common_64.c 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763
  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), "
  92. "D cache %dK (%d bytes/line)\n",
  93. edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
  94. c->x86_cache_size = (ecx>>24) + (edx>>24);
  95. /* On K8 L1 TLB is inclusive, so don't count it */
  96. c->x86_tlbsize = 0;
  97. }
  98. if (n >= 0x80000006) {
  99. cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
  100. ecx = cpuid_ecx(0x80000006);
  101. c->x86_cache_size = ecx >> 16;
  102. c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
  103. printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
  104. c->x86_cache_size, ecx & 0xFF);
  105. }
  106. }
  107. void __cpuinit detect_ht(struct cpuinfo_x86 *c)
  108. {
  109. #ifdef CONFIG_SMP
  110. u32 eax, ebx, ecx, edx;
  111. int index_msb, core_bits;
  112. cpuid(1, &eax, &ebx, &ecx, &edx);
  113. if (!cpu_has(c, X86_FEATURE_HT))
  114. return;
  115. if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
  116. goto out;
  117. smp_num_siblings = (ebx & 0xff0000) >> 16;
  118. if (smp_num_siblings == 1) {
  119. printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
  120. } else if (smp_num_siblings > 1) {
  121. if (smp_num_siblings > NR_CPUS) {
  122. printk(KERN_WARNING "CPU: Unsupported number of "
  123. "siblings %d", smp_num_siblings);
  124. smp_num_siblings = 1;
  125. return;
  126. }
  127. index_msb = get_count_order(smp_num_siblings);
  128. c->phys_proc_id = phys_pkg_id(index_msb);
  129. smp_num_siblings = smp_num_siblings / c->x86_max_cores;
  130. index_msb = get_count_order(smp_num_siblings);
  131. core_bits = get_count_order(c->x86_max_cores);
  132. c->cpu_core_id = phys_pkg_id(index_msb) &
  133. ((1 << core_bits) - 1);
  134. }
  135. out:
  136. if ((c->x86_max_cores * smp_num_siblings) > 1) {
  137. printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
  138. c->phys_proc_id);
  139. printk(KERN_INFO "CPU: Processor Core ID: %d\n",
  140. c->cpu_core_id);
  141. }
  142. #endif
  143. }
  144. static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
  145. {
  146. char *v = c->x86_vendor_id;
  147. int i;
  148. static int printed;
  149. for (i = 0; i < X86_VENDOR_NUM; i++) {
  150. if (cpu_devs[i]) {
  151. if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
  152. (cpu_devs[i]->c_ident[1] &&
  153. !strcmp(v, cpu_devs[i]->c_ident[1]))) {
  154. c->x86_vendor = i;
  155. this_cpu = cpu_devs[i];
  156. return;
  157. }
  158. }
  159. }
  160. if (!printed) {
  161. printed++;
  162. printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
  163. printk(KERN_ERR "CPU: Your system may be unstable.\n");
  164. }
  165. c->x86_vendor = X86_VENDOR_UNKNOWN;
  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. /*
  185. * The NOPL instruction is supposed to exist on all CPUs with
  186. * family >= 6, unfortunately, that's not true in practice because
  187. * of early VIA chips and (more importantly) broken virtualizers that
  188. * are not easy to detect. Hence, probe for it based on first
  189. * principles.
  190. *
  191. * Note: no 64-bit chip is known to lack these, but put the code here
  192. * for consistency with 32 bits, and to make it utterly trivial to
  193. * diagnose the problem should it ever surface.
