common_64.c 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805
  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>
  6. #include <linux/bitops.h>
  7. #include <linux/module.h>
  8. #include <linux/kgdb.h>
  9. #include <linux/topology.h>
  10. #include <linux/delay.h>
  11. #include <linux/smp.h>
  12. #include <linux/percpu.h>
  13. #include <asm/i387.h>
  14. #include <asm/msr.h>
  15. #include <asm/io.h>
  16. #include <asm/linkage.h>
  17. #include <asm/mmu_context.h>
  18. #include <asm/mtrr.h>
  19. #include <asm/mce.h>
  20. #include <asm/pat.h>
  21. #include <asm/asm.h>
  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
  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>
  37. #include "cpu.h"
  38. static struct cpu_dev *this_cpu __cpuinitdata;
  39. /* We need valid kernel segments for data and code in long mode too
  40. * IRET will check the segment types kkeil 2000/10/28
  41. * Also sysret mandates a special GDT layout
  42. */
  43. /* The TLS descriptors are currently at a different place compared to i386.
  44. Hopefully nobody expects them at a fixed place (Wine?) */
  45. DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
  46. [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
  47. [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
  48. [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
  49. [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
  50. [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
  51. [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
  52. } };
  53. EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
  54. __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
  55. /* Current gdt points %fs at the "master" per-cpu area: after this,
  56. * it's on the real one. */
  57. void switch_to_new_gdt(void)
  58. {
  59. struct desc_ptr gdt_descr;
  60. gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
  61. gdt_descr.size = GDT_SIZE - 1;
  62. load_gdt(&gdt_descr);
  63. }
  64. static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
  65. static void __cpuinit default_init(struct cpuinfo_x86 *c)
  66. {
  67. display_cacheinfo(c);
  68. }
  69. static struct cpu_dev __cpuinitdata default_cpu = {
  70. .c_init = default_init,
  71. .c_vendor = "Unknown",
  72. .c_x86_vendor = X86_VENDOR_UNKNOWN,
  73. };
  74. int __cpuinit get_model_name(struct cpuinfo_x86 *c)
  75. {
  76. unsigned int *v;
  77. char *p, *q;
  78. if (c->extended_cpuid_level < 0x80000004)
  79. return 0;
  80. v = (unsigned int *) c->x86_model_id;
  81. cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
  82. cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
  83. cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
  84. c->x86_model_id[48] = 0;
  85. /* Intel chips right-justify this string for some dumb reason;
  86. undo that brain damage */
  87. p = q = &c->x86_model_id[0];
  88. while (*p == ' ')
  89. p++;
  90. if (p != q) {
  91. while (*p)
  92. *q++ = *p++;
  93. while (q <= &c->x86_model_id[48])
  94. *q++ = '\0'; /* Zero-pad the rest */
  95. }
  96. return 1;
  97. }
  98. void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
  99. {
  100. unsigned int n, dummy, ebx, ecx, edx, l2size;
  101. n = c->extended_cpuid_level;
  102. if (n >= 0x80000005) {
  103. cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
  104. printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
  105. edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
  106. c->x86_cache_size = (ecx>>24) + (edx>>24);
  107. /* On K8 L1 TLB is inclusive, so don't count it */
  108. c->x86_tlbsize = 0;
  109. }
  110. if (n < 0x80000006) /* Some chips just has a large L1. */
  111. return;
  112. cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
