Re: [PATCH] Relax MIPS j/jal out-of-range check

[Thiemo Seufer <ths at networkno dot de>]

Daniel Jacobowitz wrote:
[snip]
> >  Daniel> Why?  Especially bearing in mind that we're in the assembler
> >  Daniel> here.  This doesn't make sense to me.
> > 
> > If you have a routine that's sometimes called from uncached code, and
> > the label is a KSEG0 address, you could call it from KSEG1 code simply
> > with
> > 	jal	label
> > and the machine will do the right thing (you end up at the KSEG1
> > address corresponding to "label").
> 
> How could you get the assembler to do this?  Normally you can't tell if
> this is going to overflow until you've linked!  Linker errors are a
> different kettle of fish.
> 
> I think the warning is just bogus, after looking at it.  We're warning
> for "jal 0x10000000".  But we don't know the location of the jal yet,
> so we haven't a clue if it's out of range, do we?

The patch is correct, but our explanation was misleading: The code
checks for absolute addresses, and warns now if it crosses a 256 MB
memory area. This allows e.g. to jump between KSEG0 and KSEG1 without
hassles.


Thiemo