]>
Commit | Line | Data |
---|---|---|
1 | #include <linux/bootmem.h> | |
2 | #include <linux/linkage.h> | |
3 | #include <linux/bitops.h> | |
4 | #include <linux/kernel.h> | |
5 | #include <linux/module.h> | |
6 | #include <linux/percpu.h> | |
7 | #include <linux/string.h> | |
8 | #include <linux/delay.h> | |
9 | #include <linux/sched.h> | |
10 | #include <linux/init.h> | |
11 | #include <linux/kgdb.h> | |
12 | #include <linux/smp.h> | |
13 | #include <linux/io.h> | |
14 | ||
15 | #include <asm/stackprotector.h> | |
16 | #include <asm/perf_event.h> | |
17 | #include <asm/mmu_context.h> | |
18 | #include <asm/hypervisor.h> | |
19 | #include <asm/processor.h> | |
20 | #include <asm/sections.h> | |
21 | #include <linux/topology.h> | |
22 | #include <linux/cpumask.h> | |
23 | #include <asm/pgtable.h> | |
24 | #include <asm/atomic.h> | |
25 | #include <asm/proto.h> | |
26 | #include <asm/setup.h> | |
27 | #include <asm/apic.h> | |
28 | #include <asm/desc.h> | |
29 | #include <asm/i387.h> | |
30 | #include <asm/mtrr.h> | |
31 | #include <linux/numa.h> | |
32 | #include <asm/asm.h> | |
33 | #include <asm/cpu.h> | |
34 | #include <asm/mce.h> | |
35 | #include <asm/msr.h> | |
36 | #include <asm/pat.h> | |
37 | ||
38 | #ifdef CONFIG_X86_LOCAL_APIC | |
39 | #include <asm/uv/uv.h> | |
40 | #endif | |
41 | ||
42 | #include "cpu.h" | |
43 | ||
44 | /* all of these masks are initialized in setup_cpu_local_masks() */ | |
45 | cpumask_var_t cpu_initialized_mask; | |
46 | cpumask_var_t cpu_callout_mask; | |
47 | cpumask_var_t cpu_callin_mask; | |
48 | ||
49 | /* representing cpus for which sibling maps can be computed */ | |
50 | cpumask_var_t cpu_sibling_setup_mask; | |
51 | ||
52 | /* correctly size the local cpu masks */ | |
53 | void __init setup_cpu_local_masks(void) | |
54 | { | |
55 | alloc_bootmem_cpumask_var(&cpu_initialized_mask); | |
56 | alloc_bootmem_cpumask_var(&cpu_callin_mask); | |
57 | alloc_bootmem_cpumask_var(&cpu_callout_mask); | |
58 | alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask); | |
59 | } | |
60 | ||
61 | static void __cpuinit default_init(struct cpuinfo_x86 *c) | |
62 | { | |
63 | #ifdef CONFIG_X86_64 | |
64 | display_cacheinfo(c); | |
65 | #else | |
66 | /* Not much we can do here... */ | |
67 | /* Check if at least it has cpuid */ | |
68 | if (c->cpuid_level == -1) { | |
69 | /* No cpuid. It must be an ancient CPU */ | |
70 | if (c->x86 == 4) | |
71 | strcpy(c->x86_model_id, "486"); | |
72 | else if (c->x86 == 3) | |
73 | strcpy(c->x86_model_id, "386"); | |
74 | } | |
75 | #endif | |
76 | } | |
77 | ||
78 | static const struct cpu_dev __cpuinitconst default_cpu = { | |
79 | .c_init = default_init, | |
80 | .c_vendor = "Unknown", | |
81 | .c_x86_vendor = X86_VENDOR_UNKNOWN, | |
82 | }; | |
83 | ||
84 | static const struct cpu_dev *this_cpu __cpuinitdata = &default_cpu; | |
85 | ||
86 | DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = { | |
87 | #ifdef CONFIG_X86_64 | |
88 | /* | |
89 | * We need valid kernel segments for data and code in long mode too | |
90 | * IRET will check the segment types kkeil 2000/10/28 | |
91 | * Also sysret mandates a special GDT layout | |
92 | * | |
93 | * TLS descriptors are currently at a different place compared to i386. | |
94 | * Hopefully nobody expects them at a fixed place (Wine?) | |
95 | */ | |
96 | [GDT_ENTRY_KERNEL32_CS] = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff), | |
97 | [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff), | |
98 | [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc093, 0, 0xfffff), | |
99 | [GDT_ENTRY_DEFAULT_USER32_CS] = GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff), | |
100 | [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff), | |
101 | [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff), | |
102 | #else | |
103 | [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xc09a, 0, 0xfffff), | |
104 | [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), | |
105 | [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff), | |
106 | [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff), | |
107 | /* | |
108 | * Segments used for calling PnP BIOS have byte granularity. | |
109 | * They code segments and data segments have fixed 64k limits, | |
110 | * the transfer segment sizes are set at run time. | |
111 | */ | |
112 | /* 32-bit code */ | |
113 | [GDT_ENTRY_PNPBIOS_CS32] = GDT_ENTRY_INIT(0x409a, 0, 0xffff), | |
114 | /* 16-bit code */ | |
115 | [GDT_ENTRY_PNPBIOS_CS16] = GDT_ENTRY_INIT(0x009a, 0, 0xffff), | |
116 | /* 16-bit data */ | |
117 | [GDT_ENTRY_PNPBIOS_DS] = GDT_ENTRY_INIT(0x0092, 0, 0xffff), | |
118 | /* 16-bit data */ | |
119 | [GDT_ENTRY_PNPBIOS_TS1] = GDT_ENTRY_INIT(0x0092, 0, 0), | |
120 | /* 16-bit data */ | |
121 | [GDT_ENTRY_PNPBIOS_TS2] = GDT_ENTRY_INIT(0x0092, 0, 0), | |
122 | /* | |
123 | * The APM segments have byte granularity and their bases | |
124 | * are set at run time. All have 64k limits. | |
125 | */ | |
126 | /* 32-bit code */ | |
127 | [GDT_ENTRY_APMBIOS_BASE] = GDT_ENTRY_INIT(0x409a, 0, 0xffff), | |
128 | /* 16-bit code */ | |
129 | [GDT_ENTRY_APMBIOS_BASE+1] = GDT_ENTRY_INIT(0x009a, 0, 0xffff), | |
