Re: [PATCH 34/34] RFC x86: introduce directio virt cap

[Wei Liu <wei.liu2@citrix.com>]

On Tue, Aug 21, 2018 at 09:40:07AM -0600, Jan Beulich wrote:
> >>> On 21.08.18 at 15:43, <wei.liu2@citrix.com> wrote:
> > On Tue, Aug 21, 2018 at 05:52:39AM -0600, Jan Beulich wrote:
> >> >>> On 17.08.18 at 17:12, <wei.liu2@citrix.com> wrote:
> >> > --- a/xen/arch/x86/sysctl.c
> >> > +++ b/xen/arch/x86/sysctl.c
> >> > @@ -92,8 +92,10 @@ void arch_do_physinfo(struct xen_sysctl_physinfo *pi)
> >> >             min(sizeof(pi->hw_cap), sizeof(boot_cpu_data.x86_capability)));
> >> >      if ( hvm_enabled )
> >> >          pi->capabilities |= XEN_SYSCTL_PHYSCAP_hvm;
> >> > -    if ( iommu_enabled )
> >> > +    if ( hvm_enabled && iommu_enabled )
> >> >          pi->capabilities |= XEN_SYSCTL_PHYSCAP_hvm_directio;
> >> > +    else if ( iommu_enabled )
> >> > +        pi->capabilities |= XEN_SYSCTL_PHYSCAP_directio;
> >> >  }
> >> 
> >> At the sysctl layer I think you can, as suggested iirc by Roger,
> >> simply replace hvm_directio with directio (or iommu). For the
> >> "xl info" output, otoh, I'm afraid this doesn't hold, as people
> >> may parse for the string. Depending on how this would best
> >> be addressed in the tool stack, replacing the sysctl names may
> >> then no longer be the most suitable solution.
> > 
> > In that case what do you think about the two flags this patch provides
> > on the toolstack level?
> > 
> > Essentially we change slightly hvm_directio's meaning to mean "hvm &&
> > iommu_enabled" while it previously mean "iommu_enabled", and then in the
> > absence of hvm_directio, add "directio" as an indication for
> > "iommu_enabled".
> 
> I think when you mean to provide two flags, retaining the meaning
> of the pre-existing one might be desirable. But as said before -
> much depends on what the tool stack means to present to the user.
> The expectations on the current meaning on "xl info" output should
> not be broken.

Previously "hvm" and "hvm_directio" always show up together. The
hvm_directio only case never showed up in practice -- I don't think
there had been systems with vt-d but not vt-x.

My plan for xl info:

pv	hvm	iommu		flags in xl info
0	0	0		n/a
0	0	1		n/a
0	1	0		hvm
0	1	1		hvm hvm_directio
1	0	0		NIL
1	0	1		directio
1	1	0		hvm
1	1	1		hvm hvm_directio directio

This retains sensible (though not completely identical) semantics for
hvm_directio.

The term "directio" can also be "iommu" if that's preferable.

Wei.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel