Re: [avr-gcc-list] avr-ld: Do linker stubs need --relax?

[Jan Waclawek]

I'd say, it's the other way round, and Joerg's remark is a *consequence* of 
PR14058, which in turn is consequence of an incomplete/untested implementation 
of the stubs. 

Oh, I already said that (in both the binutils bugtracker and the said 
avrfreaks.net) ... :-)


I'd even risk a guess that working backwards the patches we would find a point, 
where the stubs without -relax are working correctly, as PR14058 is a 
consequence of an optimisation step in the stubs handling.

Jan


----- Original Message ---------------

Subject: [avr-gcc-list] avr-ld: Do linker stubs need --relax?
   From: Georg-Johann Lay <address@hidden>
   Date: Fri, 04 May 2012 17:44:02 +0200
     To: address@hidden

>Jörg Wunsch wrote in [1]:
>
>>> The most important thing about using more than 128 KiB of flash is
>>> to not forget about the -mrelax compiler (linker) option.
>
>The question "Is using linker stubs supported without --relax at all?" came up
>recently, namely in
>
>http://sourceware.org/PR14058#c4
>
>AFAIR correct handling/generation of stubs needs --relax, but I cannot find
>it documented in binutils nor together with -mrelax documentation in avr-gcc.
>There is just a hint in AVR-LibC docs, far away from -mrelax and --relax.
>
>As far as I understand you:
>
>1) --relax is needed for linker to operate correctly
>with stubs resp. gs expression modifier.
>
>2) Binutils PR14058 and perhaps also PR13812 should be classified as "invalid".
>
>3) Binutils documentstion should document that implication.
>
>4) The avr-gcc documentation should document that implication, e.g. together
>with -mrelax and EIND policies.
>
>Johann
>
>
>[1]
>http://www.avrfreaks.net/index.php?name=PNphpBB2&file=viewtopic&p=877735#877735
>
>______