Re: [RESEND PATCH v10 04/10] mmu: spp: Add functions to operate SPP access bitmap

[Yang Weijiang]

On Fri, Jan 10, 2020 at 09:38:04AM -0800, Sean Christopherson wrote:
> On Thu, Jan 02, 2020 at 02:13:13PM +0800, Yang Weijiang wrote:
> > Create access bitmap for SPP subpages, the bitmap can
> > be accessed with a gfn. The initial access bitmap for each
> > physical page is 0xFFFFFFFF, meaning SPP is not enabled for the
> > subpages.
> 
> Wrap changelogs at ~75 chars.
> 
> Create access bitmap for SPP subpages, the bitmap can be accessed with a
> gfn.  The initial access bitmap for each physical page is 0xFFFFFFFF,
> meaning SPP is not enabled for the subpages.
> 
> There needs to be a *lot* more information provided in all of the changelogs
> for this series.  I understand the basic concepts of SPP, but nothing in the
> documentation or changelogs explains how KVM generates the SPP tables based
> on userspace input.  Essentially, explain the design in decent detail, with
> a focus on *why* KVM does what it does.
>
OK, will modify the documentation to add SPPT setup section, thanks!

> > +static int kvm_spp_level_pages(gfn_t gfn_lower, gfn_t gfn_upper, int level)
> > +{
> > +	int page_num = KVM_PAGES_PER_HPAGE(level);
> > +	gfn_t gfn_max = (gfn_lower & ~(page_num - 1)) + page_num - 1;
> > +	int ret;
> > +
> > +	if (gfn_upper <= gfn_max)
> > +		ret = gfn_upper - gfn_lower + 1;
> > +	else
> > +		ret = gfn_max - gfn_lower + 1;
> > +
> > +	return ret;
> > +}
> > +
> >  #define SPPT_ENTRY_PHA_MASK (0xFFFFFFFFFF << 12)
> 
> There's almost certainly an existing macro for this.
>
Sure, will remove it.
> >  
> >  int kvm_spp_setup_structure(struct kvm_vcpu *vcpu,
> > @@ -220,6 +249,309 @@ int kvm_spp_setup_structure(struct kvm_vcpu *vcpu,
> >  }
> >  EXPORT_SYMBOL_GPL(kvm_spp_setup_structure);
> >  
> > +int vmx_spp_flush_sppt(struct kvm *kvm, u64 gfn_base, u32 npages)
> > +{
> > +	struct kvm_shadow_walk_iterator iter;
> > +	struct kvm_vcpu *vcpu;
> > +	gfn_t gfn = gfn_base;
> > +	gfn_t gfn_max = gfn_base + npages - 1;
> 
> s/gfn_max/gfn_end.  "max" makes me think this is literally walking every
> possible gfn.
> 
Make sense, will change it.

> > +	u64 spde;
> > +	int count;
> > +	bool flush = false;
> > +
> >  /*
> >   * The bit 0 ~ bit 15 of kvm_memory_region::flags are visible for userspace,
> >   * other bits are reserved for kvm internal use which are defined in
> > -- 
> > 2.17.2
> >