for boot cpu */
542 check_enable_amd_mmconf_dmi();
543}
544
545static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
546{
547 unsigned int *v;
548
549 if (c->extended_cpuid_level < 0x80000004)
550 return 0;
551
552 v = (unsigned int *) c->x86_model_id;
553 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
554 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
555 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
556 c->x86_model_id[48] = 0;
557 return 1;
558}
559
560
561static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
562{
563 unsigned int n, dummy, eax, ebx, ecx, edx;
564
565 n = c->extended_cpuid_level;
566
567 if (n >= 0x80000005) {
568 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
569 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
570 "D cache %dK (%d bytes/line)\n",
571 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
572 c->x86_cache_size = (ecx>>24) + (edx>>24);
573 /* On K8 L1 TLB is inclusive, so don't count it */
574 c->x86_tlbsize = 0;
575 }
576
577 if (n >= 0x80000006) {
578 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
579 ecx = cpuid_ecx(0x80000006);
580 c->x86_cache_size = ecx >> 16;
581 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
582
583 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
584 c->x86_cache_size, ecx & 0xFF);
585 }
586 if (n >= 0x80000008) {
587 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
588 c->x86_virt_bits = (eax >> 8) & 0xff;
589 c->x86_phys_bits = eax & 0xff;
590 }
591}
592
593#ifdef CONFIG_NUMA
594static int __cpuinit nearby_node(int apicid)
595{
596 int i, node;
597
598 for (i = apicid - 1; i >= 0; i--) {
599 node = apicid_to_node[i];
600 if (node != NUMA_NO_NODE && node_online(node))
601 return node;
602 }
603 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
604 node = apicid_to_node[i];
605 if (node != NUMA_NO_NODE && node_online(node))
606 return node;
607 }
608 return first_node(node_online_map); /* Shouldn't happen */
609}
610#endif
611
612/*
613 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
614 * Assumes number of cores is a power of two.
615 */
616static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
617{
618#ifdef CONFIG_SMP
619 unsigned bits;
620#ifdef CONFIG_NUMA
621 int cpu = smp_processor_id();
622 int node = 0;
623 unsigned apicid = hard_smp_processor_id();
624#endif
625 bits = c->x86_coreid_bits;
626
627 /* Low order bits define the core id (index of core in socket) */
628 c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
629 /* Convert the initial APIC ID into the socket ID */
630 c->phys_proc_id = c->initial_apicid >> bits;
631
632#ifdef CONFIG_NUMA
633 node = c->phys_proc_id;
634 if (apicid_to_node[apicid] != NUMA_NO_NODE)
635 node = apicid_to_node[apicid];
636 if (!node_online(node)) {
637 /* Two possibilities here:
638 - The CPU is missing memory and no node was created.
639 In that case try picking one from a nearby CPU
640 - The APIC IDs differ from the HyperTransport node IDs
641 which the K8 northbridge parsing fills in.
642 Assume they are all increased by a constant offset,
643 but in the same order as the HT nodeids.
