arch/i386/kernel/sysenter.c

   1 /*
   2  * linux/arch/i386/kernel/sysenter.c
   3  *
   4  * (C) Copyright 2002 Linus Torvalds
   5  * Portions based on the vdso-randomization code from exec-shield:
   6  * Copyright(C) 2005-2006, Red Hat, Inc., Ingo Molnar
   7  *
   8  * This file contains the needed initializations to support sysenter.
   9  */
  10
  11 #include <linux/init.h>
  12 #include <linux/smp.h>
  13 #include <linux/thread_info.h>
  14 #include <linux/sched.h>
  15 #include <linux/gfp.h>
  16 #include <linux/string.h>
  17 #include <linux/elf.h>
  18 #include <linux/mm.h>
  19 #include <linux/module.h>
  20
  21 #include <asm/cpufeature.h>
  22 #include <asm/msr.h>
  23 #include <asm/pgtable.h>
  24 #include <asm/unistd.h>
  25
  26 /*
  27  * Should the kernel map a VDSO page into processes and pass its
  28  * address down to glibc upon exec()?
  29  */
  30 #ifdef CONFIG_PARAVIRT
  31 unsigned int __read_mostly vdso_enabled = 0;
  32 #else
  33 unsigned int __read_mostly vdso_enabled = 1;
  34 #endif
  35
  36 EXPORT_SYMBOL_GPL(vdso_enabled);
  37
  38 static int __init vdso_setup(char *s)
  39 {
  40         vdso_enabled = simple_strtoul(s, NULL, 0);
  41
  42         return 1;
  43 }
  44
  45 __setup("vdso=", vdso_setup);
  46
  47 extern asmlinkage void sysenter_entry(void);
  48
  49 void enable_sep_cpu(void)
  50 {
  51         int cpu = get_cpu();
  52         struct tss_struct *tss = &per_cpu(init_tss, cpu);
  53
  54         if (!boot_cpu_has(X86_FEATURE_SEP)) {
  55                 put_cpu();
  56                 return;
  57         }
  58
  59         tss->ss1 = __KERNEL_CS;
  60         tss->esp1 = sizeof(struct tss_struct) + (unsigned long) tss;
  61         wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
  62         wrmsr(MSR_IA32_SYSENTER_ESP, tss->esp1, 0);
  63         wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long) sysenter_entry, 0);
  64         put_cpu();
  65 }
  66
  67 /*
  68  * These symbols are defined by vsyscall.o to mark the bounds
  69  * of the ELF DSO images included therein.
  70  */
  71 extern const char vsyscall_int80_start, vsyscall_int80_end;
  72 extern const char vsyscall_sysenter_start, vsyscall_sysenter_end;
  73 static struct page *syscall_pages[1];
  74
  75 int __init sysenter_setup(void)
  76 {
  77         void *syscall_page = (void *)get_zeroed_page(GFP_ATOMIC);
  78         syscall_pages[0] = virt_to_page(syscall_page);
  79
  80 #ifdef CONFIG_COMPAT_VDSO
  81         __set_fixmap(FIX_VDSO, __pa(syscall_page), PAGE_READONLY_EXEC);
  82         printk("Compat vDSO mapped to %08lx.\n", __fix_to_virt(FIX_VDSO));
  83 #endif
  84
  85         if (!boot_cpu_has(X86_FEATURE_SEP)) {
  86                 memcpy(syscall_page,
  87                        &vsyscall_int80_start,
  88                        &vsyscall_int80_end - &vsyscall_int80_start);
  89                 return 0;
  90         }
  91
  92         memcpy(syscall_page,
  93                &vsyscall_sysenter_start,
  94                &vsyscall_sysenter_end - &vsyscall_sysenter_start);
  95
  96         return 0;
  97 }
  98
  99 #ifndef CONFIG_COMPAT_VDSO
 100 /* Defined in vsyscall-sysenter.S */
 101 extern void SYSENTER_RETURN;
 102
 103 /* Setup a VMA at program startup for the vsyscall page */
 104 int arch_setup_additional_pages(struct linux_binprm *bprm, int exstack)
 105 {
 106         struct mm_struct *mm = current->mm;
 107         unsigned long addr;
 108         int ret;
 109
 110         down_write(&mm->mmap_sem);
 111         addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
 112         if (IS_ERR_VALUE(addr)) {
 113                 ret = addr;
 114                 goto up_fail;
 115         }
 116
 117         /*
 118          * MAYWRITE to allow gdb to COW and set breakpoints
 119          *
 120          * Make sure the vDSO gets into every core dump.
 121          * Dumping its contents makes post-mortem fully interpretable later
 122          * without matching up the same kernel and hardware config to see
 123          * what PC values meant.
 124          */
 125         ret = install_special_mapping(mm, addr, PAGE_SIZE,
 126                                       VM_READ|VM_EXEC|
 127                                       VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
 128                                       VM_ALWAYSDUMP,
 129                                       syscall_pages);
 130         if (ret)
 131                 goto up_fail;
 132
 133         current->mm->context.vdso = (void *)addr;
 134         current_thread_info()->sysenter_return =
 135                                     (void *)VDSO_SYM(&SYSENTER_RETURN);
 136 up_fail:
 137         up_write(&mm->mmap_sem);
 138         return ret;
 139 }
 140
 141 const char *arch_vma_name(struct vm_area_struct *vma)
 142 {
 143         if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
 144                 return "[vdso]";
 145         return NULL;
 146 }
 147
 148 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 149 {
 150         return NULL;
 151 }
 152
 153 int in_gate_area(struct task_struct *task, unsigned long addr)
 154 {
 155         return 0;
 156 }
 157
 158 int in_gate_area_no_task(unsigned long addr)
 159 {
 160         return 0;
 161 }
 162 #endif