setup64.c 7.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293
  1. /*
  2. * X86-64 specific CPU setup.
  3. * Copyright (C) 1995 Linus Torvalds
  4. * Copyright 2001, 2002, 2003 SuSE Labs / Andi Kleen.
  5. * See setup.c for older changelog.
  6. */
  7. #include <linux/init.h>
  8. #include <linux/kernel.h>
  9. #include <linux/sched.h>
  10. #include <linux/string.h>
  11. #include <linux/bootmem.h>
  12. #include <linux/bitops.h>
  13. #include <linux/module.h>
  14. #include <asm/bootsetup.h>
  15. #include <asm/pda.h>
  16. #include <asm/pgtable.h>
  17. #include <asm/processor.h>
  18. #include <asm/desc.h>
  19. #include <asm/atomic.h>
  20. #include <asm/mmu_context.h>
  21. #include <asm/smp.h>
  22. #include <asm/i387.h>
  23. #include <asm/percpu.h>
  24. #include <asm/proto.h>
  25. #include <asm/sections.h>
  26. char x86_boot_params[BOOT_PARAM_SIZE] __initdata = {0,};
  27. cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
  28. struct x8664_pda *_cpu_pda[NR_CPUS] __read_mostly;
  29. EXPORT_SYMBOL(_cpu_pda);
  30. struct x8664_pda boot_cpu_pda[NR_CPUS] __cacheline_aligned;
  31. struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
  32. char boot_cpu_stack[IRQSTACKSIZE] __attribute__((section(".bss.page_aligned")));
  33. unsigned long __supported_pte_mask __read_mostly = ~0UL;
  34. EXPORT_SYMBOL(__supported_pte_mask);
  35. static int do_not_nx __cpuinitdata = 0;
  36. /* noexec=on|off
  37. Control non executable mappings for 64bit processes.
  38. on Enable(default)
  39. off Disable
  40. */
  41. int __init nonx_setup(char *str)
  42. {
  43. if (!strncmp(str, "on", 2)) {
  44. __supported_pte_mask |= _PAGE_NX;
  45. do_not_nx = 0;
  46. } else if (!strncmp(str, "off", 3)) {
  47. do_not_nx = 1;
  48. __supported_pte_mask &= ~_PAGE_NX;
  49. }
  50. return 1;
  51. }
  52. __setup("noexec=", nonx_setup); /* parsed early actually */
  53. int force_personality32 = 0;
  54. /* noexec32=on|off
  55. Control non executable heap for 32bit processes.
  56. To control the stack too use noexec=off
  57. on PROT_READ does not imply PROT_EXEC for 32bit processes
  58. off PROT_READ implies PROT_EXEC (default)
  59. */
  60. static int __init nonx32_setup(char *str)
  61. {
  62. if (!strcmp(str, "on"))
  63. force_personality32 &= ~READ_IMPLIES_EXEC;
  64. else if (!strcmp(str, "off"))
  65. force_personality32 |= READ_IMPLIES_EXEC;
  66. return 1;
  67. }
  68. __setup("noexec32=", nonx32_setup);
  69. /*
  70. * Great future plan:
  71. * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
  72. * Always point %gs to its beginning
  73. */
  74. void __init setup_per_cpu_areas(void)
  75. {
  76. int i;
  77. unsigned long size;
  78. #ifdef CONFIG_HOTPLUG_CPU
  79. prefill_possible_map();
  80. #endif
  81. /* Copy section for each CPU (we discard the original) */
  82. size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
  83. #ifdef CONFIG_MODULES
  84. if (size < PERCPU_ENOUGH_ROOM)
  85. size = PERCPU_ENOUGH_ROOM;
  86. #endif
  87. for_each_cpu_mask (i, cpu_possible_map) {
  88. char *ptr;
  89. if (!NODE_DATA(cpu_to_node(i))) {
  90. printk("cpu with no node %d, num_online_nodes %d\n",
  91. i, num_online_nodes());
  92. ptr = alloc_bootmem(size);
  93. } else {
  94. ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
  95. }
  96. if (!ptr)
  97. panic("Cannot allocate cpu data for CPU %d\n", i);
  98. cpu_pda(i)->data_offset = ptr - __per_cpu_start;
  99. memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
  100. }
  101. }
  102. void pda_init(int cpu)
  103. {
  104. struct x8664_pda *pda = cpu_pda(cpu);
  105. /* Setup up data that may be needed in __get_free_pages early */
  106. asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
  107. wrmsrl(MSR_GS_BASE, pda);
  108. pda->cpunumber = cpu;
  109. pda->irqcount = -1;
  110. pda->kernelstack =
  111. (unsigned long)stack_thread_info() - PDA_STACKOFFSET + THREAD_SIZE;
  112. pda->active_mm = &init_mm;
  113. pda->mmu_state = 0;
  114. if (cpu == 0) {
  115. /* others are initialized in smpboot.c */
  116. pda->pcurrent = &init_task;
  117. pda->irqstackptr = boot_cpu_stack;
  118. } else {
  119. pda->irqstackptr = (char *)
  120. __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
  121. if (!pda->irqstackptr)
  122. panic("cannot allocate irqstack for cpu %d", cpu);
  123. }
  124. pda->irqstackptr += IRQSTACKSIZE-64;
  125. }
  126. char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]
  127. __attribute__((section(".bss.page_aligned")));
  128. /* May not be marked __init: used by software suspend */
  129. void syscall_init(void)
  130. {
  131. /*
  132. * LSTAR and STAR live in a bit strange symbiosis.