644 If that doesn't result in a usable node fall back to the
645 path for the previous case. */
646
647 int ht_nodeid = c->initial_apicid;
648
649 if (ht_nodeid >= 0 &&
650 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
651 node = apicid_to_node[ht_nodeid];
652 /* Pick a nearby node */
653 if (!node_online(node))
654 node = nearby_node(apicid);
655 }
656 numa_set_node(cpu, node);
657
658 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
659#endif
660#endif
661}
662
663static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
664{
665#ifdef CONFIG_SMP
666 unsigned bits, ecx;
667
668 /* Multi core CPU? */
669 if (c->extended_cpuid_level < 0x80000008)
670 return;
671
672 ecx = cpuid_ecx(0x80000008);
673
674 c->x86_max_cores = (ecx & 0xff) + 1;
675
676 /* CPU telling us the core id bits shift? */
677 bits = (ecx >> 12) & 0xF;
678
679 /* Otherwise recompute */
680 if (bits == 0) {
681 while ((1 << bits) < c->x86_max_cores)
682 bits++;
683 }
684
685 c->x86_coreid_bits = bits;
686
687#endif
688}
689
690#define ENABLE_C1E_MASK 0x18000000
691#define CPUID_PROCESSOR_SIGNATURE 1
692#define CPUID_XFAM 0x0ff00000
693#define CPUID_XFAM_K8 0x00000000
694#define CPUID_XFAM_10H 0x00100000
695#define CPUID_XFAM_11H 0x00200000
696#define CPUID_XMOD 0x000f0000
697#define CPUID_XMOD_REV_F 0x00040000
698
699/* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
700static __cpuinit int amd_apic_timer_broken(void)
701{
702 u32 lo, hi, eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
703
704 switch (eax & CPUID_XFAM) {
705 case CPUID_XFAM_K8:
706 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
707 break;
708 case CPUID_XFAM_10H:
709 case CPUID_XFAM_11H:
710 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
711 if (lo & ENABLE_C1E_MASK)
712 return 1;
713 break;
714 default:
715 /* err on the side of caution */
716 return 1;
717 }
718 return 0;
719}
720
721static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
722{
723 early_init_amd_mc(c);
724
725 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
726 if (c->x86_power & (1<<8))
727 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
728}
729
730static void __cpuinit init_amd(struct cpuinfo_x86 *c)
731{
732 unsigned level;
733
734#ifdef CONFIG_SMP
735 unsigned long value;
736
737 /*
738 * Disable TLB flush filter by setting HWCR.FFDIS on K8
739 * bit 6 of msr C001_0015
740 *
741 * Errata 63 for SH-B3 steppings
742 * Errata 122 for all steppings (F+ have it disabled by default)
743 */
744 if (c->x86 == 15) {
745 rdmsrl(MSR_K8_HWCR, value);
746 value |= 1 << 6;
747 wrmsrl(MSR_K8_HWCR, value);
748 }
749#endif
750
751 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
752 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
753 clear_cpu_cap(c, 0*32+31);
754
755 /* On C+ stepping K8 rep microcode works well for copy/memset */
756 level = cpuid_eax(1);
757 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) ||
758 level >= 0x0f58))
759 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
760 if (c->x86 == 0x10 || c->x86 == 0x11)
761 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
762
763 /* Enable workaround for FXSAVE leak */
764 if (c->x86 >= 6)
765 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
766
767 level = get_model_name(c);
768 if (!level) {
769 switch (c->x86) {
770 case 15:
771 /* Should distinguish Models here, but this is only
772 a fallback anyways. */
773 strcpy(c->x86_model_id, "Hammer");
774 break;
775 }
776 }
777 display_cacheinfo(c);
778
779 /* Multi core CPU? */
780 if (c->extended_cpuid_level >= 0x80000008)
781 amd_detect_cmp(c);
782
783 if (c->extended_cpuid_level >= 0x80000006 &&
784 (cpuid_edx(0x80000006) & 0xf000))
785 num_cache_leaves = 4;
786 else
787 num_cache_leaves = 3;
788
789 if (c->x86 == 0xf || c->x86 == 0x10 || c->x86 == 0x11)
790 set_cpu_cap(c, X86_FEATURE_K8);
791
792 /* MFENCE stops RDTSC speculation */
793 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
794
795 if (c->x86 == 0x10)
796 fam10h_check_enable_mmcfg();
797
798 if (amd_apic_timer_broken())
799 disable_apic_timer = 1;
800
801 if (c == &boot_cpu_data && c->x86 >= 0xf && c->x86 <= 0x11) {
802 unsigned long long tseg;
803
804 /*
805 * Split up direct mapping around the TSEG SMM area.
806 * Don't do it for gbpages because there seems very little
807 * benefit in doing so.
