]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | #include <linux/init.h> |
f0fc4aff YL |
2 | #include <linux/kernel.h> |
3 | #include <linux/sched.h> | |
1da177e4 | 4 | #include <linux/string.h> |
f0fc4aff YL |
5 | #include <linux/bootmem.h> |
6 | #include <linux/bitops.h> | |
7 | #include <linux/module.h> | |
8 | #include <linux/kgdb.h> | |
9 | #include <linux/topology.h> | |
1da177e4 LT |
10 | #include <linux/delay.h> |
11 | #include <linux/smp.h> | |
1da177e4 | 12 | #include <linux/percpu.h> |
1da177e4 LT |
13 | #include <asm/i387.h> |
14 | #include <asm/msr.h> | |
15 | #include <asm/io.h> | |
f0fc4aff | 16 | #include <asm/linkage.h> |
1da177e4 | 17 | #include <asm/mmu_context.h> |
27b07da7 | 18 | #include <asm/mtrr.h> |
a03a3e28 | 19 | #include <asm/mce.h> |
8d4a4300 | 20 | #include <asm/pat.h> |
b6734c35 | 21 | #include <asm/asm.h> |
f0fc4aff | 22 | #include <asm/numa.h> |
b342797c | 23 | #include <asm/smp.h> |
1da177e4 LT |
24 | #ifdef CONFIG_X86_LOCAL_APIC |
25 | #include <asm/mpspec.h> | |
26 | #include <asm/apic.h> | |
27 | #include <mach_apic.h> | |
f0fc4aff | 28 | #include <asm/genapic.h> |
1da177e4 LT |
29 | #endif |
30 | ||
f0fc4aff YL |
31 | #include <asm/pda.h> |
32 | #include <asm/pgtable.h> | |
33 | #include <asm/processor.h> | |
34 | #include <asm/desc.h> | |
35 | #include <asm/atomic.h> | |
36 | #include <asm/proto.h> | |
37 | #include <asm/sections.h> | |
38 | #include <asm/setup.h> | |
88b094fb | 39 | #include <asm/hypervisor.h> |
f0fc4aff | 40 | |
1da177e4 LT |
41 | #include "cpu.h" |
42 | ||
c2d1cec1 MT |
43 | #ifdef CONFIG_X86_64 |
44 | ||
45 | /* all of these masks are initialized in setup_cpu_local_masks() */ | |
46 | cpumask_var_t cpu_callin_mask; | |
47 | cpumask_var_t cpu_callout_mask; | |
48 | cpumask_var_t cpu_initialized_mask; | |
49 | ||
50 | /* representing cpus for which sibling maps can be computed */ | |
51 | cpumask_var_t cpu_sibling_setup_mask; | |
52 | ||
53 | #else /* CONFIG_X86_32 */ | |
54 | ||
55 | cpumask_t cpu_callin_map; | |
56 | cpumask_t cpu_callout_map; | |
57 | cpumask_t cpu_initialized; | |
58 | cpumask_t cpu_sibling_setup_map; | |
59 | ||
60 | #endif /* CONFIG_X86_32 */ | |
61 | ||
62 | ||
0a488a53 YL |
63 | static struct cpu_dev *this_cpu __cpuinitdata; |
64 | ||
950ad7ff YL |
65 | #ifdef CONFIG_X86_64 |
66 | /* We need valid kernel segments for data and code in long mode too | |
67 | * IRET will check the segment types kkeil 2000/10/28 | |
68 | * Also sysret mandates a special GDT layout | |
69 | */ | |
70 | /* The TLS descriptors are currently at a different place compared to i386. | |
71 | Hopefully nobody expects them at a fixed place (Wine?) */ | |
7a61d35d | 72 | DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = { |
950ad7ff YL |
73 | [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } }, |
74 | [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } }, | |
75 | [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } }, | |
76 | [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } }, | |
77 | [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } }, | |
78 | [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } }, | |
79 | } }; | |
80 | #else | |
63cc8c75 | 81 | DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = { |
6842ef0e GOC |
82 | [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } }, |
83 | [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } }, | |
84 | [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } }, | |
85 | [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff200 } } }, | |
bf504672 RR |
86 | /* |
87 | * Segments used for calling PnP BIOS have byte granularity. | |
88 | * They code segments and data segments have fixed 64k limits, | |
89 | * the transfer segment sizes are set at run time. | |
90 | */ | |
6842ef0e GOC |
91 | /* 32-bit code */ |
92 | [GDT_ENTRY_PNPBIOS_CS32] = { { { 0x0000ffff, 0x00409a00 } } }, | |
93 | /* 16-bit code */ | |
94 | [GDT_ENTRY_PNPBIOS_CS16] = { { { 0x0000ffff, 0x00009a00 } } }, | |
95 | /* 16-bit data */ | |
96 | [GDT_ENTRY_PNPBIOS_DS] = { { { 0x0000ffff, 0x00009200 } } }, | |
97 | /* 16-bit data */ | |
98 | [GDT_ENTRY_PNPBIOS_TS1] = { { { 0x00000000, 0x00009200 } } }, | |
99 | /* 16-bit data */ | |
100 | [GDT_ENTRY_PNPBIOS_TS2] = { { { 0x00000000, 0x00009200 } } }, | |
bf504672 RR |
101 | /* |
102 | * The APM segments have byte granularity and their bases | |
103 | * are set at run time. All have 64k limits. | |
104 | */ | |
6842ef0e GOC |
105 | /* 32-bit code */ |
106 | [GDT_ENTRY_APMBIOS_BASE] = { { { 0x0000ffff, 0x00409a00 } } }, | |
bf504672 | 107 | /* 16-bit code */ |
6842ef0e GOC |
108 | [GDT_ENTRY_APMBIOS_BASE+1] = { { { 0x0000ffff, 0x00009a00 } } }, |
109 | /* data */ | |
110 | [GDT_ENTRY_APMBIOS_BASE+2] = { { { 0x0000ffff, 0x00409200 } } }, | |
bf504672 | 111 | |
6842ef0e GOC |
112 | [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } }, |
113 | [GDT_ENTRY_PERCPU] = { { { 0x00000000, 0x00000000 } } }, | |
7a61d35d | 114 | } }; |
950ad7ff | 115 | #endif |
7a61d35d | 116 | EXPORT_PER_CPU_SYMBOL_GPL(gdt_page); |
ae1ee11b | 117 | |
ba51dced | 118 | #ifdef CONFIG_X86_32 |
3bc9b76b | 119 | static int cachesize_override __cpuinitdata = -1; |
3bc9b76b | 120 | static int disable_x86_serial_nr __cpuinitdata = 1; |
1da177e4 | 121 | |
0a488a53 YL |
122 | static int __init cachesize_setup(char *str) |
123 | { | |
124 | get_option(&str, &cachesize_override); | |
125 | return 1; | |
126 | } | |
127 | __setup("cachesize=", cachesize_setup); | |
128 | ||
0a488a53 YL |
129 | static int __init x86_fxsr_setup(char *s) |
130 | { | |
131 | setup_clear_cpu_cap(X86_FEATURE_FXSR); | |
132 | setup_clear_cpu_cap(X86_FEATURE_XMM); | |
133 | return 1; | |
134 | } | |
135 | __setup("nofxsr", x86_fxsr_setup); | |
136 | ||
137 | static int __init x86_sep_setup(char *s) | |
138 | { | |
139 | setup_clear_cpu_cap(X86_FEATURE_SEP); | |
140 | return 1; | |
141 | } | |
142 | __setup("nosep", x86_sep_setup); | |
143 | ||
144 | /* Standard macro to see if a specific flag is changeable */ | |
145 | static inline int flag_is_changeable_p(u32 flag) | |
146 | { | |
147 | u32 f1, f2; | |
148 | ||
94f6bac1 KH |
149 | /* |
150 | * Cyrix and IDT cpus allow disabling of CPUID | |
151 | * so the code below may return different results | |
152 | * when it is executed before and after enabling | |
153 | * the CPUID. Add "volatile" to not allow gcc to | |
154 | * optimize the subsequent calls to this function. | |
155 | */ | |
156 | asm volatile ("pushfl\n\t" | |
157 | "pushfl\n\t" | |
158 | "popl %0\n\t" | |
159 | "movl %0,%1\n\t" | |
160 | "xorl %2,%0\n\t" | |
161 | "pushl %0\n\t" | |
162 | "popfl\n\t" | |
163 | "pushfl\n\t" | |
164 | "popl %0\n\t" | |
165 | "popfl\n\t" | |
166 | : "=&r" (f1), "=&r" (f2) | |
167 | : "ir" (flag)); | |
0a488a53 YL |
168 | |
169 | return ((f1^f2) & flag) != 0; | |
170 | } | |
171 | ||
172 | /* Probe for the CPUID instruction */ | |
173 | static int __cpuinit have_cpuid_p(void) | |
174 | { | |
175 | return flag_is_changeable_p(X86_EFLAGS_ID); | |
176 | } | |
177 | ||
178 | static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c) | |
179 | { | |
180 | if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) { | |
181 | /* Disable processor serial number */ | |
182 | unsigned long lo, hi; | |
183 | rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi); | |
184 | lo |= 0x200000; | |
185 | wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi); | |
186 | printk(KERN_NOTICE "CPU serial number disabled.\n"); | |
187 | clear_cpu_cap(c, X86_FEATURE_PN); | |
188 | ||
189 | /* Disabling the serial number may affect the cpuid level */ | |
190 | c->cpuid_level = cpuid_eax(0); | |
191 | } | |
192 | } | |
193 | ||
194 | static int __init x86_serial_nr_setup(char *s) | |
195 | { | |
196 | disable_x86_serial_nr = 0; | |
197 | return 1; | |
198 | } | |
199 | __setup("serialnumber", x86_serial_nr_setup); | |
ba51dced | 200 | #else |
102bbe3a YL |
201 | static inline int flag_is_changeable_p(u32 flag) |
202 | { | |
203 | return 1; | |
204 | } | |
ba51dced YL |
205 | /* Probe for the CPUID instruction */ |
206 | static inline int have_cpuid_p(void) | |
207 | { | |
208 | return 1; | |
209 | } | |
102bbe3a YL |
210 | static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c) |
211 | { | |
212 | } | |
ba51dced | 213 | #endif |
0a488a53 | 214 | |
b38b0665 PA |
215 | /* |
216 | * Some CPU features depend on higher CPUID levels, which may not always | |
217 | * be available due to CPUID level capping or broken virtualization | |
218 | * software. Add those features to this table to auto-disable them. | |
219 | */ | |
220 | struct cpuid_dependent_feature { | |
221 | u32 feature; | |
222 | u32 level; | |
223 | }; | |
224 | static const struct cpuid_dependent_feature __cpuinitconst | |
225 | cpuid_dependent_features[] = { | |
226 | { X86_FEATURE_MWAIT, 0x00000005 }, | |
227 | { X86_FEATURE_DCA, 0x00000009 }, | |
228 | { X86_FEATURE_XSAVE, 0x0000000d }, | |
229 | { 0, 0 } | |
230 | }; | |
231 | ||
232 | static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn) | |
233 | { | |
234 | const struct cpuid_dependent_feature *df; | |
235 | for (df = cpuid_dependent_features; df->feature; df++) { | |
236 | /* | |
237 | * Note: cpuid_level is set to -1 if unavailable, but | |
238 | * extended_extended_level is set to 0 if unavailable | |
239 | * and the legitimate extended levels are all negative | |
240 | * when signed; hence the weird messing around with | |
241 | * signs here... | |
242 | */ | |
243 | if (cpu_has(c, df->feature) && | |
244 | ((s32)df->feature < 0 ? | |
245 | (u32)df->feature > (u32)c->extended_cpuid_level : | |
246 | (s32)df->feature > (s32)c->cpuid_level)) { | |
247 | clear_cpu_cap(c, df->feature); | |
248 | if (warn) | |
249 | printk(KERN_WARNING | |
250 | "CPU: CPU feature %s disabled " | |
251 | "due to lack of CPUID level 0x%x\n", | |
252 | x86_cap_flags[df->feature], | |
253 | df->level); | |
254 | } | |
255 | } | |
256 | } | |
257 | ||
102bbe3a YL |
258 | /* |
259 | * Naming convention should be: <Name> [(<Codename>)] | |
260 | * This table only is used unless init_<vendor>() below doesn't set it; | |
261 | * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used | |
262 | * | |
263 | */ | |
264 | ||
265 | /* Look up CPU names by table lookup. */ | |
266 | static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c) | |
267 | { | |
268 | struct cpu_model_info *info; | |
269 | ||
270 | if (c->x86_model >= 16) | |
271 | return NULL; /* Range check */ | |
272 | ||
273 | if (!this_cpu) | |
274 | return NULL; | |
275 | ||
276 | info = this_cpu->c_models; | |
277 | ||
278 | while (info && info->family) { | |
279 | if (info->family == c->x86) | |
280 | return info->model_names[c->x86_model]; | |
281 | info++; | |
282 | } | |
283 | return NULL; /* Not found */ | |
284 | } | |
285 | ||
7d851c8d AK |
286 | __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata; |
287 | ||
9d31d35b YL |
288 | /* Current gdt points %fs at the "master" per-cpu area: after this, |
289 | * it's on the real one. */ | |
290 | void switch_to_new_gdt(void) | |
291 | { | |
292 | struct desc_ptr gdt_descr; | |
