arch/x86_64/kernel/setup64.c

   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
  27 char x86_boot_params[BOOT_PARAM_SIZE] __initdata;
  28
  29 cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
  30
  31 struct x8664_pda *_cpu_pda[NR_CPUS] __read_mostly;
  32 EXPORT_SYMBOL(_cpu_pda);
  33 struct x8664_pda boot_cpu_pda[NR_CPUS] __cacheline_aligned;
  34
  35 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
  36
  37 char boot_cpu_stack[IRQSTACKSIZE] __attribute__((section(".bss.page_aligned")));
  38
  39 unsigned long __supported_pte_mask __read_mostly = ~0UL;
  40 EXPORT_SYMBOL(__supported_pte_mask);
  41 static int do_not_nx __cpuinitdata = 0;
  42
  43 /* noexec=on|off
  44 Control non executable mappings for 64bit processes.
  45
  46 on      Enable(default)
  47 off     Disable
  48 */
  49 static int __init nonx_setup(char *str)
  50 {
  51         if (!str)
  52                 return -EINVAL;
  53         if (!strncmp(str, "on", 2)) {
  54                 __supported_pte_mask |= _PAGE_NX;
  55                 do_not_nx = 0;
  56         } else if (!strncmp(str, "off", 3)) {
  57                 do_not_nx = 1;
  58                 __supported_pte_mask &= ~_PAGE_NX;
  59         }
  60         return 0;
  61 }
  62 early_param("noexec", nonx_setup);
  63
  64 int force_personality32 = 0;
  65
  66 /* noexec32=on|off
  67 Control non executable heap for 32bit processes.
  68 To control the stack too use noexec=off
  69
  70 on      PROT_READ does not imply PROT_EXEC for 32bit processes
  71 off     PROT_READ implies PROT_EXEC (default)
  72 */
  73 static int __init nonx32_setup(char *str)
  74 {
  75         if (!strcmp(str, "on"))
  76                 force_personality32 &= ~READ_IMPLIES_EXEC;
  77         else if (!strcmp(str, "off"))
  78                 force_personality32 |= READ_IMPLIES_EXEC;
  79         return 1;
  80 }
  81 __setup("noexec32=", nonx32_setup);
  82
  83 /*
  84  * Great future plan:
  85  * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
  86  * Always point %gs to its beginning
  87  */
  88 void __init setup_per_cpu_areas(void)
  89 {
  90         int i;
  91         unsigned long size;
  92
  93 #ifdef CONFIG_HOTPLUG_CPU
  94         prefill_possible_map();
  95 #endif
  96
  97         /* Copy section for each CPU (we discard the original) */
  98         size = PERCPU_ENOUGH_ROOM;
  99
 100         printk(KERN_INFO "PERCPU: Allocating %lu bytes of per cpu data\n", size);
 101         for_each_cpu_mask (i, cpu_possible_map) {
 102                 char *ptr;
 103
 104                 if (!NODE_DATA(cpu_to_node(i))) {
 105                         printk("cpu with no node %d, num_online_nodes %d\n",
 106                                i, num_online_nodes());
 107                         ptr = alloc_bootmem(size);
 108                 } else {
 109                         ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
 110                 }
 111                 if (!ptr)
 112                         panic("Cannot allocate cpu data for CPU %d\n", i);
 113                 cpu_pda(i)->data_offset = ptr - __per_cpu_start;
 114                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 115         }
 116 }
 117
 118 void pda_init(int cpu)
 119 {
 120         struct x8664_pda *pda = cpu_pda(cpu);
 121
 122         /* Setup up data that may be needed in __get_free_pages early */
 123         asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
 124         /* Memory clobbers used to order PDA accessed */
 125         mb();
 126         wrmsrl(MSR_GS_BASE, pda);
 127         mb();
 128
 129         pda->cpunumber = cpu;
 130         pda->irqcount = -1;
 131         pda->kernelstack =
 132                 (unsigned long)stack_thread_info() - PDA_STACKOFFSET + THREAD_SIZE;
 133         pda->active_mm = &init_mm;
 134         pda->mmu_state = 0;
 135
 136         if (cpu == 0) {
 137                 /* others are initialized in smpboot.c */
 138                 pda->pcurrent = &init_task;
 139                 pda->irqstackptr = boot_cpu_stack;
 140         } else {
 141                 pda->irqstackptr = (char *)
 142                         __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
 143                 if (!pda->irqstackptr)
 144                         panic("cannot allocate irqstack for cpu %d", cpu);
 145         }
 146
 147
 148         pda->irqstackptr += IRQSTACKSIZE-64;
 149 }
 150
 151 char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]
 152 __attribute__((section(".bss.page_aligned")));
 153
 154 /* May not be marked __init: used by software suspend */
 155 void syscall_init(void)
 156 {
 157         /*
 158          * LSTAR and STAR live in a bit strange symbiosis.