808 */
809 if (!rdmsrl_safe(MSR_K8_TSEG_ADDR, &tseg) &&
810 (tseg >> PMD_SHIFT) < (max_pfn_mapped >> (PMD_SHIFT-PAGE_SHIFT)))
811 set_memory_4k((unsigned long)__va(tseg), 1);
812 }
813}
814
815void __cpuinit detect_ht(struct cpuinfo_x86 *c)
816{
817#ifdef CONFIG_SMP
818 u32 eax, ebx, ecx, edx;
819 int index_msb, core_bits;
820
821 cpuid(1, &eax, &ebx, &ecx, &edx);
822
823
824 if (!cpu_has(c, X86_FEATURE_HT))
825 return;
826 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
827 goto out;
828
829 smp_num_siblings = (ebx & 0xff0000) >> 16;
830
831 if (smp_num_siblings == 1) {
832 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
833 } else if (smp_num_siblings > 1) {
834
835 if (smp_num_siblings > NR_CPUS) {
836 printk(KERN_WARNING "CPU: Unsupported number of "
837 "siblings %d", smp_num_siblings);
838 smp_num_siblings = 1;
839 return;
840 }
841
842 index_msb = get_count_order(smp_num_siblings);
843 c->phys_proc_id = phys_pkg_id(index_msb);
844
845 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
846
847 index_msb = get_count_order(smp_num_siblings);
848
849 core_bits = get_count_order(c->x86_max_cores);
850
851 c->cpu_core_id = phys_pkg_id(index_msb) &
852 ((1 << core_bits) - 1);
853 }
854out:
855 if ((c->x86_max_cores * smp_num_siblings) > 1) {
856 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
857 c->phys_proc_id);
858 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
859 c->cpu_core_id);
860 }
861
862#endif
863}
864
865/*
866 * find out the number of processor cores on the die
867 */
868static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
869{
870 unsigned int eax, t;
871
872 if (c->cpuid_level < 4)
873 return 1;
874
875 cpuid_count(4, 0, &eax, &t, &t, &t);
876
877 if (eax & 0x1f)
878 return ((eax >> 26) + 1);
879 else
880 return 1;
881}
882
883static void __cpuinit srat_detect_node(void)
884{
885#ifdef CONFIG_NUMA
886 unsigned node;
887 int cpu = smp_processor_id();
888 int apicid = hard_smp_processor_id();
889
890 /* Don't do the funky fallback heuristics the AMD version employs
891 for now. */
892 node = apicid_to_node[apicid];
893 if (node == NUMA_NO_NODE || !node_online(node))
894 node = first_node(node_online_map);
895 numa_set_node(cpu, node);
896
897 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
898#endif
899}
900
901static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
902{
903 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
904 (c->x86 == 0x6 && c->x86_model >= 0x0e))
905 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
906}
907
908static void __cpuinit init_intel(struct cpuinfo_x86 *c)
909{
910 /* Cache sizes */
911 unsigned n;
912
913 init_intel_cacheinfo(c);
914 if (c->cpuid_level > 9) {
915 unsigned eax = cpuid_eax(10);
916 /* Check for version and the number of counters */
917 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
918 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
919 }
920
921 if (cpu_has_ds) {
922 unsigned int l1, l2;
923 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
924 if (!(l1 & (1<<11)))
925 set_cpu_cap(c, X86_FEATURE_BTS);
926 if (!(l1 & (1<<12)))
927 set_cpu_cap(c, X86_FEATURE_PEBS);
928 }
929
930
931 if (cpu_has_bts)
932 ds_init_intel(c);
933
934 n = c->extended_cpuid_level;
935 if (n >= 0x80000008) {
936 unsigned eax = cpuid_eax(0x80000008);
937 c->x86_virt_bits = (eax >> 8) & 0xff;
938 c->x86_phys_bits = eax & 0xff;
939 /* CPUID workaround for Intel 0F34 CPU */
940 if (c->x86_vendor == X86_VENDOR_INTEL &&
941 c->x86 == 0xF && c->x86_model == 0x3 &&
942 c->x86_mask == 0x4)
943 c->x86_phys_bits = 36;
944 }
945
946 if (c->x86 == 15)
947 c->x86_cache_alignment = c->x86_clflush_size * 2;
948 if (c->x86 == 6)
949 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
950 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
951 c->x86_max_cores = intel_num_cpu_cores(c);
952
953 srat_detect_node();
954}
955
956static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
957{
958 if (c->x86 == 0x6 && c->x86_model >= 0xf)
959 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
960}
961