  113. l2size = ecx >> 16;
  114. c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
  115. c->x86_cache_size = l2size;
  116. printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
  117. l2size, ecx & 0xFF);
  118. }
  119. void __cpuinit detect_ht(struct cpuinfo_x86 *c)
  120. {
  121. #ifdef CONFIG_SMP
  122. u32 eax, ebx, ecx, edx;
  123. int index_msb, core_bits;
  124. if (!cpu_has(c, X86_FEATURE_HT))
  125. return;
  126. if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
  127. goto out;
  128. cpuid(1, &eax, &ebx, &ecx, &edx);
  129. smp_num_siblings = (ebx & 0xff0000) >> 16;
  130. if (smp_num_siblings == 1) {
  131. printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
  132. } else if (smp_num_siblings > 1) {
  133. if (smp_num_siblings > NR_CPUS) {
  134. printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
  135. smp_num_siblings);
  136. smp_num_siblings = 1;
  137. return;
  138. }
  139. index_msb = get_count_order(smp_num_siblings);
  140. c->phys_proc_id = phys_pkg_id(index_msb);
  141. smp_num_siblings = smp_num_siblings / c->x86_max_cores;
  142. index_msb = get_count_order(smp_num_siblings);
  143. core_bits = get_count_order(c->x86_max_cores);
  144. c->cpu_core_id = phys_pkg_id(index_msb) &
  145. ((1 << core_bits) - 1);
  146. }
  147. out:
  148. if ((c->x86_max_cores * smp_num_siblings) > 1) {
  149. printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
  150. c->phys_proc_id);
  151. printk(KERN_INFO "CPU: Processor Core ID: %d\n",
  152. c->cpu_core_id);
  153. }
  154. #endif
  155. }
  156. static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
  157. {
  158. char *v = c->x86_vendor_id;
  159. int i;
  160. static int printed;
  161. for (i = 0; i < X86_VENDOR_NUM; i++) {
  162. if (!cpu_devs[i])
  163. break;
  164. if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
  165. (cpu_devs[i]->c_ident[1] &&
  166. !strcmp(v, cpu_devs[i]->c_ident[1]))) {
  167. this_cpu = cpu_devs[i];
  168. c->x86_vendor = this_cpu->c_x86_vendor;
  169. return;
  170. }
  171. }
  172. if (!printed) {
  173. printed++;
  174. printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
  175. printk(KERN_ERR "CPU: Your system may be unstable.\n");
  176. }
  177. c->x86_vendor = X86_VENDOR_UNKNOWN;
  178. this_cpu = &default_cpu;
  179. }
  180. void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
  181. {
  182. /* Get vendor name */
  183. cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
  184. (unsigned int *)&c->x86_vendor_id[0],
  185. (unsigned int *)&c->x86_vendor_id[8],
  186. (unsigned int *)&c->x86_vendor_id[4]);
  187. c->x86 = 4;
  188. /* Intel-defined flags: level 0x00000001 */
  189. if (c->cpuid_level >= 0x00000001) {
  190. u32 junk, tfms, cap0, misc;
  191. cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
  192. c->x86 = (tfms >> 8) & 0xf;
  193. c->x86_model = (tfms >> 4) & 0xf;
  194. c->x86_mask = tfms & 0xf;
  195. if (c->x86 == 0xf)
  196. c->x86 += (tfms >> 20) & 0xff;
  197. if (c->x86 >= 0x6)
  198. c->x86_model += ((tfms >> 16) & 0xf) << 4;
  199. if (cap0 & (1<<19)) {
  200. c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
  201. c->x86_cache_alignment = c->x86_clflush_size;
  202. }
  203. }
  204. }
  205. static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
  206. {
  207. u32 tfms, xlvl;
  208. u32 ebx;
  209. /* Intel-defined flags: level 0x00000001 */
  210. if (c->cpuid_level >= 0x00000001) {
  211. u32 capability, excap;
  212. cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
  213. c->x86_capability[0] = capability;
  214. c->x86_capability[4] = excap;
  215. }
  216. /* AMD-defined flags: level 0x80000001 */
  217. xlvl = cpuid_eax(0x80000000);