  194. */
  195. static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
  196. {
  197. const u32 nopl_signature = 0x888c53b1; /* Random number */
  198. u32 has_nopl = nopl_signature;
  199. clear_cpu_cap(c, X86_FEATURE_NOPL);
  200. if (c->x86 >= 6) {
  201. asm volatile("\n"
  202. "1: .byte 0x0f,0x1f,0xc0\n" /* nopl %eax */
  203. "2:\n"
  204. " .section .fixup,\"ax\"\n"
  205. "3: xor %0,%0\n"
  206. " jmp 2b\n"
  207. " .previous\n"
  208. _ASM_EXTABLE(1b,3b)
  209. : "+a" (has_nopl));
  210. if (has_nopl == nopl_signature)
  211. set_cpu_cap(c, X86_FEATURE_NOPL);
  212. }
  213. }
  214. static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
  215. void __init early_cpu_init(void)
  216. {
  217. struct cpu_vendor_dev *cvdev;
  218. for (cvdev = __x86cpuvendor_start ;
  219. cvdev < __x86cpuvendor_end ;
  220. cvdev++)
  221. cpu_devs[cvdev->vendor] = cvdev->cpu_dev;
  222. early_cpu_support_print();
  223. early_identify_cpu(&boot_cpu_data);
  224. }
  225. /* Do some early cpuid on the boot CPU to get some parameter that are
  226. needed before check_bugs. Everything advanced is in identify_cpu
  227. below. */
  228. static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
  229. {
  230. u32 tfms, xlvl;
  231. c->loops_per_jiffy = loops_per_jiffy;
  232. c->x86_cache_size = -1;
  233. c->x86_vendor = X86_VENDOR_UNKNOWN;
  234. c->x86_model = c->x86_mask = 0; /* So far unknown... */
  235. c->x86_vendor_id[0] = '\0'; /* Unset */
  236. c->x86_model_id[0] = '\0'; /* Unset */
  237. c->x86_clflush_size = 64;
  238. c->x86_cache_alignment = c->x86_clflush_size;
  239. c->x86_max_cores = 1;
  240. c->x86_coreid_bits = 0;
  241. c->extended_cpuid_level = 0;
  242. memset(&c->x86_capability, 0, sizeof c->x86_capability);
  243. /* Get vendor name */
  244. cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
  245. (unsigned int *)&c->x86_vendor_id[0],
  246. (unsigned int *)&c->x86_vendor_id[8],
  247. (unsigned int *)&c->x86_vendor_id[4]);
  248. get_cpu_vendor(c);
  249. /* Initialize the standard set of capabilities */
  250. /* Note that the vendor-specific code below might override */
  251. /* Intel-defined flags: level 0x00000001 */
  252. if (c->cpuid_level >= 0x00000001) {
  253. __u32 misc;
  254. cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
  255. &c->x86_capability[0]);
  256. c->x86 = (tfms >> 8) & 0xf;
  257. c->x86_model = (tfms >> 4) & 0xf;
  258. c->x86_mask = tfms & 0xf;
  259. if (c->x86 == 0xf)
  260. c->x86 += (tfms >> 20) & 0xff;
  261. if (c->x86 >= 0x6)
  262. c->x86_model += ((tfms >> 16) & 0xF) << 4;
  263. if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
  264. c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
  265. } else {
  266. /* Have CPUID level 0 only - unheard of */
  267. c->x86 = 4;
  268. }
  269. c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
  270. #ifdef CONFIG_SMP
  271. c->phys_proc_id = c->initial_apicid;
  272. #endif
  273. /* AMD-defined flags: level 0x80000001 */
  274. xlvl = cpuid_eax(0x80000000);
  275. c->extended_cpuid_level = xlvl;
  276. if ((xlvl & 0xffff0000) == 0x80000000) {
  277. if (xlvl >= 0x80000001) {
  278. c->x86_capability[1] = cpuid_edx(0x80000001);
  279. c->x86_capability[6] = cpuid_ecx(0x80000001);
  280. }
  281. if (xlvl >= 0x80000004)
  282. get_model_name(c); /* Default name */
  283. }
  284. /* Transmeta-defined flags: level 0x80860001 */
  285. xlvl = cpuid_eax(0x80860000);
  286. if ((xlvl & 0xffff0000) == 0x80860000) {
  287. /* Don't set x86_cpuid_level here for now to not confuse. */
  288. if (xlvl >= 0x80860001)
  289. c->x86_capability[2] = cpuid_edx(0x80860001);
  290. }
  291. if (c->extended_cpuid_level >= 0x80000007)
  292. c->x86_power = cpuid_edx(0x80000007);
  293. if (c->extended_cpuid_level >= 0x80000008) {
  294. u32 eax = cpuid_eax(0x80000008);
  295. c->x86_virt_bits = (eax >> 8) & 0xff;
  296. c->x86_phys_bits = eax & 0xff;
  297. }
  298. detect_nopl(c);
  299. if (c->x86_vendor != X86_VENDOR_UNKNOWN &&
  300. cpu_devs[c->x86_vendor]->c_early_init)
  301. cpu_devs[c->x86_vendor]->c_early_init(c);
  302. validate_pat_support(c);
  303. }
  304. /*
  305. * This does the hard work of actually picking apart the CPU stuff...
