Re: [PATCH v2] pc: q35: Bump max_cpus to 1024

["Michael S. Tsirkin" <mst@redhat.com>]

On Thu, Jun 01, 2023 at 11:17:30AM +0100, Daniel P. Berrangé wrote:
> On Thu, Jun 01, 2023 at 11:09:45AM +0100, Joao Martins wrote:
> > 
> > 
> > On 31/05/2023 23:51, Suravee Suthikulpanit wrote:
> > > Since KVM_MAX_VCPUS is currently defined to 1024 for x86 as shown in
> > > arch/x86/include/asm/kvm_host.h, update QEMU limits to the same number.
> > > 
> > > In case KVM could not support the specified number of vcpus, QEMU would
> > > return the following error message:
> > > 
> > >   qemu-system-x86_64: kvm_init_vcpu: kvm_get_vcpu failed (xxx): Invalid argument
> > > 
> > > Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
> > > ---
> > > 
> > > Changes from V1:
> > > (https://lore.kernel.org/all/YnkDGsIi1vFvXmiP@redhat.com/T/)
> > >  * Bump from 512 to KVM_MAX_VCPUS (per Igor's suggestion)
> > > 
> > > Note:
> > >  From the last discussion, Daniel mentioned that SMBIO 2.1 tables might
> > >  cause overflow at approx 720 CPUs, and it might require using the
> > >  SMBIO 3.0 entry point. Also, we might need to change the default for
> > >  the x86 machine type to SMBIO 3.0. However, I do not know the status
> > >  of this.
> > > 
> > I suspect smbios 3.0 (64-bit entry point) is already supported.
> > 
> > With current qemu and all the smbios fixes in the last cycle, perhaps this is
> > mainly just setting smbios_entry_point_type to SMBIOS_ENTRY_POINT_TYPE_64 if
> > MachineState::smp::max_cpus is bigger than 720 (e.g. in pc_q35_init()?)
> 
> The need for the 64-bit entry point depends on a combination of RAM config
> and CPU count. IMHO we need to unconditionally switch the latest machine
> types to use the 64-bit entry point by default, rather than trying to infer
> some special condition to dynamically change on the fly.

Yes, makes sense.

> Before doing that we need confirmation that our firmware builds in qemu.git
> (SeaBIOS and EDK2) will be happy with the 64-bit EP. I suspect they should
> be fine by now, but I've CC'd Gerd to confirm.
> 
> With regards,
> Daniel