  218. c->extended_cpuid_level = xlvl;
  219. if ((xlvl & 0xffff0000) == 0x80000000) {
  220. if (xlvl >= 0x80000001) {
  221. c->x86_capability[1] = cpuid_edx(0x80000001);
  222. c->x86_capability[6] = cpuid_ecx(0x80000001);
  223. }
  224. }
  225. /* Transmeta-defined flags: level 0x80860001 */
  226. xlvl = cpuid_eax(0x80860000);
  227. if ((xlvl & 0xffff0000) == 0x80860000) {
  228. /* Don't set x86_cpuid_level here for now to not confuse. */
  229. if (xlvl >= 0x80860001)
  230. c->x86_capability[2] = cpuid_edx(0x80860001);
  231. }
  232. if (c->extended_cpuid_level >= 0x80000007)
  233. c->x86_power = cpuid_edx(0x80000007);
  234. if (c->extended_cpuid_level >= 0x80000008) {
  235. u32 eax = cpuid_eax(0x80000008);
  236. c->x86_virt_bits = (eax >> 8) & 0xff;
  237. c->x86_phys_bits = eax & 0xff;
  238. }
  239. }
  240. /* Do some early cpuid on the boot CPU to get some parameter that are
  241. needed before check_bugs. Everything advanced is in identify_cpu
  242. below. */
  243. static void __init early_identify_cpu(struct cpuinfo_x86 *c)
  244. {
  245. c->x86_clflush_size = 64;
  246. c->x86_cache_alignment = c->x86_clflush_size;
  247. memset(&c->x86_capability, 0, sizeof c->x86_capability);
  248. c->extended_cpuid_level = 0;
  249. cpu_detect(c);
  250. get_cpu_vendor(c);
  251. get_cpu_cap(c);
  252. if (this_cpu->c_early_init)
  253. this_cpu->c_early_init(c);
  254. validate_pat_support(c);
  255. }
  256. void __init early_cpu_init(void)
  257. {
  258. struct cpu_dev **cdev;
  259. int count = 0;
  260. printk("KERNEL supported cpus:\n");
  261. for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
  262. struct cpu_dev *cpudev = *cdev;
  263. unsigned int j;
  264. if (count >= X86_VENDOR_NUM)
  265. break;
  266. cpu_devs[count] = cpudev;
  267. count++;
  268. for (j = 0; j < 2; j++) {
  269. if (!cpudev->c_ident[j])
  270. continue;
  271. printk(" %s %s\n", cpudev->c_vendor,
  272. cpudev->c_ident[j]);
  273. }
  274. }
  275. early_identify_cpu(&boot_cpu_data);
  276. }
  277. /*
  278. * The NOPL instruction is supposed to exist on all CPUs with
  279. * family >= 6, unfortunately, that's not true in practice because
  280. * of early VIA chips and (more importantly) broken virtualizers that
  281. * are not easy to detect. Hence, probe for it based on first
  282. * principles.
  283. *
  284. * Note: no 64-bit chip is known to lack these, but put the code here
  285. * for consistency with 32 bits, and to make it utterly trivial to
  286. * diagnose the problem should it ever surface.
  287. */
  288. static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
  289. {
  290. const u32 nopl_signature = 0x888c53b1; /* Random number */
  291. u32 has_nopl = nopl_signature;
  292. clear_cpu_cap(c, X86_FEATURE_NOPL);
  293. if (c->x86 >= 6) {
  294. asm volatile("\n"
  295. "1: .byte 0x0f,0x1f,0xc0\n" /* nopl %eax */
  296. "2:\n"
  297. " .section .fixup,\"ax\"\n"
  298. "3: xor %0,%0\n"
  299. " jmp 2b\n"
  300. " .previous\n"
  301. _ASM_EXTABLE(1b,3b)
  302. : "+a" (has_nopl));
  303. if (has_nopl == nopl_signature)
  304. set_cpu_cap(c, X86_FEATURE_NOPL);
  305. }
  306. }
  307. static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
  308. {
  309. c->extended_cpuid_level = 0;
  310. cpu_detect(c);
  311. get_cpu_vendor(c);
  312. get_cpu_cap(c);
  313. c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
  314. #ifdef CONFIG_SMP
  315. c->phys_proc_id = c->initial_apicid;
  316. #endif
  317. if (c->extended_cpuid_level >= 0x80000004)
  318. get_model_name(c); /* Default name */
  319. init_scattered_cpuid_features(c);
  320. detect_nopl(c);
  321. }
  322. /*
  323. * This does the hard work of actually picking apart the CPU stuff...
