common_64.c 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790
  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. /* We need valid kernel segments for data and code in long mode too
  39. * IRET will check the segment types kkeil 2000/10/28
  40. * Also sysret mandates a special GDT layout
  41. */
  42. /* The TLS descriptors are currently at a different place compared to i386.
  43. Hopefully nobody expects them at a fixed place (Wine?) */
  44. DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
  45. [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
  46. [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
  47. [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
  48. [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
  49. [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
  50. [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
  51. } };
  52. EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
  53. __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
  54. /* Current gdt points %fs at the "master" per-cpu area: after this,
  55. * it's on the real one. */
  56. void switch_to_new_gdt(void)
  57. {
  58. struct desc_ptr gdt_descr;
  59. gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
  60. gdt_descr.size = GDT_SIZE - 1;
  61. load_gdt(&gdt_descr);
  62. }
  63. struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
  64. static void __cpuinit default_init(struct cpuinfo_x86 *c)
  65. {
  66. display_cacheinfo(c);
  67. }
  68. static struct cpu_dev __cpuinitdata default_cpu = {
  69. .c_init = default_init,
  70. .c_vendor = "Unknown",
  71. };
  72. static struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
  73. int __cpuinit get_model_name(struct cpuinfo_x86 *c)
  74. {
  75. unsigned int *v;
  76. if (c->extended_cpuid_level < 0x80000004)
  77. return 0;
  78. v = (unsigned int *) c->x86_model_id;
  79. cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
  80. cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
  81. cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
  82. c->x86_model_id[48] = 0;
  83. return 1;
  84. }
  85. void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
  86. {
  87. unsigned int n, dummy, ebx, ecx, edx;
  88. n = c->extended_cpuid_level;
  89. if (n >= 0x80000005) {
  90. cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
  91. printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
  92. edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
  93. c->x86_cache_size = (ecx>>24) + (edx>>24);
  94. /* On K8 L1 TLB is inclusive, so don't count it */
  95. c->x86_tlbsize = 0;
  96. }
  97. if (n >= 0x80000006) {
  98. cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
  99. ecx = cpuid_ecx(0x80000006);
  100. c->x86_cache_size = ecx >> 16;
  101. c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
  102. printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
  103. c->x86_cache_size, ecx & 0xFF);
  104. }
  105. }
  106. void __cpuinit detect_ht(struct cpuinfo_x86 *c)
  107. {
  108. #ifdef CONFIG_SMP
  109. u32 eax, ebx, ecx, edx;
  110. int index_msb, core_bits;
  111. cpuid(1, &eax, &ebx, &ecx, &edx);
  112. if (!cpu_has(c, X86_FEATURE_HT))
  113. return;
  114. if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
  115. goto out;
  116. smp_num_siblings = (ebx & 0xff0000) >> 16;
  117. if (smp_num_siblings == 1) {
  118. printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
  119. } else if (smp_num_siblings > 1) {
  120. if (smp_num_siblings > NR_CPUS) {
  121. printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
  122. smp_num_siblings);
  123. smp_num_siblings = 1;
  124. return;
  125. }
  126. index_msb = get_count_order(smp_num_siblings);
  127. c->phys_proc_id = phys_pkg_id(index_msb);
