x86: move 32bit related functions together
[deliverable/linux.git] / arch / x86 / kernel / cpu / common_64.c
CommitLineData
f580366f 1#include <linux/init.h>
0f0124fa
YL
2#include <linux/kernel.h>
3#include <linux/sched.h>
4#include <linux/string.h>
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>
f580366f
YL
10#include <linux/delay.h>
11#include <linux/smp.h>
f580366f 12#include <linux/percpu.h>
f580366f
YL
13#include <asm/i387.h>
14#include <asm/msr.h>
15#include <asm/io.h>
cbcd79c2 16#include <asm/linkage.h>
f580366f
YL
17#include <asm/mmu_context.h>
18#include <asm/mtrr.h>
19#include <asm/mce.h>
20#include <asm/pat.h>
7e00df58 21#include <asm/asm.h>
f580366f
YL
22#include <asm/numa.h>
23#ifdef CONFIG_X86_LOCAL_APIC
24#include <asm/mpspec.h>
25#include <asm/apic.h>
26#include <mach_apic.h>
27#endif
0f0124fa
YL
28#include <asm/pda.h>
29#include <asm/pgtable.h>
30#include <asm/processor.h>
31#include <asm/desc.h>
32#include <asm/atomic.h>
33#include <asm/proto.h>
34#include <asm/sections.h>
35#include <asm/setup.h>
36#include <asm/genapic.h>
f580366f
YL
37
38#include "cpu.h"
39
0a488a53
YL
40static struct cpu_dev *this_cpu __cpuinitdata;
41
f580366f
YL
42/* We need valid kernel segments for data and code in long mode too
43 * IRET will check the segment types kkeil 2000/10/28
44 * Also sysret mandates a special GDT layout
45 */
46/* The TLS descriptors are currently at a different place compared to i386.
47 Hopefully nobody expects them at a fixed place (Wine?) */
48DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
49 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
50 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
51 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
52 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
53 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
54 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
55} };
56EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
57
58__u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
59
60/* Current gdt points %fs at the "master" per-cpu area: after this,
61 * it's on the real one. */
62void switch_to_new_gdt(void)
63{
64 struct desc_ptr gdt_descr;
65
66 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
67 gdt_descr.size = GDT_SIZE - 1;
68 load_gdt(&gdt_descr);
69}
70
10a434fc 71static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
f580366f
YL
72
73static void __cpuinit default_init(struct cpuinfo_x86 *c)
74{
75 display_cacheinfo(c);
76}
77
78static struct cpu_dev __cpuinitdata default_cpu = {
79 .c_init = default_init,
80 .c_vendor = "Unknown",
10a434fc 81 .c_x86_vendor = X86_VENDOR_UNKNOWN,
f580366f 82};
f580366f
YL
83
84int __cpuinit get_model_name(struct cpuinfo_x86 *c)
85{
86 unsigned int *v;
01b2e16a 87 char *p, *q;
f580366f
YL
88
89 if (c->extended_cpuid_level < 0x80000004)
90 return 0;
91
92 v = (unsigned int *) c->x86_model_id;
93 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
94 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
95 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
96 c->x86_model_id[48] = 0;
01b2e16a
YL
97
98 /* Intel chips right-justify this string for some dumb reason;
99 undo that brain damage */
100 p = q = &c->x86_model_id[0];
101 while (*p == ' ')
102 p++;
103 if (p != q) {
104 while (*p)
105 *q++ = *p++;
106 while (q <= &c->x86_model_id[48])
107 *q++ = '\0'; /* Zero-pad the rest */
108 }
109
f580366f
YL
110 return 1;
111}
112
113
114void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
115{
0a488a53 116 unsigned int n, dummy, ebx, ecx, edx, l2size;
f580366f
YL
117
118 n = c->extended_cpuid_level;
119
120 if (n >= 0x80000005) {
121 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
9d31d35b
YL
122 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
123 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
f580366f
YL
124 c->x86_cache_size = (ecx>>24) + (edx>>24);
