Re: [RFC PATCH-for-5.2] hw/s390x/pci: Fix endianness issue

[Cornelia Huck <cohuck@redhat.com>]

On Tue, 17 Nov 2020 14:12:00 +0100
Philippe Mathieu-Daudé <philmd@redhat.com> wrote:

> On 11/17/20 2:00 PM, Peter Maydell wrote:
> > On Tue, 17 Nov 2020 at 12:03, Philippe Mathieu-Daudé <philmd@redhat.com> wrote:  
> >>
> >> Fix an endianness issue reported by Cornelia:
> >>  
> >>> s390x tcg guest on x86, virtio-pci devices are not detected. The
> >>> relevant feature bits are visible to the guest. Same breakage with
> >>> different guest kernels.
> >>> KVM guests and s390x tcg guests on s390x are fine.  
> >>
> >> Fixes: 28dc86a0729 ("s390x/pci: use a PCI Group structure")
> >> Reported-by: Cornelia Huck <cohuck@redhat.com>
> >> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
> >> ---
> >> RFC because review-only patch, untested
> >> ---
> >>  hw/s390x/s390-pci-inst.c | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/hw/s390x/s390-pci-inst.c b/hw/s390x/s390-pci-inst.c
> >> index 58cd041d17f..cfb54b4d8ec 100644
> >> --- a/hw/s390x/s390-pci-inst.c
> >> +++ b/hw/s390x/s390-pci-inst.c
> >> @@ -305,7 +305,7 @@ int clp_service_call(S390CPU *cpu, uint8_t r2, uintptr_t ra)
> >>          ClpReqQueryPciGrp *reqgrp = (ClpReqQueryPciGrp *)reqh;
> >>          S390PCIGroup *group;
> >>
> >> -        group = s390_group_find(reqgrp->g);
> >> +        group = s390_group_find(ldl_p(&reqgrp->g));  
> > 
> > 'g' in the ClpReqQueryPciGrp struct is a uint32_t, so
> > adding the ldl_p() will have no effect unless (a) the
> > structure is not 4-aligned and (b) the host will fault on
> > unaligned accesses, which isn't the case for x86 hosts.
> > 
> > Q: is this struct really in host order, or should we
> > be using ldl_le_p() or ldl_be_p() and friends here and
> > elsewhere?  
> 
> Now than I had lunch, your comment is obvious...
> 
> IIUC we should use ldl_le_p() here and fix all the other
> uses.

I'm not sure whether it shouldn't be be -- waiting for the IBM folks to
comment on the expected layout.

> 
> > 
> > thanks
> > -- PMM
> >   
>