  128. smp_num_siblings = smp_num_siblings / c->x86_max_cores;
  129. index_msb = get_count_order(smp_num_siblings);
  130. core_bits = get_count_order(c->x86_max_cores);
  131. c->cpu_core_id = phys_pkg_id(index_msb) &
  132. ((1 << core_bits) - 1);
  133. }
  134. out:
  135. if ((c->x86_max_cores * smp_num_siblings) > 1) {
  136. printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
  137. c->phys_proc_id);
  138. printk(KERN_INFO "CPU: Processor Core ID: %d\n",
  139. c->cpu_core_id);
  140. }
  141. #endif
  142. }
  143. static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
  144. {
  145. char *v = c->x86_vendor_id;
  146. int i;
  147. static int printed;
  148. for (i = 0; i < X86_VENDOR_NUM; i++) {
  149. if (cpu_devs[i]) {
  150. if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
  151. (cpu_devs[i]->c_ident[1] &&
  152. !strcmp(v, cpu_devs[i]->c_ident[1]))) {
  153. c->x86_vendor = i;
  154. this_cpu = cpu_devs[i];
  155. return;
  156. }
  157. }
  158. }
  159. if (!printed) {
  160. printed++;
  161. printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
  162. printk(KERN_ERR "CPU: Your system may be unstable.\n");
  163. }
  164. c->x86_vendor = X86_VENDOR_UNKNOWN;
  165. this_cpu = &default_cpu;
  166. }
  167. static void __init early_cpu_support_print(void)
  168. {
  169. int i,j;
  170. struct cpu_dev *cpu_devx;
  171. printk("KERNEL supported cpus:\n");
  172. for (i = 0; i < X86_VENDOR_NUM; i++) {
  173. cpu_devx = cpu_devs[i];
  174. if (!cpu_devx)
  175. continue;
  176. for (j = 0; j < 2; j++) {
  177. if (!cpu_devx->c_ident[j])
  178. continue;
  179. printk(" %s %s\n", cpu_devx->c_vendor,
  180. cpu_devx->c_ident[j]);
  181. }
  182. }
  183. }
  184. void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
  185. {
  186. /* Get vendor name */
  187. cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
  188. (unsigned int *)&c->x86_vendor_id[0],
  189. (unsigned int *)&c->x86_vendor_id[8],
  190. (unsigned int *)&c->x86_vendor_id[4]);
  191. c->x86 = 4;
  192. /* Intel-defined flags: level 0x00000001 */
  193. if (c->cpuid_level >= 0x00000001) {
  194. u32 junk, tfms, cap0, misc;
  195. cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
  196. c->x86 = (tfms >> 8) & 0xf;
  197. c->x86_model = (tfms >> 4) & 0xf;
  198. c->x86_mask = tfms & 0xf;
  199. if (c->x86 == 0xf)
  200. c->x86 += (tfms >> 20) & 0xff;
  201. if (c->x86 >= 0x6)
  202. c->x86_model += ((tfms >> 16) & 0xf) << 4;
  203. if (cap0 & (1<<19)) {
  204. c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
  205. c->x86_cache_alignment = c->x86_clflush_size;
  206. }
  207. }
  208. }
  209. static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
  210. {
  211. u32 tfms, xlvl;
  212. u32 ebx;
  213. /* Intel-defined flags: level 0x00000001 */
  214. if (c->cpuid_level >= 0x00000001) {
  215. u32 capability, excap;
  216. cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
  217. c->x86_capability[0] = capability;
  218. c->x86_capability[4] = excap;
  219. }
  220. /* AMD-defined flags: level 0x80000001 */
  221. xlvl = cpuid_eax(0x80000000);
  222. c->extended_cpuid_level = xlvl;
  223. if ((xlvl & 0xffff0000) == 0x80000000) {
  224. if (xlvl >= 0x80000001) {
  225. c->x86_capability[1] = cpuid_edx(0x80000001);
  226. c->x86_capability[6] = cpuid_ecx(0x80000001);
  227. }
  228. }
  229. /* Transmeta-defined flags: level 0x80860001 */
  230. xlvl = cpuid_eax(0x80860000);
  231. if ((xlvl & 0xffff0000) == 0x80860000) {
  232. /* Don't set x86_cpuid_level here for now to not confuse. */
  233. if (xlvl >= 0x80860001)
  234. c->x86_capability[2] = cpuid_edx(0x80860001);
  235. }
  236. if (c->extended_cpuid_level >= 0x80000007)
  237. c->x86_power = cpuid_edx(0x80000007);
  238. if (c->extended_cpuid_level >= 0x80000008) {
  239. u32 eax = cpuid_eax(0x80000008);
  240. c->x86_virt_bits = (eax >> 8) & 0xff;
  241. c->x86_phys_bits = eax & 0xff;
  242. }
  243. }
  244. /* Do some early cpuid on the boot CPU to get some parameter that are
  245. needed before check_bugs. Everything advanced is in identify_cpu
  246. below. */
  247. static void __init early_identify_cpu(struct cpuinfo_x86 *c)
  248. {
  249. c->x86_clflush_size = 64;
  250. c->x86_cache_alignment = c->x86_clflush_size;
  251. memset(&c->x86_capability, 0, sizeof c->x86_capability);
  252. c->extended_cpuid_level = 0;
  253. cpu_detect(c);
  254. get_cpu_vendor(c);
  255. get_cpu_cap(c);
  256. if (c->x86_vendor != X86_VENDOR_UNKNOWN &&
  257. cpu_devs[c->x86_vendor]->c_early_init)
  258. cpu_devs[c->x86_vendor]->c_early_init(c);
  259. validate_pat_support(c);
  260. }
  261. void __init early_cpu_init(void)
  262. {
  263. struct cpu_vendor_dev *cvdev;
  264. for (cvdev = __x86cpuvendor_start; cvdev < __x86cpuvendor_end; cvdev++)
  265. cpu_devs[cvdev->vendor] = cvdev->cpu_dev;
  266. early_cpu_support_print();
  267. early_identify_cpu(&boot_cpu_data);
  268. }
  269. /*
  270. * The NOPL instruction is supposed to exist on all CPUs with
  271. * family >= 6, unfortunately, that's not true in practice because
  272. * of early VIA chips and (more importantly) broken virtualizers that
  273. * are not easy to detect. Hence, probe for it based on first
  274. * principles.
  275. *
  276. * Note: no 64-bit chip is known to lack these, but put the code here
  277. * for consistency with 32 bits, and to make it utterly trivial to
  278. * diagnose the problem should it ever surface.
  279. */
  280. static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
  281. {
  282. const u32 nopl_signature = 0x888c53b1; /* Random number */
  283. u32 has_nopl = nopl_signature;
  284. clear_cpu_cap(c, X86_FEATURE_NOPL);
  285. if (c->x86 >= 6) {
  286. asm volatile("\n"
  287. "1: .byte 0x0f,0x1f,0xc0\n" /* nopl %eax */
  288. "2:\n"
  289. " .section .fixup,\"ax\"\n"
  290. "3: xor %0,%0\n"
  291. " jmp 2b\n"
  292. " .previous\n"
  293. _ASM_EXTABLE(1b,3b)
  294. : "+a" (has_nopl));
  295. if (has_nopl == nopl_signature)
  296. set_cpu_cap(c, X86_FEATURE_NOPL);
  297. }
  298. }
  299. static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
  300. {
  301. c->extended_cpuid_level = 0;
  302. cpu_detect(c);
  303. get_cpu_vendor(c);
  304. get_cpu_cap(c);
  305. c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
  306. #ifdef CONFIG_SMP
  307. c->phys_proc_id = c->initial_apicid;
  308. #endif
  309. if (c->extended_cpuid_level >= 0x80000004)
  310. get_model_name(c); /* Default name */
  311. init_scattered_cpuid_features(c);
  312. detect_nopl(c);
  313. }
  314. /*
  315. * This does the hard work of actually picking apart the CPU stuff...
  316. */
  317. static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
  318. {
  319. int i;
  320. c->loops_per_jiffy = loops_per_jiffy;
  321. c->x86_cache_size = -1;
  322. c->x86_vendor = X86_VENDOR_UNKNOWN;
  323. c->x86_model = c->x86_mask = 0; /* So far unknown... */
  324. c->x86_vendor_id[0] = '\0'; /* Unset */
  325. c->x86_model_id[0] = '\0'; /* Unset */
  326. c->x86_clflush_size = 64;
  327. c->x86_cache_alignment = c->x86_clflush_size;
  328. c->x86_max_cores = 1;
  329. c->x86_coreid_bits = 0;
  330. memset(&c->x86_capability, 0, sizeof c->x86_capability);
  331. generic_identify(c);
  332. c->apicid = phys_pkg_id(0);
  333. /*
  334. * Vendor-specific initialization. In this section we
  335. * canonicalize the feature flags, meaning if there are
  336. * features a certain CPU supports which CPUID doesn't
  337. * tell us, CPUID claiming incorrect flags, or other bugs,
  338. * we handle them here.