130 | /* data */ | |
131 | [GDT_ENTRY_APMBIOS_BASE+2] = GDT_ENTRY_INIT(0x4092, 0, 0xffff), | |
132 | ||
133 | [GDT_ENTRY_ESPFIX_SS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), | |
134 | [GDT_ENTRY_PERCPU] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), | |
135 | GDT_STACK_CANARY_INIT | |
136 | #endif | |
137 | } }; | |
138 | EXPORT_PER_CPU_SYMBOL_GPL(gdt_page); | |
139 | ||
140 | static int __init x86_xsave_setup(char *s) | |
141 | { | |
142 | setup_clear_cpu_cap(X86_FEATURE_XSAVE); | |
143 | return 1; | |
144 | } | |
145 | __setup("noxsave", x86_xsave_setup); | |
146 | ||
147 | #ifdef CONFIG_X86_32 | |
148 | static int cachesize_override __cpuinitdata = -1; | |
149 | static int disable_x86_serial_nr __cpuinitdata = 1; | |
150 | ||
151 | static int __init cachesize_setup(char *str) | |
152 | { | |
153 | get_option(&str, &cachesize_override); | |
154 | return 1; | |
155 | } | |
156 | __setup("cachesize=", cachesize_setup); | |
157 | ||
158 | static int __init x86_fxsr_setup(char *s) | |
159 | { | |
160 | setup_clear_cpu_cap(X86_FEATURE_FXSR); | |
161 | setup_clear_cpu_cap(X86_FEATURE_XMM); | |
162 | return 1; | |
163 | } | |
164 | __setup("nofxsr", x86_fxsr_setup); | |
165 | ||
166 | static int __init x86_sep_setup(char *s) | |
167 | { | |
168 | setup_clear_cpu_cap(X86_FEATURE_SEP); | |
169 | return 1; | |
170 | } | |
171 | __setup("nosep", x86_sep_setup); | |
172 | ||
173 | /* Standard macro to see if a specific flag is changeable */ | |
174 | static inline int flag_is_changeable_p(u32 flag) | |
175 | { | |
176 | u32 f1, f2; | |
177 | ||
178 | /* | |
179 | * Cyrix and IDT cpus allow disabling of CPUID | |
180 | * so the code below may return different results | |
181 | * when it is executed before and after enabling | |
182 | * the CPUID. Add "volatile" to not allow gcc to | |
183 | * optimize the subsequent calls to this function. | |
184 | */ | |
185 | asm volatile ("pushfl \n\t" | |
186 | "pushfl \n\t" | |
187 | "popl %0 \n\t" | |
188 | "movl %0, %1 \n\t" | |
189 | "xorl %2, %0 \n\t" | |
190 | "pushl %0 \n\t" | |
191 | "popfl \n\t" | |
192 | "pushfl \n\t" | |
193 | "popl %0 \n\t" | |
194 | "popfl \n\t" | |
195 | ||
196 | : "=&r" (f1), "=&r" (f2) | |
197 | : "ir" (flag)); | |
198 | ||
199 | return ((f1^f2) & flag) != 0; | |
200 | } | |
201 | ||
202 | /* Probe for the CPUID instruction */ | |
203 | static int __cpuinit have_cpuid_p(void) | |
204 | { | |
205 | return flag_is_changeable_p(X86_EFLAGS_ID); | |
206 | } | |
207 | ||
208 | static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c) | |
209 | { | |
210 | unsigned long lo, hi; | |
211 | ||
212 | if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr) | |
213 | return; | |
214 | ||
215 | /* Disable processor serial number: */ | |
216 | ||
217 | rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi); | |
218 | lo |= 0x200000; | |
219 | wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi); | |
220 | ||
221 | printk(KERN_NOTICE "CPU serial number disabled.\n"); | |
222 | clear_cpu_cap(c, X86_FEATURE_PN); | |
223 | ||
224 | /* Disabling the serial number may affect the cpuid level */ | |
225 | c->cpuid_level = cpuid_eax(0); | |
226 | } | |
227 | ||
228 | static int __init x86_serial_nr_setup(char *s) | |
229 | { | |
230 | disable_x86_serial_nr = 0; | |
231 | return 1; | |
232 | } | |
233 | __setup("serialnumber", x86_serial_nr_setup); | |
234 | #else | |
235 | static inline int flag_is_changeable_p(u32 flag) | |
236 | { | |
237 | return 1; | |
238 | } | |
239 | /* Probe for the CPUID instruction */ | |
240 | static inline int have_cpuid_p(void) | |
241 | { | |
242 | return 1; | |
243 | } | |
244 | static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c) | |
245 | { | |
246 | } | |
247 | #endif | |
248 | ||
249 | /* | |
250 | * Some CPU features depend on higher CPUID levels, which may not always | |
251 | * be available due to CPUID level capping or broken virtualization | |
252 | * software. Add those features to this table to auto-disable them. | |
253 | */ | |
254 | struct cpuid_dependent_feature { | |
255 | u32 feature; | |
256 | u32 level; | |
257 | }; | |
258 | ||
259 | static const struct cpuid_dependent_feature __cpuinitconst | |
260 | cpuid_dependent_features[] = { | |
261 | { X86_FEATURE_MWAIT, 0x00000005 }, | |
262 | { X86_FEATURE_DCA, 0x00000009 }, | |
263 | { X86_FEATURE_XSAVE, 0x0000000d }, | |
264 | { 0, 0 } | |
265 | }; | |
266 | ||
267 | static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn) | |
268 | { | |
269 | const struct cpuid_dependent_feature *df; | |
270 | ||
271 | for (df = cpuid_dependent_features; df->feature; df++) { | |
272 | ||
273 | if (!cpu_has(c, df->feature)) | |
274 | continue; | |
275 | /* | |
276 | * Note: cpuid_level is set to -1 if unavailable, but | |
277 | * extended_extended_level is set to 0 if unavailable | |
278 | * and the legitimate extended levels are all negative | |
279 | * when signed; hence the weird messing around with | |
280 | * signs here... | |
281 | */ | |
282 | if (!((s32)df->level < 0 ? | |
283 | (u32)df->level > (u32)c->extended_cpuid_level : | |
284 | (s32)df->level > (s32)c->cpuid_level)) | |
285 | continue; | |
286 | ||
287 | clear_cpu_cap(c, df->feature); | |
288 | if (!warn) | |
289 | continue; | |
290 | ||
291 | printk(KERN_WARNING | |
292 | "CPU: CPU feature %s disabled, no CPUID level 0x%x\n", | |
293 | x86_cap_flags[df->feature], df->level); | |
294 | } | |
295 | } | |
296 | ||
297 | /* | |
298 | * Naming convention should be: <Name> [(<Codename>)] | |
299 | * This table only is used unless init_<vendor>() below doesn't set it; | |
300 | * in particular, if CPUID levels 0x80000002..4 are supported, this | |
301 | * isn't used | |
302 | */ | |
303 | ||
304 | /* Look up CPU names by table lookup. */ | |
305 | static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c) | |
306 | { | |
307 | const struct cpu_model_info *info; | |
308 | ||
309 | if (c->x86_model >= 16) | |
310 | return NULL; /* Range check */ | |
311 | ||
312 | if (!this_cpu) | |
313 | return NULL; | |
314 | ||
315 | info = this_cpu->c_models; | |
316 | ||
317 | while (info && info->family) { | |
318 | if (info->family == c->x86) | |
319 | return info->model_names[c->x86_model]; | |
320 | info++; | |
321 | } | |
322 | return NULL; /* Not found */ | |
323 | } | |
324 | ||
325 | __u32 cpu_caps_cleared[NCAPINTS] __cpuinitdata; | |
326 | __u32 cpu_caps_set[NCAPINTS] __cpuinitdata; | |
327 | ||
328 | void load_percpu_segment(int cpu) | |
329 | { | |
330 | #ifdef CONFIG_X86_32 | |
331 | loadsegment(fs, __KERNEL_PERCPU); | |
332 | #else | |
333 | loadsegment(gs, 0); | |
334 | wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu)); | |
335 | #endif | |
336 | load_stack_canary_segment(); | |
337 | } | |
338 | ||
339 | /* | |
340 | * Current gdt points %fs at the "master" per-cpu area: after this, | |
341 | * it's on the real one. | |
342 | */ | |
343 | void switch_to_new_gdt(int cpu) | |
344 | { | |
345 | struct desc_ptr gdt_descr; | |
346 | ||
347 | gdt_descr.address = (long)get_cpu_gdt_table(cpu); | |
348 | gdt_descr.size = GDT_SIZE - 1; | |
349 | load_gdt(&gdt_descr); | |
350 | /* Reload the per-cpu base */ | |
351 | ||
352 | load_percpu_segment(cpu); | |
353 | } | |
354 | ||
355 | static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {}; | |
356 | ||
357 | static void __cpuinit get_model_name(struct cpuinfo_x86 *c) | |
358 | { | |
359 | unsigned int *v; | |
360 | char *p, *q; | |
361 | ||
362 | if (c->extended_cpuid_level < 0x80000004) | |
363 | return; | |
364 | ||
365 | v = (unsigned int *)c->x86_model_id; | |
366 | cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]); | |
367 | cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]); | |
368 | cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]); | |
369 | c->x86_model_id[48] = 0; | |
370 | ||
371 | /* | |
372 | * Intel chips right-justify this string for some dumb reason; | |
373 | * undo that brain damage: | |
374 | */ | |
375 | p = q = &c->x86_model_id[0]; | |
376 | while (*p == ' ') | |
377 | p++; | |
378 | if (p != q) { | |
379 | while (*p) | |
380 | *q++ = *p++; | |
381 | while (q <= &c->x86_model_id[48]) | |
382 | *q++ = '\0'; /* Zero-pad the rest */ | |
383 | } | |
384 | } | |
385 | ||
386 | void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c) | |
387 | { | |
388 | unsigned int n, dummy, ebx, ecx, edx, l2size; | |
389 | ||
390 | n = c->extended_cpuid_level; | |
391 | ||
392 | if (n >= 0x80000005) { | |
393 | cpuid(0x80000005, &dummy, &ebx, &ecx, &edx); | |
394 | c->x86_cache_size = (ecx>>24) + (edx>>24); | |
395 | #ifdef CONFIG_X86_64 | |
396 | /* On K8 L1 TLB is inclusive, so don't count it */ | |
397 | c->x86_tlbsize = 0; | |
398 | #endif | |
399 | } | |
400 | ||
401 | if (n < 0x80000006) /* Some chips just has a large L1. */ | |
402 | return; | |
403 | ||
404 | cpuid(0x80000006, &dummy, &ebx, &ecx, &edx); | |
405 | l2size = ecx >> 16; | |
406 | ||
407 | #ifdef CONFIG_X86_64 | |
408 | c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff); | |
409 | #else | |
410 | /* do processor-specific cache resizing */ | |
411 | if (this_cpu->c_size_cache) | |
412 | l2size = this_cpu->c_size_cache(c, l2size); | |
413 | ||
414 | /* Allow user to override all this if necessary. */ | |
415 | if (cachesize_override != -1) | |
416 | l2size = cachesize_override; | |
417 | ||
418 | if (l2size == 0) | |
419 | return; /* Again, no L2 cache is possible */ | |
420 | #endif | |
421 | ||
422 | c->x86_cache_size = l2size; | |
423 | } | |
424 | ||
425 | void __cpuinit detect_ht(struct cpuinfo_x86 *c) | |
426 | { | |
427 | #ifdef CONFIG_X86_HT | |
428 | u32 eax, ebx, ecx, edx; | |
429 | int index_msb, core_bits; | |
430 | ||
431 | if (!cpu_has(c, X86_FEATURE_HT)) | |
432 | return; | |
433 | ||
434 | if (cpu_has(c, X86_FEATURE_CMP_LEGACY)) | |
435 | goto out; | |
436 | ||
437 | if (cpu_has(c, X86_FEATURE_XTOPOLOGY)) | |
438 | return; | |
439 | ||
440 | cpuid(1, &eax, &ebx, &ecx, &edx); | |
441 | ||
442 | smp_num_siblings = (ebx & 0xff0000) >> 16; | |
443 | ||
444 | if (smp_num_siblings == 1) { | |
445 | printk(KERN_INFO "CPU: Hyper-Threading is disabled\n"); | |
446 | goto out; | |
447 | } | |
448 | ||
449 | if (smp_num_siblings <= 1) | |
450 | goto out; | |
451 | ||
452 | if (smp_num_siblings > nr_cpu_ids) { | |
