ARM/THUMB long calls stubs

[Richard Earnshaw]

On Sat, 2004-08-07 at 10:04, Sami Kantoluoto wrote:
> Hi Nick,
> 
> On Fri, Aug 06, 2004 at 01:53:20PM +0300, Sami Kantoluoto wrote:
> > 
> > On Fri, Aug 06, 2004 at 11:25:04AM +0100, Nick Clifton wrote:
> > 
> > > The way to avoid this "feature" (I hesitate to call it a bug) is to 
> > > compile your code with the "-ffunction-sections" switch added to the GCC 
> > > command line.  This tells GCC to place each function into its own 
> > > separate section, and GCC is smart enough to know that different 
> > > sections may not end up close to each other, so it cannot optimise away 
> > > the use of the long call.
> > 
> > This helped, thanks! (But I still think that __attribute__ ((section(x)))
> > should be enough ;-)
> 
> Would it be hard to fix this? So that gcc would take __attribute__
> ((section())) in account when determining if functions are in different
> sections or not. Feels bad as all functions are called with the long call..
> 
> PS. thanks for your help (it's being very nice that someone really
>     responds)!
> 
> 
>      -sk

I think it's a waste of time trying to fix this in this way.  We should
be fixing the linker so that it can automatically insert long-call
veneers where they are needed.  This is required in any case for EABI
conformance.

R.