2 * linux/arch/i386/kernel/sysenter.c
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
8 * This file contains the needed initializations to support sysenter.
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>
19 #include <linux/err.h>
20 #include <linux/module.h>
22 #include <asm/cpufeature.h>
24 #include <asm/pgtable.h>
25 #include <asm/unistd.h>
27 #include <asm/tlbflush.h>
35 #ifdef CONFIG_COMPAT_VDSO
36 #define VDSO_DEFAULT VDSO_COMPAT
38 #define VDSO_DEFAULT VDSO_ENABLED
42 * Should the kernel map a VDSO page into processes and pass its
43 * address down to glibc upon exec()?
45 unsigned int __read_mostly vdso_enabled
= VDSO_DEFAULT
;
47 EXPORT_SYMBOL_GPL(vdso_enabled
);
49 static int __init
vdso_setup(char *s
)
51 vdso_enabled
= simple_strtoul(s
, NULL
, 0);
56 __setup("vdso=", vdso_setup
);
58 extern asmlinkage
void sysenter_entry(void);
60 static __init
void reloc_symtab(Elf32_Ehdr
*ehdr
,
61 unsigned offset
, unsigned size
)
63 Elf32_Sym
*sym
= (void *)ehdr
+ offset
;
64 unsigned nsym
= size
/ sizeof(*sym
);
67 for(i
= 0; i
< nsym
; i
++, sym
++) {
68 if (sym
->st_shndx
== SHN_UNDEF
||
69 sym
->st_shndx
== SHN_ABS
)
72 if (sym
->st_shndx
> SHN_LORESERVE
) {
73 printk(KERN_INFO
"VDSO: unexpected st_shndx %x\n",
78 switch(ELF_ST_TYPE(sym
->st_info
)) {
83 sym
->st_value
+= VDSO_HIGH_BASE
;
88 static __init
void reloc_dyn(Elf32_Ehdr
*ehdr
, unsigned offset
)
90 Elf32_Dyn
*dyn
= (void *)ehdr
+ offset
;
92 for(; dyn
->d_tag
!= DT_NULL
; dyn
++)
107 case DT_ADDRRNGLO
... DT_ADDRRNGHI
:
108 /* definitely pointers needing relocation */
109 dyn
->d_un
.d_ptr
+= VDSO_HIGH_BASE
;
112 case DT_ENCODING
... OLD_DT_LOOS
-1:
113 case DT_LOOS
... DT_HIOS
-1:
114 /* Tags above DT_ENCODING are pointers if
116 if (dyn
->d_tag
>= DT_ENCODING
&&
117 (dyn
->d_tag
& 1) == 0)
118 dyn
->d_un
.d_ptr
+= VDSO_HIGH_BASE
;
126 case DT_VALRNGLO
... DT_VALRNGHI
:
127 /* definitely not pointers */
130 case OLD_DT_LOOS
... DT_LOOS
-1:
131 case DT_HIOS
... DT_VALRNGLO
-1:
133 if (dyn
->d_tag
> DT_ENCODING
)
134 printk(KERN_INFO
"VDSO: unexpected DT_tag %x\n",
140 static __init
void relocate_vdso(Elf32_Ehdr
*ehdr
)
146 BUG_ON(memcmp(ehdr
->e_ident
, ELFMAG
, 4) != 0 ||
147 !elf_check_arch(ehdr
) ||
148 ehdr
->e_type
!= ET_DYN
);
150 ehdr
->e_entry
+= VDSO_HIGH_BASE
;
153 phdr
= (void *)ehdr
+ ehdr
->e_phoff
;
154 for (i
= 0; i
< ehdr
->e_phnum
; i
++) {
155 phdr
[i
].p_vaddr
+= VDSO_HIGH_BASE
;
157 /* relocate dynamic stuff */
158 if (phdr
[i
].p_type
== PT_DYNAMIC
)
159 reloc_dyn(ehdr
, phdr
[i
].p_offset
);
162 /* rebase sections */
163 shdr
= (void *)ehdr
+ ehdr
->e_shoff
;
164 for(i
= 0; i
< ehdr
->e_shnum
; i
++) {
165 if (!(shdr
[i
].sh_flags
& SHF_ALLOC
))
168 shdr
[i
].sh_addr
+= VDSO_HIGH_BASE
;
170 if (shdr
[i
].sh_type
== SHT_SYMTAB
||
171 shdr
[i
].sh_type
== SHT_DYNSYM
)
172 reloc_symtab(ehdr
, shdr
[i
].sh_offset
,
177 void enable_sep_cpu(void)
180 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
182 if (!boot_cpu_has(X86_FEATURE_SEP
)) {
187 tss
->x86_tss
.ss1
= __KERNEL_CS
;
188 tss
->x86_tss
.esp1
= sizeof(struct tss_struct
) + (unsigned long) tss
;
189 wrmsr(MSR_IA32_SYSENTER_CS
, __KERNEL_CS
, 0);
190 wrmsr(MSR_IA32_SYSENTER_ESP
, tss
->x86_tss
.esp1
, 0);
191 wrmsr(MSR_IA32_SYSENTER_EIP
, (unsigned long) sysenter_entry
, 0);
195 static struct vm_area_struct gate_vma
;
197 static int __init
gate_vma_init(void)
199 gate_vma
.vm_mm
= NULL
;
200 gate_vma
.vm_start
= FIXADDR_USER_START
;
201 gate_vma
.vm_end
= FIXADDR_USER_END
;
202 gate_vma
.vm_flags
= VM_READ
| VM_MAYREAD
| VM_EXEC
| VM_MAYEXEC
;
203 gate_vma
.vm_page_prot
= __P101
;
205 * Make sure the vDSO gets into every core dump.
