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 #ifndef CONFIG_DEBUG_BOOT_PARAMS
  28 struct boot_params __initdata boot_params;
  29 #else
  30 struct boot_params boot_params;
  31 #endif
  32
  33 cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
  34
  35 struct x8664_pda *_cpu_pda[NR_CPUS] __read_mostly;
  36 EXPORT_SYMBOL(_cpu_pda);
  37 struct x8664_pda boot_cpu_pda[NR_CPUS] __cacheline_aligned;
  38
  39 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
  40
  41 char boot_cpu_stack[IRQSTACKSIZE] __attribute__((section(".bss.page_aligned")));
  42
  43 unsigned long __supported_pte_mask __read_mostly = ~0UL;
  44 static int do_not_nx __cpuinitdata = 0;
  45
  46 /* noexec=on|off
  47 Control non executable mappings for 64bit processes.
  48
  49 on      Enable(default)
  50 off     Disable
  51 */
  52 static int __init nonx_setup(char *str)
  53 {
  54         if (!str)
  55                 return -EINVAL;
  56         if (!strncmp(str, "on", 2)) {
  57                 __supported_pte_mask |= _PAGE_NX;
  58                 do_not_nx = 0;
  59         } else if (!strncmp(str, "off", 3)) {
  60                 do_not_nx = 1;
  61                 __supported_pte_mask &= ~_PAGE_NX;
  62         }
  63         return 0;
  64 }
  65 early_param("noexec", nonx_setup);
  66
  67 int force_personality32 = 0;
  68
  69 /* noexec32=on|off
  70 Control non executable heap for 32bit processes.
  71 To control the stack too use noexec=off
  72
  73 on      PROT_READ does not imply PROT_EXEC for 32bit processes
  74 off     PROT_READ implies PROT_EXEC (default)
  75 */
  76 static int __init nonx32_setup(char *str)
  77 {
  78         if (!strcmp(str, "on"))
  79                 force_personality32 &= ~READ_IMPLIES_EXEC;
  80         else if (!strcmp(str, "off"))
  81                 force_personality32 |= READ_IMPLIES_EXEC;
  82         return 1;
  83 }
  84 __setup("noexec32=", nonx32_setup);
  85
  86 /*
  87  * Great future plan:
  88  * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
  89  * Always point %gs to its beginning
  90  */
  91 void __init setup_per_cpu_areas(void)
  92 {
  93         int i;
  94         unsigned long size;
  95
  96 #ifdef CONFIG_HOTPLUG_CPU
  97         prefill_possible_map();
  98 #endif
  99
 100         /* Copy section for each CPU (we discard the original) */
 101         size = PERCPU_ENOUGH_ROOM;
 102
 103         printk(KERN_INFO "PERCPU: Allocating %lu bytes of per cpu data\n", size);
 104         for_each_cpu_mask (i, cpu_possible_map) {
 105                 char *ptr;
 106
 107                 if (!NODE_DATA(cpu_to_node(i))) {
 108                         printk("cpu with no node %d, num_online_nodes %d\n",
 109                                i, num_online_nodes());
 110                         ptr = alloc_bootmem_pages(size);
 111                 } else {
 112                         ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
 113                 }
 114                 if (!ptr)
 115                         panic("Cannot allocate cpu data for CPU %d\n", i);
 116                 cpu_pda(i)->data_offset = ptr - __per_cpu_start;
 117                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 118         }
 119 }
 120
 121 void pda_init(int cpu)
 122 {
 123         struct x8664_pda *pda = cpu_pda(cpu);
 124
 125         /* Setup up data that may be needed in __get_free_pages early */
 126         asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
 127         /* Memory clobbers used to order PDA accessed */
 128         mb();
 129         wrmsrl(MSR_GS_BASE, pda);
 130         mb();
 131
 132         pda->cpunumber = cpu;
 133         pda->irqcount = -1;
 134         pda->kernelstack =
 135                 (unsigned long)stack_thread_info() - PDA_STACKOFFSET + THREAD_SIZE;
 136         pda->active_mm = &init_mm;
 137         pda->mmu_state = 0;
 138
 139         if (cpu == 0) {
 140                 /* others are initialized in smpboot.c */
 141                 pda->pcurrent = &init_task;
 142                 pda->irqstackptr = boot_cpu_stack;
 143         } else {
 144                 pda->irqstackptr = (char *)
 145                         __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
 146                 if (!pda->irqstackptr)
 147                         panic("cannot allocate irqstack for cpu %d", cpu);
 148         }
 149
 150
 151         pda->irqstackptr += IRQSTACKSIZE-64;
 152 }
 153
 154 char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]
 155 __attribute__((section(".bss.page_aligned")));
 156
 157 extern asmlinkage void ignore_sysret(void);
 158
 159 /* May not be marked __init: used by software suspend */
 160 void syscall_init(void)
 161 {
 162         /*
 163          * LSTAR and STAR live in a bit strange symbiosis.
