Why -mxgot on MIPS and not -fpic/-fPIC?

[Maciej W. Rozycki]

On Mon, 3 Oct 2016, Matthew Fortune wrote:

> > It looks like -mxgot has a similar impact on MIPS as using -fPIC
> > (instead of -fpic) on other architectures.
> > 
> > Why wasn't the -fpic/-fPIC distinction carried over to MIPS?
> 
> Much of the history here is probably folklore and guesswork but for
> my understanding is that MIPS has three levels of 'pic'.
> 
> MIPS PIC code is really controlled by -mabicalls and -mshared which
> I assume originate from the MIPS-pro days.
> 
> -fpic is simply set to imply -mabicalls and -mshared which is where
> everything gets far more confusing (but let's not get into that
> unless we have to)
> 
> The reason there is no -fPIC is probably a conscious decision as
> MIPS has 'multi-got' that allows the small -fpic model to scale to
> most situations.
> 
> -mxgot (aka big-got) is however a good match for -fPIC as it is a
> model guaranteed to work irrespective of how big the GOT is.

 For MIPS `-fPIC' is an alias to `-fpic' and the actual reason is these 
options have only been retrofitted to the MIPS target when non-PIC support 
has been added to shared-library (as opposed to bare-metal) targets.  The 
original MIPS SVR4 psABI mandated all code to be PIC, even executables, 
and the model chosen was controlled with the `-mxgot' option, hardly ever 
used, due to the code bloat implied and link-time incompatibility with the 
default `-mno-xgot' code.  The latter could be rectified to some extent by 
a specific GOT entry ordering, but with the advent of multi-GOT there was 
little incentive to actually implement it.

 So once non-PIC support has been added to shared-library targets, back in 
2008 only (i.e. 14 years from the inception of the MIPS/Linux port, almost 
twice as much time as has passed from the addition of the feature till 
now; to say nothing of IRIX support; cf. binutils commit 861fb55ab50a, GCC 
commit 139170) a choice was made not to change the existing arrangement 
for PIC code.  I think given that lots of software just blindly uses 
`-fPIC' in their Makefiles rather than switching between `-fpic' and 
`-fPIC' somehow (how?) and that code built with either of `-mno-xgot' and 
`-mxgot' is incompatible with each other it was a good choice.  I think 
`libtool' could cope with that somehow although obviously all the world is 
not `libtool'.

 NB we still have all code PIC for the new ABIs (n64 and n32), in which 
case all of `-fpic'/`-fPIC'/`-fno-pic'/`-fno-PIC' are no-ops, and GCC can 
be configured for the old ABI (o32) to default to PIC executables as well, 
in which case these are no-ops too, unless overridden with `-mplt'.

> I do have a report to deal with about -fPIC and the value it sets
> for __PIC__ because it always sets '1' due to not supporting -fPIC.
> It is tempting to just make -fPIC imply -mxgot and set __PIC__ to 2
> but it will lead to bloated code in a number of cases where
> multi-got is sufficient.

 I'd rather say it would break almost everything.

  Maciej