206 * Dumping its contents makes post-mortem fully interpretable later
207 * without matching up the same kernel and hardware config to see
208 * what PC values meant.
210 gate_vma
.vm_flags
|= VM_ALWAYSDUMP
;
215 * These symbols are defined by vsyscall.o to mark the bounds
216 * of the ELF DSO images included therein.
218 extern const char vsyscall_int80_start
, vsyscall_int80_end
;
219 extern const char vsyscall_sysenter_start
, vsyscall_sysenter_end
;
220 static struct page
*syscall_pages
[1];
222 static void map_compat_vdso(int map
)
224 static int vdso_mapped
;
226 if (map
== vdso_mapped
)
231 __set_fixmap(FIX_VDSO
, page_to_pfn(syscall_pages
[0]) << PAGE_SHIFT
,
232 map
? PAGE_READONLY_EXEC
: PAGE_NONE
);
234 /* flush stray tlbs */
238 int __init
sysenter_setup(void)
240 void *syscall_page
= (void *)get_zeroed_page(GFP_ATOMIC
);
241 const void *vsyscall
;
244 syscall_pages
[0] = virt_to_page(syscall_page
);
248 printk("Compat vDSO mapped to %08lx.\n", __fix_to_virt(FIX_VDSO
));
250 if (!boot_cpu_has(X86_FEATURE_SEP
)) {
251 vsyscall
= &vsyscall_int80_start
;
252 vsyscall_len
= &vsyscall_int80_end
- &vsyscall_int80_start
;
254 vsyscall
= &vsyscall_sysenter_start
;
255 vsyscall_len
= &vsyscall_sysenter_end
- &vsyscall_sysenter_start
;
258 memcpy(syscall_page
, vsyscall
, vsyscall_len
);
259 relocate_vdso(syscall_page
);
264 /* Defined in vsyscall-sysenter.S */
265 extern void SYSENTER_RETURN
;
267 /* Setup a VMA at program startup for the vsyscall page */
268 int arch_setup_additional_pages(struct linux_binprm
*bprm
, int exstack
)
270 struct mm_struct
*mm
= current
->mm
;
275 down_write(&mm
->mmap_sem
);
277 /* Test compat mode once here, in case someone
278 changes it via sysctl */
279 compat
= (vdso_enabled
== VDSO_COMPAT
);
281 map_compat_vdso(compat
);
284 addr
= VDSO_HIGH_BASE
;
286 addr
= get_unmapped_area(NULL
, 0, PAGE_SIZE
, 0, 0);
287 if (IS_ERR_VALUE(addr
)) {
293 * MAYWRITE to allow gdb to COW and set breakpoints
295 * Make sure the vDSO gets into every core dump.
296 * Dumping its contents makes post-mortem fully
297 * interpretable later without matching up the same
298 * kernel and hardware config to see what PC values
301 ret
= install_special_mapping(mm
, addr
, PAGE_SIZE
,
303 VM_MAYREAD
|VM_MAYWRITE
|VM_MAYEXEC
|
311 current
->mm
->context
.vdso
= (void *)addr
;
312 current_thread_info()->sysenter_return
=
313 (void *)VDSO_SYM(&SYSENTER_RETURN
);
316 up_write(&mm
->mmap_sem
);
321 const char *arch_vma_name(struct vm_area_struct
*vma
)
323 if (vma
->vm_mm
&& vma
->vm_start
== (long)vma
->vm_mm
->context
.vdso
)
328 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
330 struct mm_struct
*mm
= tsk
->mm
;
332 /* Check to see if this task was created in compat vdso mode */
333 if (mm
&& mm
->context
.vdso
== (void *)VDSO_HIGH_BASE
)
338 int in_gate_area(struct task_struct
*task
, unsigned long addr
)
340 const struct vm_area_struct
*vma
= get_gate_vma(task
);
342 return vma
&& addr
>= vma
->vm_start
&& addr
< vma
->vm_end
;
345 int in_gate_area_no_task(unsigned long addr
)