293 | ||
294 | gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id()); | |
295 | gdt_descr.size = GDT_SIZE - 1; | |
296 | load_gdt(&gdt_descr); | |
fab334c1 | 297 | #ifdef CONFIG_X86_32 |
9d31d35b | 298 | asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory"); |
fab334c1 | 299 | #endif |
9d31d35b YL |
300 | } |
301 | ||
10a434fc | 302 | static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {}; |
1da177e4 | 303 | |
34048c9e | 304 | static void __cpuinit default_init(struct cpuinfo_x86 *c) |
1da177e4 | 305 | { |
b9e67f00 YL |
306 | #ifdef CONFIG_X86_64 |
307 | display_cacheinfo(c); | |
308 | #else | |
1da177e4 LT |
309 | /* Not much we can do here... */ |
310 | /* Check if at least it has cpuid */ | |
311 | if (c->cpuid_level == -1) { | |
312 | /* No cpuid. It must be an ancient CPU */ | |
313 | if (c->x86 == 4) | |
314 | strcpy(c->x86_model_id, "486"); | |
315 | else if (c->x86 == 3) | |
316 | strcpy(c->x86_model_id, "386"); | |
317 | } | |
b9e67f00 | 318 | #endif |
1da177e4 LT |
319 | } |
320 | ||
95414930 | 321 | static struct cpu_dev __cpuinitdata default_cpu = { |
1da177e4 | 322 | .c_init = default_init, |
fe38d855 | 323 | .c_vendor = "Unknown", |
10a434fc | 324 | .c_x86_vendor = X86_VENDOR_UNKNOWN, |
1da177e4 | 325 | }; |
1da177e4 | 326 | |
1b05d60d | 327 | static void __cpuinit get_model_name(struct cpuinfo_x86 *c) |
1da177e4 LT |
328 | { |
329 | unsigned int *v; | |
330 | char *p, *q; | |
331 | ||
3da99c97 | 332 | if (c->extended_cpuid_level < 0x80000004) |
1b05d60d | 333 | return; |
1da177e4 LT |
334 | |
335 | v = (unsigned int *) c->x86_model_id; | |
336 | cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]); | |
337 | cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]); | |
338 | cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]); | |
339 | c->x86_model_id[48] = 0; | |
340 | ||
341 | /* Intel chips right-justify this string for some dumb reason; | |
342 | undo that brain damage */ | |
343 | p = q = &c->x86_model_id[0]; | |
34048c9e | 344 | while (*p == ' ') |
1da177e4 | 345 | p++; |
34048c9e PC |
346 | if (p != q) { |
347 | while (*p) | |
1da177e4 | 348 | *q++ = *p++; |
34048c9e | 349 | while (q <= &c->x86_model_id[48]) |
1da177e4 LT |
350 | *q++ = '\0'; /* Zero-pad the rest */ |
351 | } | |
1da177e4 LT |
352 | } |
353 | ||
3bc9b76b | 354 | void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c) |
1da177e4 | 355 | { |
9d31d35b | 356 | unsigned int n, dummy, ebx, ecx, edx, l2size; |
1da177e4 | 357 | |
3da99c97 | 358 | n = c->extended_cpuid_level; |
1da177e4 LT |
359 | |
360 | if (n >= 0x80000005) { | |
9d31d35b | 361 | cpuid(0x80000005, &dummy, &ebx, &ecx, &edx); |
1da177e4 | 362 | printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n", |
9d31d35b YL |
363 | edx>>24, edx&0xFF, ecx>>24, ecx&0xFF); |
364 | c->x86_cache_size = (ecx>>24) + (edx>>24); | |
140fc727 YL |
365 | #ifdef CONFIG_X86_64 |
366 | /* On K8 L1 TLB is inclusive, so don't count it */ | |
367 | c->x86_tlbsize = 0; | |
368 | #endif | |
1da177e4 LT |
369 | } |
370 | ||
371 | if (n < 0x80000006) /* Some chips just has a large L1. */ | |
372 | return; | |
373 | ||
0a488a53 | 374 | cpuid(0x80000006, &dummy, &ebx, &ecx, &edx); |
1da177e4 | 375 | l2size = ecx >> 16; |
34048c9e | 376 | |
140fc727 YL |
377 | #ifdef CONFIG_X86_64 |
378 | c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff); | |
379 | #else | |
1da177e4 LT |
380 | /* do processor-specific cache resizing */ |
381 | if (this_cpu->c_size_cache) | |
34048c9e | 382 | l2size = this_cpu->c_size_cache(c, l2size); |
1da177e4 LT |
383 | |
384 | /* Allow user to override all this if necessary. */ | |
385 | if (cachesize_override != -1) | |
386 | l2size = cachesize_override; | |
387 | ||
34048c9e | 388 | if (l2size == 0) |
1da177e4 | 389 | return; /* Again, no L2 cache is possible */ |
140fc727 | 390 | #endif |
1da177e4 LT |
391 | |
392 | c->x86_cache_size = l2size; | |
393 | ||
394 | printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n", | |
0a488a53 | 395 | l2size, ecx & 0xFF); |
1da177e4 LT |
396 | } |
397 | ||
9d31d35b | 398 | void __cpuinit detect_ht(struct cpuinfo_x86 *c) |
1da177e4 | 399 | { |
97e4db7c | 400 | #ifdef CONFIG_X86_HT |
0a488a53 YL |
401 | u32 eax, ebx, ecx, edx; |
402 | int index_msb, core_bits; | |
1da177e4 | 403 | |
0a488a53 | 404 | if (!cpu_has(c, X86_FEATURE_HT)) |
9d31d35b | 405 | return; |
1da177e4 | 406 | |
0a488a53 YL |
407 | if (cpu_has(c, X86_FEATURE_CMP_LEGACY)) |
408 | goto out; | |
1da177e4 | 409 | |
1cd78776 YL |
410 | if (cpu_has(c, X86_FEATURE_XTOPOLOGY)) |
411 | return; | |
1da177e4 | 412 | |
0a488a53 | 413 | cpuid(1, &eax, &ebx, &ecx, &edx); |
1da177e4 | 414 | |
9d31d35b YL |
415 | smp_num_siblings = (ebx & 0xff0000) >> 16; |
416 | ||
417 | if (smp_num_siblings == 1) { | |
418 | printk(KERN_INFO "CPU: Hyper-Threading is disabled\n"); | |
419 | } else if (smp_num_siblings > 1) { | |
420 | ||
9628937d | 421 | if (smp_num_siblings > nr_cpu_ids) { |
9d31d35b YL |
422 | printk(KERN_WARNING "CPU: Unsupported number of siblings %d", |
423 | smp_num_siblings); | |
424 | smp_num_siblings = 1; | |
425 | return; | |
426 | } | |
427 | ||
428 | index_msb = get_count_order(smp_num_siblings); | |
1cd78776 YL |
429 | #ifdef CONFIG_X86_64 |
430 | c->phys_proc_id = phys_pkg_id(index_msb); | |
431 | #else | |
9d31d35b | 432 | c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb); |
1cd78776 | 433 | #endif |
9d31d35b YL |
434 | |
435 | smp_num_siblings = smp_num_siblings / c->x86_max_cores; | |
436 | ||
437 | index_msb = get_count_order(smp_num_siblings); | |
438 | ||
