setup64.c 7.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289
  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;
  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. static int do_not_nx __cpuinitdata = 0;
  35. /* noexec=on|off
  36. Control non executable mappings for 64bit processes.
  37. on Enable(default)
  38. off Disable
  39. */
  40. static int __init nonx_setup(char *str)
  41. {
  42. if (!str)
  43. return -EINVAL;
  44. if (!strncmp(str, "on", 2)) {
  45. __supported_pte_mask |= _PAGE_NX;
  46. do_not_nx = 0;
  47. } else if (!strncmp(str, "off", 3)) {
  48. do_not_nx = 1;
  49. __supported_pte_mask &= ~_PAGE_NX;
  50. }
  51. return 0;
  52. }
  53. early_param("noexec", nonx_setup);
  54. int force_personality32 = 0;
  55. /* noexec32=on|off
  56. Control non executable heap for 32bit processes.
  57. To control the stack too use noexec=off
  58. on PROT_READ does not imply PROT_EXEC for 32bit processes
  59. off PROT_READ implies PROT_EXEC (default)
  60. */
  61. static int __init nonx32_setup(char *str)
  62. {
  63. if (!strcmp(str, "on"))
  64. force_personality32 &= ~READ_IMPLIES_EXEC;
  65. else if (!strcmp(str, "off"))
  66. force_personality32 |= READ_IMPLIES_EXEC;
  67. return 1;
  68. }
  69. __setup("noexec32=", nonx32_setup);
  70. /*
  71. * Great future plan:
  72. * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
  73. * Always point %gs to its beginning
  74. */
  75. void __init setup_per_cpu_areas(void)
  76. {
  77. int i;
  78. unsigned long size;
  79. #ifdef CONFIG_HOTPLUG_CPU
  80. prefill_possible_map();
  81. #endif
  82. /* Copy section for each CPU (we discard the original) */
  83. size = PERCPU_ENOUGH_ROOM;
  84. printk(KERN_INFO "PERCPU: Allocating %lu bytes of per cpu data\n", size);
  85. for_each_cpu_mask (i, cpu_possible_map) {
  86. char *ptr;
  87. if (!NODE_DATA(cpu_to_node(i))) {
  88. printk("cpu with no node %d, num_online_nodes %d\n",
  89. i, num_online_nodes());
  90. ptr = alloc_bootmem_pages(size);
  91. } else {
  92. ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
  93. }
  94. if (!ptr)
  95. panic("Cannot allocate cpu data for CPU %d\n", i);
  96. cpu_pda(i)->data_offset = ptr - __per_cpu_start;
  97. memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
  98. }
  99. }
  100. void pda_init(int cpu)
  101. {
  102. struct x8664_pda *pda = cpu_pda(cpu);
  103. /* Setup up data that may be needed in __get_free_pages early */
  104. asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
  105. /* Memory clobbers used to order PDA accessed */
  106. mb();
  107. wrmsrl(MSR_GS_BASE, pda);
  108. mb();
  109. pda->cpunumber = cpu;
  110. pda->irqcount = -1;
  111. pda->kernelstack =
  112. (unsigned long)stack_thread_info() - PDA_STACKOFFSET + THREAD_SIZE;
  113. pda->active_mm = &init_mm;
  114. pda->mmu_state = 0;
  115. if (cpu == 0) {
  116. /* others are initialized in smpboot.c */
  117. pda->pcurrent = &init_task;
  118. pda->irqstackptr = boot_cpu_stack;
  119. } else {
  120. pda->irqstackptr = (char *)
  121. __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
  122. if (!pda->irqstackptr)
  123. panic("cannot allocate irqstack for cpu %d", cpu);
  124. }
  125. pda->irqstackptr += IRQSTACKSIZE-64;
  126. }
  127. char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]
  128. __attribute__((section(".bss.page_aligned")));
  129. extern asmlinkage void ignore_sysret(void);
  130. /* May not be marked __init: used by software suspend */
  131. void syscall_init(void)
  132. {
  133. /*
  134. * LSTAR and STAR live in a bit strange symbiosis.
  135. * They both write to the same internal register. STAR allows to set CS/DS
  136. * but only a 32bit target. LSTAR sets the 64bit rip.