962static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
963{
964 /* Cache sizes */
965 unsigned n;
966
967 n = c->extended_cpuid_level;
968 if (n >= 0x80000008) {
969 unsigned eax = cpuid_eax(0x80000008);
970 c->x86_virt_bits = (eax >> 8) & 0xff;
971 c->x86_phys_bits = eax & 0xff;
972 }
973
974 if (c->x86 == 0x6 && c->x86_model >= 0xf) {
975 c->x86_cache_alignment = c->x86_clflush_size * 2;
976 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
977 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
978 }
979 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
980}
981
982static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
983{
984 char *v = c->x86_vendor_id;
985
986 if (!strcmp(v, "AuthenticAMD"))
987 c->x86_vendor = X86_VENDOR_AMD;
988 else if (!strcmp(v, "GenuineIntel"))
989 c->x86_vendor = X86_VENDOR_INTEL;
990 else if (!strcmp(v, "CentaurHauls"))
991 c->x86_vendor = X86_VENDOR_CENTAUR;
992 else
993 c->x86_vendor = X86_VENDOR_UNKNOWN;
994}
995
996/* Do some early cpuid on the boot CPU to get some parameter that are
997 needed before check_bugs. Everything advanced is in identify_cpu
998 below. */
999static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
1000{
1001 u32 tfms, xlvl;
1002
1003 c->loops_per_jiffy = loops_per_jiffy;
1004 c->x86_cache_size = -1;
1005 c->x86_vendor = X86_VENDOR_UNKNOWN;
1006 c->x86_model = c->x86_mask = 0; /* So far unknown... */
1007 c->x86_vendor_id[0] = '\0'; /* Unset */
1008 c->x86_model_id[0] = '\0'; /* Unset */
1009 c->x86_clflush_size = 64;
1010 c->x86_cache_alignment = c->x86_clflush_size;
1011 c->x86_max_cores = 1;
1012 c->x86_coreid_bits = 0;
1013 c->extended_cpuid_level = 0;
1014 memset(&c->x86_capability, 0, sizeof c->x86_capability);
1015
1016 /* Get vendor name */
1017 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
1018 (unsigned int *)&c->x86_vendor_id[0],
1019 (unsigned int *)&c->x86_vendor_id[8],
1020 (unsigned int *)&c->x86_vendor_id[4]);
1021
1022 get_cpu_vendor(c);
1023
1024 /* Initialize the standard set of capabilities */
1025 /* Note that the vendor-specific code below might override */
1026
1027 /* Intel-defined flags: level 0x00000001 */
1028 if (c->cpuid_level >= 0x00000001) {
1029 __u32 misc;
1030 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
1031 &c->x86_capability[0]);
1032 c->x86 = (tfms >> 8) & 0xf;
1033 c->x86_model = (tfms >> 4) & 0xf;
1034 c->x86_mask = tfms & 0xf;
1035 if (c->x86 == 0xf)
1036 c->x86 += (tfms >> 20) & 0xff;
1037 if (c->x86 >= 0x6)
1038 c->x86_model += ((tfms >> 16) & 0xF) << 4;
1039 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
1040 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1041 } else {
1042 /* Have CPUID level 0 only - unheard of */
1043 c->x86 = 4;
1044 }
1045
1046 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
1047#ifdef CONFIG_SMP
1048 c->phys_proc_id = c->initial_apicid;
1049#endif
1050 /* AMD-defined flags: level 0x80000001 */
1051 xlvl = cpuid_eax(0x80000000);
1052 c->extended_cpuid_level = xlvl;
1053 if ((xlvl & 0xffff0000) == 0x80000000) {
1054 if (xlvl >= 0x80000001) {
1055 c->x86_capability[1] = cpuid_edx(0x80000001);
1056 c->x86_capability[6] = cpuid_ecx(0x80000001);
1057 }
1058 if (xlvl >= 0x80000004)
1059 get_model_name(c); /* Default name */
1060 }
1061
1062 /* Transmeta-defined flags: level 0x80860001 */
1063 xlvl = cpuid_eax(0x80860000);
1064 if ((xlvl & 0xffff0000) == 0x80860000) {
1065 /* Don't set x86_cpuid_level here for now to not confuse. */
1066 if (xlvl >= 0x80860001)
1067 c->x86_capability[2] = cpuid_edx(0x80860001);
1068 }
1069
1070 c->extended_cpuid_level = cpuid_eax(0x80000000);
1071 if (c->extended_cpuid_level >= 0x80000007)
1072 c->x86_power = cpuid_edx(0x80000007);
1073
1074 switch (c->x86_vendor) {
1075 case X86_VENDOR_AMD:
1076 early_init_amd(c);
1077 break;
1078 case X86_VENDOR_INTEL:
1079 early_init_intel(c);
1080 break;
1081 case X86_VENDOR_CENTAUR:
1082 early_init_centaur(c);
1083 break;
1084 }
1085
1086 validate_pat_support(c);
1087}
1088
1089/*
1090 * This does the hard work of actually picking apart the CPU stuff...
