>-----Original Message----- >From: Paolo Bonzini <pbonzini@xxxxxxxxxx> >Sent: Thursday, April 06, 2023 15:33 >To: Yao, Yuan <yuan.yao@xxxxxxxxx>; Christopherson,, Sean <seanjc@xxxxxxxxxx>; Li, Xin3 <xin3.li@xxxxxxxxx> >Cc: kvm@xxxxxxxxxxxxxxx; Li, Xiaoyao <xiaoyao.li@xxxxxxxxx>; Dong, Eddie <eddie.dong@xxxxxxxxx>; Tian, Kevin ><kevin.tian@xxxxxxxxx>; Nakajima, Jun <jun.nakajima@xxxxxxxxx>; H.Peter Anvin <hpa@xxxxxxxxx>; Mallick, Asit K ><asit.k.mallick@xxxxxxxxx> >Subject: Re: The necessity of injecting a hardware exception reported in VMX IDT vectoring information > >On 4/6/23 08:37, Yao, Yuan wrote: >> It's definitely broken for nested case if the exception Is injected >> by L1 in the first place, but if it's injected after interception (by >> L1) for same exception, and it's trap, it can be regenerated by > ^^^^ > >if it's a fault Yes, my typo, I was thinking fault but wrote trap ... > >> re-execute the L2 code. > >You cannot know why L1 injected an exception. For example L1 could have >injected a MCE just to test the code in L1. > >This is a scary change, in a scary area of code, with unclear benefits. >It's going to be hard to convince people. :) Yes I understood, thanks Paolo. > >Paolo
