arch/x86/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/pda.h>
  15 #include <asm/pgtable.h>
  16 #include <asm/processor.h>
  17 #include <asm/desc.h>
  18 #include <asm/atomic.h>
  19 #include <asm/mmu_context.h>
  20 #include <asm/smp.h>
  21 #include <asm/i387.h>
  22 #include <asm/percpu.h>
  23 #include <asm/proto.h>
  24 #include <asm/sections.h>
  25 #include <asm/setup.h>
  26
  27 struct boot_params __initdata boot_params;
  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 static int do_not_nx __cpuinitdata = 0;
  41
  42 /* noexec=on|off
  43 Control non executable mappings for 64bit processes.
  44
  45 on      Enable(default)
  46 off     Disable
  47 */
  48 static int __init nonx_setup(char *str)
  49 {
  50         if (!str)
  51                 return -EINVAL;
  52         if (!strncmp(str, "on", 2)) {
  53                 __supported_pte_mask |= _PAGE_NX;
  54                 do_not_nx = 0;
  55         } else if (!strncmp(str, "off", 3)) {
  56                 do_not_nx = 1;
  57                 __supported_pte_mask &= ~_PAGE_NX;
  58         }
  59         return 0;
  60 }
  61 early_param("noexec", nonx_setup);
  62
  63 int force_personality32 = 0;
  64
  65 /* noexec32=on|off
  66 Control non executable heap for 32bit processes.
  67 To control the stack too use noexec=off
  68
  69 on      PROT_READ does not imply PROT_EXEC for 32bit processes
  70 off     PROT_READ implies PROT_EXEC (default)
  71 */
  72 static int __init nonx32_setup(char *str)
  73 {
  74         if (!strcmp(str, "on"))
  75                 force_personality32 &= ~READ_IMPLIES_EXEC;
  76         else if (!strcmp(str, "off"))
  77                 force_personality32 |= READ_IMPLIES_EXEC;
  78         return 1;
  79 }
  80 __setup("noexec32=", nonx32_setup);
  81
  82 /*
  83  * Great future plan:
  84  * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
  85  * Always point %gs to its beginning
  86  */
  87 void __init setup_per_cpu_areas(void)
  88 {
  89         int i;
  90         unsigned long size;
  91
  92 #ifdef CONFIG_HOTPLUG_CPU
  93         prefill_possible_map();
  94 #endif
  95
  96         /* Copy section for each CPU (we discard the original) */
  97         size = PERCPU_ENOUGH_ROOM;
  98
  99         printk(KERN_INFO "PERCPU: Allocating %lu bytes of per cpu data\n", size);
 100         for_each_cpu_mask (i, cpu_possible_map) {
 101                 char *ptr;
 102
 103                 if (!NODE_DATA(cpu_to_node(i))) {
 104                         printk("cpu with no node %d, num_online_nodes %d\n",
 105                                i, num_online_nodes());
 106                         ptr = alloc_bootmem_pages(size);
 107                 } else {
 108                         ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
 109                 }
 110                 if (!ptr)
 111                         panic("Cannot allocate cpu data for CPU %d\n", i);
 112                 cpu_pda(i)->data_offset = ptr - __per_cpu_start;
 113                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 114         }
 115 }
 116
 117 void pda_init(int cpu)
 118 {
 119         struct x8664_pda *pda = cpu_pda(cpu);
 120
 121         /* Setup up data that may be needed in __get_free_pages early */
 122         asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
 123         /* Memory clobbers used to order PDA accessed */
 124         mb();
 125         wrmsrl(MSR_GS_BASE, pda);
 126         mb();
 127
 128         pda->cpunumber = cpu;
 129         pda->irqcount = -1;
 130         pda->kernelstack =
 131                 (unsigned long)stack_thread_info() - PDA_STACKOFFSET + THREAD_SIZE;
 132         pda->active_mm = &init_mm;
 133         pda->mmu_state = 0;
 134
 135         if (cpu == 0) {
 136                 /* others are initialized in smpboot.c */
 137                 pda->pcurrent = &init_task;
 138                 pda->irqstackptr = boot_cpu_stack;
 139         } else {
 140                 pda->irqstackptr = (char *)
 141                         __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
 142                 if (!pda->irqstackptr)
 143                         panic("cannot allocate irqstack for cpu %d", cpu);
 144         }
 145
 146
 147         pda->irqstackptr += IRQSTACKSIZE-64;
 148 }
 149
 150 char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]
 151 __attribute__((section(".bss.page_aligned")));
 152
 153 extern asmlinkage void ignore_sysret(void);
 154
 155 /* May not be marked __init: used by software suspend */
 156 void syscall_init(void)
 157 {
 158         /*
 159          * LSTAR and STAR live in a bit strange symbiosis.
