Add CFI to x86_64 RTLD_START
diff mbox

Message ID CABV8kRzPpdhRsWOsU+zNEc7AcAqTqm1xdWxM0rufGAjk=yn4TA@mail.gmail.com
State New
Headers show

Commit Message

Keno Fischer July 4, 2016, 5:38 p.m. UTC
The application's entry point already has the appropriate CFI to
indicate to tools that the end of the stack has been reached. This
CFI was lacking from the dynamic linker, causing garbled backtraces
after hitting _dl_start. Fix this by adding the appropriate CFI.

Example gdb output before:

```
(gdb) bt
 #0  0x00007f10a2cee2d0 in _start () from /lib64/ld-linux-x86-64.so.2
 #1  0x0000000000000003 in ?? ()
 #2  0x00007ffd2ba54ce9 in ?? ()
 #3  0x00007ffd2ba54d04 in ?? ()
 #4  0x00007ffd2ba54d07 in ?? ()
 #5  0x0000000000000000 in ?? ()
```

after:
```
(gdb) bt
 #0  0x00007f36273551b0 in _start () from
/home/kfischer/replacement-usr/lib/ld-2.23.90.so
```

This is likely applicable to other architectures as well, but I only
have access to x86_64, so this patch is limited to that architecture.
---
 sysdeps/x86_64/dl-machine.h | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

  # See if we were run as a command with the executable file\n\
@@ -180,7 +185,8 @@ _dl_start_user:\n\
  # And make sure %rsp points to argc stored on the stack.\n\
  movq %r13, %rsp\n\
  # Jump to the user's entry point.\n\
- jmp *%r12\n\
+ jmp *%r12\n"\
+CFI_ENDPROC "\n\
 .previous\n\
 ");

Comments

Andreas Schwab July 4, 2016, 6:23 p.m. UTC | #1
Keno Fischer <keno@juliacomputing.com> writes:

> diff --git a/sysdeps/x86_64/dl-machine.h b/sysdeps/x86_64/dl-machine.h
> index ed0c1a8..7a246d4 100644
> --- a/sysdeps/x86_64/dl-machine.h
> +++ b/sysdeps/x86_64/dl-machine.h
> @@ -141,10 +141,15 @@ elf_machine_runtime_setup (struct link_map *l,
> int lazy, int profile)
>   .align 16\n\
>  .globl _start\n\
>  .globl _dl_start_user\n\
> -_start:\n\
> +_start:\n"\
> +CFI_STARTPROC "\n"\
> +CFI_UNDEFINED(%rip) "\n\

No need for backslash continuation outside of the string literals.

Andreas.
Keno Fischer July 4, 2016, 6:27 p.m. UTC | #2
I should have probably sent this patch with more diff context, but
since this is part of a preprocessor `#define`, the backslashes are
still required, no?
Andreas Schwab July 4, 2016, 7:37 p.m. UTC | #3
Keno Fischer <keno@juliacomputing.com> writes:

> I should have probably sent this patch with more diff context, but
> since this is part of a preprocessor `#define`, the backslashes are
> still required, no?

Yes, sorry, I missed that.  But please put a space before the
backslashes outside of the string literals.

Andreas.
Keno Fischer July 8, 2016, 8:45 p.m. UTC | #4
Might it make sense to change the macro to clearly
separate the string literals and the backslashes, i.e.

".globl _start\n"                 \
".globl _dl_start_user\n"   \
"_start:\n"                         \

etc.?

On Mon, Jul 4, 2016 at 3:37 PM, Andreas Schwab <schwab@linux-m68k.org> wrote:
> Keno Fischer <keno@juliacomputing.com> writes:
>
>> I should have probably sent this patch with more diff context, but
>> since this is part of a preprocessor `#define`, the backslashes are
>> still required, no?
>
> Yes, sorry, I missed that.  But please put a space before the
> backslashes outside of the string literals.
>
> Andreas.
>
> --
> Andreas Schwab, schwab@linux-m68k.org
> GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
> "And now for something completely different."
Keno Fischer Aug. 2, 2016, 3:21 a.m. UTC | #5
Hello again. I'd be happy to change this patch in whatever way you
deem best. Could you suggest what you would like me to do?

Thanks,
Keno
Keno Fischer Aug. 11, 2016, 7:38 p.m. UTC | #6
Bump again, I would like to make sure this doesn't get lost.
Please let me know how I should proceed.

On Mon, Aug 1, 2016 at 11:21 PM, Keno Fischer <keno@juliacomputing.com> wrote:
> Hello again. I'd be happy to change this patch in whatever way you
> deem best. Could you suggest what you would like me to do?
>
> Thanks,
> Keno
Keno Fischer Nov. 1, 2016, 8:48 p.m. UTC | #7
Bump. Please let me know the stylistic preferences for the assembly
code as well as any other comments
on the patch.

On Thu, Aug 11, 2016 at 3:38 PM, Keno Fischer <keno@juliacomputing.com> wrote:
> Bump again, I would like to make sure this doesn't get lost.
> Please let me know how I should proceed.
>
> On Mon, Aug 1, 2016 at 11:21 PM, Keno Fischer <keno@juliacomputing.com> wrote:
>> Hello again. I'd be happy to change this patch in whatever way you
>> deem best. Could you suggest what you would like me to do?
>>
>> Thanks,
>> Keno
H.J. Lu Nov. 7, 2016, 4:44 p.m. UTC | #8
On Tue, Nov 1, 2016 at 1:48 PM, Keno Fischer <keno@juliacomputing.com> wrote:
> Bump. Please let me know the stylistic preferences for the assembly
> code as well as any other comments
> on the patch.
>
> On Thu, Aug 11, 2016 at 3:38 PM, Keno Fischer <keno@juliacomputing.com> wrote:
>> Bump again, I would like to make sure this doesn't get lost.
>> Please let me know how I should proceed.
>>
>> On Mon, Aug 1, 2016 at 11:21 PM, Keno Fischer <keno@juliacomputing.com> wrote:
>>> Hello again. I'd be happy to change this patch in whatever way you
>>> deem best. Could you suggest what you would like me to do?
>>>
>>> Thanks,
>>> Keno

It looks good to me.
Keno Fischer Nov. 16, 2016, 3:51 a.m. UTC | #9
Thanks for taking a look H.J. What's the next step? Can somebody
commit this (or the reformatted version I sent as v2)? I'd also be
happy to do another revision, if there are any further review comments
- just want to make sure this doesn't get forgotten.

Thanks,
Keno

Patch
diff mbox

diff --git a/sysdeps/x86_64/dl-machine.h b/sysdeps/x86_64/dl-machine.h
index ed0c1a8..7a246d4 100644
--- a/sysdeps/x86_64/dl-machine.h
+++ b/sysdeps/x86_64/dl-machine.h
@@ -141,10 +141,15 @@  elf_machine_runtime_setup (struct link_map *l,
int lazy, int profile)
  .align 16\n\
 .globl _start\n\
 .globl _dl_start_user\n\
-_start:\n\
+_start:\n"\
+CFI_STARTPROC "\n"\
+CFI_UNDEFINED(%rip) "\n\
  movq %rsp, %rdi\n\
- call _dl_start\n\
-_dl_start_user:\n\
+ call _dl_start\n"\
+CFI_ENDPROC "\n\
+_dl_start_user:\n"\
+CFI_STARTPROC "\n"\
+CFI_UNDEFINED(%rip) "\n\
  # Save the user entry point address in %r12.\n\
  movq %rax, %r12\n\