453 | pr_warning("CPU: Unsupported number of siblings %d", | |
454 | smp_num_siblings); | |
455 | smp_num_siblings = 1; | |
456 | return; | |
457 | } | |
458 | ||
459 | index_msb = get_count_order(smp_num_siblings); | |
460 | c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb); | |
461 | ||
462 | smp_num_siblings = smp_num_siblings / c->x86_max_cores; | |
463 | ||
464 | index_msb = get_count_order(smp_num_siblings); | |
465 | ||
466 | core_bits = get_count_order(c->x86_max_cores); | |
467 | ||
468 | c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) & | |
469 | ((1 << core_bits) - 1); | |
470 | ||
471 | out: | |
472 | if ((c->x86_max_cores * smp_num_siblings) > 1) { | |
473 | printk(KERN_INFO "CPU: Physical Processor ID: %d\n", | |
474 | c->phys_proc_id); | |
475 | printk(KERN_INFO "CPU: Processor Core ID: %d\n", | |
476 | c->cpu_core_id); | |
477 | } | |
478 | #endif | |
479 | } | |
480 | ||
481 | static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c) | |
482 | { | |
483 | char *v = c->x86_vendor_id; | |
484 | int i; | |
485 | ||
486 | for (i = 0; i < X86_VENDOR_NUM; i++) { | |
487 | if (!cpu_devs[i]) | |
488 | break; | |
489 | ||
490 | if (!strcmp(v, cpu_devs[i]->c_ident[0]) || | |
491 | (cpu_devs[i]->c_ident[1] && | |
492 | !strcmp(v, cpu_devs[i]->c_ident[1]))) { | |
493 | ||
494 | this_cpu = cpu_devs[i]; | |
495 | c->x86_vendor = this_cpu->c_x86_vendor; | |
496 | return; | |
497 | } | |
498 | } | |
499 | ||
500 | printk_once(KERN_ERR | |
501 | "CPU: vendor_id '%s' unknown, using generic init.\n" \ | |
502 | "CPU: Your system may be unstable.\n", v); | |
503 | ||
504 | c->x86_vendor = X86_VENDOR_UNKNOWN; | |
505 | this_cpu = &default_cpu; | |
506 | } | |
507 | ||
508 | void __cpuinit cpu_detect(struct cpuinfo_x86 *c) | |
509 | { | |
510 | /* Get vendor name */ | |
511 | cpuid(0x00000000, (unsigned int *)&c->cpuid_level, | |
512 | (unsigned int *)&c->x86_vendor_id[0], | |
513 | (unsigned int *)&c->x86_vendor_id[8], | |
514 | (unsigned int *)&c->x86_vendor_id[4]); | |
515 | ||
516 | c->x86 = 4; | |
517 | /* Intel-defined flags: level 0x00000001 */ | |
518 | if (c->cpuid_level >= 0x00000001) { | |
519 | u32 junk, tfms, cap0, misc; | |
520 | ||
521 | cpuid(0x00000001, &tfms, &misc, &junk, &cap0); | |
522 | c->x86 = (tfms >> 8) & 0xf; | |
523 | c->x86_model = (tfms >> 4) & 0xf; | |
524 | c->x86_mask = tfms & 0xf; | |
525 | ||
526 | if (c->x86 == 0xf) | |
527 | c->x86 += (tfms >> 20) & 0xff; | |
528 | if (c->x86 >= 0x6) | |
529 | c->x86_model += ((tfms >> 16) & 0xf) << 4; | |
530 | ||
531 | if (cap0 & (1<<19)) { | |
532 | c->x86_clflush_size = ((misc >> 8) & 0xff) * 8; | |
533 | c->x86_cache_alignment = c->x86_clflush_size; | |
534 | } | |
535 | } | |
536 | } | |
537 | ||
538 | static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c) | |
539 | { | |
540 | u32 tfms, xlvl; | |
541 | u32 ebx; | |
542 | ||
543 | /* Intel-defined flags: level 0x00000001 */ | |
544 | if (c->cpuid_level >= 0x00000001) { | |
545 | u32 capability, excap; | |
546 | ||
547 | cpuid(0x00000001, &tfms, &ebx, &excap, &capability); | |
548 | c->x86_capability[0] = capability; | |
549 | c->x86_capability[4] = excap; | |
550 | } | |
551 | ||
552 | /* AMD-defined flags: level 0x80000001 */ | |
553 | xlvl = cpuid_eax(0x80000000); | |
554 | c->extended_cpuid_level = xlvl; | |
555 | ||
556 | if ((xlvl & 0xffff0000) == 0x80000000) { | |
557 | if (xlvl >= 0x80000001) { | |
558 | c->x86_capability[1] = cpuid_edx(0x80000001); | |
559 | c->x86_capability[6] = cpuid_ecx(0x80000001); | |
560 | } | |
561 | } | |
562 | ||
563 | if (c->extended_cpuid_level >= 0x80000008) { | |
564 | u32 eax = cpuid_eax(0x80000008); | |
565 | ||
566 | c->x86_virt_bits = (eax >> 8) & 0xff; | |
567 | c->x86_phys_bits = eax & 0xff; | |
568 | } | |
569 | #ifdef CONFIG_X86_32 | |
570 | else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36)) | |
571 | c->x86_phys_bits = 36; | |
572 | #endif | |
573 | ||
574 | if (c->extended_cpuid_level >= 0x80000007) | |
575 | c->x86_power = cpuid_edx(0x80000007); | |
576 | ||
577 | } | |
578 | ||
579 | static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c) | |
580 | { | |
581 | #ifdef CONFIG_X86_32 | |
582 | int i; | |
583 | ||
584 | /* | |
585 | * First of all, decide if this is a 486 or higher | |
586 | * It's a 486 if we can modify the AC flag | |
587 | */ | |
588 | if (flag_is_changeable_p(X86_EFLAGS_AC)) | |
589 | c->x86 = 4; | |
590 | else | |
591 | c->x86 = 3; | |
592 | ||
593 | for (i = 0; i < X86_VENDOR_NUM; i++) | |
594 | if (cpu_devs[i] && cpu_devs[i]->c_identify) { | |
595 | c->x86_vendor_id[0] = 0; | |
596 | cpu_devs[i]->c_identify(c); | |
597 | if (c->x86_vendor_id[0]) { | |
598 | get_cpu_vendor(c); | |
599 | break; | |
600 | } | |
601 | } | |
602 | #endif | |
603 | } | |
604 | ||
605 | /* | |
606 | * Do minimum CPU detection early. | |
607 | * Fields really needed: vendor, cpuid_level, family, model, mask, | |
608 | * cache alignment. | |
609 | * The others are not touched to avoid unwanted side effects. | |
610 | * | |
611 | * WARNING: this function is only called on the BP. Don't add code here | |
612 | * that is supposed to run on all CPUs. | |
613 | */ | |