 160          * They both write to the same internal register. STAR allows to set CS/DS
 161          * but only a 32bit target. LSTAR sets the 64bit rip.
 162          */
 163         wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
 164         wrmsrl(MSR_LSTAR, system_call);
 165         wrmsrl(MSR_CSTAR, ignore_sysret);
 166
 167 #ifdef CONFIG_IA32_EMULATION
 168         syscall32_cpu_init ();
 169 #endif
 170
 171         /* Flags to clear on syscall */
 172         wrmsrl(MSR_SYSCALL_MASK,
 173                X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
 174 }
 175
 176 void __cpuinit check_efer(void)
 177 {
 178         unsigned long efer;
 179
 180         rdmsrl(MSR_EFER, efer);
 181         if (!(efer & EFER_NX) || do_not_nx) {
 182                 __supported_pte_mask &= ~_PAGE_NX;
 183         }
 184 }
 185
 186 unsigned long kernel_eflags;
 187
 188 /*
 189  * Copies of the original ist values from the tss are only accessed during
 190  * debugging, no special alignment required.
 191  */
 192 DEFINE_PER_CPU(struct orig_ist, orig_ist);
 193
 194 /*
 195  * cpu_init() initializes state that is per-CPU. Some data is already
 196  * initialized (naturally) in the bootstrap process, such as the GDT
 197  * and IDT. We reload them nevertheless, this function acts as a
 198  * 'CPU state barrier', nothing should get across.
 199  * A lot of state is already set up in PDA init.
 200  */
 201 void __cpuinit cpu_init (void)
 202 {
 203         int cpu = stack_smp_processor_id();
 204         struct tss_struct *t = &per_cpu(init_tss, cpu);
 205         struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
 206         unsigned long v;
 207         char *estacks = NULL;
 208         struct task_struct *me;
 209         int i;
 210
 211         /* CPU 0 is initialised in head64.c */
 212         if (cpu != 0) {
 213                 pda_init(cpu);
 214         } else
 215                 estacks = boot_exception_stacks;
 216
 217         me = current;
 218
 219         if (cpu_test_and_set(cpu, cpu_initialized))
 220                 panic("CPU#%d already initialized!\n", cpu);
 221
 222         printk("Initializing CPU#%d\n", cpu);
 223
 224         clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
 225
 226         /*
 227          * Initialize the per-CPU GDT with the boot GDT,
 228          * and set up the GDT descriptor:
 229          */
 230         if (cpu)
 231                 memcpy(get_cpu_gdt_table(cpu), cpu_gdt_table, GDT_SIZE);
 232
 233         cpu_gdt_descr[cpu].size = GDT_SIZE;
 234         load_gdt((const struct desc_ptr *)&cpu_gdt_descr[cpu]);
 235         load_idt((const struct desc_ptr *)&idt_descr);
 236
 237         memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
 238         syscall_init();
 239
 240         wrmsrl(MSR_FS_BASE, 0);
 241         wrmsrl(MSR_KERNEL_GS_BASE, 0);
 242         barrier();
 243
 244         check_efer();
 245
 246         /*
 247          * set up and load the per-CPU TSS
 248          */
 249         for (v = 0; v < N_EXCEPTION_STACKS; v++) {
 250                 static const unsigned int order[N_EXCEPTION_STACKS] = {
 251                         [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
 252                         [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
 253                 };
 254                 if (cpu) {
 255                         estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
 256                         if (!estacks)
 257                                 panic("Cannot allocate exception stack %ld %d\n",
 258                                       v, cpu);
 259                 }
 260                 estacks += PAGE_SIZE << order[v];
 261                 orig_ist->ist[v] = t->x86_tss.ist[v] = (unsigned long)estacks;
 262         }
 263
 264         t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
 265         /*
 266          * <= is required because the CPU will access up to
 267          * 8 bits beyond the end of the IO permission bitmap.
 268          */
 269         for (i = 0; i <= IO_BITMAP_LONGS; i++)
 270                 t->io_bitmap[i] = ~0UL;
 271
 272         atomic_inc(&init_mm.mm_count);
 273         me->active_mm = &init_mm;
 274         if (me->mm)
 275                 BUG();
 276         enter_lazy_tlb(&init_mm, me);
 277
 278         set_tss_desc(cpu, t);
 279         load_TR_desc();
 280         load_LDT(&init_mm.context);
 281
 282         /*
 283          * Clear all 6 debug registers:
 284          */
 285
 286         set_debugreg(0UL, 0);
 287         set_debugreg(0UL, 1);
 288         set_debugreg(0UL, 2);
 289         set_debugreg(0UL, 3);
 290         set_debugreg(0UL, 6);
 291         set_debugreg(0UL, 7);
 292
 293         fpu_init();
 294
 295         raw_local_save_flags(kernel_eflags);
 296 }