439 | core_bits = get_count_order(c->x86_max_cores); | |
440 | ||
1cd78776 YL |
441 | #ifdef CONFIG_X86_64 |
442 | c->cpu_core_id = phys_pkg_id(index_msb) & | |
443 | ((1 << core_bits) - 1); | |
444 | #else | |
9d31d35b YL |
445 | c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) & |
446 | ((1 << core_bits) - 1); | |
1cd78776 | 447 | #endif |
1da177e4 | 448 | } |
1da177e4 | 449 | |
0a488a53 YL |
450 | out: |
451 | if ((c->x86_max_cores * smp_num_siblings) > 1) { | |
452 | printk(KERN_INFO "CPU: Physical Processor ID: %d\n", | |
453 | c->phys_proc_id); | |
454 | printk(KERN_INFO "CPU: Processor Core ID: %d\n", | |
455 | c->cpu_core_id); | |
9d31d35b | 456 | } |
9d31d35b | 457 | #endif |
97e4db7c | 458 | } |
1da177e4 | 459 | |
3da99c97 | 460 | static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c) |
1da177e4 LT |
461 | { |
462 | char *v = c->x86_vendor_id; | |
463 | int i; | |
fe38d855 | 464 | static int printed; |
1da177e4 LT |
465 | |
466 | for (i = 0; i < X86_VENDOR_NUM; i++) { | |
10a434fc YL |
467 | if (!cpu_devs[i]) |
468 | break; | |
469 | ||
470 | if (!strcmp(v, cpu_devs[i]->c_ident[0]) || | |
471 | (cpu_devs[i]->c_ident[1] && | |
472 | !strcmp(v, cpu_devs[i]->c_ident[1]))) { | |
473 | this_cpu = cpu_devs[i]; | |
474 | c->x86_vendor = this_cpu->c_x86_vendor; | |
475 | return; | |
1da177e4 LT |
476 | } |
477 | } | |
10a434fc | 478 | |
fe38d855 CE |
479 | if (!printed) { |
480 | printed++; | |
43603c8d | 481 | printk(KERN_ERR "CPU: vendor_id '%s' unknown, using generic init.\n", v); |
fe38d855 CE |
482 | printk(KERN_ERR "CPU: Your system may be unstable.\n"); |
483 | } | |
10a434fc | 484 | |
fe38d855 CE |
485 | c->x86_vendor = X86_VENDOR_UNKNOWN; |
486 | this_cpu = &default_cpu; | |
1da177e4 LT |
487 | } |
488 | ||
9d31d35b | 489 | void __cpuinit cpu_detect(struct cpuinfo_x86 *c) |
1da177e4 | 490 | { |
1da177e4 | 491 | /* Get vendor name */ |
4a148513 HH |
492 | cpuid(0x00000000, (unsigned int *)&c->cpuid_level, |
493 | (unsigned int *)&c->x86_vendor_id[0], | |
494 | (unsigned int *)&c->x86_vendor_id[8], | |
495 | (unsigned int *)&c->x86_vendor_id[4]); | |
1da177e4 | 496 | |
1da177e4 | 497 | c->x86 = 4; |
9d31d35b | 498 | /* Intel-defined flags: level 0x00000001 */ |
1da177e4 LT |
499 | if (c->cpuid_level >= 0x00000001) { |
500 | u32 junk, tfms, cap0, misc; | |
501 | cpuid(0x00000001, &tfms, &misc, &junk, &cap0); | |
9d31d35b YL |
502 | c->x86 = (tfms >> 8) & 0xf; |
503 | c->x86_model = (tfms >> 4) & 0xf; | |
504 | c->x86_mask = tfms & 0xf; | |
f5f786d0 | 505 | if (c->x86 == 0xf) |
1da177e4 | 506 | c->x86 += (tfms >> 20) & 0xff; |
f5f786d0 | 507 | if (c->x86 >= 0x6) |
9d31d35b | 508 | c->x86_model += ((tfms >> 16) & 0xf) << 4; |
d4387bd3 | 509 | if (cap0 & (1<<19)) { |
d4387bd3 | 510 | c->x86_clflush_size = ((misc >> 8) & 0xff) * 8; |
9d31d35b | 511 | c->x86_cache_alignment = c->x86_clflush_size; |
d4387bd3 | 512 | } |
1da177e4 | 513 | } |
1da177e4 | 514 | } |
3da99c97 YL |
515 | |
516 | static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c) | |
093af8d7 YL |
517 | { |
518 | u32 tfms, xlvl; | |
3da99c97 | 519 | u32 ebx; |
093af8d7 | 520 | |
3da99c97 YL |
521 | /* Intel-defined flags: level 0x00000001 */ |
522 | if (c->cpuid_level >= 0x00000001) { | |
523 | u32 capability, excap; | |
524 | cpuid(0x00000001, &tfms, &ebx, &excap, &capability); | |
525 | c->x86_capability[0] = capability; | |
526 | c->x86_capability[4] = excap; | |
527 | } | |
093af8d7 | 528 | |
3da99c97 YL |
529 | /* AMD-defined flags: level 0x80000001 */ |
530 | xlvl = cpuid_eax(0x80000000); | |
531 | c->extended_cpuid_level = xlvl; | |
532 | if ((xlvl & 0xffff0000) == 0x80000000) { | |
533 | if (xlvl >= 0x80000001) { | |
534 | c->x86_capability[1] = cpuid_edx(0x80000001); | |
535 | c->x86_capability[6] = cpuid_ecx(0x80000001); | |
093af8d7 | 536 | } |
093af8d7 | 537 | } |
093af8d7 | 538 | |
5122c890 | 539 | #ifdef CONFIG_X86_64 |
5122c890 YL |
540 | if (c->extended_cpuid_level >= 0x80000008) { |
541 | u32 eax = cpuid_eax(0x80000008); | |
542 | ||
543 | c->x86_virt_bits = (eax >> 8) & 0xff; | |
544 | c->x86_phys_bits = eax & 0xff; | |
093af8d7 | 545 | } |
5122c890 | 546 | #endif |
e3224234 YL |
547 | |
548 | if (c->extended_cpuid_level >= 0x80000007) | |
549 | c->x86_power = cpuid_edx(0x80000007); | |
093af8d7 YL |
550 | |
551 | } | |
1da177e4 | 552 | |
aef93c8b YL |
553 | static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c) |
554 | { | |
555 | #ifdef CONFIG_X86_32 | |
556 | int i; | |
557 | ||
558 | /* | |
559 | * First of all, decide if this is a 486 or higher | |
560 | * It's a 486 if we can modify the AC flag | |
561 | */ | |
562 | if (flag_is_changeable_p(X86_EFLAGS_AC)) | |
563 | c->x86 = 4; | |
564 | else | |
565 | c->x86 = 3; | |
566 | ||
567 | for (i = 0; i < X86_VENDOR_NUM; i++) | |
568 | if (cpu_devs[i] && cpu_devs[i]->c_identify) { | |
569 | c->x86_vendor_id[0] = 0; | |
570 | cpu_devs[i]->c_identify(c); | |
571 | if (c->x86_vendor_id[0]) { | |
572 | get_cpu_vendor(c); | |
573 | break; | |
574 | } | |
575 | } | |
576 | #endif | |
577 | } | |
578 | ||
34048c9e PC |
579 | /* |
580 | * Do minimum CPU detection early. | |
581 | * Fields really needed: vendor, cpuid_level, family, model, mask, | |
582 | * cache alignment. | |
583 | * The others are not touched to avoid unwanted side effects. | |
584 | * | |
585 | * WARNING: this function is only called on the BP. Don't add code here | |
586 | * that is supposed to run on all CPUs. | |
587 | */ | |
3da99c97 | 588 | static void __init early_identify_cpu(struct cpuinfo_x86 *c) |
d7cd5611 | 589 | { |
6627d242 YL |