  339. *
  340. * At the end of this section, c->x86_capability better
  341. * indicate the features this CPU genuinely supports!
  342. */
  343. if (this_cpu->c_init)
  344. this_cpu->c_init(c);
  345. detect_ht(c);
  346. /*
  347. * On SMP, boot_cpu_data holds the common feature set between
  348. * all CPUs; so make sure that we indicate which features are
  349. * common between the CPUs. The first time this routine gets
  350. * executed, c == &boot_cpu_data.
  351. */
  352. if (c != &boot_cpu_data) {
  353. /* AND the already accumulated flags with these */
  354. for (i = 0; i < NCAPINTS; i++)
  355. boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
  356. }
  357. /* Clear all flags overriden by options */
  358. for (i = 0; i < NCAPINTS; i++)
  359. c->x86_capability[i] &= ~cleared_cpu_caps[i];
  360. #ifdef CONFIG_X86_MCE
  361. mcheck_init(c);
  362. #endif
  363. select_idle_routine(c);
  364. #ifdef CONFIG_NUMA
  365. numa_add_cpu(smp_processor_id());
  366. #endif
  367. }
  368. void __init identify_boot_cpu(void)
  369. {
  370. identify_cpu(&boot_cpu_data);
  371. }
  372. void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
  373. {
  374. BUG_ON(c == &boot_cpu_data);
  375. identify_cpu(c);
  376. mtrr_ap_init();
  377. }
  378. struct msr_range {
  379. unsigned min;
  380. unsigned max;
  381. };
  382. static struct msr_range msr_range_array[] __cpuinitdata = {
  383. { 0x00000000, 0x00000418},
  384. { 0xc0000000, 0xc000040b},
  385. { 0xc0010000, 0xc0010142},
  386. { 0xc0011000, 0xc001103b},
  387. };
  388. static void __cpuinit print_cpu_msr(void)
  389. {
  390. unsigned index;
  391. u64 val;
  392. int i;
  393. unsigned index_min, index_max;
  394. for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
  395. index_min = msr_range_array[i].min;
  396. index_max = msr_range_array[i].max;
  397. for (index = index_min; index < index_max; index++) {
  398. if (rdmsrl_amd_safe(index, &val))
  399. continue;
  400. printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
  401. }
  402. }
  403. }
  404. static int show_msr __cpuinitdata;
  405. static __init int setup_show_msr(char *arg)
  406. {
  407. int num;
  408. get_option(&arg, &num);
  409. if (num > 0)
  410. show_msr = num;
  411. return 1;
  412. }
  413. __setup("show_msr=", setup_show_msr);
  414. static __init int setup_noclflush(char *arg)
  415. {
  416. setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
  417. return 1;
  418. }
  419. __setup("noclflush", setup_noclflush);
  420. void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
  421. {
  422. if (c->x86_model_id[0])
  423. printk(KERN_CONT "%s", c->x86_model_id);
  424. if (c->x86_mask || c->cpuid_level >= 0)
  425. printk(KERN_CONT " stepping %02x\n", c->x86_mask);
  426. else
  427. printk(KERN_CONT "\n");
  428. #ifdef CONFIG_SMP
  429. if (c->cpu_index < show_msr)
  430. print_cpu_msr();
  431. #else
  432. if (show_msr)
  433. print_cpu_msr();
  434. #endif
  435. }
  436. static __init int setup_disablecpuid(char *arg)
  437. {
  438. int bit;
  439. if (get_option(&arg, &bit) && bit < NCAPINTS*32)
  440. setup_clear_cpu_cap(bit);
  441. else
  442. return 0;
  443. return 1;
  444. }
  445. __setup("clearcpuid=", setup_disablecpuid);
  446. cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
  447. struct x8664_pda **_cpu_pda __read_mostly;
  448. EXPORT_SYMBOL(_cpu_pda);
  449. struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
  450. char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
  451. unsigned long __supported_pte_mask __read_mostly = ~0UL;
  452. EXPORT_SYMBOL_GPL(__supported_pte_mask);
  453. static int do_not_nx __cpuinitdata;