 164          * They both write to the same internal register. STAR allows to set CS/DS
 165          * but only a 32bit target. LSTAR sets the 64bit rip.
 166          */
 167         wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
 168         wrmsrl(MSR_LSTAR, system_call);
 169         wrmsrl(MSR_CSTAR, ignore_sysret);
 170
 171 #ifdef CONFIG_IA32_EMULATION
 172         syscall32_cpu_init ();
 173 #endif
 174
 175         /* Flags to clear on syscall */
 176         wrmsrl(MSR_SYSCALL_MASK,
 177                X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
 178 }
 179
 180 void __cpuinit check_efer(void)
 181 {
 182         unsigned long efer;
 183
 184         rdmsrl(MSR_EFER, efer);
 185         if (!(efer & EFER_NX) || do_not_nx) {
 186                 __supported_pte_mask &= ~_PAGE_NX;
 187         }
 188 }
 189
 190 unsigned long kernel_eflags;
 191
 192 /*
 193  * Copies of the original ist values from the tss are only accessed during
 194  * debugging, no special alignment required.
 195  */
 196 DEFINE_PER_CPU(struct orig_ist, orig_ist);
 197
 198 /*
 199  * cpu_init() initializes state that is per-CPU. Some data is already
 200  * initialized (naturally) in the bootstrap process, such as the GDT
 201  * and IDT. We reload them nevertheless, this function acts as a
 202  * 'CPU state barrier', nothing should get across.
 203  * A lot of state is already set up in PDA init.
 204  */
 205 void __cpuinit cpu_init (void)
 206 {
 207         int cpu = stack_smp_processor_id();
 208         struct tss_struct *t = &per_cpu(init_tss, cpu);
 209         struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
 210         unsigned long v;
 211         char *estacks = NULL;
 212         struct task_struct *me;
 213         int i;
 214
 215         /* CPU 0 is initialised in head64.c */
 216         if (cpu != 0) {
 217                 pda_init(cpu);
 218         } else
 219                 estacks = boot_exception_stacks;
 220
 221         me = current;
 222
 223         if (cpu_test_and_set(cpu, cpu_initialized))
 224                 panic("CPU#%d already initialized!\n", cpu);
 225
 226         printk("Initializing CPU#%d\n", cpu);
 227
 228         clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
 229
 230         /*
 231          * Initialize the per-CPU GDT with the boot GDT,
 232          * and set up the GDT descriptor:
 233          */
 234         if (cpu)
 235                 memcpy(get_cpu_gdt_table(cpu), cpu_gdt_table, GDT_SIZE);
 236
 237         cpu_gdt_descr[cpu].size = GDT_SIZE;
 238         load_gdt((const struct desc_ptr *)&cpu_gdt_descr[cpu]);
 239         load_idt((const struct desc_ptr *)&idt_descr);
 240
 241         memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
 242         syscall_init();
 243
 244         wrmsrl(MSR_FS_BASE, 0);
 245         wrmsrl(MSR_KERNEL_GS_BASE, 0);
 246         barrier();
 247
 248         check_efer();
 249
 250         /*
 251          * set up and load the per-CPU TSS
 252          */
 253         for (v = 0; v < N_EXCEPTION_STACKS; v++) {
 254                 static const unsigned int order[N_EXCEPTION_STACKS] = {
 255                         [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
 256                         [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
 257                 };
 258                 if (cpu) {
 259                         estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
 260                         if (!estacks)
 261                                 panic("Cannot allocate exception stack %ld %d\n",
 262                                       v, cpu);
 263                 }
 264                 estacks += PAGE_SIZE << order[v];
 265                 orig_ist->ist[v] = t->x86_tss.ist[v] = (unsigned long)estacks;
 266         }
 267
 268         t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
 269         /*
 270          * <= is required because the CPU will access up to
 271          * 8 bits beyond the end of the IO permission bitmap.
 272          */
 273         for (i = 0; i <= IO_BITMAP_LONGS; i++)
 274                 t->io_bitmap[i] = ~0UL;
 275
 276         atomic_inc(&init_mm.mm_count);
 277         me->active_mm = &init_mm;
 278         if (me->mm)
 279                 BUG();
 280         enter_lazy_tlb(&init_mm, me);
 281
 282         set_tss_desc(cpu, t);
 283         load_TR_desc();
 284         load_LDT(&init_mm.context);
 285
 286         /*
 287          * Clear all 6 debug registers:
 288          */
 289
 290         set_debugreg(0UL, 0);
 291         set_debugreg(0UL, 1);
 292         set_debugreg(0UL, 2);
 293         set_debugreg(0UL, 3);
 294         set_debugreg(0UL, 6);
 295         set_debugreg(0UL, 7);
 296
 297         fpu_init();
 298
 299         raw_local_save_flags(kernel_eflags);
 300 }