[Skiboot] [PATCH 08/12] xive: make endian-clean

[Oliver O'Halloran]

On Tue, Oct 1, 2019 at 3:41 PM Cédric Le Goater <clg at kaod.org> wrote:
>
> On 01/10/2019 07:24, Oliver O'Halloran wrote:
> > On Sun, 2019-09-29 at 17:46 +1000, Nicholas Piggin wrote:
> >> Convert xive opal calls, dt construction, and in-memory hardware tables
> >> to use explicit endian conversions.
> >
> > Cedric, ack/nak?
>
> Sorry. I was not in CC:
>
> Let's be honest. This is very valid but the result is dreadful.
>
> > I'm a bit iffy about the __prefixes since the in-memory tables are
> > homogenously BE. If we're going to be more concerned about endian
> > safety going forward I'd prefer we use the __beXX annotations and some
> > kind of automated checking rather than ad-hoc conventions.
>
> I don't mind the __be types but I don't understand the __ prefix in
> the word structures. is that a requirement ?
>
> > We might also want to do something similar to what we did with the
> > radix tables in Linux where constants we use change based on endianness
> > rather than having the code do cpu_to_beXX all over the place.
>
> I introduced helpers in QEMU to deal with endianness, as we have
> the same problem on the pseries and PowerNV machines. It does
> not cover all cases but it cleans up the code a lot.
>
> https://git.qemu.org/?p=qemu.git;a=blob;f=include/hw/ppc/xive_regs.h;h=08c8bf7172e22ec964603c2192fd462a1ae918a2;hb=HEAD#l162
>
> Are we in a hurry ? This is also completely ruining my patchset ...
>
> Nick, do you want me to take over ? I will complain less :)

I was planning on merging this after you do your v3, so no rush. I'll
let you and nick fight it out over who gets to do the rebase ;)