590 | #ifdef CONFIG_X86_64 |
591 | c->x86_clflush_size = 64; | |
592 | #else | |
d4387bd3 | 593 | c->x86_clflush_size = 32; |
6627d242 | 594 | #endif |
0a488a53 | 595 | c->x86_cache_alignment = c->x86_clflush_size; |
d7cd5611 | 596 | |
3da99c97 | 597 | memset(&c->x86_capability, 0, sizeof c->x86_capability); |
0a488a53 | 598 | c->extended_cpuid_level = 0; |
d7cd5611 | 599 | |
aef93c8b YL |
600 | if (!have_cpuid_p()) |
601 | identify_cpu_without_cpuid(c); | |
602 | ||
603 | /* cyrix could have cpuid enabled via c_identify()*/ | |
d7cd5611 RR |
604 | if (!have_cpuid_p()) |
605 | return; | |
606 | ||
607 | cpu_detect(c); | |
608 | ||
3da99c97 | 609 | get_cpu_vendor(c); |
2b16a235 | 610 | |
3da99c97 | 611 | get_cpu_cap(c); |
12cf105c | 612 | |
10a434fc YL |
613 | if (this_cpu->c_early_init) |
614 | this_cpu->c_early_init(c); | |
093af8d7 | 615 | |
1c4acdb4 | 616 | #ifdef CONFIG_SMP |
bfcb4c1b | 617 | c->cpu_index = boot_cpu_id; |
1c4acdb4 | 618 | #endif |
b38b0665 | 619 | filter_cpuid_features(c, false); |
d7cd5611 RR |
620 | } |
621 | ||
9d31d35b YL |
622 | void __init early_cpu_init(void) |
623 | { | |
10a434fc YL |
624 | struct cpu_dev **cdev; |
625 | int count = 0; | |
626 | ||
627 | printk("KERNEL supported cpus:\n"); | |
628 | for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) { | |
629 | struct cpu_dev *cpudev = *cdev; | |
630 | unsigned int j; | |
9d31d35b | 631 | |
10a434fc YL |
632 | if (count >= X86_VENDOR_NUM) |
633 | break; | |
634 | cpu_devs[count] = cpudev; | |
635 | count++; | |
636 | ||
637 | for (j = 0; j < 2; j++) { | |
638 | if (!cpudev->c_ident[j]) | |
639 | continue; | |
640 | printk(" %s %s\n", cpudev->c_vendor, | |
641 | cpudev->c_ident[j]); | |
642 | } | |
643 | } | |
9d31d35b | 644 | |
9d31d35b | 645 | early_identify_cpu(&boot_cpu_data); |
d7cd5611 | 646 | } |
093af8d7 | 647 | |
b6734c35 PA |
648 | /* |
649 | * The NOPL instruction is supposed to exist on all CPUs with | |
ba0593bf | 650 | * family >= 6; unfortunately, that's not true in practice because |
b6734c35 | 651 | * of early VIA chips and (more importantly) broken virtualizers that |
ba0593bf PA |
652 | * are not easy to detect. In the latter case it doesn't even *fail* |
653 | * reliably, so probing for it doesn't even work. Disable it completely | |
654 | * unless we can find a reliable way to detect all the broken cases. | |
b6734c35 PA |
655 | */ |
656 | static void __cpuinit detect_nopl(struct cpuinfo_x86 *c) | |
657 | { | |
b6734c35 | 658 | clear_cpu_cap(c, X86_FEATURE_NOPL); |
d7cd5611 RR |
659 | } |
660 | ||
34048c9e | 661 | static void __cpuinit generic_identify(struct cpuinfo_x86 *c) |
1da177e4 | 662 | { |
aef93c8b | 663 | c->extended_cpuid_level = 0; |
1da177e4 | 664 | |
3da99c97 | 665 | if (!have_cpuid_p()) |
aef93c8b | 666 | identify_cpu_without_cpuid(c); |
1d67953f | 667 | |
aef93c8b | 668 | /* cyrix could have cpuid enabled via c_identify()*/ |
a9853dd6 | 669 | if (!have_cpuid_p()) |
aef93c8b | 670 | return; |
1da177e4 | 671 | |
3da99c97 | 672 | cpu_detect(c); |
1da177e4 | 673 | |
3da99c97 | 674 | get_cpu_vendor(c); |
1da177e4 | 675 | |
3da99c97 | 676 | get_cpu_cap(c); |
1da177e4 | 677 | |
3da99c97 YL |
678 | if (c->cpuid_level >= 0x00000001) { |
679 | c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF; | |
b89d3b3e YL |
680 | #ifdef CONFIG_X86_32 |
681 | # ifdef CONFIG_X86_HT | |
3da99c97 | 682 | c->apicid = phys_pkg_id(c->initial_apicid, 0); |
b89d3b3e | 683 | # else |
3da99c97 | 684 | c->apicid = c->initial_apicid; |
b89d3b3e YL |
685 | # endif |
686 | #endif | |
1da177e4 | 687 | |
b89d3b3e YL |
688 | #ifdef CONFIG_X86_HT |
689 | c->phys_proc_id = c->initial_apicid; | |
1e9f28fa | 690 | #endif |
3da99c97 | 691 | } |
1da177e4 | 692 | |
1b05d60d | 693 | get_model_name(c); /* Default name */ |
1da177e4 | 694 | |
3da99c97 YL |
695 | init_scattered_cpuid_features(c); |
696 | detect_nopl(c); | |
1da177e4 | 697 | } |
1da177e4 LT |
698 | |
699 | /* | |
700 | * This does the hard work of actually picking apart the CPU stuff... | |
701 | */ | |
9a250347 | 702 | static void __cpuinit identify_cpu(struct cpuinfo_x86 *c) |
1da177e4 LT |
703 | { |
704 | int i; | |
705 | ||
706 | c->loops_per_jiffy = loops_per_jiffy; | |
707 | c->x86_cache_size = -1; | |
708 | c->x86_vendor = X86_VENDOR_UNKNOWN; | |
1da177e4 LT |
709 | c->x86_model = c->x86_mask = 0; /* So far unknown... */ |
710 | c->x86_vendor_id[0] = '\0'; /* Unset */ | |
711 | c->x86_model_id[0] = '\0'; /* Unset */ | |
94605eff | 712 | c->x86_max_cores = 1; |
102bbe3a | 713 | c->x86_coreid_bits = 0; |
11fdd252 | 714 | #ifdef CONFIG_X86_64 |
102bbe3a YL |
715 | c->x86_clflush_size = 64; |
716 | #else | |
717 | c->cpuid_level = -1; /* CPUID not detected */ | |
770d132f | 718 | c->x86_clflush_size = 32; |
102bbe3a YL |
719 | #endif |
720 | c->x86_cache_alignment = c->x86_clflush_size; | |
1da177e4 LT |
721 | memset(&c->x86_capability, 0, sizeof c->x86_capability); |
722 | ||
1da177e4 LT |
723 | generic_identify(c); |
724 | ||
3898534d | 725 | if (this_cpu->c_identify) |
1da177e4 LT |
726 | this_cpu->c_identify(c); |
727 | ||
102bbe3a YL |
728 | #ifdef CONFIG_X86_64 |
729 | c->apicid = phys_pkg_id(0); | |
730 | #endif | |
731 | ||
1da177e4 LT |
732 | /* |
733 | * Vendor-specific initialization. In this section we | |
734 | * canonicalize the feature flags, meaning if there are | |
735 | * features a certain CPU supports which CPUID doesn't | |
736 | * tell us, CPUID claiming incorrect flags, or other bugs, | |
737 | * we handle them here. | |
738 | * | |
739 | * At the end of this section, c->x86_capability better | |
