Re: [RFC PATCH v3 04/19] KVM: x86: mmu: allow to enable write tracking externally

[Sean Christopherson]

On Wed, Jul 20, 2022, Maxim Levitsky wrote:
> On Sun, 2022-05-22 at 13:22 +0300, Maxim Levitsky wrote:
> > On Thu, 2022-05-19 at 16:37 +0000, Sean Christopherson wrote:
> > > On Wed, Apr 27, 2022, Maxim Levitsky wrote:
> > > > @@ -5753,6 +5752,10 @@ int kvm_mmu_init_vm(struct kvm *kvm)
> Now for nested AVIC, this is what I would like to do:
>  
> - just like mmu, I prefer to register the write tracking notifier, when the
>   VM is created.
>
> - just like mmu, write tracking should only be enabled when nested AVIC is
>   actually used first time, so that write tracking is not always enabled when
>   you just boot a VM with nested avic supported, since the VM might not use
>   nested at all.
>  
> Thus I either need to use the __kvm_page_track_register_notifier too for AVIC
> (and thus need to export it) or I need to have a boolean
> (nested_avic_was_used_once) and register the write tracking notifier only
> when false and do it not on VM creation but on first attempt to use nested
> AVIC.
>  
> Do you think this is worth it? I mean there is some value of registering the
> notifier only when needed (this way it is not called for nothing) but it does
> complicate things a bit.

Compared to everything else that you're doing in the nested AVIC code, refcounting
the shared kvm_page_track_notifier_node object is a trivial amount of complexity.

And on that topic, do you have performance numbers to justify using a single
shared node?  E.g. if every table instance has its own notifier, then no additional
refcounting is needed.  It's not obvious that a shared node will provide better
performance, e.g. if there are only a handful of AVIC tables being shadowed, then
a linear walk of all nodes is likely fast enough, and doesn't bring the risk of
a write potentially being stalled due to having to acquire a VM-scoped mutex.

> I can also stash this boolean (like 'bool registered;') into the 'struct
> kvm_page_track_notifier_node',  and thus allow the
> kvm_page_track_register_notifier to be called more that once -  then I can
> also get rid of __kvm_page_track_register_notifier. 

No, allowing redundant registration without proper refcounting leads to pain,
e.g. X registers, Y registers, X unregisters, kaboom.