On 4/29/22 03:04, Sean Christopherson wrote:
> Clean up KVM's struct page / pfn helpers to reduce the number of
> pfn_to_page() and page_to_pfn() conversions. E.g. kvm_release_pfn_dirty()
> makes 6 (if I counted right) calls to pfn_to_page() when releasing a dirty
> pfn that backed by a vanilla struct page. That is easily trimmed down to
> a single call.
>
> And perhaps more importantly, rename and refactor kvm_is_reserved_pfn() to
> try and better reflect what it actually queries, which at this point is
> effectively whether or not the pfn is backed by a refcounted page.
>
> Sean Christopherson (10):
> KVM: Do not zero initialize 'pfn' in hva_to_pfn()
> KVM: Drop bogus "pfn != 0" guard from kvm_release_pfn()
> KVM: Don't set Accessed/Dirty bits for ZERO_PAGE
> KVM: Avoid pfn_to_page() and vice versa when releasing pages
> KVM: nVMX: Use kvm_vcpu_map() to get/pin vmcs12's APIC-access page
> KVM: Don't WARN if kvm_pfn_to_page() encounters a "reserved" pfn
> KVM: Remove kvm_vcpu_gfn_to_page() and kvm_vcpu_gpa_to_page()
> KVM: Take a 'struct page', not a pfn in kvm_is_zone_device_page()
> KVM: Rename/refactor kvm_is_reserved_pfn() to
> kvm_pfn_to_refcounted_page()
> KVM: x86/mmu: Shove refcounted page dependency into
> host_pfn_mapping_level()
>
> arch/x86/kvm/mmu/mmu.c | 26 +++++--
> arch/x86/kvm/mmu/tdp_mmu.c | 3 +-
> arch/x86/kvm/vmx/nested.c | 39 ++++-------
> arch/x86/kvm/vmx/vmx.h | 2 +-
> include/linux/kvm_host.h | 12 +---
> virt/kvm/kvm_main.c | 140 +++++++++++++++++++++++++------------
> 6 files changed, 131 insertions(+), 91 deletions(-)
>
>
> base-commit: 2a39d8b39bffdaf1a4223d0d22f07baee154c8f3
Queued, thanks.
Paolo