614 | static void __init early_identify_cpu(struct cpuinfo_x86 *c) | |
615 | { | |
616 | #ifdef CONFIG_X86_64 | |
617 | c->x86_clflush_size = 64; | |
618 | c->x86_phys_bits = 36; | |
619 | c->x86_virt_bits = 48; | |
620 | #else | |
621 | c->x86_clflush_size = 32; | |
622 | c->x86_phys_bits = 32; | |
623 | c->x86_virt_bits = 32; | |
624 | #endif | |
625 | c->x86_cache_alignment = c->x86_clflush_size; | |
626 | ||
627 | memset(&c->x86_capability, 0, sizeof c->x86_capability); | |
628 | c->extended_cpuid_level = 0; | |
629 | ||
630 | if (!have_cpuid_p()) | |
631 | identify_cpu_without_cpuid(c); | |
632 | ||
633 | /* cyrix could have cpuid enabled via c_identify()*/ | |
634 | if (!have_cpuid_p()) | |
635 | return; | |
636 | ||
637 | cpu_detect(c); | |
638 | ||
639 | get_cpu_vendor(c); | |
640 | ||
641 | get_cpu_cap(c); | |
642 | ||
643 | if (this_cpu->c_early_init) | |
644 | this_cpu->c_early_init(c); | |
645 | ||
646 | #ifdef CONFIG_SMP | |
647 | c->cpu_index = boot_cpu_id; | |
648 | #endif | |
649 | filter_cpuid_features(c, false); | |
650 | } | |
651 | ||
652 | void __init early_cpu_init(void) | |
653 | { | |
654 | #ifdef PROCESSOR_SELECT | |
655 | const struct cpu_dev *const *cdev; | |
656 | int count = 0; | |
657 | ||
658 | printk(KERN_INFO "KERNEL supported cpus:\n"); | |
659 | for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) { | |
660 | const struct cpu_dev *cpudev = *cdev; | |
661 | unsigned int j; | |
662 | ||
663 | if (count >= X86_VENDOR_NUM) | |
664 | break; | |
665 | cpu_devs[count] = cpudev; | |
666 | count++; | |
667 | ||
668 | for (j = 0; j < 2; j++) { | |
669 | if (!cpudev->c_ident[j]) | |
670 | continue; | |
671 | printk(KERN_INFO " %s %s\n", cpudev->c_vendor, | |
672 | cpudev->c_ident[j]); | |
673 | } | |
674 | } | |
675 | #endif | |
676 | early_identify_cpu(&boot_cpu_data); | |
677 | } | |
678 | ||
679 | /* | |
680 | * The NOPL instruction is supposed to exist on all CPUs with | |
681 | * family >= 6; unfortunately, that's not true in practice because | |
682 | * of early VIA chips and (more importantly) broken virtualizers that | |
683 | * are not easy to detect. In the latter case it doesn't even *fail* | |
684 | * reliably, so probing for it doesn't even work. Disable it completely | |
685 | * unless we can find a reliable way to detect all the broken cases. | |
686 | */ | |
687 | static void __cpuinit detect_nopl(struct cpuinfo_x86 *c) | |
688 | { | |
689 | clear_cpu_cap(c, X86_FEATURE_NOPL); | |
690 | } | |
691 | ||
692 | static void __cpuinit generic_identify(struct cpuinfo_x86 *c) | |
693 | { | |
694 | c->extended_cpuid_level = 0; | |
695 | ||
696 | if (!have_cpuid_p()) | |
697 | identify_cpu_without_cpuid(c); | |
698 | ||
699 | /* cyrix could have cpuid enabled via c_identify()*/ | |
700 | if (!have_cpuid_p()) | |
701 | return; | |
702 | ||
703 | cpu_detect(c); | |
704 | ||
705 | get_cpu_vendor(c); | |
706 | ||
707 | get_cpu_cap(c); | |
708 | ||
709 | if (c->cpuid_level >= 0x00000001) { | |
710 | c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF; | |
711 | #ifdef CONFIG_X86_32 | |
712 | # ifdef CONFIG_X86_HT | |
713 | c->apicid = apic->phys_pkg_id(c->initial_apicid, 0); | |
714 | # else | |
715 | c->apicid = c->initial_apicid; | |
716 | # endif | |
717 | #endif | |
718 | ||
719 | #ifdef CONFIG_X86_HT | |
720 | c->phys_proc_id = c->initial_apicid; | |
721 | #endif | |
722 | } | |
723 | ||
724 | get_model_name(c); /* Default name */ | |
725 | ||
726 | init_scattered_cpuid_features(c); | |
727 | detect_nopl(c); | |
728 | } | |
729 | ||
730 | /* | |
731 | * This does the hard work of actually picking apart the CPU stuff... | |
732 | */ | |
733 | static void __cpuinit identify_cpu(struct cpuinfo_x86 *c) | |
734 | { | |
735 | int i; | |
736 | ||
737 | c->loops_per_jiffy = loops_per_jiffy; | |
738 | c->x86_cache_size = -1; | |
739 | c->x86_vendor = X86_VENDOR_UNKNOWN; | |
740 | c->x86_model = c->x86_mask = 0; /* So far unknown... */ | |
741 | c->x86_vendor_id[0] = '\0'; /* Unset */ | |
742 | c->x86_model_id[0] = '\0'; /* Unset */ | |
743 | c->x86_max_cores = 1; | |
744 | c->x86_coreid_bits = 0; | |
745 | #ifdef CONFIG_X86_64 | |
746 | c->x86_clflush_size = 64; | |
747 | c->x86_phys_bits = 36; | |
748 | c->x86_virt_bits = 48; | |
749 | #else | |
750 | c->cpuid_level = -1; /* CPUID not detected */ | |
751 | c->x86_clflush_size = 32; | |
752 | c->x86_phys_bits = 32; | |
753 | c->x86_virt_bits = 32; | |
754 | #endif | |
755 | c->x86_cache_alignment = c->x86_clflush_size; | |
756 | memset(&c->x86_capability, 0, sizeof c->x86_capability); | |
757 | ||
758 | generic_identify(c); | |
759 | ||
760 | if (this_cpu->c_identify) | |
761 | this_cpu->c_identify(c); | |
762 | ||
763 | /* Clear/Set all flags overriden by options, after probe */ | |
764 | for (i = 0; i < NCAPINTS; i++) { | |
765 | c->x86_capability[i] &= ~cpu_caps_cleared[i]; | |
766 | c->x86_capability[i] |= cpu_caps_set[i]; | |
767 | } | |
768 | ||
769 | #ifdef CONFIG_X86_64 | |
770 | c->apicid = apic->phys_pkg_id(c->initial_apicid, 0); | |
771 | #endif | |
772 | ||
773 | /* | |
774 | * Vendor-specific initialization. In this section we | |
775 | * canonicalize the feature flags, meaning if there are | |
