On 06/28/2017 09:08 PM, David Hildenbrand wrote: > On 28.06.2017 20:59, Christian Borntraeger wrote: >> On 06/28/2017 08:06 PM, David Hildenbrand wrote: >>> On 28.06.2017 19:30, Christian Borntraeger wrote: >>>> From: QingFeng Hao <haoqf@xxxxxxxxxxxxxxxxxx> >>>> >>>> With vsie feature enabled, kvm can support nested guests (guest-3). >>>> So inject machine check to the guest-2 if it happens when the nested >>>> guest is running. And guest-2 will detect the machine check belongs >>>> to guest-3 and reinject it into guest-3. >>>> The host (guest-1) tries to inject the machine check to the picked >>>> destination vcpu if it's a floating machine check. >>> >>> The subject is confusing. We don't inject anything into the nested guest >>> here. We just catch machine checks during vsie and inject it into the >>> ordinary kvm guest. >> >> Agreed, it is confusing and maybe a leftover from an early rework due to internal >> feedback. We inject in guest-2 and rely on guest-2 to reinject to guest-3 would >> be a better wording. >> >> >> >> >>> Are there any SIE specific things to consider here, that may have to be >>> translated? >> >> As HW exits SIE before delivering the machine check, the SIE control block >> contains all saved guest3 registers and the host (guest1) registers contain >> the lazy registers (as we have already restored them) just like a normal exit. > > As mentioned in the other mail, e.g. vector register validity should > only be set if vector registers are enabled for the nested guest > (execution control enabled). Yes, but sInce we inject in the base guest (non-nested) we have to check the base guest execution control. > >>> >>>> + sie_page = container_of(scb_s, struct sie_page, sie_block); >>>> + mcck_info = &sie_page->mcck_info; >>>> + kvm_s390_reinject_machine_check(vcpu, mcck_info); >>> >>> This could be a simple >>> >>> kvm_s390_reinject_machine_check(vcpu, &vsie_page->mcck_info); >>> >>> no? >> >> Yes that would be simpler, I guess. Looks like is just a "do it like the low >> level handler". The code in nmi.c has to go back from the sie control block into >> SIE page. >> >> Do you want a respin of this patch? > > You can also send a fixup if you don't have to respin. > >