740 | * indicate the features this CPU genuinely supports! | |
741 | */ | |
742 | if (this_cpu->c_init) | |
743 | this_cpu->c_init(c); | |
744 | ||
745 | /* Disable the PN if appropriate */ | |
746 | squash_the_stupid_serial_number(c); | |
747 | ||
748 | /* | |
749 | * The vendor-specific functions might have changed features. Now | |
750 | * we do "generic changes." | |
751 | */ | |
752 | ||
b38b0665 PA |
753 | /* Filter out anything that depends on CPUID levels we don't have */ |
754 | filter_cpuid_features(c, true); | |
755 | ||
1da177e4 | 756 | /* If the model name is still unset, do table lookup. */ |
34048c9e | 757 | if (!c->x86_model_id[0]) { |
1da177e4 LT |
758 | char *p; |
759 | p = table_lookup_model(c); | |
34048c9e | 760 | if (p) |
1da177e4 LT |
761 | strcpy(c->x86_model_id, p); |
762 | else | |
763 | /* Last resort... */ | |
764 | sprintf(c->x86_model_id, "%02x/%02x", | |
54a20f8c | 765 | c->x86, c->x86_model); |
1da177e4 LT |
766 | } |
767 | ||
102bbe3a YL |
768 | #ifdef CONFIG_X86_64 |
769 | detect_ht(c); | |
770 | #endif | |
771 | ||
88b094fb | 772 | init_hypervisor(c); |
1da177e4 LT |
773 | /* |
774 | * On SMP, boot_cpu_data holds the common feature set between | |
775 | * all CPUs; so make sure that we indicate which features are | |
776 | * common between the CPUs. The first time this routine gets | |
777 | * executed, c == &boot_cpu_data. | |
778 | */ | |
34048c9e | 779 | if (c != &boot_cpu_data) { |
1da177e4 | 780 | /* AND the already accumulated flags with these */ |
9d31d35b | 781 | for (i = 0; i < NCAPINTS; i++) |
1da177e4 LT |
782 | boot_cpu_data.x86_capability[i] &= c->x86_capability[i]; |
783 | } | |
784 | ||
7d851c8d AK |
785 | /* Clear all flags overriden by options */ |
786 | for (i = 0; i < NCAPINTS; i++) | |
12c247a6 | 787 | c->x86_capability[i] &= ~cleared_cpu_caps[i]; |
7d851c8d | 788 | |
102bbe3a | 789 | #ifdef CONFIG_X86_MCE |
1da177e4 | 790 | /* Init Machine Check Exception if available. */ |
1da177e4 | 791 | mcheck_init(c); |
102bbe3a | 792 | #endif |
30d432df AK |
793 | |
794 | select_idle_routine(c); | |
102bbe3a YL |
795 | |
796 | #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64) | |
797 | numa_add_cpu(smp_processor_id()); | |
798 | #endif | |
a6c4e076 | 799 | } |
31ab269a | 800 | |
e04d645f GC |
801 | #ifdef CONFIG_X86_64 |
802 | static void vgetcpu_set_mode(void) | |
803 | { | |
804 | if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP)) | |
805 | vgetcpu_mode = VGETCPU_RDTSCP; | |
806 | else | |
807 | vgetcpu_mode = VGETCPU_LSL; | |
808 | } | |
809 | #endif | |
810 | ||
a6c4e076 JF |
811 | void __init identify_boot_cpu(void) |
812 | { | |
813 | identify_cpu(&boot_cpu_data); | |
102bbe3a | 814 | #ifdef CONFIG_X86_32 |
a6c4e076 | 815 | sysenter_setup(); |
6fe940d6 | 816 | enable_sep_cpu(); |
e04d645f GC |
817 | #else |
818 | vgetcpu_set_mode(); | |
102bbe3a | 819 | #endif |
a6c4e076 | 820 | } |
3b520b23 | 821 | |
a6c4e076 JF |
822 | void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c) |
823 | { | |
824 | BUG_ON(c == &boot_cpu_data); | |
825 | identify_cpu(c); | |
102bbe3a | 826 | #ifdef CONFIG_X86_32 |
a6c4e076 | 827 | enable_sep_cpu(); |
102bbe3a | 828 | #endif |
a6c4e076 | 829 | mtrr_ap_init(); |
1da177e4 LT |
830 | } |
831 | ||
a0854a46 YL |
832 | struct msr_range { |
833 | unsigned min; | |
834 | unsigned max; | |
835 | }; | |
1da177e4 | 836 | |
a0854a46 YL |
837 | static struct msr_range msr_range_array[] __cpuinitdata = { |
838 | { 0x00000000, 0x00000418}, | |
839 | { 0xc0000000, 0xc000040b}, | |
840 | { 0xc0010000, 0xc0010142}, | |
841 | { 0xc0011000, 0xc001103b}, | |
842 | }; | |
1da177e4 | 843 | |
a0854a46 YL |
844 | static void __cpuinit print_cpu_msr(void) |
845 | { | |
846 | unsigned index; | |
847 | u64 val; | |
848 | int i; | |
849 | unsigned index_min, index_max; | |
850 | ||
851 | for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) { | |
852 | index_min = msr_range_array[i].min; | |
853 | index_max = msr_range_array[i].max; | |
854 | for (index = index_min; index < index_max; index++) { | |
855 | if (rdmsrl_amd_safe(index, &val)) | |
856 | continue; | |
857 | printk(KERN_INFO " MSR%08x: %016llx\n", index, val); | |
1da177e4 | 858 | } |
a0854a46 YL |
859 | } |
860 | } | |
94605eff | 861 | |
a0854a46 YL |
862 | static int show_msr __cpuinitdata; |
863 | static __init int setup_show_msr(char *arg) | |
864 | { | |
865 | int num; | |
3dd9d514 | 866 | |
a0854a46 | 867 | get_option(&arg, &num); |
3dd9d514 | 868 | |
a0854a46 YL |
869 | if (num > 0) |
870 | show_msr = num; | |
871 | return 1; | |
1da177e4 | 872 | } |
a0854a46 | 873 | __setup("show_msr=", setup_show_msr); |
1da177e4 | 874 | |
191679fd AK |
875 | static __init int setup_noclflush(char *arg) |
876 | { | |
877 | setup_clear_cpu_cap(X86_FEATURE_CLFLSH); | |
878 | return 1; | |
879 | } | |
880 | __setup("noclflush", setup_noclflush); | |
881 | ||
3bc9b76b | 882 | void __cpuinit print_cpu_info(struct cpuinfo_x86 *c) |
1da177e4 LT |
883 | { |
884 | char *vendor = NULL; | |
885 | ||
886 | if (c->x86_vendor < X86_VENDOR_NUM) | |
887 | vendor = this_cpu->c_vendor; | |
888 | else if (c->cpuid_level >= 0) | |
889 | vendor = c->x86_vendor_id; | |
890 | ||
bd32a8cf | 891 | if (vendor && !strstr(c->x86_model_id, vendor)) |
9d31d35b | 892 | printk(KERN_CONT "%s ", vendor); |
1da177e4 | 893 | |
9d31d35b YL |
894 | if (c->x86_model_id[0]) |
895 | printk(KERN_CONT "%s", c->x86_model_id); | |
1da177e4 | 896 | else |
9d31d35b | 897 | printk(KERN_CONT "%d86", c->x86); |
1da177e4 | 898 | |
34048c9e | 899 | if (c->x86_mask || c->cpuid_level >= 0) |
