----- Original Message ----- > From: "Wanpeng Li" <kernellwp@xxxxxxxxx> > To: "Sean J Christopherson" <sean.j.christopherson@xxxxxxxxx> > Cc: "kvm" <kvm@xxxxxxxxxxxxxxx>, "Paolo Bonzini" <pbonzini@xxxxxxxxxx>, "Radim Krcmar" <rkrcmar@xxxxxxxxxx> > Sent: Monday, March 12, 2018 10:44:17 PM > Subject: Re: [PATCH 0/6] KVM: VMX: improve usage of unrestricted guest > > 2018-03-13 2:28 GMT+08:00 Christopherson, Sean J > <sean.j.christopherson@xxxxxxxxx>: > > On Sunday, 2018-03-11, Wanpeng Li <kernellwp@xxxxxxxxx> wrote: > >> 2018-03-06 4:04 GMT+08:00 Sean Christopherson > >> <sean.j.christopherson@xxxxxxxxx>: > >> > This series eliminates usage of legacy code that is obsolete when > >> > unrestricted guest is enabled. > >> > > >> > Because unrestricted guest (Westmere) was introduced one generation > >> > later than EPT (Nehalem), KVM's initial support for EPT by necessity > >> > retained/introduced soon-to-be-legacy code for handling a guest with > >> > EPT enabled and CR0.PG/PE==0. Some of this now-legacy code/behavior > >> > wasn't correctly deprecated when KVM added support for unrestricted > >> > guest, e.g. KVM still creates and loads identity mapped IA32 page > >> > tables when the guest disables paging, even though CR0.PG is cleared > >> > in hardware. > >> > >> Thanks for the patchset, in addition, I think A20 emulation is still > >> missing for vm8086. > > > > Missing in the sense that you want to be able to emulate A20M behavior > > in KVM, i.e. run a guest with A20M asserted? > > I mean the guest maybe access the address > 1MB in vm8086 and the kvm > doesn't support address-wraparound that occurs at 1MByte. A20 is gone in processors newer than Haswell, KVM need not emulate it. Paolo
