aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel/cpu/common.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/kernel/cpu/common.c')
-rw-r--r--arch/x86/kernel/cpu/common.c396
1 files changed, 224 insertions, 172 deletions
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index 826d5c876278..e2962cc1e27b 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -1,52 +1,52 @@
1#include <linux/init.h>
2#include <linux/kernel.h>
3#include <linux/sched.h>
4#include <linux/string.h>
5#include <linux/bootmem.h> 1#include <linux/bootmem.h>
2#include <linux/linkage.h>
6#include <linux/bitops.h> 3#include <linux/bitops.h>
4#include <linux/kernel.h>
7#include <linux/module.h> 5#include <linux/module.h>
8#include <linux/kgdb.h> 6#include <linux/percpu.h>
9#include <linux/topology.h> 7#include <linux/string.h>
10#include <linux/delay.h> 8#include <linux/delay.h>
9#include <linux/sched.h>
10#include <linux/init.h>
11#include <linux/kgdb.h>
11#include <linux/smp.h> 12#include <linux/smp.h>
12#include <linux/percpu.h> 13#include <linux/io.h>
13#include <asm/i387.h> 14
14#include <asm/msr.h> 15#include <asm/stackprotector.h>
15#include <asm/io.h>
16#include <asm/linkage.h>
17#include <asm/mmu_context.h> 16#include <asm/mmu_context.h>
17#include <asm/hypervisor.h>
18#include <asm/processor.h>
19#include <asm/sections.h>
20#include <asm/topology.h>
21#include <asm/cpumask.h>
22#include <asm/pgtable.h>
23#include <asm/atomic.h>
24#include <asm/proto.h>
25#include <asm/setup.h>
26#include <asm/apic.h>
27#include <asm/desc.h>
28#include <asm/i387.h>
18#include <asm/mtrr.h> 29#include <asm/mtrr.h>
30#include <asm/numa.h>
31#include <asm/asm.h>
32#include <asm/cpu.h>
19#include <asm/mce.h> 33#include <asm/mce.h>
34#include <asm/msr.h>
20#include <asm/pat.h> 35#include <asm/pat.h>
21#include <asm/asm.h>
22#include <asm/numa.h>
23#include <asm/smp.h> 36#include <asm/smp.h>
24#include <asm/cpu.h>
25#include <asm/cpumask.h>
26#include <asm/apic.h>
27 37
28#ifdef CONFIG_X86_LOCAL_APIC 38#ifdef CONFIG_X86_LOCAL_APIC
29#include <asm/uv/uv.h> 39#include <asm/uv/uv.h>
30#endif 40#endif
31 41
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>
39#include <asm/hypervisor.h>
40#include <asm/stackprotector.h>
41
42#include "cpu.h" 42#include "cpu.h"
43 43
44#ifdef CONFIG_X86_64 44#ifdef CONFIG_X86_64
45 45
46/* all of these masks are initialized in setup_cpu_local_masks() */ 46/* all of these masks are initialized in setup_cpu_local_masks() */
47cpumask_var_t cpu_callin_mask;
48cpumask_var_t cpu_callout_mask;
49cpumask_var_t cpu_initialized_mask; 47cpumask_var_t cpu_initialized_mask;
48cpumask_var_t cpu_callout_mask;
49cpumask_var_t cpu_callin_mask;
50 50
51/* representing cpus for which sibling maps can be computed */ 51/* representing cpus for which sibling maps can be computed */
52cpumask_var_t cpu_sibling_setup_mask; 52cpumask_var_t cpu_sibling_setup_mask;
@@ -62,15 +62,15 @@ void __init setup_cpu_local_masks(void)
62 62
63#else /* CONFIG_X86_32 */ 63#else /* CONFIG_X86_32 */
64 64
65cpumask_t cpu_callin_map; 65cpumask_t cpu_sibling_setup_map;
66cpumask_t cpu_callout_map; 66cpumask_t cpu_callout_map;
67cpumask_t cpu_initialized; 67cpumask_t cpu_initialized;
68cpumask_t cpu_sibling_setup_map; 68cpumask_t cpu_callin_map;
69 69
70#endif /* CONFIG_X86_32 */ 70#endif /* CONFIG_X86_32 */
71 71
72 72
73static struct cpu_dev *this_cpu __cpuinitdata; 73static const struct cpu_dev *this_cpu __cpuinitdata;
74 74
75DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = { 75DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
76#ifdef CONFIG_X86_64 76#ifdef CONFIG_X86_64
@@ -79,48 +79,48 @@ DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
79 * IRET will check the segment types kkeil 2000/10/28 79 * IRET will check the segment types kkeil 2000/10/28
80 * Also sysret mandates a special GDT layout 80 * Also sysret mandates a special GDT layout
81 * 81 *
82 * The TLS descriptors are currently at a different place compared to i386. 82 * TLS descriptors are currently at a different place compared to i386.
83 * Hopefully nobody expects them at a fixed place (Wine?) 83 * Hopefully nobody expects them at a fixed place (Wine?)