1091 */
1092void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1093{
1094 int i;
1095
1096 early_identify_cpu(c);
1097
1098 init_scattered_cpuid_features(c);
1099
1100 c->apicid = phys_pkg_id(0);
1101
1102 /*
1103 * Vendor-specific initialization. In this section we
1104 * canonicalize the feature flags, meaning if there are
1105 * features a certain CPU supports which CPUID doesn't
1106 * tell us, CPUID claiming incorrect flags, or other bugs,
1107 * we handle them here.
1108 *
1109 * At the end of this section, c->x86_capability better
1110 * indicate the features this CPU genuinely supports!
1111 */
1112 switch (c->x86_vendor) {
1113 case X86_VENDOR_AMD:
1114 init_amd(c);
1115 break;
1116
1117 case X86_VENDOR_INTEL:
1118 init_intel(c);
1119 break;
1120
1121 case X86_VENDOR_CENTAUR:
1122 init_centaur(c);
1123 break;
1124
1125 case X86_VENDOR_UNKNOWN:
1126 default:
1127 display_cacheinfo(c);
1128 break;
1129 }
1130
1131 detect_ht(c);
1132
1133 /*
1134 * On SMP, boot_cpu_data holds the common feature set between
1135 * all CPUs; so make sure that we indicate which features are
1136 * common between the CPUs. The first time this routine gets
1137 * executed, c == &boot_cpu_data.
1138 */
1139 if (c != &boot_cpu_data) {
1140 /* AND the already accumulated flags with these */
1141 for (i = 0; i < NCAPINTS; i++)
1142 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1143 }
1144
1145 /* Clear all flags overriden by options */
1146 for (i = 0; i < NCAPINTS; i++)
1147 c->x86_capability[i] &= ~cleared_cpu_caps[i];
1148
1149#ifdef CONFIG_X86_MCE
1150 mcheck_init(c);
1151#endif
1152 select_idle_routine(c);
1153
1154#ifdef CONFIG_NUMA
1155 numa_add_cpu(smp_processor_id());
1156#endif
1157
1158}
1159
1160void __cpuinit identify_boot_cpu(void)
1161{
1162 identify_cpu(&boot_cpu_data);
1163}
1164
1165void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
1166{
1167 BUG_ON(c == &boot_cpu_data);
1168 identify_cpu(c);
1169 mtrr_ap_init();
1170}
1171
1172static __init int setup_noclflush(char *arg)
1173{
1174 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
1175 return 1;
1176}
1177__setup("noclflush", setup_noclflush);
1178
1179void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1180{
1181 if (c->x86_model_id[0])
1182 printk(KERN_CONT "%s", c->x86_model_id);
1183
1184 if (c->x86_mask || c->cpuid_level >= 0)
1185 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
1186 else
1187 printk(KERN_CONT "\n");
1188}
1189
1190static __init int setup_disablecpuid(char *arg)
1191{
1192 int bit;
1193 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1194 setup_clear_cpu_cap(bit);
1195 else
1196 return 0;
1197 return 1;
1198}
1199__setup("clearcpuid=", setup_disablecpuid);