776 | * features a certain CPU supports which CPUID doesn't | |
777 | * tell us, CPUID claiming incorrect flags, or other bugs, | |
778 | * we handle them here. | |
779 | * | |
780 | * At the end of this section, c->x86_capability better | |
781 | * indicate the features this CPU genuinely supports! | |
782 | */ | |
783 | if (this_cpu->c_init) | |
784 | this_cpu->c_init(c); | |
785 | ||
786 | /* Disable the PN if appropriate */ | |
787 | squash_the_stupid_serial_number(c); | |
788 | ||
789 | /* | |
790 | * The vendor-specific functions might have changed features. | |
791 | * Now we do "generic changes." | |
792 | */ | |
793 | ||
794 | /* Filter out anything that depends on CPUID levels we don't have */ | |
795 | filter_cpuid_features(c, true); | |
796 | ||
797 | /* If the model name is still unset, do table lookup. */ | |
798 | if (!c->x86_model_id[0]) { | |
799 | const char *p; | |
800 | p = table_lookup_model(c); | |
801 | if (p) | |
802 | strcpy(c->x86_model_id, p); | |
803 | else | |
804 | /* Last resort... */ | |
805 | sprintf(c->x86_model_id, "%02x/%02x", | |
806 | c->x86, c->x86_model); | |
807 | } | |
808 | ||
809 | #ifdef CONFIG_X86_64 | |
810 | detect_ht(c); | |
811 | #endif | |
812 | ||
813 | init_hypervisor(c); | |
814 | ||
815 | /* | |
816 | * Clear/Set all flags overriden by options, need do it | |
817 | * before following smp all cpus cap AND. | |
818 | */ | |
819 | for (i = 0; i < NCAPINTS; i++) { | |
820 | c->x86_capability[i] &= ~cpu_caps_cleared[i]; | |
821 | c->x86_capability[i] |= cpu_caps_set[i]; | |
822 | } | |
823 | ||
824 | /* | |
825 | * On SMP, boot_cpu_data holds the common feature set between | |
826 | * all CPUs; so make sure that we indicate which features are | |
827 | * common between the CPUs. The first time this routine gets | |
828 | * executed, c == &boot_cpu_data. | |
829 | */ | |
830 | if (c != &boot_cpu_data) { | |
831 | /* AND the already accumulated flags with these */ | |
832 | for (i = 0; i < NCAPINTS; i++) | |
833 | boot_cpu_data.x86_capability[i] &= c->x86_capability[i]; | |
834 | } | |
835 | ||
836 | #ifdef CONFIG_X86_MCE | |
837 | /* Init Machine Check Exception if available. */ | |
838 | mcheck_init(c); | |
839 | #endif | |
840 | ||
841 | select_idle_routine(c); | |
842 | ||
843 | #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64) | |
844 | numa_add_cpu(smp_processor_id()); | |
845 | #endif | |
846 | } | |
847 | ||
848 | #ifdef CONFIG_X86_64 | |
849 | static void vgetcpu_set_mode(void) | |
850 | { | |
851 | if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP)) | |
852 | vgetcpu_mode = VGETCPU_RDTSCP; | |
853 | else | |
854 | vgetcpu_mode = VGETCPU_LSL; | |
855 | } | |
856 | #endif | |
857 | ||
858 | void __init identify_boot_cpu(void) | |
859 | { | |
860 | identify_cpu(&boot_cpu_data); | |
861 | init_c1e_mask(); | |
862 | #ifdef CONFIG_X86_32 | |
863 | sysenter_setup(); | |
864 | enable_sep_cpu(); | |
865 | #else | |
866 | vgetcpu_set_mode(); | |
867 | #endif | |
868 | init_hw_perf_events(); | |
869 | } | |
870 | ||
871 | void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c) | |
872 | { | |
873 | BUG_ON(c == &boot_cpu_data); | |
874 | identify_cpu(c); | |
875 | #ifdef CONFIG_X86_32 | |
876 | enable_sep_cpu(); | |
877 | #endif | |
878 | mtrr_ap_init(); | |
879 | } | |
880 | ||
881 | struct msr_range { | |
882 | unsigned min; | |
883 | unsigned max; | |
884 | }; | |
885 | ||
886 | static const struct msr_range msr_range_array[] __cpuinitconst = { | |
887 | { 0x00000000, 0x00000418}, | |
888 | { 0xc0000000, 0xc000040b}, | |
889 | { 0xc0010000, 0xc0010142}, | |
890 | { 0xc0011000, 0xc001103b}, | |
891 | }; | |
892 | ||
893 | static void __cpuinit print_cpu_msr(void) | |
894 | { | |
895 | unsigned index_min, index_max; | |
896 | unsigned index; | |
897 | u64 val; | |
898 | int i; | |
899 | ||
900 | for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) { | |
901 | index_min = msr_range_array[i].min; | |
902 | index_max = msr_range_array[i].max; | |
903 | ||
904 | for (index = index_min; index < index_max; index++) { | |
905 | if (rdmsrl_amd_safe(index, &val)) | |
906 | continue; | |
907 | printk(KERN_INFO " MSR%08x: %016llx\n", index, val); | |
908 | } | |
909 | } | |
910 | } | |
911 | ||
912 | static int show_msr __cpuinitdata; | |
913 | ||
914 | static __init int setup_show_msr(char *arg) | |
915 | { | |
916 | int num; | |
917 | ||
918 | get_option(&arg, &num); | |
919 | ||
920 | if (num > 0) | |
921 | show_msr = num; | |
922 | return 1; | |
923 | } | |
924 | __setup("show_msr=", setup_show_msr); | |
925 | ||
926 | static __init int setup_noclflush(char *arg) | |
927 | { | |
928 | setup_clear_cpu_cap(X86_FEATURE_CLFLSH); | |
929 | return 1; | |
930 | } | |
931 | __setup("noclflush", setup_noclflush); | |
932 | ||
933 | void __cpuinit print_cpu_info(struct cpuinfo_x86 *c) | |
934 | { | |
935 | const char *vendor = NULL; | |
936 | ||
937 | if (c->x86_vendor < X86_VENDOR_NUM) { | |
938 | vendor = this_cpu->c_vendor; | |
939 | } else { | |
940 | if (c->cpuid_level >= 0) | |
941 | vendor = c->x86_vendor_id; | |
942 | } | |
943 | ||
944 | if (vendor && !strstr(c->x86_model_id, vendor)) | |