84 */ 84 */
85 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } }, 85 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
86 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } }, 86 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
87 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } }, 87 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
88 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } }, 88 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
89 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } }, 89 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
90 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } }, 90 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
91#else 91#else
92 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } }, 92 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } },
93 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } }, 93 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } },
94 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } }, 94 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } },
95 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff200 } } }, 95 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff200 } } },
96 /* 96 /*
97 * Segments used for calling PnP BIOS have byte granularity. 97 * Segments used for calling PnP BIOS have byte granularity.
98 * They code segments and data segments have fixed 64k limits, 98 * They code segments and data segments have fixed 64k limits,
99 * the transfer segment sizes are set at run time. 99 * the transfer segment sizes are set at run time.
100 */ 100 */
101 /* 32-bit code */ 101 /* 32-bit code */
102 [GDT_ENTRY_PNPBIOS_CS32] = { { { 0x0000ffff, 0x00409a00 } } }, 102 [GDT_ENTRY_PNPBIOS_CS32] = { { { 0x0000ffff, 0x00409a00 } } },
103 /* 16-bit code */ 103 /* 16-bit code */
104 [GDT_ENTRY_PNPBIOS_CS16] = { { { 0x0000ffff, 0x00009a00 } } }, 104 [GDT_ENTRY_PNPBIOS_CS16] = { { { 0x0000ffff, 0x00009a00 } } },
105 /* 16-bit data */ 105 /* 16-bit data */
106 [GDT_ENTRY_PNPBIOS_DS] = { { { 0x0000ffff, 0x00009200 } } }, 106 [GDT_ENTRY_PNPBIOS_DS] = { { { 0x0000ffff, 0x00009200 } } },
107 /* 16-bit data */ 107 /* 16-bit data */
108 [GDT_ENTRY_PNPBIOS_TS1] = { { { 0x00000000, 0x00009200 } } }, 108 [GDT_ENTRY_PNPBIOS_TS1] = { { { 0x00000000, 0x00009200 } } },
109 /* 16-bit data */ 109 /* 16-bit data */
110 [GDT_ENTRY_PNPBIOS_TS2] = { { { 0x00000000, 0x00009200 } } }, 110 [GDT_ENTRY_PNPBIOS_TS2] = { { { 0x00000000, 0x00009200 } } },
111 /* 111 /*
112 * The APM segments have byte granularity and their bases 112 * The APM segments have byte granularity and their bases
113 * are set at run time. All have 64k limits. 113 * are set at run time. All have 64k limits.
114 */ 114 */
115 /* 32-bit code */ 115 /* 32-bit code */
116 [GDT_ENTRY_APMBIOS_BASE] = { { { 0x0000ffff, 0x00409a00 } } }, 116 [GDT_ENTRY_APMBIOS_BASE] = { { { 0x0000ffff, 0x00409a00 } } },
117 /* 16-bit code */ 117 /* 16-bit code */
118 [GDT_ENTRY_APMBIOS_BASE+1] = { { { 0x0000ffff, 0x00009a00 } } }, 118 [GDT_ENTRY_APMBIOS_BASE+1] = { { { 0x0000ffff, 0x00009a00 } } },
119 /* data */ 119 /* data */
120 [GDT_ENTRY_APMBIOS_BASE+2] = { { { 0x0000ffff, 0x00409200 } } }, 120 [GDT_ENTRY_APMBIOS_BASE+2] = { { { 0x0000ffff, 0x00409200 } } },
121 121
122 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } }, 122 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } },
123 [GDT_ENTRY_PERCPU] = { { { 0x0000ffff, 0x00cf9200 } } }, 123 [GDT_ENTRY_PERCPU] = { { { 0x0000ffff, 0x00cf9200 } } },
124 GDT_STACK_CANARY_INIT 124 GDT_STACK_CANARY_INIT
125#endif 125#endif
126} }; 126} };
@@ -164,16 +164,17 @@ static inline int flag_is_changeable_p(u32 flag)
164 * the CPUID. Add "volatile" to not allow gcc to 164 * the CPUID. Add "volatile" to not allow gcc to
165 * optimize the subsequent calls to this function. 165 * optimize the subsequent calls to this function.