125 /* On K8 L1 TLB is inclusive, so don't count it */
126 c->x86_tlbsize = 0;
127 }
128
0a488a53
YL
129 if (n < 0x80000006) /* Some chips just has a large L1. */
130 return;
f580366f 131
0a488a53
YL
132 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
133 l2size = ecx >> 16;
134 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
135
136 c->x86_cache_size = l2size;
137
138 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
139 l2size, ecx & 0xFF);
f580366f
YL
140}
141
142void __cpuinit detect_ht(struct cpuinfo_x86 *c)
143{
144#ifdef CONFIG_SMP
145 u32 eax, ebx, ecx, edx;
146 int index_msb, core_bits;
147
f580366f
YL
148 if (!cpu_has(c, X86_FEATURE_HT))
149 return;
150 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
151 goto out;
152
0a488a53
YL
153 cpuid(1, &eax, &ebx, &ecx, &edx);
154
f580366f
YL
155 smp_num_siblings = (ebx & 0xff0000) >> 16;
156
157 if (smp_num_siblings == 1) {
158 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
159 } else if (smp_num_siblings > 1) {
160
161 if (smp_num_siblings > NR_CPUS) {
9d31d35b
YL
162 printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
163 smp_num_siblings);
f580366f
YL
164 smp_num_siblings = 1;
165 return;
166 }
167
168 index_msb = get_count_order(smp_num_siblings);
169 c->phys_proc_id = phys_pkg_id(index_msb);
170
171 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
172
173 index_msb = get_count_order(smp_num_siblings);
174
175 core_bits = get_count_order(c->x86_max_cores);
176
177 c->cpu_core_id = phys_pkg_id(index_msb) &
178 ((1 << core_bits) - 1);
179 }
0a488a53 180
f580366f
YL
181out:
182 if ((c->x86_max_cores * smp_num_siblings) > 1) {
183 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
184 c->phys_proc_id);
185 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
186 c->cpu_core_id);
187 }
f580366f
YL
188#endif
189}
190
191static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
192{
193 char *v = c->x86_vendor_id;
194 int i;
195 static int printed;
196
197 for (i = 0; i < X86_VENDOR_NUM; i++) {
10a434fc
YL
198 if (!cpu_devs[i])
199 break;
200
201 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
202 (cpu_devs[i]->c_ident[1] &&
203 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
204 this_cpu = cpu_devs[i];
205 c->x86_vendor = this_cpu->c_x86_vendor;
206 return;
f580366f
YL
207 }
208 }
10a434fc 209
f580366f
YL
210 if (!printed) {
211 printed++;
212 printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
213 printk(KERN_ERR "CPU: Your system may be unstable.\n");
214 }
10a434fc 215
f580366f 216 c->x86_vendor = X86_VENDOR_UNKNOWN;
3da99c97 217 this_cpu = &default_cpu;
f580366f
YL
218}
219
3da99c97 220void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
f580366f 221{
f580366f
YL
222 /* Get vendor name */
223 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
224 (unsigned int *)&c->x86_vendor_id[0],
225 (unsigned int *)&c->x86_vendor_id[8],
226 (unsigned int *)&c->x86_vendor_id[4]);
227
9d31d35b 228 c->x86 = 4;
f580366f
YL
229 /* Intel-defined flags: level 0x00000001 */
230 if (c->cpuid_level >= 0x00000001) {
3da99c97
YL
231 u32 junk, tfms, cap0, misc;
232 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
f580366f
YL
233 c->x86 = (tfms >> 8) & 0xf;
234 c->x86_model = (tfms >> 4) & 0xf;
235 c->x86_mask = tfms & 0xf;
236 if (c->x86 == 0xf)
237 c->x86 += (tfms >> 20) & 0xff;
238 if (c->x86 >= 0x6)
9d31d35b
YL
239 c->x86_model += ((tfms >> 16) & 0xf) << 4;
240 if (cap0 & (1<<19)) {
f580366f 241 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
9d31d35b
YL
242 c->x86_cache_alignment = c->x86_clflush_size;
243 }
f580366f 244 }
3da99c97
YL
245}
246
247
248static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
249{
250 u32 tfms, xlvl;
251 u32 ebx;
252
3da99c97
YL
253 /* Intel-defined flags: level 0x00000001 */
254 if (c->cpuid_level >= 0x00000001) {
255 u32 capability, excap;
256
257 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
