On Tue, 2022-04-26 at 18:47 +0200, Paolo Bonzini wrote: > On 4/26/22 18:43, Maxim Levitsky wrote: > > Actually for vmrun, that #GP is I think sort of correct now - that is > > what AMD cpus do on 'invalid physical address', but for VM exit, we > > just need to have the vmcb mapped instead of mapping it again - > > injecting #GP at that point which will go to the nested guest is just > > wrong. > > Yeah, in that case I can't think of anything better than triple fault. > > Paolo > But do you think that it would be better to keep the vmcb12 page mapped/pinned while doing the nested run which will both solve the problem of calling sleeping function and allow us to remove the case of map failing on vm exit? If I remember correctly, vmx does something like that. Best regards, Maxim Levitsky