166 */ 166 */
167 asm volatile ("pushfl\n\t" 167 asm volatile ("pushfl \n\t"
168 "pushfl\n\t" 168 "pushfl \n\t"
169 "popl %0\n\t" 169 "popl %0 \n\t"
170 "movl %0,%1\n\t" 170 "movl %0, %1 \n\t"
171 "xorl %2,%0\n\t" 171 "xorl %2, %0 \n\t"
172 "pushl %0\n\t" 172 "pushl %0 \n\t"
173 "popfl\n\t" 173 "popfl \n\t"
174 "pushfl\n\t" 174 "pushfl \n\t"
175 "popl %0\n\t" 175 "popl %0 \n\t"
176 "popfl\n\t" 176 "popfl \n\t"
177
177 : "=&r" (f1), "=&r" (f2) 178 : "=&r" (f1), "=&r" (f2)
178 : "ir" (flag)); 179 : "ir" (flag));
179 180
@@ -188,18 +189,22 @@ static int __cpuinit have_cpuid_p(void)
188 189
189static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c) 190static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
190{ 191{
191 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) { 192 unsigned long lo, hi;
192 /* Disable processor serial number */ 193
193 unsigned long lo, hi; 194 if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
194 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi); 195 return;
195 lo |= 0x200000; 196
196 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi); 197 /* Disable processor serial number: */
197 printk(KERN_NOTICE "CPU serial number disabled.\n"); 198
198 clear_cpu_cap(c, X86_FEATURE_PN); 199 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
199 200 lo |= 0x200000;
200 /* Disabling the serial number may affect the cpuid level */ 201 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
201 c->cpuid_level = cpuid_eax(0); 202
202 } 203 printk(KERN_NOTICE "CPU serial number disabled.\n");
204 clear_cpu_cap(c, X86_FEATURE_PN);
205
206 /* Disabling the serial number may affect the cpuid level */
207 c->cpuid_level = cpuid_eax(0);
203} 208}
204 209
205static int __init x86_serial_nr_setup(char *s) 210static int __init x86_serial_nr_setup(char *s)
@@ -232,6 +237,7 @@ struct cpuid_dependent_feature {
232 u32 feature; 237 u32 feature;
233 u32 level; 238 u32 level;
234}; 239};
240
235static const struct cpuid_dependent_feature __cpuinitconst 241static const struct cpuid_dependent_feature __cpuinitconst
236cpuid_dependent_features[] = { 242cpuid_dependent_features[] = {
237 { X86_FEATURE_MWAIT, 0x00000005 }, 243 { X86_FEATURE_MWAIT, 0x00000005 },
@@ -243,7 +249,11 @@ cpuid_dependent_features[] = {
243static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn) 249static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
244{ 250{
245 const struct cpuid_dependent_feature *df; 251 const struct cpuid_dependent_feature *df;
252
246 for (df = cpuid_dependent_features; df->feature; df++) { 253 for (df = cpuid_dependent_features; df->feature; df++) {
254
255 if (!cpu_has(c, df->feature))
256 continue;
247 /* 257 /*
248 * Note: cpuid_level is set to -1 if unavailable, but 258 * Note: cpuid_level is set to -1 if unavailable, but
249 * extended_extended_level is set to 0 if unavailable 259 * extended_extended_level is set to 0 if unavailable
@@ -251,32 +261,32 @@ static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
251 * when signed; hence the weird messing around with 261 * when signed; hence the weird messing around with
252 * signs here... 262 * signs here...
253 */ 263 */
254 if (cpu_has(c, df->feature) && 264 if (!((s32)df->level < 0 ?
255 ((s32)df->level < 0 ?
256 (u32)df->level > (u32)c->extended_cpuid_level : 265 (u32)df->level > (u32)c->extended_cpuid_level :
257 (s32)df->level > (s32)c->cpuid_level)) { 266 (s32)df->level > (s32)c->cpuid_level))
258 clear_cpu_cap(c, df->feature); 267 continue;
259 if (warn) 268
260 printk(KERN_WARNING 269 clear_cpu_cap(c, df->feature);
261 "CPU: CPU feature %s disabled " 270 if (!warn)
262 "due to lack of CPUID level 0x%x\n", 271 continue;
263 x86_cap_flags[df->feature], 272
264 df->level); 273 printk(KERN_WARNING
265 } 274 "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
275 x86_cap_flags[df->feature], df->level);
266 } 276 }
267} 277}
268 278
269/* 279/*
270 * Naming convention should be: <Name> [(<Codename>)] 280 * Naming convention should be: <Name> [(<Codename>)]
271 * This table only is used unless init_<vendor>() below doesn't set it; 281 * This table only is used unless init_<vendor>() below doesn't set it;
272 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used 282 * in particular, if CPUID levels 0x80000002..4 are supported, this
273 * 283 * isn't used
274 */ 284 */
275 285
276/* Look up CPU names by table lookup. */ 286/* Look up CPU names by table lookup. */
277static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c) 287static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
278{ 288{
279 struct cpu_model_info *info; 289 const struct cpu_model_info *info;
280 290
281 if (c->x86_model >= 16) 291 if (c->x86_model >= 16)
282 return NULL; /* Range check */ 292 return NULL; /* Range check */
@@ -307,8 +317,10 @@ void load_percpu_segment(int cpu)
307 load_stack_canary_segment(); 317 load_stack_canary_segment();
308} 318}
309 319
310/* Current gdt points %fs at the "master" per-cpu area: after this, 320/*
311 * it's on the real one. */ 321 * Current gdt points %fs at the "master" per-cpu area: after this,
322 * it's on the real one.
