Re: annual howto rampage

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Fri, Jul 12, 2024 at 09:52:53AM -0400, Greg Troxel wrote:
> Thanks for the detailed reply to my queries.  I have applied your
> comments to the howto, and when that was straightforward I have pruned
> the reply below.
> 
> Manuel Bouyer <bouyer%antioche.eu.org@localhost> writes:
> 
> >>   - Is the problem with i386 PV guests under 4.18 about all PV, or just
> >>     NetBSD?
> >
> > What problem are you talking about ?
> > It this is the requirement to use pvhsim, it's for all i386 PV guests
> 
> I meant "i386 pv guest does not work".  Ack, and edited.
> 
> >>   - Is it true that PVH and PVHVM both also require VT-X/SVM just like HVM?
> >
> > Yes, the requirements are the same as HVM
> 
> I have a machine without VT-X (yes, I know it's from 2010):
> 
>   cpu0: "Pentium(R) Dual-Core  CPU      E5700  @ 3.00GHz"
>   cpu0: Intel Xeon 31xx, 33xx, 52xx, 54xx, Core 2 Quad 8xxx and 9xxx (686-class)
>   cpu0: family 0x6 model 0x17 stepping 0xa (id 0x1067a)
>   cpu0: features 0x1fc9cbf5<FPU,DE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MCA,CMOV,PAT>
>   cpu0: features 0x1fc9cbf5<CLFSH,ACPI,MMX,FXSR,SSE,SSE2,SS,HTT>
>   cpu0: features1 0x4002201<SSE3,SSSE3,CX16,XSAVE>
>   cpu0: features2 0x20100800<SYSCALL/SYSRET,XD,EM64T>
>   cpu0: features3 0x1<LAHF>
> 
> It runs PV guests fine.  Interestingly, it runs a PV netbsd-9 i386 guest
> as pvh/pvshim apparently fine.  But flipping a netbsd-9 amd64 guest to
> pvh/pvhsim (from just pv) makes it at least 100x slow (guessing, but it
> goes from normal to "did I get teleported back to a VAX 11/750 with 1 MB
> of RAM that is paging constantly")?
> 
> I wonder what's going on with i386 pvshim, if it somehow doesn't provoke
> the slowness.
> 
> I wonder then if trying to run pvh/hvm/pvhvm "needing" VT-X is that it
> won't work at all, or that it is so painful that nobody would ever do
> it.

Well, AFAIK PVH uses the HVM hardware extention to virtualise the pagetables.
I don't have x86 host without VT-X anymore so I can't really test this.


> 
> Regarding your tests:
> 
>   I don't see PVHVM mode.  Is that because it doesn't work, or because
>   you just don't, or ?

HVM tests are really PVHVM, as GENERIC includes the PV drivers and there's no
real way to disable the PV drivers at the xl level (AFAIK)

> 
>   The test code to run various modes is great howto fodder.  Is that in
>   a repo I can point to?

No it's just a script calling anita. I can share it if it's usefull.

> I see I can find the anita output when it
>   fails from the log, but if it works, there is only the ATF output.  It
>   would be nice to link anita output always.

I don't keep it the text succeeds. I'll have to check what 
happends after anita returns (I don't remember)

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--