David Daney writes:Ralf,
> Andreas Tobler wrote:
> > > David Daney wrote:
> >
> >
> >> I know next to nothing about PPC ABIs, but are any of these floating > >> point registers?
> >
> >
> > There are, yes.
> >
> >> Are there any call saved FP registers in this ABI? and if so are you > >> restoring them. Although I don't think that the unwinder uses > >> floating point, it seems that restoring call saved FP registers is a > >> good idea if you are not already doing it.
> >
> >
> > Well, here I expect the advise from the experts, I have floats around > > and I may try to restore them.
> >
> > But, I need some guidance here.
> > When I did the MD_FALLBACK_FRAME_STATE_FOR for mips/linux I did not > handle floating point either as the problem did not occur to me until > after I checked in the code.
> > However after thinking about it and posting:
> > http://gcc.gnu.org/ml/gcc/2003-10/msg00972.html
> > I learned that this is a real issue.
> > I may be about ready to do some more mips/linux work soon and may > revisit MD_FALLBACK_FRAME_STATE_FOR. Because in its current state it > seems to be incomplete.
You only need to restore what has been saved. Looking at /usr/src/linux-2.4/arch/mips/kernel/signal.c, it seems that there is a call to save_fp_context(). However, this is only executed if (current->used_math) is set; you mustn't restore any fp registers if the process hasn't saved the fp state.
There is a field called sc_used_math in the sigcontext struct. I think this tells you what you need to know. But I am not a kernel hacker...
Andrew.
Is this all true?
Perhaps you could shed some light on what really needs to be done in the MIPS/linux case.
Also what should be done in the case of mips 4Kc core where there is only software floating point?
David Daney.