323 */
312void switch_to_new_gdt(int cpu) 324void switch_to_new_gdt(int cpu)
313{ 325{
314 struct desc_ptr gdt_descr; 326 struct desc_ptr gdt_descr;
@@ -321,7 +333,7 @@ void switch_to_new_gdt(int cpu)
321 load_percpu_segment(cpu); 333 load_percpu_segment(cpu);
322} 334}
323 335
324static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {}; 336static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
325 337
326static void __cpuinit default_init(struct cpuinfo_x86 *c) 338static void __cpuinit default_init(struct cpuinfo_x86 *c)
327{ 339{
@@ -340,7 +352,7 @@ static void __cpuinit default_init(struct cpuinfo_x86 *c)
340#endif 352#endif
341} 353}
342 354
343static struct cpu_dev __cpuinitdata default_cpu = { 355static const struct cpu_dev __cpuinitconst default_cpu = {
344 .c_init = default_init, 356 .c_init = default_init,
345 .c_vendor = "Unknown", 357 .c_vendor = "Unknown",
346 .c_x86_vendor = X86_VENDOR_UNKNOWN, 358 .c_x86_vendor = X86_VENDOR_UNKNOWN,
@@ -354,22 +366,24 @@ static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
354 if (c->extended_cpuid_level < 0x80000004) 366 if (c->extended_cpuid_level < 0x80000004)
355 return; 367 return;
356 368
357 v = (unsigned int *) c->x86_model_id; 369 v = (unsigned int *)c->x86_model_id;
358 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]); 370 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
359 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]); 371 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
360 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]); 372 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
361 c->x86_model_id[48] = 0; 373 c->x86_model_id[48] = 0;
362 374
363 /* Intel chips right-justify this string for some dumb reason; 375 /*
364 undo that brain damage */ 376 * Intel chips right-justify this string for some dumb reason;
377 * undo that brain damage:
378 */
365 p = q = &c->x86_model_id[0]; 379 p = q = &c->x86_model_id[0];
366 while (*p == ' ') 380 while (*p == ' ')
367 p++; 381 p++;
368 if (p != q) { 382 if (p != q) {
369 while (*p) 383 while (*p)
370 *q++ = *p++; 384 *q++ = *p++;
371 while (q <= &c->x86_model_id[48]) 385 while (q <= &c->x86_model_id[48])
372 *q++ = '\0'; /* Zero-pad the rest */ 386 *q++ = '\0'; /* Zero-pad the rest */
373 } 387 }
374} 388}
375 389
@@ -438,27 +452,30 @@ void __cpuinit detect_ht(struct cpuinfo_x86 *c)
438 452
439 if (smp_num_siblings == 1) { 453 if (smp_num_siblings == 1) {
440 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n"); 454 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
441 } else if (smp_num_siblings > 1) { 455 goto out;
456 }
442 457
443 if (smp_num_siblings > nr_cpu_ids) { 458 if (smp_num_siblings <= 1)
444 printk(KERN_WARNING "CPU: Unsupported number of siblings %d", 459 goto out;
445 smp_num_siblings); 460
446 smp_num_siblings = 1; 461 if (smp_num_siblings > nr_cpu_ids) {
447 return; 462 pr_warning("CPU: Unsupported number of siblings %d",
448 } 463 smp_num_siblings);
464 smp_num_siblings = 1;
465 return;
466 }
449 467
450 index_msb = get_count_order(smp_num_siblings); 468 index_msb = get_count_order(smp_num_siblings);
451 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb); 469 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
452 470
453 smp_num_siblings = smp_num_siblings / c->x86_max_cores; 471 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
454 472
455 index_msb = get_count_order(smp_num_siblings); 473 index_msb = get_count_order(smp_num_siblings);
456 474
457 core_bits = get_count_order(c->x86_max_cores); 475 core_bits = get_count_order(c->x86_max_cores);
458 476
459 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) & 477 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
460 ((1 << core_bits) - 1); 478 ((1 << core_bits) - 1);
461 }
462 479
463out: 480out:
464 if ((c->x86_max_cores * smp_num_siblings) > 1) { 481 if ((c->x86_max_cores * smp_num_siblings) > 1) {
@@ -473,8 +490,8 @@ out:
473static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c) 490static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
474{ 491{
475 char *v = c->x86_vendor_id; 492 char *v = c->x86_vendor_id;
476 int i;
477 static int printed; 493 static int printed;
494 int i;
478 495
479 for (i = 0; i < X86_VENDOR_NUM; i++) { 496 for (i = 0; i < X86_VENDOR_NUM; i++) {
480 if (!cpu_devs[i]) 497 if (!cpu_devs[i])
@@ -483,6 +500,7 @@ static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
