Re: [Qemu-devel] nvram and boot order

[Benjamin Herrenschmidt]

On Fri, 2012-10-19 at 09:21 -0500, Anthony Liguori wrote:

> > That's true to an extent.  However, I vehemently disagree that it's
> > arbitrary which one gets the new option.  Neither -boot nor bootindex=
> > alter any persistent data now and they should not suddenly start doing
> > so.
> 
> That's not true.
> 
> For the PC, -boot modifies the CMOS memory.  If we persisted CMOS, then
> -boot would cause a persistent modification in the boot order.

Come on Anthony, you are talking like a politician now :-) "If we
were...". The fact is that from a user perspective today, the effect is
not persistent. Thus making it so would be a chance in perceived
behaviour. Not a big deal but I think your "correction" of David
statement was unnecessary :-)

As a matter of fact, I would even argue that if you make the CMOS
persistent on x86, you should change the effect of -boot to write to a
part of that CMOS that is "for the next boot only" in order to preserve
the user perceived (and probably expected behaviour).

In any case, I think we have an agreement that we want both methods.

What I think David explained fairly well and I share his opinion here is
that it's fairly easy and probably higher on the priority list, to
provide the methods to do things in a volatile way. So let's do that
now, which is going to be in line with the current perceived behaviour
and generally makes sense.

We can *also* look at putting together a mechanism for setting the boot
device permanently. But that's non trivial for several reasons and so we
shouldn't gate the first one which is immediately useful because we are
still debating the second one.

Cheers,
Ben.