Re: [PATCH v2 22/49] KVM: x86: Add a macro to precisely handle aliased 0x1.EDX CPUID features

[Maxim Levitsky]

On Mon, 2024-08-05 at 15:00 -0700, Sean Christopherson wrote:
> On Mon, Aug 05, 2024, mlevitsk@xxxxxxxxxx wrote:
> > У чт, 2024-07-25 у 11:39 -0700, Sean Christopherson пише:
> > > > On Wed, Jul 24, 2024, Maxim Levitsky wrote:
> > > > > > On Mon, 2024-07-08 at 14:08 -0700, Sean Christopherson wrote:
> > > > > > > > On Thu, Jul 04, 2024, Maxim Levitsky wrote:
> > > > > > > > > > What if we defined the aliased features instead.
> > > > > > > > > > Something like this:
> > > > > > > > > > 
> > > > > > > > > > #define __X86_FEATURE_8000_0001_ALIAS(feature) \
> > > > > > > > > >         (feature + (CPUID_8000_0001_EDX - CPUID_1_EDX) * 32)
> > > > > > > > > > 
> > > > > > > > > > #define KVM_X86_FEATURE_FPU_ALIAS       __X86_FEATURE_8000_0001_ALIAS(KVM_X86_FEATURE_FPU)
> > > > > > > > > > #define KVM_X86_FEATURE_VME_ALIAS       __X86_FEATURE_8000_0001_ALIAS(KVM_X86_FEATURE_VME)
> > > > > > > > > > 
> > > > > > > > > > And then just use for example the 'F(FPU_ALIAS)' in the CPUID_8000_0001_EDX
> > > > > > > > 
> > > > > > > > At first glance, I really liked this idea, but after working through the
> > > > > > > > ramifications, I think I prefer "converting" the flag when passing it to
> > > > > > > > kvm_cpu_cap_init().  In-place conversion makes it all but impossible for KVM to
> > > > > > > > check the alias, e.g. via guest_cpu_cap_has(), especially since the AF() macro
> > > > > > > > doesn't set the bits in kvm_known_cpu_caps (if/when a non-hacky validation of
> > > > > > > > usage becomes reality).
> > > > > > 
> > > > > > Could you elaborate on this as well?
> > > > > > 
> > > > > > My suggestion was that we can just treat aliases as completely independent
> > > > > > and dummy features, say KVM_X86_FEATURE_FPU_ALIAS, and pass them as is to the
> > > > > > guest, which means that if an alias is present in host cpuid, it appears in
> > > > > > kvm caps, and thus qemu can then set it in guest cpuid.
> > > > > > 
> > > > > > I don't think that we need any special treatment for them if you look at it
> > > > > > this way.  If you don't agree, can you give me an example?
> > > > 
> > > > KVM doesn't honor the aliases beyond telling userspace they can be set (see below
> > > > for all the aliased features that KVM _should_ be checking).  The APM clearly
> > > > states that the features are the same as their CPUID.0x1 counterparts, but Intel
> > > > CPUs don't support the aliases.  So, as you also note below, I think we could
> > > > unequivocally say that enumerating the aliases but not the "real" features is a
> > > > bogus CPUID model, but we can't say the opposite, i.e. the real features can
> > > > exists without the aliases.
> > > > 
> > > > And that means that KVM must never query the aliases, e.g. should never do
> > > > guest_cpu_cap_has(KVM_X86_FEATURE_FPU_ALIAS), because the result is essentially
> > > > meaningless.  It's a small thing, but if KVM_X86_FEATURE_FPU_ALIAS simply doesn't
> > > > exist, i.e. we do in-place conversion, then it's impossible to feed the aliases
> > > > into things like guest_cpu_cap_has().
> > 
> > This only makes my case stronger - treating the aliases as just features will
> > allow us to avoid adding more logic to code which is already too complex IMHO.
> > 
> > If your concern is that features could be queried by guest_cpu_cap_has()
> > that is easy to fix, we can (and should) put them into a separate file and
> > #include them only in cpuid.c.
> > 
> > We can even #undef the __X86_FEATURE_8000_0001_ALIAS macro after the kvm_set_cpu_caps,
> > then if I understand the macro pre-processor correctly, any use of feature alias
> > macros will not fully evaluate and cause a compile error.
> 
> I don't see how that's less code.  Either way, KVM needs a macro to handle aliases,
> e.g. either we end up with ALIAS_F() or __X86_FEATURE_8000_0001_ALIAS().  For the
> macros themselves, IMO they carry the same amount of complexity.
> 
> If we go with ALIASED_F() (or ALIASED_8000_0001_F()), then that macro is all that
> is needed, and it's bulletproof.  E.g. there is no KVM_X86_FEATURE_FPU_ALIAS that
> can be queried, and thus no need to be ensure it's defined in cpuid.c and #undef'd
> after its use.
> 
> Hmm, I supposed we could harden the aliased feature usage in the same way as the
> ALIASED_F(), e.g.
> 
>   #define __X86_FEATURE_8000_0001_ALIAS(feature)				\
>   ({										\
> 	BUILD_BUG_ON(__feature_leaf(X86_FEATURE_##name) != CPUID_1_EDX);	\
> 	BUILD_BUG_ON(kvm_cpu_cap_init_in_progress != CPUID_8000_0001_EDX);	\
> 	(feature + (CPUID_8000_0001_EDX - CPUID_1_EDX) * 32);			\
>   })
> 
> If something tries to use an X86_FEATURE_*_ALIAS outside if kvm_cpu_cap_init(),
> it would need to define and set kvm_cpu_cap_init_in_progress, i.e. would really
> have to try to mess up.
> 
> Effectively the only differences are that KVM would have ~10 or so more lines of
> code to define the X86_FEATURE_*_ALIAS macros, and that the usage would look like:
> 
> 	VIRTUALIZED_F(FPU_ALIAS)
> 
> versus
> 
> 	ALIASED_F(FPU)


This is exactly my point. I want to avoid profiliation of the _F macros, because
later, we will need to figure out what each of them (e.g ALIASED_F) does.

A whole leaf alias, is once in x86 arch life misfeature, and it is very likely that
Intel/AMD won't add more such aliases.

Why VIRTUALIZED_F though, it wasn't in the patch series? Normal F() should be enough
IMHO.


> 
> At that point, I'm ok with defining each alias, though I honestly still don't
> understand the motivation for defining single-use macros.
> 

The idea is that nobody will need to look at these macros (e.g__X86_FEATURE_8000_0001_ALIAS() and its usages), 
because it's clear what they do, they just define few extra CPUID features 
that nobody really cares about.

ALIASED_F() on the other hand is yet another _F macro() and we will need,
once again and again to figure out why it is there, what it does, etc.

Best regards,
	Maxim Levitsky