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. * $Id: setup64.c,v 1.12 2002/03/21 10:09:17 ak Exp $
  7. */
  8. #include <linux/config.h>
  9. #include <linux/init.h>
  10. #include <linux/kernel.h>
  11. #include <linux/sched.h>
  12. #include <linux/string.h>
  13. #include <linux/bootmem.h>
  14. #include <linux/bitops.h>
  15. #include <asm/bootsetup.h>
  16. #include <asm/pda.h>
  17. #include <asm/pgtable.h>
  18. #include <asm/processor.h>
  19. #include <asm/desc.h>
  20. #include <asm/atomic.h>
  21. #include <asm/mmu_context.h>
  22. #include <asm/smp.h>
  23. #include <asm/i387.h>
  24. #include <asm/percpu.h>
  25. #include <asm/mtrr.h>
  26. #include <asm/proto.h>
  27. #include <asm/mman.h>
  28. #include <asm/numa.h>
  29. char x86_boot_params[BOOT_PARAM_SIZE] __initdata = {0,};
  30. cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
  31. struct x8664_pda cpu_pda[NR_CPUS] __cacheline_aligned;
  32. extern struct task_struct init_task;
  33. extern unsigned char __per_cpu_start[], __per_cpu_end[];
  34. extern struct desc_ptr cpu_gdt_descr[];
  35. struct desc_ptr idt_descr = { 256 * 16, (unsigned long) idt_table };
  36. char boot_cpu_stack[IRQSTACKSIZE] __attribute__((section(".bss.page_aligned")));
  37. unsigned long __supported_pte_mask = ~0UL;
  38. static int do_not_nx __initdata = 0;
  39. /* noexec=on|off
  40. Control non executable mappings for 64bit processes.
  41. on Enable(default)
  42. off Disable
  43. */
  44. int __init nonx_setup(char *str)
  45. {
  46. if (!strncmp(str, "on", 2)) {
  47. __supported_pte_mask |= _PAGE_NX;
  48. do_not_nx = 0;
  49. } else if (!strncmp(str, "off", 3)) {
  50. do_not_nx = 1;
  51. __supported_pte_mask &= ~_PAGE_NX;
  52. }
  53. return 0;
  54. }
  55. __setup("noexec=", nonx_setup); /* parsed early actually */
  56. int force_personality32 = READ_IMPLIES_EXEC;
  57. /* noexec32=on|off
  58. Control non executable heap for 32bit processes.
  59. To control the stack too use noexec=off
  60. on PROT_READ does not imply PROT_EXEC for 32bit processes
  61. off PROT_READ implies PROT_EXEC (default)
  62. */
  63. static int __init nonx32_setup(char *str)
  64. {
  65. if (!strcmp(str, "on"))
  66. force_personality32 &= ~READ_IMPLIES_EXEC;
  67. else if (!strcmp(str, "off"))
  68. force_personality32 |= READ_IMPLIES_EXEC;
  69. return 0;
  70. }
  71. __setup("noexec32=", nonx32_setup);
  72. /*
  73. * Great future plan:
  74. * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
  75. * Always point %gs to its beginning
  76. */
  77. void __init setup_per_cpu_areas(void)
  78. {
  79. int i;
  80. unsigned long size;
  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 (i = 0; i < NR_CPUS; i++) {
  88. unsigned 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, cpu_pda + cpu);
  108. pda->me = pda;
  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. asm volatile("movq %0,%%cr3" :: "r" (__pa_symbol(&init_level4_pgt)));
  126. pda->irqstackptr += IRQSTACKSIZE-64;
  127. }
  128. char boot_exception_stacks[N_EXCEPTION_STACKS * EXCEPTION_STKSZ]
  129. __attribute__((section(".bss.page_aligned")));
  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. #ifdef CONFIG_IA32_EMULATION
  141. syscall32_cpu_init ();
  142. #endif
  143. /* Flags to clear on syscall */
  144. wrmsrl(MSR_SYSCALL_MASK, EF_TF|EF_DF|EF_IE|0x3000);
  145. }
  146. void __cpuinit check_efer(void)
  147. {
  148. unsigned long efer;
  149. rdmsrl(MSR_EFER, efer);
  150. if (!(efer & EFER_NX) || do_not_nx) {
  151. __supported_pte_mask &= ~_PAGE_NX;
  152. }
  153. }
  154. /*
  155. * cpu_init() initializes state that is per-CPU. Some data is already
  156. * initialized (naturally) in the bootstrap process, such as the GDT
  157. * and IDT. We reload them nevertheless, this function acts as a
  158. * 'CPU state barrier', nothing should get across.