  454. /* noexec=on|off
  455. Control non executable mappings for 64bit processes.
  456. on Enable(default)
  457. off Disable
  458. */
  459. static int __init nonx_setup(char *str)
  460. {
  461. if (!str)
  462. return -EINVAL;
  463. if (!strncmp(str, "on", 2)) {
  464. __supported_pte_mask |= _PAGE_NX;
  465. do_not_nx = 0;
  466. } else if (!strncmp(str, "off", 3)) {
  467. do_not_nx = 1;
  468. __supported_pte_mask &= ~_PAGE_NX;
  469. }
  470. return 0;
  471. }
  472. early_param("noexec", nonx_setup);
  473. int force_personality32;
  474. /* noexec32=on|off
  475. Control non executable heap for 32bit processes.
  476. To control the stack too use noexec=off
  477. on PROT_READ does not imply PROT_EXEC for 32bit processes (default)
  478. off PROT_READ implies PROT_EXEC
  479. */
  480. static int __init nonx32_setup(char *str)
  481. {
  482. if (!strcmp(str, "on"))
  483. force_personality32 &= ~READ_IMPLIES_EXEC;
  484. else if (!strcmp(str, "off"))
  485. force_personality32 |= READ_IMPLIES_EXEC;
  486. return 1;
  487. }
  488. __setup("noexec32=", nonx32_setup);
  489. void pda_init(int cpu)
  490. {
  491. struct x8664_pda *pda = cpu_pda(cpu);
  492. /* Setup up data that may be needed in __get_free_pages early */
  493. loadsegment(fs, 0);
  494. loadsegment(gs, 0);
  495. /* Memory clobbers used to order PDA accessed */
  496. mb();
  497. wrmsrl(MSR_GS_BASE, pda);
  498. mb();
  499. pda->cpunumber = cpu;
  500. pda->irqcount = -1;
  501. pda->kernelstack = (unsigned long)stack_thread_info() -
  502. PDA_STACKOFFSET + THREAD_SIZE;
  503. pda->active_mm = &init_mm;
  504. pda->mmu_state = 0;
  505. if (cpu == 0) {
  506. /* others are initialized in smpboot.c */
  507. pda->pcurrent = &init_task;
  508. pda->irqstackptr = boot_cpu_stack;
  509. } else {
  510. pda->irqstackptr = (char *)
  511. __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
  512. if (!pda->irqstackptr)
  513. panic("cannot allocate irqstack for cpu %d", cpu);
  514. if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
  515. pda->nodenumber = cpu_to_node(cpu);
  516. }
  517. pda->irqstackptr += IRQSTACKSIZE-64;
  518. }
  519. char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
  520. DEBUG_STKSZ] __page_aligned_bss;
  521. extern asmlinkage void ignore_sysret(void);
  522. /* May not be marked __init: used by software suspend */
  523. void syscall_init(void)
  524. {
  525. /*
  526. * LSTAR and STAR live in a bit strange symbiosis.
  527. * They both write to the same internal register. STAR allows to
  528. * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
  529. */
  530. wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
  531. wrmsrl(MSR_LSTAR, system_call);
  532. wrmsrl(MSR_CSTAR, ignore_sysret);
  533. #ifdef CONFIG_IA32_EMULATION
  534. syscall32_cpu_init();
  535. #endif
  536. /* Flags to clear on syscall */
  537. wrmsrl(MSR_SYSCALL_MASK,
  538. X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
  539. }
  540. void __cpuinit check_efer(void)
  541. {
  542. unsigned long efer;
  543. rdmsrl(MSR_EFER, efer);
  544. if (!(efer & EFER_NX) || do_not_nx)
  545. __supported_pte_mask &= ~_PAGE_NX;
  546. }
  547. unsigned long kernel_eflags;
  548. /*
  549. * Copies of the original ist values from the tss are only accessed during
  550. * debugging, no special alignment required.