9d31d35b | 900 | printk(KERN_CONT " stepping %02x\n", c->x86_mask); |
1da177e4 | 901 | else |
9d31d35b | 902 | printk(KERN_CONT "\n"); |
a0854a46 YL |
903 | |
904 | #ifdef CONFIG_SMP | |
905 | if (c->cpu_index < show_msr) | |
906 | print_cpu_msr(); | |
907 | #else | |
908 | if (show_msr) | |
909 | print_cpu_msr(); | |
910 | #endif | |
1da177e4 LT |
911 | } |
912 | ||
ac72e788 AK |
913 | static __init int setup_disablecpuid(char *arg) |
914 | { | |
915 | int bit; | |
916 | if (get_option(&arg, &bit) && bit < NCAPINTS*32) | |
917 | setup_clear_cpu_cap(bit); | |
918 | else | |
919 | return 0; | |
920 | return 1; | |
921 | } | |
922 | __setup("clearcpuid=", setup_disablecpuid); | |
923 | ||
d5494d4f YL |
924 | #ifdef CONFIG_X86_64 |
925 | struct x8664_pda **_cpu_pda __read_mostly; | |
926 | EXPORT_SYMBOL(_cpu_pda); | |
927 | ||
928 | struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table }; | |
929 | ||
34945ede | 930 | static char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss; |
d5494d4f | 931 | |
2d9cd6c2 | 932 | void __cpuinit pda_init(int cpu) |
d5494d4f YL |
933 | { |
934 | struct x8664_pda *pda = cpu_pda(cpu); | |
935 | ||
936 | /* Setup up data that may be needed in __get_free_pages early */ | |
937 | loadsegment(fs, 0); | |
938 | loadsegment(gs, 0); | |
939 | /* Memory clobbers used to order PDA accessed */ | |
940 | mb(); | |
941 | wrmsrl(MSR_GS_BASE, pda); | |
942 | mb(); | |
943 | ||
944 | pda->cpunumber = cpu; | |
945 | pda->irqcount = -1; | |
946 | pda->kernelstack = (unsigned long)stack_thread_info() - | |
947 | PDA_STACKOFFSET + THREAD_SIZE; | |
948 | pda->active_mm = &init_mm; | |
949 | pda->mmu_state = 0; | |
950 | ||
951 | if (cpu == 0) { | |
952 | /* others are initialized in smpboot.c */ | |
953 | pda->pcurrent = &init_task; | |
954 | pda->irqstackptr = boot_cpu_stack; | |
955 | pda->irqstackptr += IRQSTACKSIZE - 64; | |
956 | } else { | |
957 | if (!pda->irqstackptr) { | |
958 | pda->irqstackptr = (char *) | |
959 | __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER); | |
960 | if (!pda->irqstackptr) | |
961 | panic("cannot allocate irqstack for cpu %d", | |
962 | cpu); | |
963 | pda->irqstackptr += IRQSTACKSIZE - 64; | |
964 | } | |
965 | ||
966 | if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE) | |
967 | pda->nodenumber = cpu_to_node(cpu); | |
968 | } | |
969 | } | |
970 | ||
34945ede JS |
971 | static char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + |
972 | DEBUG_STKSZ] __page_aligned_bss; | |
d5494d4f YL |
973 | |
974 | extern asmlinkage void ignore_sysret(void); | |
975 | ||
976 | /* May not be marked __init: used by software suspend */ | |
977 | void syscall_init(void) | |
1da177e4 | 978 | { |
d5494d4f YL |
979 | /* |
980 | * LSTAR and STAR live in a bit strange symbiosis. | |
981 | * They both write to the same internal register. STAR allows to | |
982 | * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip. | |
983 | */ | |
984 | wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32); | |
985 | wrmsrl(MSR_LSTAR, system_call); | |
986 | wrmsrl(MSR_CSTAR, ignore_sysret); | |
03ae5768 | 987 | |
d5494d4f YL |
988 | #ifdef CONFIG_IA32_EMULATION |
989 | syscall32_cpu_init(); | |
990 | #endif | |
03ae5768 | 991 | |
d5494d4f YL |
992 | /* Flags to clear on syscall */ |
993 | wrmsrl(MSR_SYSCALL_MASK, | |
994 | X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL); | |
1da177e4 | 995 | } |
62111195 | 996 | |
d5494d4f YL |
997 | unsigned long kernel_eflags; |
998 | ||
999 | /* | |
1000 | * Copies of the original ist values from the tss are only accessed during | |
1001 | * debugging, no special alignment required. | |
1002 | */ | |
1003 | DEFINE_PER_CPU(struct orig_ist, orig_ist); | |
1004 | ||
1005 | #else | |
1006 | ||
7c3576d2 | 1007 | /* Make sure %fs is initialized properly in idle threads */ |
6b2fb3c6 | 1008 | struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs) |
f95d47ca JF |
1009 | { |
1010 | memset(regs, 0, sizeof(struct pt_regs)); | |
65ea5b03 | 1011 | regs->fs = __KERNEL_PERCPU; |
f95d47ca JF |
1012 | return regs; |
1013 | } | |
d5494d4f | 1014 | #endif |
c5413fbe | 1015 | |
d2cbcc49 RR |
1016 | /* |
1017 | * cpu_init() initializes state that is per-CPU. Some data is already | |
1018 | * initialized (naturally) in the bootstrap process, such as the GDT | |
1019 | * and IDT. We reload them nevertheless, this function acts as a | |
1020 | * 'CPU state barrier', nothing should get across. | |
1ba76586 | 1021 | * A lot of state is already set up in PDA init for 64 bit |
d2cbcc49 | 1022 | */ |
1ba76586 YL |
1023 | #ifdef CONFIG_X86_64 |
1024 | void __cpuinit cpu_init(void) | |
1025 | { | |
1026 | int cpu = stack_smp_processor_id(); | |
1027 | struct tss_struct *t = &per_cpu(init_tss, cpu); | |
1028 | struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu); | |
1029 | unsigned long v; | |
1030 | char *estacks = NULL; | |
1031 | struct task_struct *me; | |
1032 | int i; | |
1033 | ||
1034 | /* CPU 0 is initialised in head64.c */ | |
1035 | if (cpu != 0) | |
1036 | pda_init(cpu); | |
1037 | else | |
1038 | estacks = boot_exception_stacks; | |
1039 | ||
1040 | me = current; | |
1041 | ||
c2d1cec1 | 1042 | if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) |
1ba76586 YL |
1043 | panic("CPU#%d already initialized!\n", cpu); |
1044 | ||
1045 | printk(KERN_INFO "Initializing CPU#%d\n", cpu); | |
1046 | ||
1047 | clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE); | |
1048 | ||
1049 | /* | |
1050 | * Initialize the per-CPU GDT with the boot GDT, | |
