avoid the confusion in call trace because of the lack of padding at the tail of function.
Signed-off-by: jia zhang <[email protected]>
---
When do_exit get call, the return address behind call instruction is pushed into stack. If something get wrong in do_exit, for x86_64, the entry "kernel_execve +0x00/0xXX" rather than "child_rip +0xYY/0xZZ" is remained in call trace. It looks confused.
b/arch/x86/kernel/entry_32.S | 1 +
b/arch/x86/kernel/entry_64.S | 1 +
2 files changed, 2 insertions(+)
diff --git a/arch/x86/kernel/entry_32.S b/arch/x86/kernel/entry_32.S
index 28b597e..3290819 100644
--- a/arch/x86/kernel/entry_32.S
+++ b/arch/x86/kernel/entry_32.S
@@ -1051,6 +1051,7 @@ ENTRY(kernel_thread_helper)
push %eax
CFI_ADJUST_CFA_OFFSET 4
call do_exit
+ nop # padding for call trace
CFI_ENDPROC
ENDPROC(kernel_thread_helper)
diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
index b86f332..6f677a6 100644
--- a/arch/x86/kernel/entry_64.S
+++ b/arch/x86/kernel/entry_64.S
@@ -1170,6 +1170,7 @@ child_rip:
# exit
mov %eax, %edi
call do_exit
+ nop # padding for call trace
CFI_ENDPROC
ENDPROC(child_rip)
On Sun, Nov 23, 2008 at 10:47:10PM +0800, jia zhang wrote:
>
> avoid the confusion in call trace because of the lack of padding at the tail of
> function.
>
> Signed-off-by: jia zhang <[email protected]> --- When do_exit get call,
> the return address behind call instruction is pushed into stack. If something
> get wrong in do_exit, for x86_64, the entry "kernel_execve +0x00/0xXX" rather
> than "child_rip +0xYY/0xZZ" is remained in call trace. It looks confused.
It's valuable to know from where do_exit is called, so that's the
only reason why using "call" is important. Otherwise it could just
be changed to a jmp as do_exit does not return. It's maybe a bit
ugly to add the padding, but it does give better traces. Maybe
a 'ret', 'int3', or 'ud2' would be slightly less offensive?
Anyhow, I don't have a problem with the added padding, so:
Acked-by: Alexander van Heukelum <[email protected]>
> b/arch/x86/kernel/entry_32.S | 1 +
> b/arch/x86/kernel/entry_64.S | 1 +
> 2 files changed, 2 insertions(+)
>
> diff --git a/arch/x86/kernel/entry_32.S b/arch/x86/kernel/entry_32.S
> index 28b597e..3290819 100644
> --- a/arch/x86/kernel/entry_32.S
> +++ b/arch/x86/kernel/entry_32.S
> @@ -1051,6 +1051,7 @@ ENTRY(kernel_thread_helper)
> push %eax
> CFI_ADJUST_CFA_OFFSET 4
> call do_exit
> + nop # padding for call trace
> CFI_ENDPROC
> ENDPROC(kernel_thread_helper)
>
> diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S
> index b86f332..6f677a6 100644
> --- a/arch/x86/kernel/entry_64.S
> +++ b/arch/x86/kernel/entry_64.S
> @@ -1170,6 +1170,7 @@ child_rip:
> # exit
> mov %eax, %edi
> call do_exit
> + nop # padding for call trace
> CFI_ENDPROC
> ENDPROC(child_rip)
* Alexander van Heukelum <[email protected]> wrote:
> On Sun, Nov 23, 2008 at 10:47:10PM +0800, jia zhang wrote:
> >
> > avoid the confusion in call trace because of the lack of padding at the tail of
> > function.
> >
> > Signed-off-by: jia zhang <[email protected]> --- When do_exit get call,
> > the return address behind call instruction is pushed into stack. If something
> > get wrong in do_exit, for x86_64, the entry "kernel_execve +0x00/0xXX" rather
> > than "child_rip +0xYY/0xZZ" is remained in call trace. It looks confused.
>
> It's valuable to know from where do_exit is called, so that's the
> only reason why using "call" is important. Otherwise it could just
> be changed to a jmp as do_exit does not return. It's maybe a bit
> ugly to add the padding, but it does give better traces. Maybe
> a 'ret', 'int3', or 'ud2' would be slightly less offensive?
>
> Anyhow, I don't have a problem with the added padding, so:
>
> Acked-by: Alexander van Heukelum <[email protected]>
applied to tip/x86/debug, thanks guys!
i changed it to u2d based on your suggestion - that makes it not only
easier to understand, but also more robust, should do_exit() return.
(which it should never in practice)
Ingo