  306. */
  307. static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
  308. {
  309. int i;
  310. early_identify_cpu(c);
  311. init_scattered_cpuid_features(c);
  312. c->apicid = phys_pkg_id(0);
  313. /*
  314. * Vendor-specific initialization. In this section we
  315. * canonicalize the feature flags, meaning if there are
  316. * features a certain CPU supports which CPUID doesn't
  317. * tell us, CPUID claiming incorrect flags, or other bugs,
  318. * we handle them here.
  319. *
  320. * At the end of this section, c->x86_capability better
  321. * indicate the features this CPU genuinely supports!
  322. */
  323. if (this_cpu->c_init)
  324. this_cpu->c_init(c);
  325. detect_ht(c);
  326. /*
  327. * On SMP, boot_cpu_data holds the common feature set between
  328. * all CPUs; so make sure that we indicate which features are
  329. * common between the CPUs. The first time this routine gets
  330. * executed, c == &boot_cpu_data.
  331. */
  332. if (c != &boot_cpu_data) {
  333. /* AND the already accumulated flags with these */
  334. for (i = 0; i < NCAPINTS; i++)
  335. boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
  336. }
  337. /* Clear all flags overriden by options */
  338. for (i = 0; i < NCAPINTS; i++)
  339. c->x86_capability[i] &= ~cleared_cpu_caps[i];
  340. #ifdef CONFIG_X86_MCE
  341. mcheck_init(c);
  342. #endif
  343. select_idle_routine(c);
  344. #ifdef CONFIG_NUMA
  345. numa_add_cpu(smp_processor_id());
  346. #endif
  347. }
  348. void __cpuinit identify_boot_cpu(void)
  349. {
  350. identify_cpu(&boot_cpu_data);
  351. }
  352. void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
  353. {
  354. BUG_ON(c == &boot_cpu_data);
  355. identify_cpu(c);
  356. mtrr_ap_init();
  357. }
  358. static __init int setup_noclflush(char *arg)
  359. {
  360. setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
  361. return 1;
  362. }
  363. __setup("noclflush", setup_noclflush);
  364. struct msr_range {
  365. unsigned min;
  366. unsigned max;
  367. };
  368. static struct msr_range msr_range_array[] __cpuinitdata = {
  369. { 0x00000000, 0x00000418},
  370. { 0xc0000000, 0xc000040b},
  371. { 0xc0010000, 0xc0010142},
  372. { 0xc0011000, 0xc001103b},
  373. };
  374. static void __cpuinit print_cpu_msr(void)
  375. {
  376. unsigned index;
  377. u64 val;
  378. int i;
  379. unsigned index_min, index_max;
  380. for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
  381. index_min = msr_range_array[i].min;
  382. index_max = msr_range_array[i].max;
  383. for (index = index_min; index < index_max; index++) {
  384. if (rdmsrl_amd_safe(index, &val))
  385. continue;
  386. printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
  387. }
  388. }
  389. }
  390. static int show_msr __cpuinitdata;
  391. static __init int setup_show_msr(char *arg)
  392. {
  393. int num;
  394. get_option(&arg, &num);
  395. if (num > 0)
  396. show_msr = num;
  397. return 1;
  398. }
  399. __setup("show_msr=", setup_show_msr);
  400. void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
  401. {
  402. if (c->x86_model_id[0])
  403. printk(KERN_CONT "%s", c->x86_model_id);
  404. if (c->x86_mask || c->cpuid_level >= 0)
  405. printk(KERN_CONT " stepping %02x\n", c->x86_mask);
  406. else
  407. printk(KERN_CONT "\n");
  408. #ifdef CONFIG_SMP
  409. if (c->cpu_index < show_msr)
  410. print_cpu_msr();
  411. #else
  412. if (show_msr)
  413. print_cpu_msr();
  414. #endif
  415. }
  416. static __init int setup_disablecpuid(char *arg)
  417. {
  418. int bit;
  419. if (get_option(&arg, &bit) && bit < NCAPINTS*32)
  420. setup_clear_cpu_cap(bit);
  421. else
  422. return 0;
  423. return 1;
  424. }
  425. __setup("clearcpuid=", setup_disablecpuid);
  426. cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
  427. struct x8664_pda **_cpu_pda __read_mostly;
  428. EXPORT_SYMBOL(_cpu_pda);
  429. struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
  430. char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
  431. unsigned long __supported_pte_mask __read_mostly = ~0UL;
  432. EXPORT_SYMBOL_GPL(__supported_pte_mask);
  433. static int do_not_nx __cpuinitdata;
