Re: [PATCH] pc: q35: Bump max_cpus to 512

[Igor Mammedov]

On Mon, 9 May 2022 13:03:38 +0100
Daniel P. Berrangé <berrange@redhat.com> wrote:

> On Mon, May 09, 2022 at 09:12:49AM +0200, Igor Mammedov wrote:
> > On Wed, 4 May 2022 08:16:39 -0500
> > Suravee Suthikulpanit <suravee.suthikulpanit@amd.com> wrote:
> >   
> > > This is the maximum number of vCPU supported by
> > > the AMD x2APIC virtualization.
> > > 
> > > Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
> > > ---
> > >  hw/i386/pc_q35.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/hw/i386/pc_q35.c b/hw/i386/pc_q35.c
> > > index 302288342a..e82b1c690d 100644
> > > --- a/hw/i386/pc_q35.c
> > > +++ b/hw/i386/pc_q35.c
> > > @@ -357,7 +357,7 @@ static void pc_q35_machine_options(MachineClass *m)
> > >      machine_class_allow_dynamic_sysbus_dev(m, TYPE_INTEL_IOMMU_DEVICE);
> > >      machine_class_allow_dynamic_sysbus_dev(m, TYPE_RAMFB_DEVICE);
> > >      machine_class_allow_dynamic_sysbus_dev(m, TYPE_VMBUS_BRIDGE);
> > > -    m->max_cpus = 288;
> > > +    m->max_cpus = 512;  
> > 
> > Maybe we should bump it to KVM VCPU maximum,
> > and make sure we error out if asked for combination of
> > hardware/irqchip is not usable.  
> 
> In RHEL downstream we currently bump this to 710 CPUs, because you
> overflow the SMBIOS 2.1 tables at approx 720 CPUs (give/take a little
> depending on other config options).

Also there were some testing done with 1024,
but my main reason for matching KVM's limit is unblock upstream
testing so it would be easier to push limits for others.
Downstream can clamp that value down to whatever it deems as supported.

> Going beyond 710 CPUs value requires using the SMBIOS 3 entry point.
> 
> AFAIK, the x86 machine types still default to SMBIOS 2.1, so that
> would need changing too.

Yep, we can change default to SMBIOS 3 starting with new machine type (7.1?)
or conditionally depending on requested number of CPUs,
though I'd prefer machine type approach.

As for SMBIOS 3, we still have to update CPU structures to support more than
255 vcpus (Julia was working on it). It's long standing bug, but that doesn't
seem to be critical, as guest boots fine with old structures.

Other than that I don't recall any outright bugs related to large number of 
VCPUs.

> With regards,
> Daniel