tunables vs osxsave vs checkpointing vs _dl_runtime_resolve

[DJ Delorie]

Szabolcs Nagy <szabolcs.nagy@arm.com> writes:
> so are we supposed to handle migrations to machines with
> different arch extensions?

No, that's not the intent.  There's a lot of "caveat user" in what our
customer wants, and if the x86 tunables didn't already *almost* do what
they wanted, we would have just said no.  I mean, x86 has code to do
what our customer wants, and has code to let tunables control it, it
just didn't actually work.  So in this case, it's more of a bug fix than
a new feature.

> cpu_features based decisions can break across different
> machines and there is no reliable (future proof) way to
> request baseline arch features.

That's a separate problem, which we've solved with ifuncs, but others
solve with multilibs.  X86 chooses to let tunables override ifunc
choices, and exposes CPU features, setting an expectation.  If other
arches don't do that, so be it.