483 if (!strcmp(v, cpu_devs[i]->c_ident[0]) || 500 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
484 (cpu_devs[i]->c_ident[1] && 501 (cpu_devs[i]->c_ident[1] &&
485 !strcmp(v, cpu_devs[i]->c_ident[1]))) { 502 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
503
486 this_cpu = cpu_devs[i]; 504 this_cpu = cpu_devs[i];
487 c->x86_vendor = this_cpu->c_x86_vendor; 505 c->x86_vendor = this_cpu->c_x86_vendor;
488 return; 506 return;
@@ -491,7 +509,9 @@ static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
491 509
492 if (!printed) { 510 if (!printed) {
493 printed++; 511 printed++;
494 printk(KERN_ERR "CPU: vendor_id '%s' unknown, using generic init.\n", v); 512 printk(KERN_ERR
513 "CPU: vendor_id '%s' unknown, using generic init.\n", v);
514
495 printk(KERN_ERR "CPU: Your system may be unstable.\n"); 515 printk(KERN_ERR "CPU: Your system may be unstable.\n");
496 } 516 }
497 517
@@ -511,14 +531,17 @@ void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
511 /* Intel-defined flags: level 0x00000001 */ 531 /* Intel-defined flags: level 0x00000001 */
512 if (c->cpuid_level >= 0x00000001) { 532 if (c->cpuid_level >= 0x00000001) {
513 u32 junk, tfms, cap0, misc; 533 u32 junk, tfms, cap0, misc;
534
514 cpuid(0x00000001, &tfms, &misc, &junk, &cap0); 535 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
515 c->x86 = (tfms >> 8) & 0xf; 536 c->x86 = (tfms >> 8) & 0xf;
516 c->x86_model = (tfms >> 4) & 0xf; 537 c->x86_model = (tfms >> 4) & 0xf;
517 c->x86_mask = tfms & 0xf; 538 c->x86_mask = tfms & 0xf;
539
518 if (c->x86 == 0xf) 540 if (c->x86 == 0xf)
519 c->x86 += (tfms >> 20) & 0xff; 541 c->x86 += (tfms >> 20) & 0xff;
520 if (c->x86 >= 0x6) 542 if (c->x86 >= 0x6)
521 c->x86_model += ((tfms >> 16) & 0xf) << 4; 543 c->x86_model += ((tfms >> 16) & 0xf) << 4;
544
522 if (cap0 & (1<<19)) { 545 if (cap0 & (1<<19)) {
523 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8; 546 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
524 c->x86_cache_alignment = c->x86_clflush_size; 547 c->x86_cache_alignment = c->x86_clflush_size;
@@ -534,6 +557,7 @@ static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
534 /* Intel-defined flags: level 0x00000001 */ 557 /* Intel-defined flags: level 0x00000001 */
535 if (c->cpuid_level >= 0x00000001) { 558 if (c->cpuid_level >= 0x00000001) {
536 u32 capability, excap; 559 u32 capability, excap;
560
537 cpuid(0x00000001, &tfms, &ebx, &excap, &capability); 561 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
538 c->x86_capability[0] = capability; 562 c->x86_capability[0] = capability;
539 c->x86_capability[4] = excap; 563 c->x86_capability[4] = excap;
@@ -542,6 +566,7 @@ static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
542 /* AMD-defined flags: level 0x80000001 */ 566 /* AMD-defined flags: level 0x80000001 */
543 xlvl = cpuid_eax(0x80000000); 567 xlvl = cpuid_eax(0x80000000);
544 c->extended_cpuid_level = xlvl; 568 c->extended_cpuid_level = xlvl;
569
545 if ((xlvl & 0xffff0000) == 0x80000000) { 570 if ((xlvl & 0xffff0000) == 0x80000000) {
546 if (xlvl >= 0x80000001) { 571 if (xlvl >= 0x80000001) {
547 c->x86_capability[1] = cpuid_edx(0x80000001); 572 c->x86_capability[1] = cpuid_edx(0x80000001);
@@ -549,13 +574,15 @@ static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
549 } 574 }
550 } 575 }
551 576
552#ifdef CONFIG_X86_64
553 if (c->extended_cpuid_level >= 0x80000008) { 577 if (c->extended_cpuid_level >= 0x80000008) {
554 u32 eax = cpuid_eax(0x80000008); 578 u32 eax = cpuid_eax(0x80000008);
555 579
556 c->x86_virt_bits = (eax >> 8) & 0xff; 580 c->x86_virt_bits = (eax >> 8) & 0xff;
557 c->x86_phys_bits = eax & 0xff; 581 c->x86_phys_bits = eax & 0xff;
558 } 582 }
583#ifdef CONFIG_X86_32
584 else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
585 c->x86_phys_bits = 36;
559#endif 586#endif
560 587
561 if (c->extended_cpuid_level >= 0x80000007) 588 if (c->extended_cpuid_level >= 0x80000007)
@@ -602,8 +629,12 @@ static void __init early_identify_cpu(struct cpuinfo_x86 *c)
602{ 629{
603#ifdef CONFIG_X86_64 630#ifdef CONFIG_X86_64
604 c->x86_clflush_size = 64; 631 c->x86_clflush_size = 64;
632 c->x86_phys_bits = 36;
633 c->x86_virt_bits = 48;
605#else 634#else
606 c->x86_clflush_size = 32; 635 c->x86_clflush_size = 32;
636 c->x86_phys_bits = 32;
637 c->x86_virt_bits = 32;
607#endif 638#endif
608 c->x86_cache_alignment = c->x86_clflush_size; 639 c->x86_cache_alignment = c->x86_clflush_size;
609 640