258 c->x86_capability[0] = capability;
259 c->x86_capability[4] = excap;
260 }
f580366f 261
f580366f
YL
262 /* AMD-defined flags: level 0x80000001 */
263 xlvl = cpuid_eax(0x80000000);
264 c->extended_cpuid_level = xlvl;
265 if ((xlvl & 0xffff0000) == 0x80000000) {
266 if (xlvl >= 0x80000001) {
267 c->x86_capability[1] = cpuid_edx(0x80000001);
268 c->x86_capability[6] = cpuid_ecx(0x80000001);
269 }
f580366f
YL
270 }
271
272 /* Transmeta-defined flags: level 0x80860001 */
273 xlvl = cpuid_eax(0x80860000);
274 if ((xlvl & 0xffff0000) == 0x80860000) {
275 /* Don't set x86_cpuid_level here for now to not confuse. */
276 if (xlvl >= 0x80860001)
277 c->x86_capability[2] = cpuid_edx(0x80860001);
278 }
279
f580366f
YL
280 if (c->extended_cpuid_level >= 0x80000007)
281 c->x86_power = cpuid_edx(0x80000007);
282
87a1c441
YL
283 if (c->extended_cpuid_level >= 0x80000008) {
284 u32 eax = cpuid_eax(0x80000008);
285
286 c->x86_virt_bits = (eax >> 8) & 0xff;
287 c->x86_phys_bits = eax & 0xff;
288 }
3da99c97 289}
87a1c441 290
3da99c97
YL
291/* Do some early cpuid on the boot CPU to get some parameter that are
292 needed before check_bugs. Everything advanced is in identify_cpu
293 below. */
294static void __init early_identify_cpu(struct cpuinfo_x86 *c)
295{
296
297 c->x86_clflush_size = 64;
298 c->x86_cache_alignment = c->x86_clflush_size;
299
300 memset(&c->x86_capability, 0, sizeof c->x86_capability);
301
302 c->extended_cpuid_level = 0;
303
304 cpu_detect(c);
305
306 get_cpu_vendor(c);
307
308 get_cpu_cap(c);
7e00df58 309
10a434fc
YL
310 if (this_cpu->c_early_init)
311 this_cpu->c_early_init(c);
f580366f
YL
312
313 validate_pat_support(c);
f580366f
YL
314}
315
3da99c97
YL
316void __init early_cpu_init(void)
317{
10a434fc
YL
318 struct cpu_dev **cdev;
319 int count = 0;
320
321 printk("KERNEL supported cpus:\n");
322 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
323 struct cpu_dev *cpudev = *cdev;
324 unsigned int j;
3da99c97 325
10a434fc
YL
326 if (count >= X86_VENDOR_NUM)
327 break;
328 cpu_devs[count] = cpudev;
329 count++;
330
331 for (j = 0; j < 2; j++) {
332 if (!cpudev->c_ident[j])
333 continue;
334 printk(" %s %s\n", cpudev->c_vendor,
335 cpudev->c_ident[j]);
336 }
337 }
3da99c97 338
3da99c97
YL
339 early_identify_cpu(&boot_cpu_data);
340}
341
9d31d35b
YL
342/*
343 * The NOPL instruction is supposed to exist on all CPUs with
344 * family >= 6, unfortunately, that's not true in practice because
345 * of early VIA chips and (more importantly) broken virtualizers that
346 * are not easy to detect. Hence, probe for it based on first
347 * principles.
348 *
349 * Note: no 64-bit chip is known to lack these, but put the code here
350 * for consistency with 32 bits, and to make it utterly trivial to
351 * diagnose the problem should it ever surface.
352 */
353static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
354{
355 const u32 nopl_signature = 0x888c53b1; /* Random number */
356 u32 has_nopl = nopl_signature;
357
358 clear_cpu_cap(c, X86_FEATURE_NOPL);
359 if (c->x86 >= 6) {
360 asm volatile("\n"
361 "1: .byte 0x0f,0x1f,0xc0\n" /* nopl %eax */
362 "2:\n"
363 " .section .fixup,\"ax\"\n"
364 "3: xor %0,%0\n"
365 " jmp 2b\n"
366 " .previous\n"
367 _ASM_EXTABLE(1b,3b)
368 : "+a" (has_nopl));
369
370 if (has_nopl == nopl_signature)
371 set_cpu_cap(c, X86_FEATURE_NOPL);
372 }
373}
374
3da99c97
YL
375static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
376{
377 c->extended_cpuid_level = 0;
378
379 cpu_detect(c);
380
381 get_cpu_vendor(c);
382
383 get_cpu_cap(c);
384
385 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
386#ifdef CONFIG_SMP
387 c->phys_proc_id = c->initial_apicid;
388#endif
389
390 if (c->extended_cpuid_level >= 0x80000004)
391 get_model_name(c); /* Default name */
392
393 init_scattered_cpuid_features(c);
394 detect_nopl(c);
395}
396
f580366f
YL
397/*
398 * This does the hard work of actually picking apart the CPU stuff...
