Re: [PATCH RFCv2 2/4] i386/pc: relocate 4g start to 1T where applicable

[Dr. David Alan Gilbert]

* Igor Mammedov (imammedo@redhat.com) wrote:
> On Tue, 22 Feb 2022 10:42:55 +0100
> Gerd Hoffmann <kraxel@redhat.com> wrote:
> 
> >   Hi,
> > 
> > > > And the upstream code is now pretty much identical except for the
> > > > default;  note that for TCG you do need to keep to 40 I think.  
> > > 
> > > will TCG work with 40bits on host that supports less than that?  
> > 
> > When I understand things correctly the problem is that the phys-bits
> > limit applies to the npt/ept tables too, effectively restricting guest
> > physical address space to host physical address space.
> > 
> > TCG is not affected by that and should work just fine.
> > 
> > Not sure what happens if you turn off npt/ept and run on softmmu.
> > Possibly that works fine too.
> > 
> > > Also quick look at host-phys-bits shows that it affects only 'host'
> > > cpu model and is NOP for all other models.  
> > 
> > I don't think so.  microvm forces host-phys-bits=on and that works with
> > all cpu models.
> 
> I just don't see how host-phys-bits can work for other than 'host' cpu model.
> It's true that property is available for all cpu models, but the field it sets
> is only used in target/i386/host-cpu.c, the same applies to 
> host-phys-bits-limit.
> Am I missing something?

The hook in kvm/kvm-cpu.c kvm_cpu_realizefn:

    /*
     * The realize order is important, since x86_cpu_realize() checks if
     * nothing else has been set by the user (or by accelerators) in
     * cpu->ucode_rev and cpu->phys_bits, and updates the CPUID results in
     * mwait.ecx.
     * This accel realization code also assumes cpu features are already 
expanded.
     *
     * realize order:
     *
     * x86_cpu_realize():
     *  -> x86_cpu_expand_features()
     *  -> cpu_exec_realizefn():
     *            -> accel_cpu_realizefn()
     *               kvm_cpu_realizefn() -> host_cpu_realizefn()
     *  -> check/update ucode_rev, phys_bits, mwait
     */

Dave

> > 
> > take care,
> >   Gerd
> > 
> 
-- 
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK