]>
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/archrandom.h> | |
19 | #include <asm/hypervisor.h> | |
20 | #include <asm/processor.h> | |
21 | #include <asm/debugreg.h> | |
22 | #include <asm/sections.h> | |
23 | #include <linux/topology.h> | |
24 | #include <linux/cpumask.h> | |
25 | #include <asm/pgtable.h> | |
26 | #include <linux/atomic.h> | |
27 | #include <asm/proto.h> | |
28 | #include <asm/setup.h> | |
29 | #include <asm/apic.h> | |
30 | #include <asm/desc.h> | |
31 | #include <asm/i387.h> | |
32 | #include <asm/fpu-internal.h> | |
33 | #include <asm/mtrr.h> | |
34 | #include <linux/numa.h> | |
35 | #include <asm/asm.h> | |
36 | #include <asm/cpu.h> | |
37 | #include <asm/mce.h> | |
38 | #include <asm/msr.h> | |
39 | #include <asm/pat.h> | |
40 | #include <asm/microcode.h> | |
41 | #include <asm/microcode_intel.h> | |
42 | ||
43 | #ifdef CONFIG_X86_LOCAL_APIC | |
44 | #include <asm/uv/uv.h> | |
45 | #endif | |
46 | ||
47 | #include "cpu.h" | |
48 | ||
49 | /* all of these masks are initialized in setup_cpu_local_masks() */ | |
50 | cpumask_var_t cpu_initialized_mask; | |
51 | cpumask_var_t cpu_callout_mask; | |
52 | cpumask_var_t cpu_callin_mask; | |
53 | ||
54 | /* representing cpus for which sibling maps can be computed */ | |
55 | cpumask_var_t cpu_sibling_setup_mask; | |
56 | ||
57 | /* correctly size the local cpu masks */ | |
58 | void __init setup_cpu_local_masks(void) | |
59 | { | |
60 | alloc_bootmem_cpumask_var(&cpu_initialized_mask); | |
61 | alloc_bootmem_cpumask_var(&cpu_callin_mask); | |
62 | alloc_bootmem_cpumask_var(&cpu_callout_mask); | |
63 | alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask); | |
64 | } | |
65 | ||
66 | static void __cpuinit default_init(struct cpuinfo_x86 *c) | |
67 | { | |
68 | #ifdef CONFIG_X86_64 | |
69 | cpu_detect_cache_sizes(c); | |
70 | #else | |
71 | /* Not much we can do here... */ | |
72 | /* Check if at least it has cpuid */ | |
73 | if (c->cpuid_level == -1) { | |
74 | /* No cpuid. It must be an ancient CPU */ | |
75 | if (c->x86 == 4) | |
76 | strcpy(c->x86_model_id, "486"); | |
77 | else if (c->x86 == 3) | |
78 | strcpy(c->x86_model_id, "386"); | |
79 | } | |
80 | #endif | |
81 | } | |
82 | ||
83 | static const struct cpu_dev __cpuinitconst default_cpu = { | |
84 | .c_init = default_init, | |
85 | .c_vendor = "Unknown", | |
86 | .c_x86_vendor = X86_VENDOR_UNKNOWN, | |
87 | }; | |
88 | ||
89 | static const struct cpu_dev *this_cpu __cpuinitdata = &default_cpu; | |
90 | ||
91 | DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = { | |
92 | #ifdef CONFIG_X86_64 | |
93 | /* | |
94 | * We need valid kernel segments for data and code in long mode too | |
95 | * IRET will check the segment types kkeil 2000/10/28 | |
96 | * Also sysret mandates a special GDT layout | |
97 | * | |
98 | * TLS descriptors are currently at a different place compared to i386. | |
99 | * Hopefully nobody expects them at a fixed place (Wine?) | |
100 | */ | |
101 | [GDT_ENTRY_KERNEL32_CS] = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff), | |
102 | [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff), | |
103 | [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc093, 0, 0xfffff), | |
104 | [GDT_ENTRY_DEFAULT_USER32_CS] = GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff), | |
105 | [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff), | |
106 | [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff), | |
107 | #else | |
108 | [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xc09a, 0, 0xfffff), | |
109 | [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), | |
110 | [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff), | |
111 | [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff), | |
112 | /* | |
113 | * Segments used for calling PnP BIOS have byte granularity. | |
114 | * They code segments and data segments have fixed 64k limits, | |
115 | * the transfer segment sizes are set at run time. | |
116 | */ | |
117 | /* 32-bit code */ | |
118 | [GDT_ENTRY_PNPBIOS_CS32] = GDT_ENTRY_INIT(0x409a, 0, 0xffff), | |
119 | /* 16-bit code */ | |
120 | [GDT_ENTRY_PNPBIOS_CS16] = GDT_ENTRY_INIT(0x009a, 0, 0xffff), | |
121 | /* 16-bit data */ | |
122 | [GDT_ENTRY_PNPBIOS_DS] = GDT_ENTRY_INIT(0x0092, 0, 0xffff), | |
123 | /* 16-bit data */ | |
124 | [GDT_ENTRY_PNPBIOS_TS1] = GDT_ENTRY_INIT(0x0092, 0, 0), | |
125 | /* 16-bit data */ | |
126 | [GDT_ENTRY_PNPBIOS_TS2] = GDT_ENTRY_INIT(0x0092, 0, 0), | |
127 | /* | |
128 | * The APM segments have byte granularity and their bases | |
129 | * are set at run time. All have 64k limits. | |
130 | */ | |
131 | /* 32-bit code */ | |
132 | [GDT_ENTRY_APMBIOS_BASE] = GDT_ENTRY_INIT(0x409a, 0, 0xffff), | |
133 | /* 16-bit code */ | |
134 | [GDT_ENTRY_APMBIOS_BASE+1] = GDT_ENTRY_INIT(0x009a, 0, 0xffff), | |
135 | /* data */ | |
136 | [GDT_ENTRY_APMBIOS_BASE+2] = GDT_ENTRY_INIT(0x4092, 0, 0xffff), | |
137 | ||
138 | [GDT_ENTRY_ESPFIX_SS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), | |
139 | [GDT_ENTRY_PERCPU] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff), | |
140 | GDT_STACK_CANARY_INIT | |
141 | #endif | |
142 | } }; | |
143 | EXPORT_PER_CPU_SYMBOL_GPL(gdt_page); | |
144 | ||
145 | static int __init x86_xsave_setup(char *s) | |
146 | { | |
147 | setup_clear_cpu_cap(X86_FEATURE_XSAVE); | |
148 | setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT); | |
149 | setup_clear_cpu_cap(X86_FEATURE_AVX); | |
150 | setup_clear_cpu_cap(X86_FEATURE_AVX2); | |
151 | return 1; | |
152 | } | |
153 | __setup("noxsave", x86_xsave_setup); | |
154 | ||
155 | static int __init x86_xsaveopt_setup(char *s) | |
156 | { | |
157 | setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT); | |
158 | return 1; | |
159 | } | |
160 | __setup("noxsaveopt", x86_xsaveopt_setup); | |
161 | ||
162 | #ifdef CONFIG_X86_32 | |
163 | static int cachesize_override __cpuinitdata = -1; | |
164 | static int disable_x86_serial_nr __cpuinitdata = 1; | |
165 | ||
166 | static int __init cachesize_setup(char *str) | |
167 | { | |
168 | get_option(&str, &cachesize_override); | |
169 | return 1; | |
170 | } | |
171 | __setup("cachesize=", cachesize_setup); | |
172 | ||
173 | static int __init x86_fxsr_setup(char *s) | |
174 | { | |
175 | setup_clear_cpu_cap(X86_FEATURE_FXSR); | |
176 | setup_clear_cpu_cap(X86_FEATURE_XMM); | |
177 | return 1; | |
178 | } | |
179 | __setup("nofxsr", x86_fxsr_setup); | |
180 | ||
181 | static int __init x86_sep_setup(char *s) | |
182 | { | |
183 | setup_clear_cpu_cap(X86_FEATURE_SEP); | |
184 | return 1; | |
185 | } | |
186 | __setup("nosep", x86_sep_setup); | |
187 | ||
188 | /* Standard macro to see if a specific flag is changeable */ | |
189 | static inline int flag_is_changeable_p(u32 flag) | |
190 | { | |
191 | u32 f1, f2; | |
192 | ||
193 | /* | |
194 | * Cyrix and IDT cpus allow disabling of CPUID | |
195 | * so the code below may return different results | |
196 | * when it is executed before and after enabling | |
197 | * the CPUID. Add "volatile" to not allow gcc to | |
198 | * optimize the subsequent calls to this function. | |
199 | */ | |
200 | asm volatile ("pushfl \n\t" | |
201 | "pushfl \n\t" | |
202 | "popl %0 \n\t" | |
203 | "movl %0, %1 \n\t" | |
204 | "xorl %2, %0 \n\t" | |
205 | "pushl %0 \n\t" | |
206 | "popfl \n\t" | |
207 | "pushfl \n\t" | |
208 | "popl %0 \n\t" | |
209 | "popfl \n\t" | |
210 | ||
211 | : "=&r" (f1), "=&r" (f2) | |
212 | : "ir" (flag)); | |
213 | ||
214 | return ((f1^f2) & flag) != 0; | |
215 | } | |
216 | ||
217 | /* Probe for the CPUID instruction */ | |
218 | int __cpuinit have_cpuid_p(void) | |
219 | { | |
220 | return flag_is_changeable_p(X86_EFLAGS_ID); | |
221 | } | |
222 | ||
223 | static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c) | |
224 | { | |
225 | unsigned long lo, hi; | |
226 | ||
227 | if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr) | |
228 | return; | |
229 | ||
230 | /* Disable processor serial number: */ | |
231 | ||
232 | rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi); | |
233 | lo |= 0x200000; | |
234 | wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi); | |
235 | ||
236 | printk(KERN_NOTICE "CPU serial number disabled.\n"); | |
237 | clear_cpu_cap(c, X86_FEATURE_PN); | |
238 | ||
239 | /* Disabling the serial number may affect the cpuid level */ | |
240 | c->cpuid_level = cpuid_eax(0); | |
241 | } | |
242 | ||
243 | static int __init x86_serial_nr_setup(char *s) | |
244 | { | |
245 | disable_x86_serial_nr = 0; | |
246 | return 1; | |
247 | } | |
248 | __setup("serialnumber", x86_serial_nr_setup); | |
249 | #else | |
250 | static inline int flag_is_changeable_p(u32 flag) | |
251 | { | |
252 | return 1; | |
253 | } | |
254 | static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c) | |
255 | { | |
256 | } | |
257 | #endif | |
258 | ||
259 | static __init int setup_disable_smep(char *arg) | |
260 | { | |
261 | setup_clear_cpu_cap(X86_FEATURE_SMEP); | |
262 | return 1; | |
263 | } | |
264 | __setup("nosmep", setup_disable_smep); | |
265 | ||
266 | static __always_inline void setup_smep(struct cpuinfo_x86 *c) | |
267 | { | |
268 | if (cpu_has(c, X86_FEATURE_SMEP)) | |
269 | set_in_cr4(X86_CR4_SMEP); | |
270 | } | |
271 | ||
272 | static __init int setup_disable_smap(char *arg) | |
273 | { | |
274 | setup_clear_cpu_cap(X86_FEATURE_SMAP); | |
275 | return 1; | |
276 | } | |
277 | __setup("nosmap", setup_disable_smap); | |
278 | ||
279 | static __always_inline void setup_smap(struct cpuinfo_x86 *c) | |
280 | { | |
281 | unsigned long eflags; | |
282 | ||
283 | /* This should have been cleared long ago */ | |
284 | raw_local_save_flags(eflags); | |
285 | BUG_ON(eflags & X86_EFLAGS_AC); | |
286 | ||
287 | if (cpu_has(c, X86_FEATURE_SMAP)) | |
288 | set_in_cr4(X86_CR4_SMAP); | |
289 | } | |
290 | ||
291 | /* | |
292 | * Some CPU features depend on higher CPUID levels, which may not always | |
293 | * be available due to CPUID level capping or broken virtualization | |
294 | * software. Add those features to this table to auto-disable them. | |
295 | */ | |
296 | struct cpuid_dependent_feature { | |
297 | u32 feature; | |
298 | u32 level; | |
299 | }; | |
300 | ||
301 | static const struct cpuid_dependent_feature __cpuinitconst | |
302 | cpuid_dependent_features[] = { | |
303 | { X86_FEATURE_MWAIT, 0x00000005 }, | |
304 | { X86_FEATURE_DCA, 0x00000009 }, | |
305 | { X86_FEATURE_XSAVE, 0x0000000d }, | |
306 | { 0, 0 } | |
307 | }; | |
308 | ||
309 | static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn) | |
310 | { | |
311 | const struct cpuid_dependent_feature *df; | |
312 | ||
313 | for (df = cpuid_dependent_features; df->feature; df++) { | |
314 | ||
315 | if (!cpu_has(c, df->feature)) | |
316 | continue; | |
317 | /* | |
318 | * Note: cpuid_level is set to -1 if unavailable, but | |
319 | * extended_extended_level is set to 0 if unavailable | |
320 | * and the legitimate extended levels are all negative | |
321 | * when signed; hence the weird messing around with | |
322 | * signs here... | |
323 | */ | |
324 | if (!((s32)df->level < 0 ? | |
325 | (u32)df->level > (u32)c->extended_cpuid_level : | |
326 | (s32)df->level > (s32)c->cpuid_level)) | |
327 | continue; | |
328 | ||
329 | clear_cpu_cap(c, df->feature); | |
330 | if (!warn) | |
331 | continue; | |
332 | ||
333 | printk(KERN_WARNING | |
334 | "CPU: CPU feature %s disabled, no CPUID level 0x%x\n", | |
335 | x86_cap_flags[df->feature], df->level); | |
336 | } | |
337 | } | |
338 | ||
339 | /* | |
340 | * Naming convention should be: <Name> [(<Codename>)] | |
341 | * This table only is used unless init_<vendor>() below doesn't set it; | |
342 | * in particular, if CPUID levels 0x80000002..4 are supported, this | |
343 | * isn't used | |
344 | */ | |
345 | ||
346 | /* Look up CPU names by table lookup. */ | |
347 | static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c) | |
348 | { | |
349 | const struct cpu_model_info *info; | |
350 | ||
351 | if (c->x86_model >= 16) | |
352 | return NULL; /* Range check */ | |
353 | ||
354 | if (!this_cpu) | |
355 | return NULL; | |
356 | ||
357 | info = this_cpu->c_models; | |
358 | ||
359 | while (info && info->family) { | |
360 | if (info->family == c->x86) | |
361 | return info->model_names[c->x86_model]; | |
362 | info++; | |
363 | } | |
364 | return NULL; /* Not found */ | |
365 | } | |
366 | ||
367 | __u32 cpu_caps_cleared[NCAPINTS] __cpuinitdata; | |
368 | __u32 cpu_caps_set[NCAPINTS] __cpuinitdata; | |
369 | ||
370 | void load_percpu_segment(int cpu) | |
371 | { | |
372 | #ifdef CONFIG_X86_32 | |
373 | loadsegment(fs, __KERNEL_PERCPU); | |
374 | #else | |
375 | loadsegment(gs, 0); | |
376 | wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu)); | |
377 | #endif | |
378 | load_stack_canary_segment(); | |
379 | } | |
380 | ||
381 | /* | |
382 | * Current gdt points %fs at the "master" per-cpu area: after this, | |
383 | * it's on the real one. | |
384 | */ | |
385 | void switch_to_new_gdt(int cpu) | |
386 | { | |
387 | struct desc_ptr gdt_descr; | |
388 | ||
389 | gdt_descr.address = (long)get_cpu_gdt_table(cpu); | |
390 | gdt_descr.size = GDT_SIZE - 1; | |
391 | load_gdt(&gdt_descr); | |
392 | /* Reload the per-cpu base */ | |
393 | ||
394 | load_percpu_segment(cpu); | |
395 | } | |
396 | ||
397 | static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {}; | |
398 | ||
399 | static void __cpuinit get_model_name(struct cpuinfo_x86 *c) | |
400 | { | |
401 | unsigned int *v; | |
402 | char *p, *q; | |
403 | ||
404 | if (c->extended_cpuid_level < 0x80000004) | |
405 | return; | |
406 | ||
407 | v = (unsigned int *)c->x86_model_id; | |
408 | cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]); | |
409 | cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]); | |
410 | cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]); | |
411 | c->x86_model_id[48] = 0; | |
412 | ||
413 | /* | |
414 | * Intel chips right-justify this string for some dumb reason; | |
415 | * undo that brain damage: | |
416 | */ | |
417 | p = q = &c->x86_model_id[0]; | |
418 | while (*p == ' ') | |
419 | p++; | |
420 | if (p != q) { | |
421 | while (*p) | |
422 | *q++ = *p++; | |
423 | while (q <= &c->x86_model_id[48]) | |
424 | *q++ = '\0'; /* Zero-pad the rest */ | |
425 | } | |
426 | } | |
427 | ||
428 | void __cpuinit cpu_detect_cache_sizes(struct cpuinfo_x86 *c) | |
429 | { | |
430 | unsigned int n, dummy, ebx, ecx, edx, l2size; | |
431 | ||
432 | n = c->extended_cpuid_level; | |
433 | ||
434 | if (n >= 0x80000005) { | |
435 | cpuid(0x80000005, &dummy, &ebx, &ecx, &edx); | |
436 | c->x86_cache_size = (ecx>>24) + (edx>>24); | |
437 | #ifdef CONFIG_X86_64 | |
438 | /* On K8 L1 TLB is inclusive, so don't count it */ | |
439 | c->x86_tlbsize = 0; | |
440 | #endif | |
441 | } | |
442 | ||
443 | if (n < 0x80000006) /* Some chips just has a large L1. */ | |
444 | return; | |
445 | ||
446 | cpuid(0x80000006, &dummy, &ebx, &ecx, &edx); | |
447 | l2size = ecx >> 16; | |
448 | ||
449 | #ifdef CONFIG_X86_64 | |
450 | c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff); | |
451 | #else | |
452 | /* do processor-specific cache resizing */ | |
453 | if (this_cpu->c_size_cache) | |
454 | l2size = this_cpu->c_size_cache(c, l2size); | |
455 | ||
456 | /* Allow user to override all this if necessary. */ | |
457 | if (cachesize_override != -1) | |
458 | l2size = cachesize_override; | |
459 | ||
460 | if (l2size == 0) | |
461 | return; /* Again, no L2 cache is possible */ | |
462 | #endif | |
463 | ||
464 | c->x86_cache_size = l2size; | |
465 | } | |
466 | ||
467 | u16 __read_mostly tlb_lli_4k[NR_INFO]; | |
468 | u16 __read_mostly tlb_lli_2m[NR_INFO]; | |
469 | u16 __read_mostly tlb_lli_4m[NR_INFO]; | |
470 | u16 __read_mostly tlb_lld_4k[NR_INFO]; | |
471 | u16 __read_mostly tlb_lld_2m[NR_INFO]; | |
472 | u16 __read_mostly tlb_lld_4m[NR_INFO]; | |
473 | ||
474 | /* | |
475 | * tlb_flushall_shift shows the balance point in replacing cr3 write | |
476 | * with multiple 'invlpg'. It will do this replacement when | |
477 | * flush_tlb_lines <= active_lines/2^tlb_flushall_shift. | |
478 | * If tlb_flushall_shift is -1, means the replacement will be disabled. | |
479 | */ | |
480 | s8 __read_mostly tlb_flushall_shift = -1; | |
481 | ||
482 | void __cpuinit cpu_detect_tlb(struct cpuinfo_x86 *c) | |
483 | { | |
484 | if (this_cpu->c_detect_tlb) | |
485 | this_cpu->c_detect_tlb(c); | |
486 | ||
487 | printk(KERN_INFO "Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n" \ | |
488 | "Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d\n" \ | |
489 | "tlb_flushall_shift: %d\n", | |
490 | tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES], | |
491 | tlb_lli_4m[ENTRIES], tlb_lld_4k[ENTRIES], | |
492 | tlb_lld_2m[ENTRIES], tlb_lld_4m[ENTRIES], | |
493 | tlb_flushall_shift); | |
494 | } | |
495 | ||
496 | void __cpuinit detect_ht(struct cpuinfo_x86 *c) | |
497 | { | |
498 | #ifdef CONFIG_X86_HT | |
499 | u32 eax, ebx, ecx, edx; | |
500 | int index_msb, core_bits; | |
501 | static bool printed; | |
502 | ||
503 | if (!cpu_has(c, X86_FEATURE_HT)) | |
504 | return; | |
505 | ||
506 | if (cpu_has(c, X86_FEATURE_CMP_LEGACY)) | |
507 | goto out; | |
508 | ||
509 | if (cpu_has(c, X86_FEATURE_XTOPOLOGY)) | |
510 | return; | |
511 | ||
512 | cpuid(1, &eax, &ebx, &ecx, &edx); | |
513 | ||
514 | smp_num_siblings = (ebx & 0xff0000) >> 16; | |
515 | ||
516 | if (smp_num_siblings == 1) { | |
517 | printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n"); | |
518 | goto out; | |
519 | } | |
520 | ||
521 | if (smp_num_siblings <= 1) | |
522 | goto out; | |
523 | ||
524 | index_msb = get_count_order(smp_num_siblings); | |
525 | c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb); | |
526 | ||
527 | smp_num_siblings = smp_num_siblings / c->x86_max_cores; | |
528 | ||
529 | index_msb = get_count_order(smp_num_siblings); | |
530 | ||
531 | core_bits = get_count_order(c->x86_max_cores); | |
532 | ||
533 | c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) & | |
534 | ((1 << core_bits) - 1); | |
535 | ||
536 | out: | |
537 | if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) { | |
538 | printk(KERN_INFO "CPU: Physical Processor ID: %d\n", | |
539 | c->phys_proc_id); | |
540 | printk(KERN_INFO "CPU: Processor Core ID: %d\n", | |
541 | c->cpu_core_id); | |
542 | printed = 1; | |
543 | } | |
544 | #endif | |
545 | } | |
546 | ||
547 | static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c) | |
548 | { | |
549 | char *v = c->x86_vendor_id; | |
550 | int i; | |
551 | ||
552 | for (i = 0; i < X86_VENDOR_NUM; i++) { | |
553 | if (!cpu_devs[i]) | |
554 | break; | |
555 | ||
556 | if (!strcmp(v, cpu_devs[i]->c_ident[0]) || | |
557 | (cpu_devs[i]->c_ident[1] && | |
558 | !strcmp(v, cpu_devs[i]->c_ident[1]))) { | |
559 | ||
560 | this_cpu = cpu_devs[i]; | |
561 | c->x86_vendor = this_cpu->c_x86_vendor; | |
562 | return; | |
563 | } | |
564 | } | |
565 | ||
566 | printk_once(KERN_ERR | |
567 | "CPU: vendor_id '%s' unknown, using generic init.\n" \ | |
568 | "CPU: Your system may be unstable.\n", v); | |
569 | ||
570 | c->x86_vendor = X86_VENDOR_UNKNOWN; | |
571 | this_cpu = &default_cpu; | |
572 | } | |
573 | ||
574 | void __cpuinit cpu_detect(struct cpuinfo_x86 *c) | |
575 | { | |
576 | /* Get vendor name */ | |
577 | cpuid(0x00000000, (unsigned int *)&c->cpuid_level, | |
578 | (unsigned int *)&c->x86_vendor_id[0], | |
579 | (unsigned int *)&c->x86_vendor_id[8], | |
580 | (unsigned int *)&c->x86_vendor_id[4]); | |
581 | ||
582 | c->x86 = 4; | |
583 | /* Intel-defined flags: level 0x00000001 */ | |
584 | if (c->cpuid_level >= 0x00000001) { | |
585 | u32 junk, tfms, cap0, misc; | |
586 | ||
587 | cpuid(0x00000001, &tfms, &misc, &junk, &cap0); | |
588 | c->x86 = (tfms >> 8) & 0xf; | |
589 | c->x86_model = (tfms >> 4) & 0xf; | |
590 | c->x86_mask = tfms & 0xf; | |
591 | ||
592 | if (c->x86 == 0xf) | |
593 | c->x86 += (tfms >> 20) & 0xff; | |
594 | if (c->x86 >= 0x6) | |
595 | c->x86_model += ((tfms >> 16) & 0xf) << 4; | |
596 | ||
597 | if (cap0 & (1<<19)) { | |
598 | c->x86_clflush_size = ((misc >> 8) & 0xff) * 8; | |
599 | c->x86_cache_alignment = c->x86_clflush_size; | |
600 | } | |
601 | } | |
602 | } | |
603 | ||
604 | void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c) | |
605 | { | |
606 | u32 tfms, xlvl; | |
607 | u32 ebx; | |
608 | ||
609 | /* Intel-defined flags: level 0x00000001 */ | |
610 | if (c->cpuid_level >= 0x00000001) { | |
611 | u32 capability, excap; | |
612 | ||
613 | cpuid(0x00000001, &tfms, &ebx, &excap, &capability); | |
614 | c->x86_capability[0] = capability; | |
615 | c->x86_capability[4] = excap; | |
616 | } | |
617 | ||
618 | /* Additional Intel-defined flags: level 0x00000007 */ | |
619 | if (c->cpuid_level >= 0x00000007) { | |
620 | u32 eax, ebx, ecx, edx; | |
621 | ||
622 | cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx); | |
623 | ||
624 | c->x86_capability[9] = ebx; | |
625 | } | |
626 | ||
627 | /* AMD-defined flags: level 0x80000001 */ | |
628 | xlvl = cpuid_eax(0x80000000); | |
629 | c->extended_cpuid_level = xlvl; | |
630 | ||
631 | if ((xlvl & 0xffff0000) == 0x80000000) { | |
632 | if (xlvl >= 0x80000001) { | |
633 | c->x86_capability[1] = cpuid_edx(0x80000001); | |
634 | c->x86_capability[6] = cpuid_ecx(0x80000001); | |
635 | } | |
636 | } | |
637 | ||
638 | if (c->extended_cpuid_level >= 0x80000008) { | |
639 | u32 eax = cpuid_eax(0x80000008); | |
640 | ||
641 | c->x86_virt_bits = (eax >> 8) & 0xff; | |
642 | c->x86_phys_bits = eax & 0xff; | |
643 | } | |
644 | #ifdef CONFIG_X86_32 | |
645 | else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36)) | |
646 | c->x86_phys_bits = 36; | |
647 | #endif | |
648 | ||
649 | if (c->extended_cpuid_level >= 0x80000007) | |
650 | c->x86_power = cpuid_edx(0x80000007); | |
651 | ||
652 | init_scattered_cpuid_features(c); | |
653 | } | |
654 | ||
655 | static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c) | |
656 | { | |
657 | #ifdef CONFIG_X86_32 | |
658 | int i; | |
659 | ||
660 | /* | |
661 | * First of all, decide if this is a 486 or higher | |
662 | * It's a 486 if we can modify the AC flag | |
663 | */ | |
664 | if (flag_is_changeable_p(X86_EFLAGS_AC)) | |
665 | c->x86 = 4; | |
666 | else | |
667 | c->x86 = 3; | |
668 | ||
669 | for (i = 0; i < X86_VENDOR_NUM; i++) | |
670 | if (cpu_devs[i] && cpu_devs[i]->c_identify) { | |
671 | c->x86_vendor_id[0] = 0; | |
672 | cpu_devs[i]->c_identify(c); | |
673 | if (c->x86_vendor_id[0]) { | |
674 | get_cpu_vendor(c); | |
675 | break; | |
676 | } | |
677 | } | |
678 | #endif | |
679 | } | |
680 | ||
681 | /* | |
682 | * Do minimum CPU detection early. | |
683 | * Fields really needed: vendor, cpuid_level, family, model, mask, | |
684 | * cache alignment. | |
685 | * The others are not touched to avoid unwanted side effects. | |
686 | * | |
687 | * WARNING: this function is only called on the BP. Don't add code here | |
688 | * that is supposed to run on all CPUs. | |
689 | */ | |
690 | static void __init early_identify_cpu(struct cpuinfo_x86 *c) | |
691 | { | |
692 | #ifdef CONFIG_X86_64 | |
693 | c->x86_clflush_size = 64; | |
694 | c->x86_phys_bits = 36; | |
695 | c->x86_virt_bits = 48; | |
696 | #else | |
697 | c->x86_clflush_size = 32; | |
698 | c->x86_phys_bits = 32; | |
699 | c->x86_virt_bits = 32; | |
700 | #endif | |
701 | c->x86_cache_alignment = c->x86_clflush_size; | |
702 | ||
703 | memset(&c->x86_capability, 0, sizeof c->x86_capability); | |
704 | c->extended_cpuid_level = 0; | |
705 | ||
706 | if (!have_cpuid_p()) | |
707 | identify_cpu_without_cpuid(c); | |
708 | ||
709 | /* cyrix could have cpuid enabled via c_identify()*/ | |
710 | if (!have_cpuid_p()) | |
711 | return; | |
712 | ||
713 | cpu_detect(c); | |
714 | ||
715 | get_cpu_vendor(c); | |
716 | ||
717 | get_cpu_cap(c); | |
718 | ||
719 | if (this_cpu->c_early_init) | |
720 | this_cpu->c_early_init(c); | |
721 | ||
722 | c->cpu_index = 0; | |
723 | filter_cpuid_features(c, false); | |
724 | ||
725 | if (this_cpu->c_bsp_init) | |
726 | this_cpu->c_bsp_init(c); | |
727 | } | |
728 | ||
729 | void __init early_cpu_init(void) | |
730 | { | |
731 | const struct cpu_dev *const *cdev; | |
732 | int count = 0; | |
733 | ||
734 | #ifdef CONFIG_PROCESSOR_SELECT | |
735 | printk(KERN_INFO "KERNEL supported cpus:\n"); | |
736 | #endif | |
737 | ||
738 | for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) { | |
739 | const struct cpu_dev *cpudev = *cdev; | |
740 | ||
741 | if (count >= X86_VENDOR_NUM) | |
742 | break; | |
743 | cpu_devs[count] = cpudev; | |
744 | count++; | |
745 | ||
746 | #ifdef CONFIG_PROCESSOR_SELECT | |
747 | { | |
748 | unsigned int j; | |
749 | ||
750 | for (j = 0; j < 2; j++) { | |
751 | if (!cpudev->c_ident[j]) | |
752 | continue; | |
753 | printk(KERN_INFO " %s %s\n", cpudev->c_vendor, | |
754 | cpudev->c_ident[j]); | |
755 | } | |
756 | } | |
757 | #endif | |
758 | } | |
759 | early_identify_cpu(&boot_cpu_data); | |
760 | } | |
761 | ||
762 | /* | |
763 | * The NOPL instruction is supposed to exist on all CPUs of family >= 6; | |
764 | * unfortunately, that's not true in practice because of early VIA | |
765 | * chips and (more importantly) broken virtualizers that are not easy | |
766 | * to detect. In the latter case it doesn't even *fail* reliably, so | |
767 | * probing for it doesn't even work. Disable it completely on 32-bit | |
768 | * unless we can find a reliable way to detect all the broken cases. | |
769 | * Enable it explicitly on 64-bit for non-constant inputs of cpu_has(). | |
770 | */ | |
771 | static void __cpuinit detect_nopl(struct cpuinfo_x86 *c) | |
772 | { | |
773 | #ifdef CONFIG_X86_32 | |
774 | clear_cpu_cap(c, X86_FEATURE_NOPL); | |
775 | #else | |
776 | set_cpu_cap(c, X86_FEATURE_NOPL); | |
777 | #endif | |
778 | } | |
779 | ||
780 | static void __cpuinit generic_identify(struct cpuinfo_x86 *c) | |
781 | { | |
782 | c->extended_cpuid_level = 0; | |
783 | ||
784 | if (!have_cpuid_p()) | |
785 | identify_cpu_without_cpuid(c); | |
786 | ||
787 | /* cyrix could have cpuid enabled via c_identify()*/ | |
788 | if (!have_cpuid_p()) | |
789 | return; | |
790 | ||
791 | cpu_detect(c); | |
792 | ||
793 | get_cpu_vendor(c); | |
794 | ||
795 | get_cpu_cap(c); | |
796 | ||
797 | if (c->cpuid_level >= 0x00000001) { | |
798 | c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF; | |
799 | #ifdef CONFIG_X86_32 | |
800 | # ifdef CONFIG_X86_HT | |
801 | c->apicid = apic->phys_pkg_id(c->initial_apicid, 0); | |
802 | # else | |
803 | c->apicid = c->initial_apicid; | |
804 | # endif | |
805 | #endif | |
806 | c->phys_proc_id = c->initial_apicid; | |
807 | } | |
808 | ||
809 | get_model_name(c); /* Default name */ | |
810 | ||
811 | detect_nopl(c); | |
812 | } | |
813 | ||
814 | /* | |
815 | * This does the hard work of actually picking apart the CPU stuff... | |
816 | */ | |
817 | static void __cpuinit identify_cpu(struct cpuinfo_x86 *c) | |
818 | { | |
819 | int i; | |
820 | ||
821 | c->loops_per_jiffy = loops_per_jiffy; | |
822 | c->x86_cache_size = -1; | |
823 | c->x86_vendor = X86_VENDOR_UNKNOWN; | |
824 | c->x86_model = c->x86_mask = 0; /* So far unknown... */ | |
825 | c->x86_vendor_id[0] = '\0'; /* Unset */ | |
826 | c->x86_model_id[0] = '\0'; /* Unset */ | |
827 | c->x86_max_cores = 1; | |
828 | c->x86_coreid_bits = 0; | |
829 | #ifdef CONFIG_X86_64 | |
830 | c->x86_clflush_size = 64; | |
831 | c->x86_phys_bits = 36; | |
832 | c->x86_virt_bits = 48; | |
833 | #else | |
834 | c->cpuid_level = -1; /* CPUID not detected */ | |
835 | c->x86_clflush_size = 32; | |
836 | c->x86_phys_bits = 32; | |
837 | c->x86_virt_bits = 32; | |
838 | #endif | |
839 | c->x86_cache_alignment = c->x86_clflush_size; | |
840 | memset(&c->x86_capability, 0, sizeof c->x86_capability); | |
841 | ||
842 | generic_identify(c); | |
843 | ||
844 | if (this_cpu->c_identify) | |
845 | this_cpu->c_identify(c); | |
846 | ||
847 | /* Clear/Set all flags overriden by options, after probe */ | |
848 | for (i = 0; i < NCAPINTS; i++) { | |
849 | c->x86_capability[i] &= ~cpu_caps_cleared[i]; | |
