Re: [PATCH 2/5] s390x/pv: Implement CGS check handler

[Claudio Imbrenda <imbrenda@linux.ibm.com>]

On Thu, 5 Jan 2023 12:42:54 +0100
Thomas Huth <thuth@redhat.com> wrote:

> On 04/01/2023 12.51, Cédric Le Goater wrote:
> > From: Cédric Le Goater <clg@redhat.com>
> > 
> > When a protected VM is started with the maximum number of CPUs (248),
> > the service call providing information on the CPUs requires more
> > buffer space than allocated and QEMU disgracefully aborts :
> > 
> >      LOADPARM=[........]
> >      Using virtio-blk.
> >      Using SCSI scheme.
> >      ...................................................................................
> >      qemu-system-s390x: KVM_S390_MEM_OP failed: Argument list too long
> > 
> > Implement a test for this limitation in the ConfidentialGuestSupportClass
> > check handler and provide some valid information to the user before the
> > machine starts.
> > 
> > Signed-off-by: Cédric Le Goater <clg@redhat.com>
> > ---
> >   hw/s390x/pv.c | 23 +++++++++++++++++++++++
> >   1 file changed, 23 insertions(+)
> > 
> > diff --git a/hw/s390x/pv.c b/hw/s390x/pv.c
> > index 8dfe92d8df..3a7ec70634 100644
> > --- a/hw/s390x/pv.c
> > +++ b/hw/s390x/pv.c
> > @@ -266,6 +266,26 @@ int s390_pv_kvm_init(ConfidentialGuestSupport *cgs, Error **errp)
> >       return 0;
> >   }
> >   
> > +static bool s390_pv_check_cpus(Error **errp)
> > +{
> > +    MachineState *ms = MACHINE(qdev_get_machine());
> > +    MachineClass *mc = MACHINE_GET_CLASS(ms);
> > +    uint32_t pv_max_cpus = mc->max_cpus - 1;  
> 
> Not sure whether "mc->max_cpus - 1" is the right approach here. I think it 
> would be better to calculate the amount of CPUs that we can support.
> 
> So AFAIK the problem is that SCLP information that is gathered during 
> read_SCP_info() in hw/s390x/sclp.c. If protected virtualization is enabled, 
> everything has to fit in one page (i.e. 4096 bytes) there.
> 
> So we have space for
> 
>   (TARGET_PAGE_SIZE - offset_cpu) / sizeof(CPUEntry)
> 
> CPUs.
> 
> With S390_FEAT_EXTENDED_LENGTH_SCCB enabled, offset_cpu is 144 (see struct 
> ReadInfo in sclp.h), otherwise it is 128.
> 
> That means, with S390_FEAT_EXTENDED_LENGTH_SCCB we can have a maximum of:
> 
>   (4096 - 144) / 16 = 247 CPUs
> 
> which is what you were trying to check with the mc->max_cpus - 1 here.
> 
> But with "-cpu els=off", it sounds like we could fit all 248 also with 
> protected VMs? Could you please give it a try?
> 
> Anyway, instead of using "pv_max_cpus = mc->max_cpus - 1" I'd suggest to use 
> something like this instead:
> 
>   int offset_cpu = s390_has_feat(S390_FEAT_EXTENDED_LENGTH_SCCB) ?
>                       offsetof(ReadInfo, entries) :
>                       SCLP_READ_SCP_INFO_FIXED_CPU_OFFSET;
>   pv_max_cpus = (TARGET_PAGE_SIZE - offset_cpu) /sizeof(CPUEntry);

I agree with Thomas here

> 
>    Thomas
> 
>