399 */
9a250347 400static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
f580366f
YL
401{
402 int i;
403
3da99c97
YL
404 c->loops_per_jiffy = loops_per_jiffy;
405 c->x86_cache_size = -1;
406 c->x86_vendor = X86_VENDOR_UNKNOWN;
407 c->x86_model = c->x86_mask = 0; /* So far unknown... */
408 c->x86_vendor_id[0] = '\0'; /* Unset */
409 c->x86_model_id[0] = '\0'; /* Unset */
3da99c97
YL
410 c->x86_max_cores = 1;
411 c->x86_coreid_bits = 0;
0a488a53
YL
412 c->x86_clflush_size = 64;
413 c->x86_cache_alignment = c->x86_clflush_size;
3da99c97 414 memset(&c->x86_capability, 0, sizeof c->x86_capability);
f580366f 415
3da99c97 416 generic_identify(c);
f580366f
YL
417
418 c->apicid = phys_pkg_id(0);
419
420 /*
421 * Vendor-specific initialization. In this section we
422 * canonicalize the feature flags, meaning if there are
423 * features a certain CPU supports which CPUID doesn't
424 * tell us, CPUID claiming incorrect flags, or other bugs,
425 * we handle them here.
426 *
427 * At the end of this section, c->x86_capability better
428 * indicate the features this CPU genuinely supports!
429 */
430 if (this_cpu->c_init)
431 this_cpu->c_init(c);
432
433 detect_ht(c);
434
435 /*
436 * On SMP, boot_cpu_data holds the common feature set between
437 * all CPUs; so make sure that we indicate which features are
438 * common between the CPUs. The first time this routine gets
439 * executed, c == &boot_cpu_data.
440 */
441 if (c != &boot_cpu_data) {
442 /* AND the already accumulated flags with these */
443 for (i = 0; i < NCAPINTS; i++)
444 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
445 }
446
447 /* Clear all flags overriden by options */
448 for (i = 0; i < NCAPINTS; i++)
449 c->x86_capability[i] &= ~cleared_cpu_caps[i];
450
451#ifdef CONFIG_X86_MCE
452 mcheck_init(c);
453#endif
454 select_idle_routine(c);
455
456#ifdef CONFIG_NUMA
457 numa_add_cpu(smp_processor_id());
458#endif
459
460}
461
9d31d35b 462void __init identify_boot_cpu(void)
f580366f
YL
463{
464 identify_cpu(&boot_cpu_data);
465}
466
467void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
468{
469 BUG_ON(c == &boot_cpu_data);
470 identify_cpu(c);
471 mtrr_ap_init();
472}
473
b05f78f5
YL
474struct msr_range {
475 unsigned min;
476 unsigned max;
477};
478
479static struct msr_range msr_range_array[] __cpuinitdata = {
480 { 0x00000000, 0x00000418},
481 { 0xc0000000, 0xc000040b},
482 { 0xc0010000, 0xc0010142},
483 { 0xc0011000, 0xc001103b},
484};
485
486static void __cpuinit print_cpu_msr(void)
487{
488 unsigned index;
489 u64 val;
490 int i;
491 unsigned index_min, index_max;
492
493 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
494 index_min = msr_range_array[i].min;
495 index_max = msr_range_array[i].max;
496 for (index = index_min; index < index_max; index++) {
497 if (rdmsrl_amd_safe(index, &val))
498 continue;
499 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
500 }
501 }
502}
503
504static int show_msr __cpuinitdata;
505static __init int setup_show_msr(char *arg)
506{
507 int num;
508
509 get_option(&arg, &num);
510
511 if (num > 0)
512 show_msr = num;
513 return 1;
514}
515__setup("show_msr=", setup_show_msr);
516
9d31d35b
YL
517static __init int setup_noclflush(char *arg)
518{
519 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
520 return 1;
521}
522__setup("noclflush", setup_noclflush);
523
f580366f
YL
524void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