  434. /* noexec=on|off
  435. Control non executable mappings for 64bit processes.
  436. on Enable(default)
  437. off Disable
  438. */
  439. static int __init nonx_setup(char *str)
  440. {
  441. if (!str)
  442. return -EINVAL;
  443. if (!strncmp(str, "on", 2)) {
  444. __supported_pte_mask |= _PAGE_NX;
  445. do_not_nx = 0;
  446. } else if (!strncmp(str, "off", 3)) {
  447. do_not_nx = 1;
  448. __supported_pte_mask &= ~_PAGE_NX;
  449. }
  450. return 0;
  451. }
  452. early_param("noexec", nonx_setup);
  453. int force_personality32;
  454. /* noexec32=on|off
  455. Control non executable heap for 32bit processes.
  456. To control the stack too use noexec=off
  457. on PROT_READ does not imply PROT_EXEC for 32bit processes (default)
  458. off PROT_READ implies PROT_EXEC
  459. */
  460. static int __init nonx32_setup(char *str)
  461. {
  462. if (!strcmp(str, "on"))
  463. force_personality32 &= ~READ_IMPLIES_EXEC;
  464. else if (!strcmp(str, "off"))
  465. force_personality32 |= READ_IMPLIES_EXEC;
  466. return 1;
  467. }
  468. __setup("noexec32=", nonx32_setup);
  469. void pda_init(int cpu)
  470. {
  471. struct x8664_pda *pda = cpu_pda(cpu);
  472. /* Setup up data that may be needed in __get_free_pages early */
  473. loadsegment(fs, 0);
  474. loadsegment(gs, 0);
  475. /* Memory clobbers used to order PDA accessed */
  476. mb();
  477. wrmsrl(MSR_GS_BASE, pda);
  478. mb();
  479. pda->cpunumber = cpu;
  480. pda->irqcount = -1;
  481. pda->kernelstack = (unsigned long)stack_thread_info() -
  482. PDA_STACKOFFSET + THREAD_SIZE;
  483. pda->active_mm = &init_mm;
  484. pda->mmu_state = 0;
  485. if (cpu == 0) {
  486. /* others are initialized in smpboot.c */
  487. pda->pcurrent = &init_task;
  488. pda->irqstackptr = boot_cpu_stack;
  489. pda->irqstackptr += IRQSTACKSIZE - 64;
  490. } else {
  491. if (!pda->irqstackptr) {
  492. pda->irqstackptr = (char *)
  493. __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
  494. if (!pda->irqstackptr)
  495. panic("cannot allocate irqstack for cpu %d",
  496. cpu);
  497. pda->irqstackptr += IRQSTACKSIZE - 64;
  498. }
  499. if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
  500. pda->nodenumber = cpu_to_node(cpu);
  501. }
  502. }
  503. char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
  504. DEBUG_STKSZ] __page_aligned_bss;
  505. extern asmlinkage void ignore_sysret(void);
  506. /* May not be marked __init: used by software suspend */
  507. void syscall_init(void)
  508. {
  509. /*
  510. * LSTAR and STAR live in a bit strange symbiosis.
  511. * They both write to the same internal register. STAR allows to
  512. * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
  513. */
  514. wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
  515. wrmsrl(MSR_LSTAR, system_call);
  516. wrmsrl(MSR_CSTAR, ignore_sysret);
  517. #ifdef CONFIG_IA32_EMULATION
  518. syscall32_cpu_init();
  519. #endif
  520. /* Flags to clear on syscall */
  521. wrmsrl(MSR_SYSCALL_MASK,
  522. X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
  523. }
  524. void __cpuinit check_efer(void)
  525. {
  526. unsigned long efer;
  527. rdmsrl(MSR_EFER, efer);
  528. if (!(efer & EFER_NX) || do_not_nx)
  529. __supported_pte_mask &= ~_PAGE_NX;
  530. }
  531. unsigned long kernel_eflags;
  532. /*
  533. * Copies of the original ist values from the tss are only accessed during
  534. * debugging, no special alignment required.
