Re: [GIT PULL] x86/build changes for v4.17

From: Ingo Molnar
Date: Tue Apr 03 2018 - 05:51:26 EST



* Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:

> On Mon, Apr 02, 2018 at 02:44:48PM -0700, Linus Torvalds wrote:
> > On Mon, Apr 2, 2018 at 2:50 AM, Ingo Molnar <mingo@xxxxxxxxxx> wrote:
> > >
> > > The biggest change is the forcing of asm-goto support on x86, which effectively
> > > increases the GCC minimum supported version to gcc-4.5 (on x86).
> >
> > So my biggest worry isn't gcc-4.5 (anybody who hasn't updated deserves
> > to be forced, or can stay with old kernels).
> >
> > No, my biggest worry is clang. What's the status there?
> >
> > I've pulled this, and honestly, the disaster with
> > -fmerge-all-constants makes me think that clang isn't that good a
> > compiler choice anyway, but it's sad if this undoes a lot of clang
> > work just because of the worries about Spectre and mis-speculated
> > branches.
>
> It's not just spectre, I believe you yourself wanted to use asm-goto
> somewhere in the x86 code:
>
> http://lkml.kernel.org/r/CA+55aFyCp-9Qqjcn9wp=VDp2KO7tfYuUMJxVKC75Xxu0wEB5Cw@xxxxxxxxxxxxxx
>
> There was some KVM talk of relying on it here:
>
> http://lkml.kernel.org/r/6a5f2453-cf51-d491-db54-5f239caa29bc@xxxxxxxxxx
>
> And there's the comment here:
>
> https://elixir.bootlin.com/linux/v4.16-rc7/source/arch/x86/kvm/emulate.c#L457
>
> As to the suitablility of using clang, there's also this unresolved
> issue:
>
> http://lkml.kernel.org/r/20180321211931.GA111711@xxxxxxxxxx
>
> The fact that even without asm-goto they cannot correctly compile a
> kernel and have sat on their hands regarding asm-goto for the past 7 odd
> years makes me care very little.
>
> And since they need to spin a new version of the compiler with all the
> various bugs fixed, they might as well include asm-goto in that and be
> done with it.

So there's really two questions here:

- This asm-goto change only impacts x86, is there any production x86 kernel being
built with Clang? I think the Pixel kernel is built with Clang, but that's ARM.

- Is there anyone on the Clang side _actually_ bending metal and working on
asm-goto support, with something like very early alpha test patches available,
etc.? Last I saw the communicated Clang POV was still that they wanted to do
something "better" (and incompatible to ...) asm-goto. Has this changed?

If it's being relied on, or if there's actually something firmly planned,
which we could track, then I'd have no problem with reverting this change
and waiting one more kernel cycle or so.

Thanks,

Ingo