945 | printk(KERN_CONT "%s ", vendor); | |
946 | ||
947 | if (c->x86_model_id[0]) | |
948 | printk(KERN_CONT "%s", c->x86_model_id); | |
949 | else | |
950 | printk(KERN_CONT "%d86", c->x86); | |
951 | ||
952 | if (c->x86_mask || c->cpuid_level >= 0) | |
953 | printk(KERN_CONT " stepping %02x\n", c->x86_mask); | |
954 | else | |
955 | printk(KERN_CONT "\n"); | |
956 | ||
957 | #ifdef CONFIG_SMP | |
958 | if (c->cpu_index < show_msr) | |
959 | print_cpu_msr(); | |
960 | #else | |
961 | if (show_msr) | |
962 | print_cpu_msr(); | |
963 | #endif | |
964 | } | |
965 | ||
966 | static __init int setup_disablecpuid(char *arg) | |
967 | { | |
968 | int bit; | |
969 | ||
970 | if (get_option(&arg, &bit) && bit < NCAPINTS*32) | |
971 | setup_clear_cpu_cap(bit); | |
972 | else | |
973 | return 0; | |
974 | ||
975 | return 1; | |
976 | } | |
977 | __setup("clearcpuid=", setup_disablecpuid); | |
978 | ||
979 | #ifdef CONFIG_X86_64 | |
980 | struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table }; | |
981 | ||
982 | DEFINE_PER_CPU_FIRST(union irq_stack_union, | |
983 | irq_stack_union) __aligned(PAGE_SIZE); | |
984 | ||
985 | /* | |
986 | * The following four percpu variables are hot. Align current_task to | |
987 | * cacheline size such that all four fall in the same cacheline. | |
988 | */ | |
989 | DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned = | |
990 | &init_task; | |
991 | EXPORT_PER_CPU_SYMBOL(current_task); | |
992 | ||
993 | DEFINE_PER_CPU(unsigned long, kernel_stack) = | |
994 | (unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE; | |
995 | EXPORT_PER_CPU_SYMBOL(kernel_stack); | |
996 | ||
997 | DEFINE_PER_CPU(char *, irq_stack_ptr) = | |
998 | init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64; | |
999 | ||
1000 | DEFINE_PER_CPU(unsigned int, irq_count) = -1; | |
1001 | ||
1002 | /* | |
1003 | * Special IST stacks which the CPU switches to when it calls | |
1004 | * an IST-marked descriptor entry. Up to 7 stacks (hardware | |
1005 | * limit), all of them are 4K, except the debug stack which | |
1006 | * is 8K. | |
1007 | */ | |
1008 | static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = { | |
1009 | [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ, | |
1010 | [DEBUG_STACK - 1] = DEBUG_STKSZ | |
1011 | }; | |
1012 | ||
1013 | static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks | |
1014 | [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]); | |
1015 | ||
1016 | /* May not be marked __init: used by software suspend */ | |
1017 | void syscall_init(void) | |
1018 | { | |
1019 | /* | |
1020 | * LSTAR and STAR live in a bit strange symbiosis. | |
1021 | * They both write to the same internal register. STAR allows to | |
1022 | * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip. | |
1023 | */ | |
1024 | wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32); | |
1025 | wrmsrl(MSR_LSTAR, system_call); | |
1026 | wrmsrl(MSR_CSTAR, ignore_sysret); | |
1027 | ||
1028 | #ifdef CONFIG_IA32_EMULATION | |
1029 | syscall32_cpu_init(); | |
1030 | #endif | |
1031 | ||
1032 | /* Flags to clear on syscall */ | |
1033 | wrmsrl(MSR_SYSCALL_MASK, | |
1034 | X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL); | |
1035 | } | |
1036 | ||
1037 | unsigned long kernel_eflags; | |
1038 | ||
1039 | /* | |
1040 | * Copies of the original ist values from the tss are only accessed during | |
1041 | * debugging, no special alignment required. | |
1042 | */ | |
1043 | DEFINE_PER_CPU(struct orig_ist, orig_ist); | |
1044 | ||
1045 | #else /* CONFIG_X86_64 */ | |
1046 | ||
1047 | DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task; | |
1048 | EXPORT_PER_CPU_SYMBOL(current_task); | |
1049 | ||
1050 | #ifdef CONFIG_CC_STACKPROTECTOR | |
1051 | DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary); | |
1052 | #endif | |
1053 | ||
1054 | /* Make sure %fs and %gs are initialized properly in idle threads */ | |
1055 | struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs) | |
1056 | { | |
1057 | memset(regs, 0, sizeof(struct pt_regs)); | |
1058 | regs->fs = __KERNEL_PERCPU; | |
1059 | regs->gs = __KERNEL_STACK_CANARY; | |
1060 | ||
1061 | return regs; | |
1062 | } | |
1063 | #endif /* CONFIG_X86_64 */ | |
1064 | ||
1065 | /* | |
1066 | * Clear all 6 debug registers: | |
1067 | */ | |
1068 | static void clear_all_debug_regs(void) | |
1069 | { | |
1070 | int i; | |
1071 | ||
1072 | for (i = 0; i < 8; i++) { | |
1073 | /* Ignore db4, db5 */ | |
1074 | if ((i == 4) || (i == 5)) | |
1075 | continue; | |
1076 | ||
1077 | set_debugreg(0, i); | |
1078 | } | |
1079 | } | |
1080 | ||
1081 | /* | |
1082 | * cpu_init() initializes state that is per-CPU. Some data is already | |
1083 | * initialized (naturally) in the bootstrap process, such as the GDT | |
1084 | * and IDT. We reload them nevertheless, this function acts as a | |
1085 | * 'CPU state barrier', nothing should get across. | |
1086 | * A lot of state is already set up in PDA init for 64 bit | |
1087 | */ | |
1088 | #ifdef CONFIG_X86_64 | |
1089 | ||
1090 | void __cpuinit cpu_init(void) | |
1091 | { | |
1092 | struct orig_ist *orig_ist; | |
1093 | struct task_struct *me; | |
