On Tue, Sep 13, 2022, Alexey Kardashevskiy wrote:
> On 12/9/22 19:36, Sean Christopherson wrote:
> > On Mon, Sep 12, 2022, Alexey Kardashevskiy wrote:
> > > A recent renaming patch missed 1 spot, fix it.
> > >
> > > This should cause no behavioural change.
> > >
> > > Fixes: 23e5092b6e2a ("KVM: SVM: Rename hook implementations to conform to kvm_x86_ops' names")
> > > Signed-off-by: Alexey Kardashevskiy <aik@xxxxxxx>
> > > ---
> > > arch/x86/kvm/svm/sev.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
> > > index 28064060413a..3b99a690b60d 100644
> > > --- a/arch/x86/kvm/svm/sev.c
> > > +++ b/arch/x86/kvm/svm/sev.c
> > > @@ -3015,7 +3015,7 @@ void sev_es_prepare_switch_to_guest(struct sev_es_save_area *hostsa)
> > > /*
> > > * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
> > > * of which one step is to perform a VMLOAD. KVM performs the
> > > - * corresponding VMSAVE in svm_prepare_guest_switch for both
> > > + * corresponding VMSAVE in svm_prepare_switch_to_guest for both
> > > * traditional and SEV-ES guests.
> > > */
> >
> > Rather than match the rename, what about tweaking the wording to not tie the comment
> > to the function name, e.g. "VMSAVE in common SVM code".
>
> Although I kinda like the pointer to the caller, it is not that useful with
> a single caller and working cscope :)
Yeah, having exact function names is nice, but we always seem to end up with goofs
like this where a comment gets left behind and then they become stale and confusing.
> > Even better, This would be a good opportunity to reword this comment to make it more
> > clear why SEV-ES needs a hook, and to absorb the somewhat useless comments below.
> >
> > Would something like this be accurate? Please modify and/or add details as necessary.
> >
> > diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
> > index 3b99a690b60d..c50c6851aedb 100644
> > --- a/arch/x86/kvm/svm/sev.c
> > +++ b/arch/x86/kvm/svm/sev.c
> > @@ -3013,19 +3013,14 @@ void sev_es_vcpu_reset(struct vcpu_svm *svm)
> > void sev_es_prepare_switch_to_guest(struct sev_es_save_area *hostsa)
> > {
> > /*
> > - * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
> > - * of which one step is to perform a VMLOAD. KVM performs the
> > - * corresponding VMSAVE in svm_prepare_switch_to_guest for both
> > - * traditional and SEV-ES guests.
> > + * Manually save host state that is automatically loaded by hardware on
> > + * VM-Exit from SEV-ES guests, but that is not saved by VMSAVE (which is
> > + * performed by common SVM code). Hardware unconditionally restores
> > + * host state, and so KVM skips manually restoring this state in common
> > + * code.
>
> I am new to this arch so not sure :) The AMD spec calls these three "Type B
> swaps" from the VMSA's "Table B-3. Swap Types" so may be just say:
>
> ===
> These are Type B swaps which are not saved by VMSAVE (performed by common
> SVM code) but restored by VMEXIT unconditionally.
Weird consistency nit: KVM refers to VM-Exit as an event and not a thing/action,
whereas the APM tends to refer to VMEXIT as a thing, thus the "on VM-Exit" stylization
versus "by VMEXIT". Similarly, when talking about the broader event of entering
the guest, KVM uses "VM-Enter".
VMRUN and VMSAVE on the other hand are instructions and so are "things" in KVM's world.
Using the VM-Enter/VM-Exit terminology consistently throughout KVM x86 makes it easy
to talk about KVM x86 behavior that is common to both SVM and VMX without getting
tripped up on naming differences between the two. So even though it's a little odd
odd when looking only at SVM code, using "on VM-Exit" instead of "by VMEXIT" is
preferred.
> ===
I want to avoid relying on the APM's arbitrary "Type B" classification. Having to
dig up and look at a manual to understand something that's conceptually quite simple
is frustrating. Providing references to "Type B" and the table in the changelog is
definitely welcome, e.g. so that someone who wants more details/background can easily
find that info via via git blame. But for a comment, providing all the information
in the comment itself is usually preferable.
How about this?
Save state that is loaded unconditionally by hardware on VM-Exit for SEV-ES
guests, but is not saved via VMRUN or VMSAVE (performed by common SVM code).