  551. */
  552. DEFINE_PER_CPU(struct orig_ist, orig_ist);
  553. /*
  554. * cpu_init() initializes state that is per-CPU. Some data is already
  555. * initialized (naturally) in the bootstrap process, such as the GDT
  556. * and IDT. We reload them nevertheless, this function acts as a
  557. * 'CPU state barrier', nothing should get across.
  558. * A lot of state is already set up in PDA init.
  559. */
  560. void __cpuinit cpu_init(void)
  561. {
  562. int cpu = stack_smp_processor_id();
  563. struct tss_struct *t = &per_cpu(init_tss, cpu);
  564. struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
  565. unsigned long v;
  566. char *estacks = NULL;
  567. struct task_struct *me;
  568. int i;
  569. /* CPU 0 is initialised in head64.c */
  570. if (cpu != 0)
  571. pda_init(cpu);
  572. else
  573. estacks = boot_exception_stacks;
  574. me = current;
  575. if (cpu_test_and_set(cpu, cpu_initialized))
  576. panic("CPU#%d already initialized!\n", cpu);
  577. printk(KERN_INFO "Initializing CPU#%d\n", cpu);
  578. clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
  579. /*
  580. * Initialize the per-CPU GDT with the boot GDT,
  581. * and set up the GDT descriptor:
  582. */
  583. switch_to_new_gdt();
  584. load_idt((const struct desc_ptr *)&idt_descr);
  585. memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
  586. syscall_init();
  587. wrmsrl(MSR_FS_BASE, 0);
  588. wrmsrl(MSR_KERNEL_GS_BASE, 0);
  589. barrier();
  590. check_efer();
  591. /*
  592. * set up and load the per-CPU TSS
  593. */
  594. for (v = 0; v < N_EXCEPTION_STACKS; v++) {
  595. static const unsigned int order[N_EXCEPTION_STACKS] = {
  596. [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
  597. [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
  598. };
  599. if (cpu) {
  600. estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
  601. if (!estacks)
  602. panic("Cannot allocate exception stack %ld %d\n",
  603. v, cpu);
  604. }
  605. estacks += PAGE_SIZE << order[v];
  606. orig_ist->ist[v] = t->x86_tss.ist[v] = (unsigned long)estacks;
  607. }
  608. t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
  609. /*
  610. * <= is required because the CPU will access up to
  611. * 8 bits beyond the end of the IO permission bitmap.
  612. */
  613. for (i = 0; i <= IO_BITMAP_LONGS; i++)
  614. t->io_bitmap[i] = ~0UL;
  615. atomic_inc(&init_mm.mm_count);
  616. me->active_mm = &init_mm;
  617. if (me->mm)
  618. BUG();
  619. enter_lazy_tlb(&init_mm, me);
  620. load_sp0(t, &current->thread);
  621. set_tss_desc(cpu, t);
  622. load_TR_desc();
  623. load_LDT(&init_mm.context);
  624. #ifdef CONFIG_KGDB
  625. /*
  626. * If the kgdb is connected no debug regs should be altered. This
  627. * is only applicable when KGDB and a KGDB I/O module are built
  628. * into the kernel and you are using early debugging with
  629. * kgdbwait. KGDB will control the kernel HW breakpoint registers.
  630. */
  631. if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
  632. arch_kgdb_ops.correct_hw_break();
  633. else {
  634. #endif
  635. /*
  636. * Clear all 6 debug registers:
  637. */
  638. set_debugreg(0UL, 0);
  639. set_debugreg(0UL, 1);
  640. set_debugreg(0UL, 2);
  641. set_debugreg(0UL, 3);
  642. set_debugreg(0UL, 6);
  643. set_debugreg(0UL, 7);
  644. #ifdef CONFIG_KGDB
  645. /* If the kgdb is connected no debug regs should be altered. */
  646. }
  647. #endif
  648. fpu_init();
  649. raw_local_save_flags(kernel_eflags);
  650. if (is_uv_system())
  651. uv_cpu_init();
  652. }