@@ -634,12 +665,12 @@ static void __init early_identify_cpu(struct cpuinfo_x86 *c)
634 665
635void __init early_cpu_init(void) 666void __init early_cpu_init(void)
636{ 667{
637 struct cpu_dev **cdev; 668 const struct cpu_dev *const *cdev;
638 int count = 0; 669 int count = 0;
639 670
640 printk("KERNEL supported cpus:\n"); 671 printk(KERN_INFO "KERNEL supported cpus:\n");
641 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) { 672 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
642 struct cpu_dev *cpudev = *cdev; 673 const struct cpu_dev *cpudev = *cdev;
643 unsigned int j; 674 unsigned int j;
644 675
645 if (count >= X86_VENDOR_NUM) 676 if (count >= X86_VENDOR_NUM)
@@ -650,7 +681,7 @@ void __init early_cpu_init(void)
650 for (j = 0; j < 2; j++) { 681 for (j = 0; j < 2; j++) {
651 if (!cpudev->c_ident[j]) 682 if (!cpudev->c_ident[j])
652 continue; 683 continue;
653 printk(" %s %s\n", cpudev->c_vendor, 684 printk(KERN_INFO " %s %s\n", cpudev->c_vendor,
654 cpudev->c_ident[j]); 685 cpudev->c_ident[j]);
655 } 686 }
656 } 687 }
@@ -726,9 +757,13 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
726 c->x86_coreid_bits = 0; 757 c->x86_coreid_bits = 0;
727#ifdef CONFIG_X86_64 758#ifdef CONFIG_X86_64
728 c->x86_clflush_size = 64; 759 c->x86_clflush_size = 64;
760 c->x86_phys_bits = 36;
761 c->x86_virt_bits = 48;
729#else 762#else
730 c->cpuid_level = -1; /* CPUID not detected */ 763 c->cpuid_level = -1; /* CPUID not detected */
731 c->x86_clflush_size = 32; 764 c->x86_clflush_size = 32;
765 c->x86_phys_bits = 32;
766 c->x86_virt_bits = 32;
732#endif 767#endif
733 c->x86_cache_alignment = c->x86_clflush_size; 768 c->x86_cache_alignment = c->x86_clflush_size;
734 memset(&c->x86_capability, 0, sizeof c->x86_capability); 769 memset(&c->x86_capability, 0, sizeof c->x86_capability);
@@ -759,8 +794,8 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
759 squash_the_stupid_serial_number(c); 794 squash_the_stupid_serial_number(c);
760 795
761 /* 796 /*
762 * The vendor-specific functions might have changed features. Now 797 * The vendor-specific functions might have changed features.
763 * we do "generic changes." 798 * Now we do "generic changes."
764 */ 799 */
765 800
766 /* Filter out anything that depends on CPUID levels we don't have */ 801 /* Filter out anything that depends on CPUID levels we don't have */
@@ -768,7 +803,7 @@ static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
768 803
769 /* If the model name is still unset, do table lookup. */ 804 /* If the model name is still unset, do table lookup. */
770 if (!c->x86_model_id[0]) { 805 if (!c->x86_model_id[0]) {
771 char *p; 806 const char *p;
772 p = table_lookup_model(c); 807 p = table_lookup_model(c);
773 if (p) 808 if (p)
774 strcpy(c->x86_model_id, p); 809 strcpy(c->x86_model_id, p);
@@ -843,11 +878,11 @@ void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
843} 878}
844 879
845struct msr_range { 880struct msr_range {
846 unsigned min; 881 unsigned min;
847 unsigned max; 882 unsigned max;
848}; 883};
849 884
850static struct msr_range msr_range_array[] __cpuinitdata = { 885static const struct msr_range msr_range_array[] __cpuinitconst = {
851 { 0x00000000, 0x00000418}, 886 { 0x00000000, 0x00000418},
852 { 0xc0000000, 0xc000040b}, 887 { 0xc0000000, 0xc000040b},
853 { 0xc0010000, 0xc0010142}, 888 { 0xc0010000, 0xc0010142},
@@ -856,14 +891,15 @@ static struct msr_range msr_range_array[] __cpuinitdata = {
856 891
857static void __cpuinit print_cpu_msr(void) 892static void __cpuinit print_cpu_msr(void)
858{ 893{
894 unsigned index_min, index_max;
859 unsigned index; 895 unsigned index;
860 u64 val; 896 u64 val;
861 int i; 897 int i;
862 unsigned index_min, index_max;
863 898
864 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) { 899 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
865 index_min = msr_range_array[i].min; 900 index_min = msr_range_array[i].min;
866 index_max = msr_range_array[i].max; 901 index_max = msr_range_array[i].max;
902
867 for (index = index_min; index < index_max; index++) { 903 for (index = index_min; index < index_max; index++) {
868 if (rdmsrl_amd_safe(index, &val)) 904 if (rdmsrl_amd_safe(index, &val))
869 continue; 905 continue;
@@ -873,6 +909,7 @@ static void __cpuinit print_cpu_msr(void)
873} 909}
874 910
875static int show_msr __cpuinitdata; 911static int show_msr __cpuinitdata;
912
876static __init int setup_show_msr(char *arg) 913static __init int setup_show_msr(char *arg)
877{ 914{
