Re: compiling glibc using -fasynchronous-unwind-tables for arm

[Adhemerval Zanella Netto <adhemerval.zanella@linaro.org>]

Hi Abhijit,

I am not sure why you need to build all glibc with -fasynchronous-unwind-tables, my
understanding you need either to improve debugability or performance profiling on
workloads with signals.

And I think this is similar to https://sourceware.org/bugzilla/show_bug.cgi?id=29621 ,
and if it were the case where libgcc was built with -fexceptions but without optimization
we can't support it on glibc at the moment (as Florian has added on last comment).

On 21/03/23 08:30, Abhijit Ray Chaudhury wrote:
> Hi Adhemerval,
> 
> In my build environment gcc toolchain is installed in docker through apt. I will try to find the source code and build .
> 
> Can you please let me know why you are concentrating on the resolution of " __aeabi_unwind_cpp_pr0" symbol ? I thought if you give "-fasynchronous-unwind-tables -fno-unwind-tables", gcc should not even generate the " __aeabi_unwind_cpp_pr0" symbol. My toolchain wrongly generates it .
> 
> Is my understanding correct?
> 
> Kindly advise,
> -Abhijit
> 
> On Mon, Mar 20, 2023 at 10:30 PM Adhemerval Zanella Netto <adhemerval.zanella@linaro.org <mailto:adhemerval.zanella@linaro.org>> wrote:
> 
>     The -fasynchronous-unwind-table usually is used so backtraces works through signal handlers.
>     Could you check if you build your toolchain with optimization flags (so libgcc is built with
>     -O2) you can also build with -fasynchronous-unwind-table?
> 
>     On 17/03/23 10:37, Abhijit Ray Chaudhury wrote:
>     > Hello Adhemerval,
>     >
>     > I could compile glibc with -funwind-table instead of -fasynchronous-unwind-table. I noticed glibc uses one of the flags for different files. As far as stack unwinding is concerned , what is the difference between -fasynchronous-unwind-table and  -funwind-table ?
>     >
>     > Thanking You,
>     > -Abhijit
>     >
>     > On Thu, Mar 16, 2023 at 4:29 PM Abhijit Ray Chaudhury <abhijit.ray.chaudhury@gmail.com <mailto:abhijit.ray.chaudhury@gmail.com> <mailto:abhijit.ray.chaudhury@gmail.com <mailto:abhijit.ray.chaudhury@gmail.com>>> wrote:
>     >
>     >     Hi,
>     >
>     >     I think the problem lies with the precedence of seemingly conflicting CFLAGS "-fasynchronous-unwind-tables -fno-unwind-tables" supplied to my cross toolchain gcc. I compiled a simple C file with " -fasynchronous-unwind-tables" , it generated an undefined symbol " `__aeabi_unwind_cpp_pr0'" although compilation was successful without -static flag. Then I compiled with " -fasynchronous-unwind-tables -fno-unwind-tables", it still generated the undefined symbol " `__aeabi_unwind_cpp_pr0'". It suggests " -fno-unwind-tables" does not stop unwind information in the ELF when I have " -fasynchronous-unwind-tables" . I then compiled with " -funwind-tables -fno-unwind-table", I did not find  undefined symbol " `__aeabi_unwind_cpp_pr0'"in the generated ELF.
>     >
>     >     So I will compile glibc with " -funwind-table" instead of " -fasynchronous-unwind-table" and let you know the result.
>     >
>     >     Do you see any problem with this approach ?
>     >
>     >     Thanking You,
>     >     -Abhijit
>     >
>     >     On Thu, Mar 16, 2023 at 1:22 AM Adhemerval Zanella Netto <adhemerval.zanella@linaro.org <mailto:adhemerval.zanella@linaro.org> <mailto:adhemerval.zanella@linaro.org <mailto:adhemerval.zanella@linaro.org>>> wrote:
>     >
>     >
>     >
>     >         On 15/03/23 11:20, Florian Weimer wrote:
>     >         > * Adhemerval Zanella Netto via Libc-help:
>     >         >
>     >         >> And I have tested with both:
>     >         >>
>     >         >> CC="arm-glibc-linux-gnueabihf-gcc -fasynchronous-unwind-tables"
>     >         >>
>     >         >> and
>     >         >>
>     >         >> CC=arm-glibc-linux-gnueabihf-gcc CFLAGS="-O2 -fasynchronous-unwind-tables"
>     >         >>
>     >         >> I also tested with gcc from ubuntu22 to check if this might something
>     >         >> I haven't enabled on my toolchain (since I don't build all languages),
>     >         >> but again I did not see any issue.
>     >         >
>     >         > It probably depends on whether libgcc.a has been built with
>     >         > optimizations or not.
>     >
>     >         Indeed this looks like https://sourceware.org/bugzilla/show_bug.cgi?id=29621 <https://sourceware.org/bugzilla/show_bug.cgi?id=29621> <https://sourceware.org/bugzilla/show_bug.cgi?id=29621 <https://sourceware.org/bugzilla/show_bug.cgi?id=29621>> .
>     >         Does the build work if you manually adds the -fasynchronous-unwind-tables
>     >         on config.make?
>     >
>