Re: [Qemu-ppc] [Qemu-devel] [PULL 57/60] target/ppc: add HV support for POWER9

[David Gibson]

On Tue, Mar 12, 2019 at 11:01:15AM -0400, Cleber Rosa wrote:
> On Sun, Mar 10, 2019 at 07:27:00PM +1100, David Gibson wrote:
> > From: Cédric Le Goater <address@hidden>
> > 
> > We now have enough support to boot a PowerNV machine with a POWER9
> > processor. Allow HV mode on POWER9.
> > 
> > Signed-off-by: Cédric Le Goater <address@hidden>
> > Message-Id: <address@hidden>
> > Signed-off-by: David Gibson <address@hidden>
> > ---
> >  target/ppc/translate_init.inc.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/target/ppc/translate_init.inc.c 
> > b/target/ppc/translate_init.inc.c
> > index af70a3b78c..0bd555eb19 100644
> > --- a/target/ppc/translate_init.inc.c
> > +++ b/target/ppc/translate_init.inc.c
> > @@ -8895,7 +8895,7 @@ POWERPC_FAMILY(POWER9)(ObjectClass *oc, void *data)
> >                         PPC_CACHE | PPC_CACHE_ICBI | PPC_CACHE_DCBZ |
> >                         PPC_MEM_SYNC | PPC_MEM_EIEIO |
> >                         PPC_MEM_TLBSYNC |
> > -                       PPC_64B | PPC_64BX | PPC_ALTIVEC |
> > +                       PPC_64B | PPC_64H | PPC_64BX | PPC_ALTIVEC |
> >                         PPC_SEGMENT_64B | PPC_SLBI |
> >                         PPC_POPCNTB | PPC_POPCNTWD |
> >                         PPC_CILDST;
> > @@ -8907,6 +8907,7 @@ POWERPC_FAMILY(POWER9)(ObjectClass *oc, void *data)
> >                          PPC2_ISA205 | PPC2_ISA207S | PPC2_FP_CVT_S64 |
> >                          PPC2_TM | PPC2_ISA300 | PPC2_PRCNTL;
> >      pcc->msr_mask = (1ull << MSR_SF) |
> > +                    (1ull << MSR_SHV) |
> >                      (1ull << MSR_TM) |
> >                      (1ull << MSR_VR) |
> >                      (1ull << MSR_VSX) |
> 
> This change prevents a Fedora 29 kernel[1] from booting... is this
> intended or a known limitation of the Fedora 29 kernel?
> 
> FIY, by using a "power8" CPU the Fedora 29 kernel boots successfully.
> For a reproducer, please refer to [2].

Ouch.  I don't know how, but this has introduced a nasty regression.
I've reproduced it myself, here's what I've observed so far.  Nothing
very surprising, but helps narrow it down a bit.

 * It works with KVM
 * It works with -cpu power8
 * It also fails with the current RHEL8 kernel
 * It also fails on a ppc64le host (using TCG)
 * Also fails with a fresh build upstream kernel

It's very strange because a) the flags it sets are already enabled for
power8, which seems to work and b) literally the only thing that
PPC_64H seems to control is presence of the hrfid instruction, which
the guest kernel shouldn't be even attempting to use.

It seems to be hitting an explicit panic() in the guest code (no
message because it's before we have a working console).

Continuing to investigate...

-- 
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