850 | c->x86_capability[i] |= cpu_caps_set[i]; | |
851 | } | |
852 | ||
853 | #ifdef CONFIG_X86_64 | |
854 | c->apicid = apic->phys_pkg_id(c->initial_apicid, 0); | |
855 | #endif | |
856 | ||
857 | /* | |
858 | * Vendor-specific initialization. In this section we | |
859 | * canonicalize the feature flags, meaning if there are | |
860 | * features a certain CPU supports which CPUID doesn't | |
861 | * tell us, CPUID claiming incorrect flags, or other bugs, | |
862 | * we handle them here. | |
863 | * | |
864 | * At the end of this section, c->x86_capability better | |
865 | * indicate the features this CPU genuinely supports! | |
866 | */ | |
867 | if (this_cpu->c_init) | |
868 | this_cpu->c_init(c); | |
869 | ||
870 | /* Disable the PN if appropriate */ | |
871 | squash_the_stupid_serial_number(c); | |
872 | ||
873 | /* Set up SMEP/SMAP */ | |
874 | setup_smep(c); | |
875 | setup_smap(c); | |
876 | ||
877 | /* | |
878 | * The vendor-specific functions might have changed features. | |
879 | * Now we do "generic changes." | |
880 | */ | |
881 | ||
882 | /* Filter out anything that depends on CPUID levels we don't have */ | |
883 | filter_cpuid_features(c, true); | |
884 | ||
885 | /* If the model name is still unset, do table lookup. */ | |
886 | if (!c->x86_model_id[0]) { | |
887 | const char *p; | |
888 | p = table_lookup_model(c); | |
889 | if (p) | |
890 | strcpy(c->x86_model_id, p); | |
891 | else | |
892 | /* Last resort... */ | |
893 | sprintf(c->x86_model_id, "%02x/%02x", | |
894 | c->x86, c->x86_model); | |
895 | } | |
896 | ||
897 | #ifdef CONFIG_X86_64 | |
898 | detect_ht(c); | |
899 | #endif | |
900 | ||
901 | init_hypervisor(c); | |
902 | x86_init_rdrand(c); | |
903 | ||
904 | /* | |
905 | * Clear/Set all flags overriden by options, need do it | |
906 | * before following smp all cpus cap AND. | |
907 | */ | |
908 | for (i = 0; i < NCAPINTS; i++) { | |
909 | c->x86_capability[i] &= ~cpu_caps_cleared[i]; | |
910 | c->x86_capability[i] |= cpu_caps_set[i]; | |
911 | } | |
912 | ||
913 | /* | |
914 | * On SMP, boot_cpu_data holds the common feature set between | |
915 | * all CPUs; so make sure that we indicate which features are | |
916 | * common between the CPUs. The first time this routine gets | |
917 | * executed, c == &boot_cpu_data. | |
918 | */ | |
919 | if (c != &boot_cpu_data) { | |
920 | /* AND the already accumulated flags with these */ | |
921 | for (i = 0; i < NCAPINTS; i++) | |
922 | boot_cpu_data.x86_capability[i] &= c->x86_capability[i]; | |
923 | ||
924 | /* OR, i.e. replicate the bug flags */ | |
925 | for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++) | |
926 | c->x86_capability[i] |= boot_cpu_data.x86_capability[i]; | |
927 | } | |
928 | ||
929 | /* Init Machine Check Exception if available. */ | |
930 | mcheck_cpu_init(c); | |
931 | ||
932 | select_idle_routine(c); | |
933 | ||
934 | #ifdef CONFIG_NUMA | |
935 | numa_add_cpu(smp_processor_id()); | |
936 | #endif | |
937 | } | |
938 | ||
939 | #ifdef CONFIG_X86_64 | |
940 | static void vgetcpu_set_mode(void) | |
941 | { | |
942 | if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP)) | |
943 | vgetcpu_mode = VGETCPU_RDTSCP; | |
944 | else | |
945 | vgetcpu_mode = VGETCPU_LSL; | |
946 | } | |
947 | #endif | |
948 | ||
949 | void __init identify_boot_cpu(void) | |
950 | { | |
951 | identify_cpu(&boot_cpu_data); | |
952 | init_amd_e400_c1e_mask(); | |
953 | #ifdef CONFIG_X86_32 | |
954 | sysenter_setup(); | |
955 | enable_sep_cpu(); | |
956 | #else | |
957 | vgetcpu_set_mode(); | |
958 | #endif | |
959 | cpu_detect_tlb(&boot_cpu_data); | |
960 | } | |
961 | ||
962 | void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c) | |
963 | { | |
964 | BUG_ON(c == &boot_cpu_data); | |
965 | identify_cpu(c); | |
966 | #ifdef CONFIG_X86_32 | |
967 | enable_sep_cpu(); | |
968 | #endif | |
969 | mtrr_ap_init(); | |
970 | } | |
971 | ||
972 | struct msr_range { | |
973 | unsigned min; | |
974 | unsigned max; | |
975 | }; | |
976 | ||
977 | static const struct msr_range msr_range_array[] __cpuinitconst = { | |
978 | { 0x00000000, 0x00000418}, | |
979 | { 0xc0000000, 0xc000040b}, | |
980 | { 0xc0010000, 0xc0010142}, | |
981 | { 0xc0011000, 0xc001103b}, | |
982 | }; | |
983 | ||
984 | static void __cpuinit __print_cpu_msr(void) | |
985 | { | |
986 | unsigned index_min, index_max; | |
987 | unsigned index; | |
988 | u64 val; | |
989 | int i; | |
990 | ||
991 | for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) { | |
992 | index_min = msr_range_array[i].min; | |
993 | index_max = msr_range_array[i].max; | |
994 | ||
995 | for (index = index_min; index < index_max; index++) { | |
996 | if (rdmsrl_safe(index, &val)) | |
997 | continue; | |
998 | printk(KERN_INFO " MSR%08x: %016llx\n", index, val); | |
999 | } | |
1000 | } | |
1001 | } | |
1002 | ||
1003 | static int show_msr __cpuinitdata; | |
1004 | ||
1005 | static __init int setup_show_msr(char *arg) | |
1006 | { | |
1007 | int num; | |
1008 | ||
1009 | get_option(&arg, &num); | |
1010 | ||
1011 | if (num > 0) | |
1012 | show_msr = num; | |
1013 | return 1; | |
1014 | } | |
1015 | __setup("show_msr=", setup_show_msr); | |
1016 | ||
1017 | static __init int setup_noclflush(char *arg) | |
1018 | { | |
1019 | setup_clear_cpu_cap(X86_FEATURE_CLFLSH); | |
1020 | return 1; | |
1021 | } | |
1022 | __setup("noclflush", setup_noclflush); | |
1023 | ||
1024 | void __cpuinit print_cpu_info(struct cpuinfo_x86 *c) | |
1025 | { | |
1026 | const char *vendor = NULL; | |
1027 | ||
1028 | if (c->x86_vendor < X86_VENDOR_NUM) { | |
1029 | vendor = this_cpu->c_vendor; | |
1030 | } else { | |
1031 | if (c->cpuid_level >= 0) | |
1032 | vendor = c->x86_vendor_id; | |
1033 | } | |
1034 | ||
1035 | if (vendor && !strstr(c->x86_model_id, vendor)) | |
1036 | printk(KERN_CONT "%s ", vendor); | |
1037 | ||
1038 | if (c->x86_model_id[0]) | |
1039 | printk(KERN_CONT "%s", strim(c->x86_model_id)); | |
1040 | else | |
1041 | printk(KERN_CONT "%d86", c->x86); | |
1042 | ||
1043 | printk(KERN_CONT " (fam: %02x, model: %02x", c->x86, c->x86_model); | |
1044 | ||
1045 | if (c->x86_mask || c->cpuid_level >= 0) | |
1046 | printk(KERN_CONT ", stepping: %02x)\n", c->x86_mask); | |
1047 | else | |
1048 | printk(KERN_CONT ")\n"); | |
1049 | ||
1050 | print_cpu_msr(c); | |
1051 | } | |
1052 | ||
1053 | void __cpuinit print_cpu_msr(struct cpuinfo_x86 *c) | |
1054 | { | |
1055 | if (c->cpu_index < show_msr) | |
1056 | __print_cpu_msr(); | |
1057 | } | |
1058 | ||
1059 | static __init int setup_disablecpuid(char *arg) | |
1060 | { | |
1061 | int bit; | |
1062 | ||
1063 | if (get_option(&arg, &bit) && bit < NCAPINTS*32) | |
1064 | setup_clear_cpu_cap(bit); | |
1065 | else | |
1066 | return 0; | |
1067 | ||
1068 | return 1; | |
1069 | } | |
1070 | __setup("clearcpuid=", setup_disablecpuid); | |
1071 | ||
1072 | #ifdef CONFIG_X86_64 | |
1073 | struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table }; | |
1074 | struct desc_ptr debug_idt_descr = { NR_VECTORS * 16 - 1, | |
1075 | (unsigned long) debug_idt_table }; | |
1076 | ||
1077 | DEFINE_PER_CPU_FIRST(union irq_stack_union, | |
1078 | irq_stack_union) __aligned(PAGE_SIZE); | |
1079 | ||
1080 | /* | |
1081 | * The following four percpu variables are hot. Align current_task to | |
1082 | * cacheline size such that all four fall in the same cacheline. | |
1083 | */ | |
1084 | DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned = | |
1085 | &init_task; | |
1086 | EXPORT_PER_CPU_SYMBOL(current_task); | |
1087 | ||
1088 | DEFINE_PER_CPU(unsigned long, kernel_stack) = | |
1089 | (unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE; | |
1090 | EXPORT_PER_CPU_SYMBOL(kernel_stack); | |
1091 | ||
1092 | DEFINE_PER_CPU(char *, irq_stack_ptr) = | |
1093 | init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64; | |
1094 | ||
1095 | DEFINE_PER_CPU(unsigned int, irq_count) = -1; | |
1096 | ||
1097 | DEFINE_PER_CPU(struct task_struct *, fpu_owner_task); | |
1098 | ||
1099 | /* | |
1100 | * Special IST stacks which the CPU switches to when it calls | |
1101 | * an IST-marked descriptor entry. Up to 7 stacks (hardware | |
1102 | * limit), all of them are 4K, except the debug stack which | |
1103 | * is 8K. | |
1104 | */ | |
1105 | static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = { | |
1106 | [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ, | |
1107 | [DEBUG_STACK - 1] = DEBUG_STKSZ | |
1108 | }; | |
1109 | ||
1110 | static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks | |
1111 | [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]); | |
1112 | ||
1113 | /* May not be marked __init: used by software suspend */ | |
1114 | void syscall_init(void) | |
1115 | { | |
1116 | /* | |
1117 | * LSTAR and STAR live in a bit strange symbiosis. | |
1118 | * They both write to the same internal register. STAR allows to | |
1119 | * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip. | |
1120 | */ | |
1121 | wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32); | |
1122 | wrmsrl(MSR_LSTAR, system_call); | |
1123 | wrmsrl(MSR_CSTAR, ignore_sysret); | |
1124 | ||
1125 | #ifdef CONFIG_IA32_EMULATION | |
1126 | syscall32_cpu_init(); | |
1127 | #endif | |
1128 | ||
1129 | /* Flags to clear on syscall */ | |
1130 | wrmsrl(MSR_SYSCALL_MASK, | |
1131 | X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF| | |
1132 | X86_EFLAGS_IOPL|X86_EFLAGS_AC); | |
1133 | } | |
1134 | ||
1135 | /* | |
1136 | * Copies of the original ist values from the tss are only accessed during | |
1137 | * debugging, no special alignment required. | |
1138 | */ | |
1139 | DEFINE_PER_CPU(struct orig_ist, orig_ist); | |
1140 | ||
1141 | static DEFINE_PER_CPU(unsigned long, debug_stack_addr); | |
1142 | DEFINE_PER_CPU(int, debug_stack_usage); | |
1143 | ||
1144 | int is_debug_stack(unsigned long addr) | |
1145 | { | |
1146 | return __get_cpu_var(debug_stack_usage) || | |
1147 | (addr <= __get_cpu_var(debug_stack_addr) && | |
1148 | addr > (__get_cpu_var(debug_stack_addr) - DEBUG_STKSZ)); | |
1149 | } | |
1150 | ||
1151 | DEFINE_PER_CPU(u32, debug_idt_ctr); | |
1152 | ||
1153 | void debug_stack_set_zero(void) | |
1154 | { | |
1155 | this_cpu_inc(debug_idt_ctr); | |
1156 | load_current_idt(); | |
1157 | } | |
1158 | ||
1159 | void debug_stack_reset(void) | |
1160 | { | |
1161 | if (WARN_ON(!this_cpu_read(debug_idt_ctr))) | |
1162 | return; | |
1163 | if (this_cpu_dec_return(debug_idt_ctr) == 0) | |
1164 | load_current_idt(); | |
1165 | } | |
1166 | ||
1167 | #else /* CONFIG_X86_64 */ | |
1168 | ||
1169 | DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task; | |
1170 | EXPORT_PER_CPU_SYMBOL(current_task); | |
1171 | DEFINE_PER_CPU(struct task_struct *, fpu_owner_task); | |
1172 | ||
1173 | #ifdef CONFIG_CC_STACKPROTECTOR | |
1174 | DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary); | |
1175 | #endif | |
1176 | ||
1177 | #endif /* CONFIG_X86_64 */ | |
1178 | ||
1179 | /* | |
1180 | * Clear all 6 debug registers: | |
1181 | */ | |
1182 | static void clear_all_debug_regs(void) | |
1183 | { | |
1184 | int i; | |
1185 | ||
1186 | for (i = 0; i < 8; i++) { | |
1187 | /* Ignore db4, db5 */ | |
1188 | if ((i == 4) || (i == 5)) | |
1189 | continue; | |
1190 | ||
1191 | set_debugreg(0, i); | |
1192 | } | |
1193 | } | |
1194 | ||
1195 | #ifdef CONFIG_KGDB | |
1196 | /* | |
1197 | * Restore debug regs if using kgdbwait and you have a kernel debugger | |
1198 | * connection established. | |
1199 | */ | |
1200 | static void dbg_restore_debug_regs(void) | |
1201 | { | |
1202 | if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break)) | |
1203 | arch_kgdb_ops.correct_hw_break(); | |
1204 | } | |
1205 | #else /* ! CONFIG_KGDB */ | |
1206 | #define dbg_restore_debug_regs() | |
1207 | #endif /* ! CONFIG_KGDB */ | |
1208 | ||
1209 | /* | |
1210 | * cpu_init() initializes state that is per-CPU. Some data is already | |
1211 | * initialized (naturally) in the bootstrap process, such as the GDT | |
1212 | * and IDT. We reload them nevertheless, this function acts as a | |
1213 | * 'CPU state barrier', nothing should get across. | |
1214 | * A lot of state is already set up in PDA init for 64 bit | |
1215 | */ | |
1216 | #ifdef CONFIG_X86_64 | |
1217 | ||
1218 | void __cpuinit cpu_init(void) | |
1219 | { | |
1220 | struct orig_ist *oist; | |
1221 | struct task_struct *me; | |
1222 | struct tss_struct *t; | |
1223 | unsigned long v; | |
1224 | int cpu; | |
1225 | int i; | |
1226 | ||
1227 | /* | |
1228 | * Load microcode on this cpu if a valid microcode is available. | |
1229 | * This is early microcode loading procedure. | |
1230 | */ | |
1231 | load_ucode_ap(); | |
1232 | ||
1233 | cpu = stack_smp_processor_id(); | |
1234 | t = &per_cpu(init_tss, cpu); | |
1235 | oist = &per_cpu(orig_ist, cpu); | |
1236 | ||
1237 | #ifdef CONFIG_NUMA | |
1238 | if (this_cpu_read(numa_node) == 0 && | |
1239 | early_cpu_to_node(cpu) != NUMA_NO_NODE) | |
1240 | set_numa_node(early_cpu_to_node(cpu)); | |
1241 | #endif | |
1242 | ||
1243 | me = current; | |
1244 | ||
1245 | if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) | |
1246 | panic("CPU#%d already initialized!\n", cpu); | |
1247 | ||
1248 | pr_debug("Initializing CPU#%d\n", cpu); | |
1249 | ||
1250 | clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE); | |
1251 | ||
1252 | /* | |
1253 | * Initialize the per-CPU GDT with the boot GDT, | |
1254 | * and set up the GDT descriptor: | |
1255 | */ | |
1256 | ||
1257 | switch_to_new_gdt(cpu); | |
1258 | loadsegment(fs, 0); | |
1259 | ||
1260 | load_idt((const struct desc_ptr *)&idt_descr); | |
1261 | ||
1262 | memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8); | |
1263 | syscall_init(); | |
1264 | ||
1265 | wrmsrl(MSR_FS_BASE, 0); | |
1266 | wrmsrl(MSR_KERNEL_GS_BASE, 0); | |
1267 | barrier(); | |
1268 | ||
1269 | x86_configure_nx(); | |
1270 | enable_x2apic(); | |
1271 | ||
1272 | /* | |
1273 | * set up and load the per-CPU TSS | |
1274 | */ | |
1275 | if (!oist->ist[0]) { | |
1276 | char *estacks = per_cpu(exception_stacks, cpu); | |
1277 | ||
1278 | for (v = 0; v < N_EXCEPTION_STACKS; v++) { | |
1279 | estacks += exception_stack_sizes[v]; | |
1280 | oist->ist[v] = t->x86_tss.ist[v] = | |
1281 | (unsigned long)estacks; | |
1282 | if (v == DEBUG_STACK-1) | |
1283 | per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks; | |
1284 | } | |
1285 | } | |
1286 | ||
1287 | t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap); | |
1288 | ||
1289 | /* | |
1290 | * <= is required because the CPU will access up to | |
1291 | * 8 bits beyond the end of the IO permission bitmap. | |
1292 | */ | |
1293 | for (i = 0; i <= IO_BITMAP_LONGS; i++) | |
1294 | t->io_bitmap[i] = ~0UL; | |
1295 | ||
1296 | atomic_inc(&init_mm.mm_count); | |
1297 | me->active_mm = &init_mm; | |
1298 | BUG_ON(me->mm); | |
1299 | enter_lazy_tlb(&init_mm, me); | |
1300 | ||
1301 | load_sp0(t, ¤t->thread); | |
1302 | set_tss_desc(cpu, t); | |
1303 | load_TR_desc(); | |
1304 | load_LDT(&init_mm.context); | |
1305 | ||
1306 | clear_all_debug_regs(); | |
1307 | dbg_restore_debug_regs(); | |
1308 | ||
1309 | fpu_init(); | |
1310 | ||
1311 | if (is_uv_system()) | |
1312 | uv_cpu_init(); | |
1313 | } | |
1314 | ||
1315 | #else | |
1316 | ||
1317 | void __cpuinit cpu_init(void) | |
1318 | { | |
1319 | int cpu = smp_processor_id(); | |
1320 | struct task_struct *curr = current; | |
1321 | struct tss_struct *t = &per_cpu(init_tss, cpu); | |
1322 | struct thread_struct *thread = &curr->thread; | |
1323 | ||
1324 | show_ucode_info_early(); | |
1325 | ||
1326 | if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) { | |
1327 | printk(KERN_WARNING "CPU#%d already initialized!\n", cpu); | |
1328 | for (;;) | |
1329 | local_irq_enable(); | |
1330 | } | |
1331 | ||
1332 | printk(KERN_INFO "Initializing CPU#%d\n", cpu); | |
1333 | ||
1334 | if (cpu_has_vme || cpu_has_tsc || cpu_has_de) | |
1335 | clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE); | |
1336 | ||
1337 | load_idt(&idt_descr); | |
1338 | switch_to_new_gdt(cpu); | |
1339 | ||
1340 | /* | |
1341 | * Set up and load the per-CPU TSS and LDT | |
1342 | */ | |
1343 | atomic_inc(&init_mm.mm_count); | |
1344 | curr->active_mm = &init_mm; | |
1345 | BUG_ON(curr->mm); | |
1346 | enter_lazy_tlb(&init_mm, curr); | |
1347 | ||
1348 | load_sp0(t, thread); | |
1349 | set_tss_desc(cpu, t); | |
1350 | load_TR_desc(); | |
1351 | load_LDT(&init_mm.context); | |
1352 | ||
1353 | t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap); | |
1354 | ||
1355 | #ifdef CONFIG_DOUBLEFAULT | |
1356 | /* Set up doublefault TSS pointer in the GDT */ | |
1357 | __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss); | |
1358 | #endif | |
1359 | ||
1360 | clear_all_debug_regs(); | |
1361 | dbg_restore_debug_regs(); | |
1362 | ||
1363 | fpu_init(); | |
1364 | } | |
1365 | #endif |