x86-64, mm: Initialize VDSO earlier on 64 bits

When initrd is in use and a driver does request_module() in its
module_init (i.e. __initcall or device_initcall), a modprobe process
is created with VDSO mapping. But VDSO is inited even in __initcall,
i.e. on the same level (at the same time), so it may not be inited
yet (link order matters).

Move the VDSO initialization code earlier by switching to something
before rootfs_initcall where initrd is loaded as rootfs. Specifically
to subsys_initcall. Do it for standard 64-bit path (init_vdso_vars)
and for compat (sysenter_setup), just in case people have 32-bit
initrd and ia32 emulation built-in.

i386 (pure 32-bit) is not affected, since sysenter_setup() is called
from check_bugs()->identify_boot_cpu() in start_kernel() before
rest_init()->kernel_thread(kernel_init) where even kernel_init() calls
do_basic_setup()->do_initcalls().

What this patch fixes are early modprobe crashes such as:
Unpacking initramfs...
Freeing initrd memory: 9324k freed
modprobe[368]: segfault at 7fff4429c020 ip 00007fef397e160c \
sp 00007fff442795c0 error 4 in ld-2.11.2.so[7fef397df000+1f000]

Signed-off-by: Jiri Slaby <[e-mail]>
LKML-Reference: <[e-mail]>
Signed-off-by: H. Peter Anvin <[e-mail]>

arch/x86/vdso/vdso32-setup.c

···
 
 #ifdef CONFIG_X86_64
 
-__initcall(sysenter_setup);
+subsys_initcall(sysenter_setup);
 
 #ifdef CONFIG_SYSCTL
 /* Register vsyscall32 into the ABI table */

arch/x86/vdso/vma.c

···
 	vdso_enabled = 0;
 	return -ENOMEM;
 }
-__initcall(init_vdso_vars);
+subsys_initcall(init_vdso_vars);
 
 struct linux_binprm;