Re: [Libunwind-devel] libunwind with LD_PRELOAD option

[Ken Werner]

On 09/07/2011 03:37 PM, Shan Shan wrote:
>         Hi,
>
>         Sorry for being late. My laptop died on Monday morning and I'm
>         in the
>         middle of setting up a new one.
>
>         Libunwind on ARM can unwind the stack via:
>         1) DWARF information (-g)
>         2) ARM specific unwind tables (-f-unwind-tables)
>         3) APCS frame parsing (-mapcs-frame)
>
> Thanks. If none of these options are specified, will it still be able to
> unwind. I guess this is the case with my main application.

If there is insufficient unwind data available the result is undefined I 
guess (I wouldn't expect libunwind to crash though). On modern ARM-Linux 
systems you'll need either DWARF (-g) or the ARM specific unwind tables 
(-funwind-tables) to be able to backtrace. The latter takes less space 
but is not as accurate as DWARF in some corner cases (for example in 
case an async signal interrupted a prologue or epilogue).

> Also, is there any clue on why the app crashes just with linking and no
> attempts made to get backtrace?

No, not from the top of my head. It could be a symbol collision as Arun 
mentioned.

>         By default libunwind looks at the DWARF info first because it's more
>         accurate and if that fails libunwind will use the ARM specific
>         unwind
>         tables. In case everything fails it falls back on APCS frame
>         parsing.
>         Since modern ARM-Linux systems are using the ARM EABI rather
>         than the
>         APCS this code won't help there. But from looking at the sources
>         I could
>         imagine that the APCS parsing could cause crashes (just a wild
>         guess).
>         Could you try running with the environment variable
>         UNW_ARM_METHOD_DWARF
>         set to "5" and check if it still crashes?
>
>
>     Sorry, I meant:
>     UNW_ARM_UNWIND_METHOD=5
>
> I just tried this option and it still crashes :-(. I am yet to try the
> option of having "-funwind-tables" with my main app because with this
> option, my unit test app worked fine. I will update you all soon on
> that. Got tied up with another issue :(

Ok, adding -funwind-tables sounds like a good plan if you want to 
backtrace but I'm not sure it'll fix your crashes. : )

Regards
Ken