  535. */
  536. DEFINE_PER_CPU(struct orig_ist, orig_ist);
  537. /*
  538. * cpu_init() initializes state that is per-CPU. Some data is already
  539. * initialized (naturally) in the bootstrap process, such as the GDT
  540. * and IDT. We reload them nevertheless, this function acts as a
  541. * 'CPU state barrier', nothing should get across.
  542. * A lot of state is already set up in PDA init.
  543. */
  544. void __cpuinit cpu_init(void)
  545. {
  546. int cpu = stack_smp_processor_id();
  547. struct tss_struct *t = &per_cpu(init_tss, cpu);
  548. struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
  549. unsigned long v;
  550. char *estacks = NULL;
  551. struct task_struct *me;
  552. int i;
  553. /* CPU 0 is initialised in head64.c */
  554. if (cpu != 0)
  555. pda_init(cpu);
  556. else
  557. estacks = boot_exception_stacks;
  558. me = current;
  559. if (cpu_test_and_set(cpu, cpu_initialized))
  560. panic("CPU#%d already initialized!\n", cpu);
  561. printk(KERN_INFO "Initializing CPU#%d\n", cpu);
  562. clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
  563. /*
  564. * Initialize the per-CPU GDT with the boot GDT,
  565. * and set up the GDT descriptor:
  566. */
  567. switch_to_new_gdt();
  568. load_idt((const struct desc_ptr *)&idt_descr);
  569. memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
  570. syscall_init();
  571. wrmsrl(MSR_FS_BASE, 0);
  572. wrmsrl(MSR_KERNEL_GS_BASE, 0);
  573. barrier();
  574. check_efer();
  575. /*
  576. * set up and load the per-CPU TSS
  577. */
  578. if (!orig_ist->ist[0]) {
  579. static const unsigned int order[N_EXCEPTION_STACKS] = {
  580. [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
  581. [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
  582. };
  583. for (v = 0; v < N_EXCEPTION_STACKS; v++) {
  584. if (cpu) {
  585. estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
  586. if (!estacks)
  587. panic("Cannot allocate exception "
  588. "stack %ld %d\n", v, cpu);
  589. }
  590. estacks += PAGE_SIZE << order[v];
  591. orig_ist->ist[v] = t->x86_tss.ist[v] =
  592. (unsigned long)estacks;
  593. }
  594. }
  595. t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
  596. /*
  597. * <= is required because the CPU will access up to
  598. * 8 bits beyond the end of the IO permission bitmap.
  599. */
  600. for (i = 0; i <= IO_BITMAP_LONGS; i++)
  601. t->io_bitmap[i] = ~0UL;
  602. atomic_inc(&init_mm.mm_count);
  603. me->active_mm = &init_mm;
  604. if (me->mm)
  605. BUG();
  606. enter_lazy_tlb(&init_mm, me);
  607. load_sp0(t, &current->thread);
  608. set_tss_desc(cpu, t);
  609. load_TR_desc();
  610. load_LDT(&init_mm.context);
  611. #ifdef CONFIG_KGDB
  612. /*
  613. * If the kgdb is connected no debug regs should be altered. This
  614. * is only applicable when KGDB and a KGDB I/O module are built
  615. * into the kernel and you are using early debugging with
  616. * kgdbwait. KGDB will control the kernel HW breakpoint registers.
  617. */
  618. if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
  619. arch_kgdb_ops.correct_hw_break();
  620. else {
  621. #endif
  622. /*
  623. * Clear all 6 debug registers:
  624. */
  625. set_debugreg(0UL, 0);
  626. set_debugreg(0UL, 1);
  627. set_debugreg(0UL, 2);
  628. set_debugreg(0UL, 3);
  629. set_debugreg(0UL, 6);
  630. set_debugreg(0UL, 7);
  631. #ifdef CONFIG_KGDB
  632. /* If the kgdb is connected no debug regs should be altered. */
  633. }
  634. #endif
  635. fpu_init();
  636. raw_local_save_flags(kernel_eflags);
  637. if (is_uv_system())
  638. uv_cpu_init();
  639. }