  159. * A lot of state is already set up in PDA init.
  160. */
  161. void __cpuinit cpu_init (void)
  162. {
  163. #ifdef CONFIG_SMP
  164. int cpu = stack_smp_processor_id();
  165. #else
  166. int cpu = smp_processor_id();
  167. #endif
  168. struct tss_struct *t = &per_cpu(init_tss, cpu);
  169. unsigned long v;
  170. char *estacks = NULL;
  171. struct task_struct *me;
  172. int i;
  173. /* CPU 0 is initialised in head64.c */
  174. if (cpu != 0) {
  175. pda_init(cpu);
  176. } else
  177. estacks = boot_exception_stacks;
  178. me = current;
  179. if (cpu_test_and_set(cpu, cpu_initialized))
  180. panic("CPU#%d already initialized!\n", cpu);
  181. printk("Initializing CPU#%d\n", cpu);
  182. clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
  183. /*
  184. * Initialize the per-CPU GDT with the boot GDT,
  185. * and set up the GDT descriptor:
  186. */
  187. if (cpu) {
  188. memcpy(cpu_gdt_table[cpu], cpu_gdt_table[0], GDT_SIZE);
  189. }
  190. cpu_gdt_descr[cpu].size = GDT_SIZE;
  191. cpu_gdt_descr[cpu].address = (unsigned long)cpu_gdt_table[cpu];
  192. asm volatile("lgdt %0" :: "m" (cpu_gdt_descr[cpu]));
  193. asm volatile("lidt %0" :: "m" (idt_descr));
  194. memcpy(me->thread.tls_array, cpu_gdt_table[cpu], GDT_ENTRY_TLS_ENTRIES * 8);
  195. /*
  196. * Delete NT
  197. */
  198. asm volatile("pushfq ; popq %%rax ; btr $14,%%rax ; pushq %%rax ; popfq" ::: "eax");
  199. syscall_init();
  200. wrmsrl(MSR_FS_BASE, 0);
  201. wrmsrl(MSR_KERNEL_GS_BASE, 0);
  202. barrier();
  203. check_efer();
  204. /*
  205. * set up and load the per-CPU TSS
  206. */
  207. for (v = 0; v < N_EXCEPTION_STACKS; v++) {
  208. if (cpu) {
  209. estacks = (char *)__get_free_pages(GFP_ATOMIC,
  210. EXCEPTION_STACK_ORDER);
  211. if (!estacks)
  212. panic("Cannot allocate exception stack %ld %d\n",
  213. v, cpu);
  214. }
  215. estacks += EXCEPTION_STKSZ;
  216. t->ist[v] = (unsigned long)estacks;
  217. }
  218. t->io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
  219. /*
  220. * <= is required because the CPU will access up to
  221. * 8 bits beyond the end of the IO permission bitmap.
  222. */
  223. for (i = 0; i <= IO_BITMAP_LONGS; i++)
  224. t->io_bitmap[i] = ~0UL;
  225. atomic_inc(&init_mm.mm_count);
  226. me->active_mm = &init_mm;
  227. if (me->mm)
  228. BUG();
  229. enter_lazy_tlb(&init_mm, me);
  230. set_tss_desc(cpu, t);
  231. load_TR_desc();
  232. load_LDT(&init_mm.context);
  233. /*
  234. * Clear all 6 debug registers:
  235. */
  236. set_debug(0UL, 0);
  237. set_debug(0UL, 1);
  238. set_debug(0UL, 2);
  239. set_debug(0UL, 3);
  240. set_debug(0UL, 6);
  241. set_debug(0UL, 7);
  242. fpu_init();
  243. }