  324. */
  325. static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
  326. {
  327. int i;
  328. c->loops_per_jiffy = loops_per_jiffy;
  329. c->x86_cache_size = -1;
  330. c->x86_vendor = X86_VENDOR_UNKNOWN;
  331. c->x86_model = c->x86_mask = 0; /* So far unknown... */
  332. c->x86_vendor_id[0] = '\0'; /* Unset */
  333. c->x86_model_id[0] = '\0'; /* Unset */
  334. c->x86_max_cores = 1;
  335. c->x86_coreid_bits = 0;
  336. c->x86_clflush_size = 64;
  337. c->x86_cache_alignment = c->x86_clflush_size;
  338. memset(&c->x86_capability, 0, sizeof c->x86_capability);
  339. generic_identify(c);
  340. c->apicid = phys_pkg_id(0);
  341. /*
  342. * Vendor-specific initialization. In this section we
  343. * canonicalize the feature flags, meaning if there are
  344. * features a certain CPU supports which CPUID doesn't
  345. * tell us, CPUID claiming incorrect flags, or other bugs,
  346. * we handle them here.
  347. *
  348. * At the end of this section, c->x86_capability better
  349. * indicate the features this CPU genuinely supports!
  350. */
  351. if (this_cpu->c_init)
  352. this_cpu->c_init(c);
  353. detect_ht(c);
  354. /*
  355. * On SMP, boot_cpu_data holds the common feature set between
  356. * all CPUs; so make sure that we indicate which features are
  357. * common between the CPUs. The first time this routine gets
  358. * executed, c == &boot_cpu_data.
  359. */
  360. if (c != &boot_cpu_data) {
  361. /* AND the already accumulated flags with these */
  362. for (i = 0; i < NCAPINTS; i++)
  363. boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
  364. }
  365. /* Clear all flags overriden by options */
  366. for (i = 0; i < NCAPINTS; i++)
  367. c->x86_capability[i] &= ~cleared_cpu_caps[i];
  368. #ifdef CONFIG_X86_MCE
  369. mcheck_init(c);
  370. #endif
  371. select_idle_routine(c);
  372. #ifdef CONFIG_NUMA
  373. numa_add_cpu(smp_processor_id());
  374. #endif
  375. }
  376. void __init identify_boot_cpu(void)
  377. {
  378. identify_cpu(&boot_cpu_data);
  379. }
  380. void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
  381. {
  382. BUG_ON(c == &boot_cpu_data);
  383. identify_cpu(c);
  384. mtrr_ap_init();
  385. }
  386. struct msr_range {
  387. unsigned min;
  388. unsigned max;
  389. };
  390. static struct msr_range msr_range_array[] __cpuinitdata = {
  391. { 0x00000000, 0x00000418},
  392. { 0xc0000000, 0xc000040b},
  393. { 0xc0010000, 0xc0010142},
  394. { 0xc0011000, 0xc001103b},
  395. };
  396. static void __cpuinit print_cpu_msr(void)
  397. {
  398. unsigned index;
  399. u64 val;
  400. int i;
  401. unsigned index_min, index_max;
  402. for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
  403. index_min = msr_range_array[i].min;
  404. index_max = msr_range_array[i].max;
  405. for (index = index_min; index < index_max; index++) {
  406. if (rdmsrl_amd_safe(index, &val))
  407. continue;
  408. printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
  409. }
  410. }
  411. }
  412. static int show_msr __cpuinitdata;
  413. static __init int setup_show_msr(char *arg)
  414. {
  415. int num;
  416. get_option(&arg, &num);
  417. if (num > 0)
  418. show_msr = num;
  419. return 1;
  420. }
  421. __setup("show_msr=", setup_show_msr);
  422. static __init int setup_noclflush(char *arg)
  423. {
  424. setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
  425. return 1;
  426. }
  427. __setup("noclflush", setup_noclflush);
  428. void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
  429. {
  430. if (c->x86_model_id[0])
  431. printk(KERN_CONT "%s", c->x86_model_id);
  432. if (c->x86_mask || c->cpuid_level >= 0)
  433. printk(KERN_CONT " stepping %02x\n", c->x86_mask);
  434. else
  435. printk(KERN_CONT "\n");
  436. #ifdef CONFIG_SMP
  437. if (c->cpu_index < show_msr)
  438. print_cpu_msr();
  439. #else
  440. if (show_msr)
  441. print_cpu_msr();
  442. #endif
  443. }
  444. static __init int setup_disablecpuid(char *arg)
  445. {
  446. int bit;
  447. if (get_option(&arg, &bit) && bit < NCAPINTS*32)
  448. setup_clear_cpu_cap(bit);
  449. else
  450. return 0;
  451. return 1;
  452. }
  453. __setup("clearcpuid=", setup_disablecpuid);
  454. cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
  455. struct x8664_pda **_cpu_pda __read_mostly;
  456. EXPORT_SYMBOL(_cpu_pda);
  457. struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
  458. char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
  459. unsigned long __supported_pte_mask __read_mostly = ~0UL;
  460. EXPORT_SYMBOL_GPL(__supported_pte_mask);
  461. static int do_not_nx __cpuinitdata;
