Re: [PATCH qemu v14] spapr: Implement Open Firmware client interface

[David Gibson]

On Tue, Mar 02, 2021 at 10:37:59AM +0100, BALATON Zoltan wrote:
> On Tue, 2 Mar 2021, Alexey Kardashevskiy wrote:
> > On 02/03/2021 14:35, David Gibson wrote:
> > > Alexey or Zoltan, any thoughts on how non-PAPR versions of this would
> > > call into qemu to get the non-guest parts of VOF to execute?
> > 
> > Non-PAPR could do it as we do it for soft breakpoints in KVM - some
> > predefined illegal instruction which KVM knows that it is used for soft
> > breakpoints.
> 
> So far I've thought I'd need to implement TYPE_PPC_VIRTUAL_HYPERVISOR
> interface in the machine for the code in target/ppc/excp_helper.c to forward
> client syscalls to QEMU where I could handle the VOF related calls but not
> sure this is the best way.

I'm not sure that will work on its own.  VIRTUAL_HYPERVISOR just traps
the "sc 1" (hypecall instruction).  If that's an illegal instruction
for CPU, we can theoretically still use it, but I suspect TCG will
flag it as an illegal instruction and trap before we even get to the
VIRTUAL_HYPERVISOR dispatch point.  You'll need to investigate.

> This could be OK for the initial boot when
> nothing else should use syscalls yet but with RTAS this may not work as the
> guest OS could also use syscalls so to avoid conflicts we may need to shut
> down the virtual hypervisor on quiesce

Uh.. no.  VIRTUAL_HYPERVISOR doesn't intercept normal system calls,
only "level 1" system calls which are explicitly designated for
hypercalls.

> which means I may need a minimal
> guest only rtas for pegasos2 (which would be OK as I think it's only used
> for shutdown/reboot anyway). Also may need some changes to allow empty
> callbacks in vhyp to be ignored when I only want to implement hypercall
> method but that's just adding checks to only call non-NULL callbacks in
> PPCVirtualHypervisorClass.
> 
> There's also an old patch from Benjamin Herrenschmidt to add MOL OSI which
> is a similar hypercall interface:
> https://github.com/ozbenh/qemu/commit/6dc8803641e323030ffd01ad8ce0dcf081896698
> This might also be useful later to use MOL paravirtual drivers to speed up
> MacOSX emulation. but I haven't looked at the details yet.
> 
> Any other ideas?
> 
> Regards,
> BALATON Zoltan
> 

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

signature.asc
Description: PGP signature