 159          * They both write to the same internal register. STAR allows to set CS/DS
 160          * but only a 32bit target. LSTAR sets the 64bit rip.
 161          */
 162         wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
 163         wrmsrl(MSR_LSTAR, system_call);
 164
 165 #ifdef CONFIG_IA32_EMULATION
 166         syscall32_cpu_init ();
 167 #endif
 168
 169         /* Flags to clear on syscall */
 170         wrmsrl(MSR_SYSCALL_MASK, EF_TF|EF_DF|EF_IE|0x3000);
 171 }
 172
 173 void __cpuinit check_efer(void)
 174 {
 175         unsigned long efer;
 176
 177         rdmsrl(MSR_EFER, efer);
 178         if (!(efer & EFER_NX) || do_not_nx) {
 179                 __supported_pte_mask &= ~_PAGE_NX;
 180         }
 181 }
 182
 183 unsigned long kernel_eflags;
 184
 185 /*
 186  * cpu_init() initializes state that is per-CPU. Some data is already
 187  * initialized (naturally) in the bootstrap process, such as the GDT
 188  * and IDT. We reload them nevertheless, this function acts as a
 189  * 'CPU state barrier', nothing should get across.
 190  * A lot of state is already set up in PDA init.
 191  */
 192 void __cpuinit cpu_init (void)
 193 {
 194         int cpu = stack_smp_processor_id();
 195         struct tss_struct *t = &per_cpu(init_tss, cpu);
 196         struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
 197         unsigned long v;
 198         char *estacks = NULL;
 199         struct task_struct *me;
 200         int i;
 201
 202         /* CPU 0 is initialised in head64.c */
 203         if (cpu != 0) {
 204                 pda_init(cpu);
 205                 zap_low_mappings(cpu);
 206         } else
 207                 estacks = boot_exception_stacks;
 208
 209         me = current;
 210
 211         if (cpu_test_and_set(cpu, cpu_initialized))
 212                 panic("CPU#%d already initialized!\n", cpu);
 213
 214         printk("Initializing CPU#%d\n", cpu);
 215
 216         clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
 217
 218         /*
 219          * Initialize the per-CPU GDT with the boot GDT,
 220          * and set up the GDT descriptor:
 221          */
 222         if (cpu)
 223                 memcpy(cpu_gdt(cpu), cpu_gdt_table, GDT_SIZE);
 224
 225         cpu_gdt_descr[cpu].size = GDT_SIZE;
 226         asm volatile("lgdt %0" :: "m" (cpu_gdt_descr[cpu]));
 227         asm volatile("lidt %0" :: "m" (idt_descr));
 228
 229         memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
 230         syscall_init();
 231
 232         wrmsrl(MSR_FS_BASE, 0);
 233         wrmsrl(MSR_KERNEL_GS_BASE, 0);
 234         barrier();
 235
 236         check_efer();
 237
 238         /*
 239          * set up and load the per-CPU TSS
 240          */
 241         for (v = 0; v < N_EXCEPTION_STACKS; v++) {
 242                 static const unsigned int order[N_EXCEPTION_STACKS] = {
 243                         [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
 244                         [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
 245                 };
 246                 if (cpu) {
 247                         estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
 248                         if (!estacks)
 249                                 panic("Cannot allocate exception stack %ld %d\n",
 250                                       v, cpu);
 251                 }
 252                 estacks += PAGE_SIZE << order[v];
 253                 orig_ist->ist[v] = t->ist[v] = (unsigned long)estacks;
 254         }
 255
 256         t->io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
 257         /*
 258          * <= is required because the CPU will access up to
 259          * 8 bits beyond the end of the IO permission bitmap.
 260          */
 261         for (i = 0; i <= IO_BITMAP_LONGS; i++)
 262                 t->io_bitmap[i] = ~0UL;
 263
 264         atomic_inc(&init_mm.mm_count);
 265         me->active_mm = &init_mm;
 266         if (me->mm)
 267                 BUG();
 268         enter_lazy_tlb(&init_mm, me);
 269
 270         set_tss_desc(cpu, t);
 271         load_TR_desc();
 272         load_LDT(&init_mm.context);
 273
 274         /*
 275          * Clear all 6 debug registers:
 276          */
 277
 278         set_debugreg(0UL, 0);
 279         set_debugreg(0UL, 1);
 280         set_debugreg(0UL, 2);
 281         set_debugreg(0UL, 3);
 282         set_debugreg(0UL, 6);
 283         set_debugreg(0UL, 7);
 284
 285         fpu_init();
 286
 287         raw_local_save_flags(kernel_eflags);
 288 }