On Thu, 2024-04-25 at 07:42 -0700, Sean Christopherson wrote:
> On Thu, Apr 25, 2024, Kai Huang wrote:
> > On Wed, 2024-04-24 at 13:06 -0700, Sean Christopherson wrote:
> > > > > static inline u32 vmx_basic_vmcs_mem_type(u64 vmx_basic)
> > > > > {
> > > > > return (vmx_basic & GENMASK_ULL(53, 50)) >>
> > > > > VMX_BASIC_MEM_TYPE_SHIFT;
> > > > > }
> > > > >
> > > > > looks not intuitive than original patch.
> > > >
> > > > Yeah, agreed, that's taking the worst of both worlds. I'll update patch 5 to drop
> > > > VMX_BASIC_MEM_TYPE_SHIFT when effectively "moving" it into vmx_basic_vmcs_mem_type().
> > >
> > > Drat. Finally getting back to this, dropping VMX_BASIC_MEM_TYPE_SHIFT doesn't
> > > work because it's used by nested_vmx_setup_basic(), as is VMX_BASIC_VMCS_SIZE_SHIFT,
> > > which is presumably why past me kept them around.
> > >
> > > I'm leaning towards keeping things as proposed in this series. I don't see us
> > > gaining a third copy, or even a third user, i.e. I don't think we are creating a
> > > future problem by open coding the shift in vmx_basic_vmcs_mem_type(). And IMO
> > > code like this
> > >
> > > return (vmx_basic & VMX_BASIC_MEM_TYPE_MASK) >>
> > > VMX_BASIC_MEM_TYPE_SHIFT;
> > >
> > > is an unnecessary obfuscation when there is literally one user (the accessor).
> > >
> > > Another idea would be to delete VMX_BASIC_MEM_TYPE_SHIFT and VMX_BASIC_VMCS_SIZE_SHIFT,
> > > and either open code the values or use local const variables, but that also seems
> > > like a net negative, e.g. splits the effective definitions over too many locations.
> >
> > Alternatively, we can add macros like below to <asm/vmx.h> close to
> > vmx_basic_vmcs_size() etc, so it's straightforward to see.
> >
> > +#define VMX_BSAIC_VMCS12_SIZE ((u64)VMCS12_SIZE << 32)
> > +#define VMX_BASIC_MEM_TYPE_WB (MEM_TYPE_WB << 50)
>
> Hmm, it's a bit hard to see it's specifically VMCS12 size, and given that prior
> to this series, VMX_BASIC_MEM_TYPE_WB = 6, I'm hesitant to re-introduce/redefine
> that macro with a different value.
>
> What if we add a helper in vmx.h to encode the VMCS info? Then the #defines for
> the shifts can go away because the open coded shifts are colocated and more
> obviously related. E.g.
>
> static inline u64 vmx_basic_encode_vmcs_info(u32 revision, u16 size, u8 memtype)
> {
> return revision | ((u64)size << 32) | ((u64)memtype << 50);
> }
>
>
> and
>
> static void nested_vmx_setup_basic(struct nested_vmx_msrs *msrs)
> {
> /*
> * This MSR reports some information about VMX support. We
> * should return information about the VMX we emulate for the
> * guest, and the VMCS structure we give it - not about the
> * VMX support of the underlying hardware.
> */
> msrs->basic = vmx_basic_encode_vmcs_info(VMCS12_REVISION, VMCS12_SIZE,
> X86_MEMTYPE_WB);
>
> msrs->basic |= VMX_BASIC_TRUE_CTLS
> if (cpu_has_vmx_basic_inout())
> msrs->basic |= VMX_BASIC_INOUT;
> }
Yeah this is better. Thanks.
