Re: [PATCH] powerpc/64s: POWER10 CPU Kconfig build option

["Nicholas Piggin" <npiggin@gmail.com>]

On Fri Oct 7, 2022 at 9:23 AM AEST, Segher Boessenkool wrote:
> On Fri, Oct 07, 2022 at 07:56:09AM +1000, Nicholas Piggin wrote:
> > On Fri Oct 7, 2022 at 5:54 AM AEST, Segher Boessenkool wrote:
> > > On Fri, Sep 23, 2022 at 01:30:04PM +1000, Nicholas Piggin wrote:
> > > > This adds basic POWER10_CPU option, which builds with -mcpu=power10.
> > >
> > > > +# No prefix or pcrel
> > > > +KBUILD_CFLAGS += $(call cc-option,-mno-prefixed)
> > > > +KBUILD_CFLAGS += $(call cc-option,-mno-pcrel)
> > >
> > > Why do you disable all prefixed insns?  What goes wrong if you don't?
> > 
> > Potentially things like kprobes.
>
> So mention that?  "This patch is due to an abundance of caution".

Well it's in next now. I did say *basic*, I'm sure not changing the ABI
or adding prefix instructions isn't too mysterious.

>
> What I meant to ask is if you *saw* something going wrong, not if you
> can imagine something going wrong.  I can imagine ten gazillion things
> going wrong, that is not why I asked :-)
>
> > > Same question for pcrel.  I'm sure you want to optimise it better, but
> > > it's not clear to me how it fails now?
> > 
> > For pcrel addressing? Bootstrapping the C environment is one, the
> > module dynamic linker is another.
>
> I don't know what either of those mean.

arch/powerpc/kernel/head_64.S and arch/powerpc/kernel/module_64.c

Can discuss in the pcrel patch series thread if you would like to know
more.

>
> > Some details in this series.
> > 
> > https://lists.ozlabs.org/pipermail/linuxppc-dev/2022-September/248521.html
>
> I've watched that series with great interest, but I don't remember
> anything like that?  Are you refering to the commentary in 7/7?
> "Definitely ftrace and probes, possibly BPF and KVM have some breakage.
> I haven't looked closely yet."...  This doesn't mean much does it :-)
> It can be a triviality or two.  Or it could be a massive roadblock.
>
> Just say in a comment where you disable stuff that it is to prevent
> possible problems, this is a WIP, that kind of thing?  Otherwise other
> people (like me :-) ) will read it and think there must be some deeper
> reason.  Like, changing code to work with pcrel is hard or a lot of
> work -- it isn't :-)  As you say in 0/7 yourself btw!
>

I will describe limitations and issues a bit more in changelog of patches
to enable prefix and pcrel when I submit as non-RFC candidate. It would
probably not be too hard to get things to a workable state that could be
merged.

> > > > +# No AltiVec or VSX or MMA instructions when building kernel
> > > >  KBUILD_CFLAGS += $(call cc-option,-mno-altivec)
> > > >  KBUILD_CFLAGS += $(call cc-option,-mno-vsx)
> > > > +KBUILD_CFLAGS += $(call cc-option,-mno-mma)
> > >
> > > MMA code is never generated unless the code asks for it explicitly.
> > > This is fundamental, not just an implementations side effect.
> > 
> > Well, now it double won't be generated :)
>
> Yeah, but there are many other things you can unnecessarily disable as
> well!  :-)
>
> VMX and VSX are disabled here because the compiler *will* use those
> registers if it feels like it (that is, if it thinks that will be
> faster).  MMA is a very different beast: the compiler can never know if
> it will be faster, to start with.

True, but now I don't have to find the exact clause and have my lawyer
confirm that it definitely probably won't change in future and break
things.

Thanks,
Nick