1051 | * and set up the GDT descriptor: | |
1052 | */ | |
1053 | ||
1054 | switch_to_new_gdt(); | |
1055 | load_idt((const struct desc_ptr *)&idt_descr); | |
1056 | ||
1057 | memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8); | |
1058 | syscall_init(); | |
1059 | ||
1060 | wrmsrl(MSR_FS_BASE, 0); | |
1061 | wrmsrl(MSR_KERNEL_GS_BASE, 0); | |
1062 | barrier(); | |
1063 | ||
1064 | check_efer(); | |
1065 | if (cpu != 0 && x2apic) | |
1066 | enable_x2apic(); | |
1067 | ||
1068 | /* | |
1069 | * set up and load the per-CPU TSS | |
1070 | */ | |
1071 | if (!orig_ist->ist[0]) { | |
1072 | static const unsigned int order[N_EXCEPTION_STACKS] = { | |
1073 | [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER, | |
1074 | [DEBUG_STACK - 1] = DEBUG_STACK_ORDER | |
1075 | }; | |
1076 | for (v = 0; v < N_EXCEPTION_STACKS; v++) { | |
1077 | if (cpu) { | |
1078 | estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]); | |
1079 | if (!estacks) | |
1080 | panic("Cannot allocate exception " | |
1081 | "stack %ld %d\n", v, cpu); | |
1082 | } | |
1083 | estacks += PAGE_SIZE << order[v]; | |
1084 | orig_ist->ist[v] = t->x86_tss.ist[v] = | |
1085 | (unsigned long)estacks; | |
1086 | } | |
1087 | } | |
1088 | ||
1089 | t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap); | |
1090 | /* | |
1091 | * <= is required because the CPU will access up to | |
1092 | * 8 bits beyond the end of the IO permission bitmap. | |
1093 | */ | |
1094 | for (i = 0; i <= IO_BITMAP_LONGS; i++) | |
1095 | t->io_bitmap[i] = ~0UL; | |
1096 | ||
1097 | atomic_inc(&init_mm.mm_count); | |
1098 | me->active_mm = &init_mm; | |
1099 | if (me->mm) | |
1100 | BUG(); | |
1101 | enter_lazy_tlb(&init_mm, me); | |
1102 | ||
1103 | load_sp0(t, ¤t->thread); | |
1104 | set_tss_desc(cpu, t); | |
1105 | load_TR_desc(); | |
1106 | load_LDT(&init_mm.context); | |
1107 | ||
1108 | #ifdef CONFIG_KGDB | |
1109 | /* | |
1110 | * If the kgdb is connected no debug regs should be altered. This | |
1111 | * is only applicable when KGDB and a KGDB I/O module are built | |
1112 | * into the kernel and you are using early debugging with | |
1113 | * kgdbwait. KGDB will control the kernel HW breakpoint registers. | |
1114 | */ | |
1115 | if (kgdb_connected && arch_kgdb_ops.correct_hw_break) | |
1116 | arch_kgdb_ops.correct_hw_break(); | |
1117 | else { | |
1118 | #endif | |
1119 | /* | |
1120 | * Clear all 6 debug registers: | |
1121 | */ | |
1122 | ||
1123 | set_debugreg(0UL, 0); | |
1124 | set_debugreg(0UL, 1); | |
1125 | set_debugreg(0UL, 2); | |
1126 | set_debugreg(0UL, 3); | |
1127 | set_debugreg(0UL, 6); | |
1128 | set_debugreg(0UL, 7); | |
1129 | #ifdef CONFIG_KGDB | |
1130 | /* If the kgdb is connected no debug regs should be altered. */ | |
1131 | } | |
1132 | #endif | |
1133 | ||
1134 | fpu_init(); | |
1135 | ||
1136 | raw_local_save_flags(kernel_eflags); | |
1137 | ||
1138 | if (is_uv_system()) | |
1139 | uv_cpu_init(); | |
1140 | } | |
1141 | ||
1142 | #else | |
1143 | ||
d2cbcc49 | 1144 | void __cpuinit cpu_init(void) |
9ee79a3d | 1145 | { |
d2cbcc49 RR |
1146 | int cpu = smp_processor_id(); |
1147 | struct task_struct *curr = current; | |
34048c9e | 1148 | struct tss_struct *t = &per_cpu(init_tss, cpu); |
9ee79a3d | 1149 | struct thread_struct *thread = &curr->thread; |
62111195 | 1150 | |
c2d1cec1 | 1151 | if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) { |
62111195 JF |
1152 | printk(KERN_WARNING "CPU#%d already initialized!\n", cpu); |
1153 | for (;;) local_irq_enable(); | |
1154 | } | |
1155 | ||
1156 | printk(KERN_INFO "Initializing CPU#%d\n", cpu); | |
1157 | ||
1158 | if (cpu_has_vme || cpu_has_tsc || cpu_has_de) | |
1159 | clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE); | |
62111195 | 1160 | |
4d37e7e3 | 1161 | load_idt(&idt_descr); |
c5413fbe | 1162 | switch_to_new_gdt(); |
1da177e4 | 1163 | |
1da177e4 LT |
1164 | /* |
1165 | * Set up and load the per-CPU TSS and LDT | |
1166 | */ | |
1167 | atomic_inc(&init_mm.mm_count); | |
62111195 JF |
1168 | curr->active_mm = &init_mm; |
1169 | if (curr->mm) | |
1170 | BUG(); | |
1171 | enter_lazy_tlb(&init_mm, curr); | |
1da177e4 | 1172 | |
faca6227 | 1173 | load_sp0(t, thread); |
34048c9e | 1174 | set_tss_desc(cpu, t); |
1da177e4 LT |
1175 | load_TR_desc(); |
1176 | load_LDT(&init_mm.context); | |
1177 | ||
22c4e308 | 1178 | #ifdef CONFIG_DOUBLEFAULT |
1da177e4 LT |
1179 | /* Set up doublefault TSS pointer in the GDT */ |
1180 | __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss); | |
22c4e308 | 1181 | #endif |
1da177e4 | 1182 | |
464d1a78 JF |
1183 | /* Clear %gs. */ |
1184 | asm volatile ("mov %0, %%gs" : : "r" (0)); | |
1da177e4 LT |
1185 | |
1186 | /* Clear all 6 debug registers: */ | |
4bb0d3ec ZA |
1187 | set_debugreg(0, 0); |
1188 | set_debugreg(0, 1); | |
1189 | set_debugreg(0, 2); | |
1190 | set_debugreg(0, 3); | |
1191 | set_debugreg(0, 6); | |
1192 | set_debugreg(0, 7); | |
1da177e4 LT |
1193 | |
1194 | /* | |
1195 | * Force FPU initialization: | |
1196 | */ | |
b359e8a4 SS |
1197 | if (cpu_has_xsave) |
1198 | current_thread_info()->status = TS_XSAVE; | |
1199 | else | |
1200 | current_thread_info()->status = 0; | |
1da177e4 LT |
1201 | clear_used_math(); |
1202 | mxcsr_feature_mask_init(); | |
dc1e35c6 SS |
1203 | |
1204 | /* | |
1205 | * Boot processor to setup the FP and extended state context info. | |
1206 | */ | |
b3572e36 | 1207 | if (smp_processor_id() == boot_cpu_id) |
dc1e35c6 SS |
1208 | init_thread_xstate(); |
1209 | ||
1210 | xsave_init(); | |
1da177e4 | 1211 | } |
e1367daf | 1212 | |
1ba76586 YL |
1213 | |
1214 | #endif |