  137. */
  138. wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
  139. wrmsrl(MSR_LSTAR, system_call);
  140. wrmsrl(MSR_CSTAR, ignore_sysret);
  141. #ifdef CONFIG_IA32_EMULATION
  142. syscall32_cpu_init ();
  143. #endif
  144. /* Flags to clear on syscall */
  145. wrmsrl(MSR_SYSCALL_MASK, EF_TF|EF_DF|EF_IE|0x3000);
  146. }
  147. void __cpuinit check_efer(void)
  148. {
  149. unsigned long efer;
  150. rdmsrl(MSR_EFER, efer);
  151. if (!(efer & EFER_NX) || do_not_nx) {
  152. __supported_pte_mask &= ~_PAGE_NX;
  153. }
  154. }
  155. unsigned long kernel_eflags;
  156. /*
  157. * cpu_init() initializes state that is per-CPU. Some data is already
  158. * initialized (naturally) in the bootstrap process, such as the GDT
  159. * and IDT. We reload them nevertheless, this function acts as a
  160. * 'CPU state barrier', nothing should get across.
  161. * A lot of state is already set up in PDA init.
  162. */
  163. void __cpuinit cpu_init (void)
  164. {
  165. int cpu = stack_smp_processor_id();
  166. struct tss_struct *t = &per_cpu(init_tss, cpu);
  167. struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
  168. unsigned long v;
  169. char *estacks = NULL;
  170. struct task_struct *me;
  171. int i;
  172. /* CPU 0 is initialised in head64.c */
  173. if (cpu != 0) {
  174. pda_init(cpu);
  175. } else
  176. estacks = boot_exception_stacks;
  177. me = current;
  178. if (cpu_test_and_set(cpu, cpu_initialized))
  179. panic("CPU#%d already initialized!\n", cpu);
  180. printk("Initializing CPU#%d\n", cpu);
  181. clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
  182. /*
  183. * Initialize the per-CPU GDT with the boot GDT,
  184. * and set up the GDT descriptor:
  185. */
  186. if (cpu)
  187. memcpy(cpu_gdt(cpu), cpu_gdt_table, GDT_SIZE);
  188. cpu_gdt_descr[cpu].size = GDT_SIZE;
  189. asm volatile("lgdt %0" :: "m" (cpu_gdt_descr[cpu]));
  190. asm volatile("lidt %0" :: "m" (idt_descr));
  191. memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
  192. syscall_init();
  193. wrmsrl(MSR_FS_BASE, 0);
  194. wrmsrl(MSR_KERNEL_GS_BASE, 0);
  195. barrier();
  196. check_efer();
  197. /*
  198. * set up and load the per-CPU TSS
  199. */
  200. for (v = 0; v < N_EXCEPTION_STACKS; v++) {
  201. static const unsigned int order[N_EXCEPTION_STACKS] = {
  202. [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
  203. [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
  204. };
  205. if (cpu) {
  206. estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
  207. if (!estacks)
  208. panic("Cannot allocate exception stack %ld %d\n",
  209. v, cpu);
  210. }
  211. estacks += PAGE_SIZE << order[v];
  212. orig_ist->ist[v] = t->ist[v] = (unsigned long)estacks;
  213. }
  214. t->io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
  215. /*
  216. * <= is required because the CPU will access up to
  217. * 8 bits beyond the end of the IO permission bitmap.
  218. */
  219. for (i = 0; i <= IO_BITMAP_LONGS; i++)
  220. t->io_bitmap[i] = ~0UL;
  221. atomic_inc(&init_mm.mm_count);
  222. me->active_mm = &init_mm;
  223. if (me->mm)
  224. BUG();
  225. enter_lazy_tlb(&init_mm, me);
  226. set_tss_desc(cpu, t);
  227. load_TR_desc();
  228. load_LDT(&init_mm.context);
  229. /*
  230. * Clear all 6 debug registers:
  231. */
  232. set_debugreg(0UL, 0);
  233. set_debugreg(0UL, 1);
  234. set_debugreg(0UL, 2);
  235. set_debugreg(0UL, 3);
  236. set_debugreg(0UL, 6);
  237. set_debugreg(0UL, 7);
  238. fpu_init();
  239. raw_local_save_flags(kernel_eflags);
  240. }