878 int num; 915 int num;
@@ -894,12 +931,14 @@ __setup("noclflush", setup_noclflush);
894 931
895void __cpuinit print_cpu_info(struct cpuinfo_x86 *c) 932void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
896{ 933{
897 char *vendor = NULL; 934 const char *vendor = NULL;
898 935
899 if (c->x86_vendor < X86_VENDOR_NUM) 936 if (c->x86_vendor < X86_VENDOR_NUM) {
900 vendor = this_cpu->c_vendor; 937 vendor = this_cpu->c_vendor;
901 else if (c->cpuid_level >= 0) 938 } else {
902 vendor = c->x86_vendor_id; 939 if (c->cpuid_level >= 0)
940 vendor = c->x86_vendor_id;
941 }
903 942
904 if (vendor && !strstr(c->x86_model_id, vendor)) 943 if (vendor && !strstr(c->x86_model_id, vendor))
905 printk(KERN_CONT "%s ", vendor); 944 printk(KERN_CONT "%s ", vendor);
@@ -926,10 +965,12 @@ void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
926static __init int setup_disablecpuid(char *arg) 965static __init int setup_disablecpuid(char *arg)
927{ 966{
928 int bit; 967 int bit;
968
929 if (get_option(&arg, &bit) && bit < NCAPINTS*32) 969 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
930 setup_clear_cpu_cap(bit); 970 setup_clear_cpu_cap(bit);
931 else 971 else
932 return 0; 972 return 0;
973
933 return 1; 974 return 1;
934} 975}
935__setup("clearcpuid=", setup_disablecpuid); 976__setup("clearcpuid=", setup_disablecpuid);
@@ -939,6 +980,7 @@ struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
939 980
940DEFINE_PER_CPU_FIRST(union irq_stack_union, 981DEFINE_PER_CPU_FIRST(union irq_stack_union,
941 irq_stack_union) __aligned(PAGE_SIZE); 982 irq_stack_union) __aligned(PAGE_SIZE);
983
942DEFINE_PER_CPU(char *, irq_stack_ptr) = 984DEFINE_PER_CPU(char *, irq_stack_ptr) =
943 init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64; 985 init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
944 986
@@ -948,12 +990,21 @@ EXPORT_PER_CPU_SYMBOL(kernel_stack);
948 990
949DEFINE_PER_CPU(unsigned int, irq_count) = -1; 991DEFINE_PER_CPU(unsigned int, irq_count) = -1;
950 992
993/*
994 * Special IST stacks which the CPU switches to when it calls
995 * an IST-marked descriptor entry. Up to 7 stacks (hardware
996 * limit), all of them are 4K, except the debug stack which
997 * is 8K.
998 */
999static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1000 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1001 [DEBUG_STACK - 1] = DEBUG_STKSZ
1002};
1003
951static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks 1004static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
952 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]) 1005 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ])
953 __aligned(PAGE_SIZE); 1006 __aligned(PAGE_SIZE);
954 1007
955extern asmlinkage void ignore_sysret(void);
956
957/* May not be marked __init: used by software suspend */ 1008/* May not be marked __init: used by software suspend */
958void syscall_init(void) 1009void syscall_init(void)
959{ 1010{
@@ -983,7 +1034,7 @@ unsigned long kernel_eflags;
983 */ 1034 */
984DEFINE_PER_CPU(struct orig_ist, orig_ist); 1035DEFINE_PER_CPU(struct orig_ist, orig_ist);
985 1036
986#else /* x86_64 */ 1037#else /* CONFIG_X86_64 */
987 1038
988#ifdef CONFIG_CC_STACKPROTECTOR 1039#ifdef CONFIG_CC_STACKPROTECTOR
989DEFINE_PER_CPU(unsigned long, stack_canary); 1040DEFINE_PER_CPU(unsigned long, stack_canary);
@@ -995,9 +1046,26 @@ struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
995 memset(regs, 0, sizeof(struct pt_regs)); 1046 memset(regs, 0, sizeof(struct pt_regs));
996 regs->fs = __KERNEL_PERCPU; 1047 regs->fs = __KERNEL_PERCPU;
997 regs->gs = __KERNEL_STACK_CANARY; 1048 regs->gs = __KERNEL_STACK_CANARY;
1049
998 return regs; 1050 return regs;
999} 1051}
1000#endif /* x86_64 */ 1052#endif /* CONFIG_X86_64 */
1053
1054/*
1055 * Clear all 6 debug registers:
1056 */
1057static void clear_all_debug_regs(void)
1058{
1059 int i;
1060
1061 for (i = 0; i < 8; i++) {
1062 /* Ignore db4, db5 */
1063 if ((i == 4) || (i == 5))
1064 continue;
1065
1066 set_debugreg(0, i);
1067 }
1068}
1001 1069
1002/* 1070/*
1003 * cpu_init() initializes state that is per-CPU. Some data is already 1071 * cpu_init() initializes state that is per-CPU. Some data is already
@@ -1007,15 +1075,20 @@ struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
1007 * A lot of state is already set up in PDA init for 64 bit 1075 * A lot of state is already set up in PDA init for 64 bit
1008 */ 1076 */
1009#ifdef CONFIG_X86_64 1077#ifdef CONFIG_X86_64