  133. * They both write to the same internal register. STAR allows to set CS/DS
  134. * but only a 32bit target. LSTAR sets the 64bit rip.
  135. */
  136. wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
  137. wrmsrl(MSR_LSTAR, system_call);
  138. #ifdef CONFIG_IA32_EMULATION
  139. syscall32_cpu_init ();
  140. #endif
  141. /* Flags to clear on syscall */
  142. wrmsrl(MSR_SYSCALL_MASK, EF_TF|EF_DF|EF_IE|0x3000);
  143. }
  144. void __cpuinit check_efer(void)
  145. {
  146. unsigned long efer;
  147. rdmsrl(MSR_EFER, efer);
  148. if (!(efer & EFER_NX) || do_not_nx) {
  149. __supported_pte_mask &= ~_PAGE_NX;
  150. }
  151. }
  152. /*
  153. * cpu_init() initializes state that is per-CPU. Some data is already
  154. * initialized (naturally) in the bootstrap process, such as the GDT
  155. * and IDT. We reload them nevertheless, this function acts as a
  156. * 'CPU state barrier', nothing should get across.
  157. * A lot of state is already set up in PDA init.
  158. */
  159. void __cpuinit cpu_init (void)
  160. {
  161. int cpu = stack_smp_processor_id();
  162. struct tss_struct *t = &per_cpu(init_tss, cpu);
  163. struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
  164. unsigned long v;
  165. char *estacks = NULL;
  166. struct task_struct *me;
  167. int i;
  168. /* CPU 0 is initialised in head64.c */
  169. if (cpu != 0) {
  170. pda_init(cpu);
  171. zap_low_mappings(cpu);
  172. } else
  173. estacks = boot_exception_stacks;
  174. me = current;
  175. if (cpu_test_and_set(cpu, cpu_initialized))
  176. panic("CPU#%d already initialized!\n", cpu);
  177. printk("Initializing CPU#%d\n", cpu);
  178. clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
  179. /*
  180. * Initialize the per-CPU GDT with the boot GDT,
  181. * and set up the GDT descriptor:
  182. */
  183. if (cpu)
  184. memcpy(cpu_gdt(cpu), cpu_gdt_table, GDT_SIZE);
  185. cpu_gdt_descr[cpu].size = GDT_SIZE;
  186. asm volatile("lgdt %0" :: "m" (cpu_gdt_descr[cpu]));
  187. asm volatile("lidt %0" :: "m" (idt_descr));
  188. memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
  189. syscall_init();
  190. wrmsrl(MSR_FS_BASE, 0);
  191. wrmsrl(MSR_KERNEL_GS_BASE, 0);
  192. barrier();
  193. check_efer();
  194. /*
  195. * set up and load the per-CPU TSS
  196. */
  197. for (v = 0; v < N_EXCEPTION_STACKS; v++) {
  198. if (cpu) {
  199. static const unsigned int order[N_EXCEPTION_STACKS] = {
  200. [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
  201. [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
  202. };
  203. estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
  204. if (!estacks)
  205. panic("Cannot allocate exception stack %ld %d\n",
  206. v, cpu);
  207. }
  208. switch (v + 1) {
  209. #if DEBUG_STKSZ > EXCEPTION_STKSZ
  210. case DEBUG_STACK:
  211. cpu_pda(cpu)->debugstack = (unsigned long)estacks;
  212. estacks += DEBUG_STKSZ;
  213. break;
  214. #endif
  215. default:
  216. estacks += EXCEPTION_STKSZ;
  217. break;
  218. }
  219. orig_ist->ist[v] = t->ist[v] = (unsigned long)estacks;
  220. }
  221. t->io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
  222. /*
  223. * <= is required because the CPU will access up to
  224. * 8 bits beyond the end of the IO permission bitmap.
  225. */
  226. for (i = 0; i <= IO_BITMAP_LONGS; i++)
  227. t->io_bitmap[i] = ~0UL;
  228. atomic_inc(&init_mm.mm_count);
  229. me->active_mm = &init_mm;
  230. if (me->mm)
  231. BUG();
  232. enter_lazy_tlb(&init_mm, me);
  233. set_tss_desc(cpu, t);
  234. load_TR_desc();
  235. load_LDT(&init_mm.context);
  236. /*
  237. * Clear all 6 debug registers:
  238. */
  239. set_debugreg(0UL, 0);
  240. set_debugreg(0UL, 1);
  241. set_debugreg(0UL, 2);
  242. set_debugreg(0UL, 3);
  243. set_debugreg(0UL, 6);
  244. set_debugreg(0UL, 7);
  245. fpu_init();
  246. }