525{
526 if (c->x86_model_id[0])
527 printk(KERN_CONT "%s", c->x86_model_id);
528
529 if (c->x86_mask || c->cpuid_level >= 0)
530 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
531 else
532 printk(KERN_CONT "\n");
b05f78f5
YL
533
534#ifdef CONFIG_SMP
535 if (c->cpu_index < show_msr)
536 print_cpu_msr();
537#else
538 if (show_msr)
539 print_cpu_msr();
540#endif
f580366f
YL
541}
542
543static __init int setup_disablecpuid(char *arg)
544{
545 int bit;
546 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
547 setup_clear_cpu_cap(bit);
548 else
549 return 0;
550 return 1;
551}
552__setup("clearcpuid=", setup_disablecpuid);
0f0124fa 553
0f0124fa
YL
554cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
555
556struct x8664_pda **_cpu_pda __read_mostly;
557EXPORT_SYMBOL(_cpu_pda);
558
559struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
560
561char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
562
563unsigned long __supported_pte_mask __read_mostly = ~0UL;
564EXPORT_SYMBOL_GPL(__supported_pte_mask);
565
566static int do_not_nx __cpuinitdata;
567
568/* noexec=on|off
569Control non executable mappings for 64bit processes.
570
571on Enable(default)
572off Disable
573*/
574static int __init nonx_setup(char *str)
575{
576 if (!str)
577 return -EINVAL;
578 if (!strncmp(str, "on", 2)) {
579 __supported_pte_mask |= _PAGE_NX;
580 do_not_nx = 0;
581 } else if (!strncmp(str, "off", 3)) {
582 do_not_nx = 1;
583 __supported_pte_mask &= ~_PAGE_NX;
584 }
585 return 0;
586}
587early_param("noexec", nonx_setup);
588
589int force_personality32;
590
591/* noexec32=on|off
592Control non executable heap for 32bit processes.
593To control the stack too use noexec=off
594
595on PROT_READ does not imply PROT_EXEC for 32bit processes (default)
596off PROT_READ implies PROT_EXEC
597*/
598static int __init nonx32_setup(char *str)
599{
600 if (!strcmp(str, "on"))
601 force_personality32 &= ~READ_IMPLIES_EXEC;
602 else if (!strcmp(str, "off"))
603 force_personality32 |= READ_IMPLIES_EXEC;
604 return 1;
605}
606__setup("noexec32=", nonx32_setup);
607
608void pda_init(int cpu)
609{
610 struct x8664_pda *pda = cpu_pda(cpu);
611
612 /* Setup up data that may be needed in __get_free_pages early */
ada85708
JF
613 loadsegment(fs, 0);
614 loadsegment(gs, 0);
0f0124fa
YL
615 /* Memory clobbers used to order PDA accessed */
616 mb();
617 wrmsrl(MSR_GS_BASE, pda);
618 mb();
619
620 pda->cpunumber = cpu;
621 pda->irqcount = -1;
622 pda->kernelstack = (unsigned long)stack_thread_info() -
623 PDA_STACKOFFSET + THREAD_SIZE;
624 pda->active_mm = &init_mm;
625 pda->mmu_state = 0;
626
627 if (cpu == 0) {
628 /* others are initialized in smpboot.c */
629 pda->pcurrent = &init_task;
630 pda->irqstackptr = boot_cpu_stack;
631 } else {
632 pda->irqstackptr = (char *)
633 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
634 if (!pda->irqstackptr)
635 panic("cannot allocate irqstack for cpu %d", cpu);
636
637 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
638 pda->nodenumber = cpu_to_node(cpu);
639 }
640
641 pda->irqstackptr += IRQSTACKSIZE-64;
642}
643
644char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
cbcd79c2 645 DEBUG_STKSZ] __page_aligned_bss;
0f0124fa
YL
646
647extern asmlinkage void ignore_sysret(void);
648
649/* May not be marked __init: used by software suspend */
650void syscall_init(void)
651{
652 /*
653 * LSTAR and STAR live in a bit strange symbiosis.
