x86, vdso: Fix an OOPS accessing the HPET mapping w/o an HPET
authorAndy Lutomirski <luto@amacapital.net>
Wed, 14 May 2014 23:23:13 +0000 (16:23 -0700)
committerH. Peter Anvin <hpa@linux.intel.com>
Wed, 21 May 2014 23:14:04 +0000 (16:14 -0700)
The oops can be triggered in qemu using -no-hpet (but not nohpet) by
running a 32-bit program and reading a couple of pages before the vdso.
This should send SIGBUS instead of OOPSing.

The bug was introduced by:

commit 7a59ed415f5b57469e22e41fc4188d5399e0b194
Author: Stefani Seibold <stefani@seibold.net>
Date:   Mon Mar 17 23:22:09 2014 +0100

    x86, vdso: Add 32 bit VDSO time support for 32 bit kernel

which is new in 3.15.

Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Link: http://lkml.kernel.org/r/e99025d887d6670b6c4d81e6ccfeeb83770b21e9.1400109621.git.luto@amacapital.net
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
arch/x86/vdso/vdso32-setup.c

index e1f220e..310c5f0 100644 (file)
@@ -155,6 +155,7 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
        unsigned long addr;
        int ret = 0;
        struct vm_area_struct *vma;
+       static struct page *no_pages[] = {NULL};
 
 #ifdef CONFIG_X86_X32_ABI
        if (test_thread_flag(TIF_X32))
@@ -193,7 +194,7 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
                        addr -  VDSO_OFFSET(VDSO_PREV_PAGES),
                        VDSO_OFFSET(VDSO_PREV_PAGES),
                        VM_READ,
-                       NULL);
+                       no_pages);
 
        if (IS_ERR(vma)) {
                ret = PTR_ERR(vma);