  462. /* noexec=on|off
  463. Control non executable mappings for 64bit processes.
  464. on Enable(default)
  465. off Disable
  466. */
  467. static int __init nonx_setup(char *str)
  468. {
  469. if (!str)
  470. return -EINVAL;
  471. if (!strncmp(str, "on", 2)) {
  472. __supported_pte_mask |= _PAGE_NX;
  473. do_not_nx = 0;
  474. } else if (!strncmp(str, "off", 3)) {
  475. do_not_nx = 1;
  476. __supported_pte_mask &= ~_PAGE_NX;
  477. }
  478. return 0;
  479. }
  480. early_param("noexec", nonx_setup);
  481. int force_personality32;
  482. /* noexec32=on|off
  483. Control non executable heap for 32bit processes.
  484. To control the stack too use noexec=off
  485. on PROT_READ does not imply PROT_EXEC for 32bit processes (default)
  486. off PROT_READ implies PROT_EXEC
  487. */
  488. static int __init nonx32_setup(char *str)
  489. {
  490. if (!strcmp(str, "on"))
  491. force_personality32 &= ~READ_IMPLIES_EXEC;
  492. else if (!strcmp(str, "off"))
  493. force_personality32 |= READ_IMPLIES_EXEC;
  494. return 1;
  495. }
  496. __setup("noexec32=", nonx32_setup);
  497. void pda_init(int cpu)
  498. {
  499. struct x8664_pda *pda = cpu_pda(cpu);
  500. /* Setup up data that may be needed in __get_free_pages early */
  501. loadsegment(fs, 0);
  502. loadsegment(gs, 0);
  503. /* Memory clobbers used to order PDA accessed */
  504. mb();
  505. wrmsrl(MSR_GS_BASE, pda);
  506. mb();
  507. pda->cpunumber = cpu;
  508. pda->irqcount = -1;
  509. pda->kernelstack = (unsigned long)stack_thread_info() -
  510. PDA_STACKOFFSET + THREAD_SIZE;
  511. pda->active_mm = &init_mm;
  512. pda->mmu_state = 0;
  513. if (cpu == 0) {
  514. /* others are initialized in smpboot.c */
  515. pda->pcurrent = &init_task;
  516. pda->irqstackptr = boot_cpu_stack;
  517. } else {
  518. pda->irqstackptr = (char *)
  519. __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
  520. if (!pda->irqstackptr)
  521. panic("cannot allocate irqstack for cpu %d", cpu);
  522. if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
  523. pda->nodenumber = cpu_to_node(cpu);
  524. }
  525. pda->irqstackptr += IRQSTACKSIZE-64;
  526. }
  527. char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
  528. DEBUG_STKSZ] __page_aligned_bss;
  529. extern asmlinkage void ignore_sysret(void);
  530. /* May not be marked __init: used by software suspend */
  531. void syscall_init(void)
  532. {
  533. /*
  534. * LSTAR and STAR live in a bit strange symbiosis.
  535. * They both write to the same internal register. STAR allows to
  536. * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
  537. */
  538. wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
  539. wrmsrl(MSR_LSTAR, system_call);
  540. wrmsrl(MSR_CSTAR, ignore_sysret);
  541. #ifdef CONFIG_IA32_EMULATION
  542. syscall32_cpu_init();
  543. #endif
  544. /* Flags to clear on syscall */
  545. wrmsrl(MSR_SYSCALL_MASK,
  546. X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
  547. }
  548. void __cpuinit check_efer(void)
  549. {
  550. unsigned long efer;
  551. rdmsrl(MSR_EFER, efer);
  552. if (!(efer & EFER_NX) || do_not_nx)
  553. __supported_pte_mask &= ~_PAGE_NX;
  554. }
  555. unsigned long kernel_eflags;
  556. /*
  557. * Copies of the original ist values from the tss are only accessed during
  558. * debugging, no special alignment required.