654 * They both write to the same internal register. STAR allows to
655 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
656 */
657 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
658 wrmsrl(MSR_LSTAR, system_call);
659 wrmsrl(MSR_CSTAR, ignore_sysret);
660
661#ifdef CONFIG_IA32_EMULATION
662 syscall32_cpu_init();
663#endif
664
665 /* Flags to clear on syscall */
666 wrmsrl(MSR_SYSCALL_MASK,
667 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
668}
669
670void __cpuinit check_efer(void)
671{
672 unsigned long efer;
673
674 rdmsrl(MSR_EFER, efer);
675 if (!(efer & EFER_NX) || do_not_nx)
676 __supported_pte_mask &= ~_PAGE_NX;
677}
678
679unsigned long kernel_eflags;
680
681/*
682 * Copies of the original ist values from the tss are only accessed during
683 * debugging, no special alignment required.
684 */
685DEFINE_PER_CPU(struct orig_ist, orig_ist);
686
687/*
688 * cpu_init() initializes state that is per-CPU. Some data is already
689 * initialized (naturally) in the bootstrap process, such as the GDT
690 * and IDT. We reload them nevertheless, this function acts as a
691 * 'CPU state barrier', nothing should get across.
692 * A lot of state is already set up in PDA init.
693 */
694void __cpuinit cpu_init(void)
695{
696 int cpu = stack_smp_processor_id();
697 struct tss_struct *t = &per_cpu(init_tss, cpu);
698 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
699 unsigned long v;
700 char *estacks = NULL;
701 struct task_struct *me;
702 int i;
703
704 /* CPU 0 is initialised in head64.c */
705 if (cpu != 0)
706 pda_init(cpu);
707 else
708 estacks = boot_exception_stacks;
709
710 me = current;
711
712 if (cpu_test_and_set(cpu, cpu_initialized))
713 panic("CPU#%d already initialized!\n", cpu);
714
715 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
716
717 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
718
719 /*
720 * Initialize the per-CPU GDT with the boot GDT,
721 * and set up the GDT descriptor:
722 */
723
724 switch_to_new_gdt();
725 load_idt((const struct desc_ptr *)&idt_descr);
726
727 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
728 syscall_init();
729
730 wrmsrl(MSR_FS_BASE, 0);
731 wrmsrl(MSR_KERNEL_GS_BASE, 0);
732 barrier();
733
734 check_efer();
735
736 /*
737 * set up and load the per-CPU TSS
738 */
739 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
740 static const unsigned int order[N_EXCEPTION_STACKS] = {
741 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
742 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
743 };
744 if (cpu) {
745 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
746 if (!estacks)
747 panic("Cannot allocate exception stack %ld %d\n",
748 v, cpu);
749 }
750 estacks += PAGE_SIZE << order[v];
751 orig_ist->ist[v] = t->x86_tss.ist[v] = (unsigned long)estacks;
752 }
753
754 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
755 /*
756 * <= is required because the CPU will access up to
757 * 8 bits beyond the end of the IO permission bitmap.
758 */
759 for (i = 0; i <= IO_BITMAP_LONGS; i++)
760 t->io_bitmap[i] = ~0UL;
761
762 atomic_inc(&init_mm.mm_count);
763 me->active_mm = &init_mm;
764 if (me->mm)
765 BUG();
766 enter_lazy_tlb(&init_mm, me);
767
768 load_sp0(t, &current->thread);
769 set_tss_desc(cpu, t);
770 load_TR_desc();
771 load_LDT(&init_mm.context);
772
773#ifdef CONFIG_KGDB
774 /*
775 * If the kgdb is connected no debug regs should be altered. This
776 * is only applicable when KGDB and a KGDB I/O module are built
777 * into the kernel and you are using early debugging with
778 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
779 */
780 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
781 arch_kgdb_ops.correct_hw_break();
782 else {
783#endif
784 /*
785 * Clear all 6 debug registers:
786 */
787
788 set_debugreg(0UL, 0);
789 set_debugreg(0UL, 1);
790 set_debugreg(0UL, 2);
791 set_debugreg(0UL, 3);
792 set_debugreg(0UL, 6);
793 set_debugreg(0UL, 7);
794#ifdef CONFIG_KGDB
795 /* If the kgdb is connected no debug regs should be altered. */
796 }
797#endif
798
799 fpu_init();
800
801 raw_local_save_flags(kernel_eflags);
802
803 if (is_uv_system())
804 uv_cpu_init();
805}
This page took 0.171133 seconds and 5 git commands to generate.