1078
1010void __cpuinit cpu_init(void) 1079void __cpuinit cpu_init(void)
1011{ 1080{
1012 int cpu = stack_smp_processor_id(); 1081 struct orig_ist *orig_ist;
1013 struct tss_struct *t = &per_cpu(init_tss, cpu);
1014 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
1015 unsigned long v;
1016 struct task_struct *me; 1082 struct task_struct *me;
1083 struct tss_struct *t;
1084 unsigned long v;
1085 int cpu;
1017 int i; 1086 int i;
1018 1087
1088 cpu = stack_smp_processor_id();
1089 t = &per_cpu(init_tss, cpu);
1090 orig_ist = &per_cpu(orig_ist, cpu);
1091
1019#ifdef CONFIG_NUMA 1092#ifdef CONFIG_NUMA
1020 if (cpu != 0 && percpu_read(node_number) == 0 && 1093 if (cpu != 0 && percpu_read(node_number) == 0 &&
1021 cpu_to_node(cpu) != NUMA_NO_NODE) 1094 cpu_to_node(cpu) != NUMA_NO_NODE)
@@ -1056,19 +1129,17 @@ void __cpuinit cpu_init(void)
1056 * set up and load the per-CPU TSS 1129 * set up and load the per-CPU TSS
1057 */ 1130 */
1058 if (!orig_ist->ist[0]) { 1131 if (!orig_ist->ist[0]) {
1059 static const unsigned int sizes[N_EXCEPTION_STACKS] = {
1060 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1061 [DEBUG_STACK - 1] = DEBUG_STKSZ
1062 };
1063 char *estacks = per_cpu(exception_stacks, cpu); 1132 char *estacks = per_cpu(exception_stacks, cpu);
1133
1064 for (v = 0; v < N_EXCEPTION_STACKS; v++) { 1134 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1065 estacks += sizes[v]; 1135 estacks += exception_stack_sizes[v];
1066 orig_ist->ist[v] = t->x86_tss.ist[v] = 1136 orig_ist->ist[v] = t->x86_tss.ist[v] =
1067 (unsigned long)estacks; 1137 (unsigned long)estacks;
1068 } 1138 }
1069 } 1139 }
1070 1140
1071 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap); 1141 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1142
1072 /* 1143 /*
1073 * <= is required because the CPU will access up to 1144 * <= is required because the CPU will access up to
1074 * 8 bits beyond the end of the IO permission bitmap. 1145 * 8 bits beyond the end of the IO permission bitmap.
@@ -1078,8 +1149,7 @@ void __cpuinit cpu_init(void)
1078 1149
1079 atomic_inc(&init_mm.mm_count); 1150 atomic_inc(&init_mm.mm_count);
1080 me->active_mm = &init_mm; 1151 me->active_mm = &init_mm;
1081 if (me->mm) 1152 BUG_ON(me->mm);
1082 BUG();
1083 enter_lazy_tlb(&init_mm, me); 1153 enter_lazy_tlb(&init_mm, me);
1084 1154
1085 load_sp0(t, &current->thread); 1155 load_sp0(t, &current->thread);
@@ -1098,17 +1168,7 @@ void __cpuinit cpu_init(void)
1098 arch_kgdb_ops.correct_hw_break(); 1168 arch_kgdb_ops.correct_hw_break();
1099 else 1169 else
1100#endif 1170#endif
1101 { 1171 clear_all_debug_regs();
1102 /*
1103 * Clear all 6 debug registers:
1104 */
1105 set_debugreg(0UL, 0);
1106 set_debugreg(0UL, 1);
1107 set_debugreg(0UL, 2);
1108 set_debugreg(0UL, 3);
1109 set_debugreg(0UL, 6);
1110 set_debugreg(0UL, 7);
1111 }
1112 1172
1113 fpu_init(); 1173 fpu_init();
1114 1174
@@ -1129,7 +1189,8 @@ void __cpuinit cpu_init(void)
1129 1189
1130 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) { 1190 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1131 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu); 1191 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1132 for (;;) local_irq_enable(); 1192 for (;;)
1193 local_irq_enable();
1133 } 1194 }
1134 1195
1135 printk(KERN_INFO "Initializing CPU#%d\n", cpu); 1196 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
@@ -1145,8 +1206,7 @@ void __cpuinit cpu_init(void)
1145 */ 1206 */
1146 atomic_inc(&init_mm.mm_count); 1207 atomic_inc(&init_mm.mm_count);
1147 curr->active_mm = &init_mm; 1208 curr->active_mm = &init_mm;
1148 if (curr->mm) 1209 BUG_ON(curr->mm);
1149 BUG();
1150 enter_lazy_tlb(&init_mm, curr); 1210 enter_lazy_tlb(&init_mm, curr);
1151 1211
1152 load_sp0(t, thread); 1212 load_sp0(t, thread);
@@ -1159,13 +1219,7 @@ void __cpuinit cpu_init(void)
1159 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss); 1219 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1160#endif 1220#endif
1161 1221
1162 /* Clear all 6 debug registers: */ 1222 clear_all_debug_regs();
1163 set_debugreg(0, 0);
1164 set_debugreg(0, 1);
1165 set_debugreg(0, 2);
1166 set_debugreg(0, 3);
1167 set_debugreg(0, 6);
1168 set_debugreg(0, 7);
1169 1223
1170 /* 1224 /*
1171 * Force FPU initialization: 1225 * Force FPU initialization:
@@ -1185,6 +1239,4 @@ void __cpuinit cpu_init(void)
1185 1239
1186 xsave_init(); 1240 xsave_init();
1187} 1241}
1188
1189
1190#endif 1242#endif