1094 | struct tss_struct *t; | |
1095 | unsigned long v; | |
1096 | int cpu; | |
1097 | int i; | |
1098 | ||
1099 | cpu = stack_smp_processor_id(); | |
1100 | t = &per_cpu(init_tss, cpu); | |
1101 | orig_ist = &per_cpu(orig_ist, cpu); | |
1102 | ||
1103 | #ifdef CONFIG_NUMA | |
1104 | if (cpu != 0 && percpu_read(node_number) == 0 && | |
1105 | cpu_to_node(cpu) != NUMA_NO_NODE) | |
1106 | percpu_write(node_number, cpu_to_node(cpu)); | |
1107 | #endif | |
1108 | ||
1109 | me = current; | |
1110 | ||
1111 | if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) | |
1112 | panic("CPU#%d already initialized!\n", cpu); | |
1113 | ||
1114 | printk(KERN_INFO "Initializing CPU#%d\n", cpu); | |
1115 | ||
1116 | clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE); | |
1117 | ||
1118 | /* | |
1119 | * Initialize the per-CPU GDT with the boot GDT, | |
1120 | * and set up the GDT descriptor: | |
1121 | */ | |
1122 | ||
1123 | switch_to_new_gdt(cpu); | |
1124 | loadsegment(fs, 0); | |
1125 | ||
1126 | load_idt((const struct desc_ptr *)&idt_descr); | |
1127 | ||
1128 | memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8); | |
1129 | syscall_init(); | |
1130 | ||
1131 | wrmsrl(MSR_FS_BASE, 0); | |
1132 | wrmsrl(MSR_KERNEL_GS_BASE, 0); | |
1133 | barrier(); | |
1134 | ||
1135 | check_efer(); | |
1136 | if (cpu != 0) | |
1137 | enable_x2apic(); | |
1138 | ||
1139 | /* | |
1140 | * set up and load the per-CPU TSS | |
1141 | */ | |
1142 | if (!orig_ist->ist[0]) { | |
1143 | char *estacks = per_cpu(exception_stacks, cpu); | |
1144 | ||
1145 | for (v = 0; v < N_EXCEPTION_STACKS; v++) { | |
1146 | estacks += exception_stack_sizes[v]; | |
1147 | orig_ist->ist[v] = t->x86_tss.ist[v] = | |
1148 | (unsigned long)estacks; | |
1149 | } | |
1150 | } | |
1151 | ||
1152 | t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap); | |
1153 | ||
1154 | /* | |
1155 | * <= is required because the CPU will access up to | |
1156 | * 8 bits beyond the end of the IO permission bitmap. | |
1157 | */ | |
1158 | for (i = 0; i <= IO_BITMAP_LONGS; i++) | |
1159 | t->io_bitmap[i] = ~0UL; | |
1160 | ||
1161 | atomic_inc(&init_mm.mm_count); | |
1162 | me->active_mm = &init_mm; | |
1163 | BUG_ON(me->mm); | |
1164 | enter_lazy_tlb(&init_mm, me); | |
1165 | ||
1166 | load_sp0(t, ¤t->thread); | |
1167 | set_tss_desc(cpu, t); | |
1168 | load_TR_desc(); | |
1169 | load_LDT(&init_mm.context); | |
1170 | ||
1171 | #ifdef CONFIG_KGDB | |
1172 | /* | |
1173 | * If the kgdb is connected no debug regs should be altered. This | |
1174 | * is only applicable when KGDB and a KGDB I/O module are built | |
1175 | * into the kernel and you are using early debugging with | |
1176 | * kgdbwait. KGDB will control the kernel HW breakpoint registers. | |
1177 | */ | |
1178 | if (kgdb_connected && arch_kgdb_ops.correct_hw_break) | |
1179 | arch_kgdb_ops.correct_hw_break(); | |
1180 | else | |
1181 | #endif | |
1182 | clear_all_debug_regs(); | |
1183 | ||
1184 | fpu_init(); | |
1185 | ||
1186 | raw_local_save_flags(kernel_eflags); | |
1187 | ||
1188 | if (is_uv_system()) | |
1189 | uv_cpu_init(); | |
1190 | } | |
1191 | ||
1192 | #else | |
1193 | ||
1194 | void __cpuinit cpu_init(void) | |
1195 | { | |
1196 | int cpu = smp_processor_id(); | |
1197 | struct task_struct *curr = current; | |
1198 | struct tss_struct *t = &per_cpu(init_tss, cpu); | |
1199 | struct thread_struct *thread = &curr->thread; | |
1200 | ||
1201 | if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) { | |
1202 | printk(KERN_WARNING "CPU#%d already initialized!\n", cpu); | |
1203 | for (;;) | |
1204 | local_irq_enable(); | |
1205 | } | |
1206 | ||
1207 | printk(KERN_INFO "Initializing CPU#%d\n", cpu); | |
1208 | ||
1209 | if (cpu_has_vme || cpu_has_tsc || cpu_has_de) | |
1210 | clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE); | |
1211 | ||
1212 | load_idt(&idt_descr); | |
1213 | switch_to_new_gdt(cpu); | |
1214 | ||
1215 | /* | |
1216 | * Set up and load the per-CPU TSS and LDT | |
1217 | */ | |
1218 | atomic_inc(&init_mm.mm_count); | |
1219 | curr->active_mm = &init_mm; | |
1220 | BUG_ON(curr->mm); | |
1221 | enter_lazy_tlb(&init_mm, curr); | |
1222 | ||
1223 | load_sp0(t, thread); | |
1224 | set_tss_desc(cpu, t); | |
1225 | load_TR_desc(); | |
1226 | load_LDT(&init_mm.context); | |
1227 | ||
1228 | t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap); | |
1229 | ||
1230 | #ifdef CONFIG_DOUBLEFAULT | |
1231 | /* Set up doublefault TSS pointer in the GDT */ | |
1232 | __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss); | |
1233 | #endif | |
1234 | ||
1235 | clear_all_debug_regs(); | |
1236 | ||
1237 | /* | |
1238 | * Force FPU initialization: | |
1239 | */ | |
1240 | if (cpu_has_xsave) | |
1241 | current_thread_info()->status = TS_XSAVE; | |
1242 | else | |
1243 | current_thread_info()->status = 0; | |
1244 | clear_used_math(); | |
1245 | mxcsr_feature_mask_init(); | |
1246 | ||
1247 | /* | |
1248 | * Boot processor to setup the FP and extended state context info. | |
1249 | */ | |
1250 | if (smp_processor_id() == boot_cpu_id) | |
1251 | init_thread_xstate(); | |
1252 | ||
1253 | xsave_init(); | |
1254 | } | |
1255 | #endif |