  559. */
  560. DEFINE_PER_CPU(struct orig_ist, orig_ist);
  561. /*
  562. * cpu_init() initializes state that is per-CPU. Some data is already
  563. * initialized (naturally) in the bootstrap process, such as the GDT
  564. * and IDT. We reload them nevertheless, this function acts as a
  565. * 'CPU state barrier', nothing should get across.
  566. * A lot of state is already set up in PDA init.
  567. */
  568. void __cpuinit cpu_init(void)
  569. {
  570. int cpu = stack_smp_processor_id();
  571. struct tss_struct *t = &per_cpu(init_tss, cpu);
  572. struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
  573. unsigned long v;
  574. char *estacks = NULL;
  575. struct task_struct *me;
  576. int i;
  577. /* CPU 0 is initialised in head64.c */
  578. if (cpu != 0)
  579. pda_init(cpu);
  580. else
  581. estacks = boot_exception_stacks;
  582. me = current;
  583. if (cpu_test_and_set(cpu, cpu_initialized))
  584. panic("CPU#%d already initialized!\n", cpu);
  585. printk(KERN_INFO "Initializing CPU#%d\n", cpu);
  586. clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
  587. /*
  588. * Initialize the per-CPU GDT with the boot GDT,
  589. * and set up the GDT descriptor:
  590. */
  591. switch_to_new_gdt();
  592. load_idt((const struct desc_ptr *)&idt_descr);
  593. memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
  594. syscall_init();
  595. wrmsrl(MSR_FS_BASE, 0);
  596. wrmsrl(MSR_KERNEL_GS_BASE, 0);
  597. barrier();
  598. check_efer();
  599. /*
  600. * set up and load the per-CPU TSS
  601. */
  602. for (v = 0; v < N_EXCEPTION_STACKS; v++) {
  603. static const unsigned int order[N_EXCEPTION_STACKS] = {
  604. [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
  605. [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
  606. };
  607. if (cpu) {
  608. estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
  609. if (!estacks)
  610. panic("Cannot allocate exception stack %ld %d\n",
  611. v, cpu);
  612. }
  613. estacks += PAGE_SIZE << order[v];
  614. orig_ist->ist[v] = t->x86_tss.ist[v] = (unsigned long)estacks;
  615. }
  616. t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
  617. /*
  618. * <= is required because the CPU will access up to
  619. * 8 bits beyond the end of the IO permission bitmap.
  620. */
  621. for (i = 0; i <= IO_BITMAP_LONGS; i++)
  622. t->io_bitmap[i] = ~0UL;
  623. atomic_inc(&init_mm.mm_count);
  624. me->active_mm = &init_mm;
  625. if (me->mm)
  626. BUG();
  627. enter_lazy_tlb(&init_mm, me);
  628. load_sp0(t, &current->thread);
  629. set_tss_desc(cpu, t);
  630. load_TR_desc();
  631. load_LDT(&init_mm.context);
  632. #ifdef CONFIG_KGDB
  633. /*
  634. * If the kgdb is connected no debug regs should be altered. This
  635. * is only applicable when KGDB and a KGDB I/O module are built
  636. * into the kernel and you are using early debugging with
  637. * kgdbwait. KGDB will control the kernel HW breakpoint registers.
  638. */
  639. if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
  640. arch_kgdb_ops.correct_hw_break();
  641. else {
  642. #endif
  643. /*
  644. * Clear all 6 debug registers:
  645. */
  646. set_debugreg(0UL, 0);
  647. set_debugreg(0UL, 1);
  648. set_debugreg(0UL, 2);
  649. set_debugreg(0UL, 3);
  650. set_debugreg(0UL, 6);
  651. set_debugreg(0UL, 7);
  652. #ifdef CONFIG_KGDB
  653. /* If the kgdb is connected no debug regs should be altered. */
  654. }
  655. #endif
  656. fpu_init();
  657. raw_local_save_flags(kernel_eflags);
  658. if (is_uv_system())
  659. uv_cpu_init();
  660. }