Re: [PATCH] hw/core/machine-smp: Remove deprecated "parameter=0" SMP configurations

[Zhao Liu]

Hi Prasad,

On Mon, Mar 04, 2024 at 11:23:58AM +0530, Prasad Pandit wrote:
> Date: Mon, 4 Mar 2024 11:23:58 +0530
> From: Prasad Pandit <ppandit@redhat.com>
> Subject: Re: [PATCH] hw/core/machine-smp: Remove deprecated "parameter=0"
>  SMP configurations
> 
> On Mon, 4 Mar 2024 at 10:02, Zhao Liu <zhao1.liu@linux.intel.com> wrote:
> > diff --git a/hw/core/machine-smp.c b/hw/core/machine-smp.c
> > index 25019c91ee36..96533886b14e 100644
> > --- a/hw/core/machine-smp.c
> > +++ b/hw/core/machine-smp.c
> > @@ -105,8 +105,9 @@ void machine_parse_smp_config(MachineState *ms,
> >          (config->has_cores && config->cores == 0) ||
> >          (config->has_threads && config->threads == 0) ||
> >          (config->has_maxcpus && config->maxcpus == 0)) {
> > -        warn_report("Deprecated CPU topology (considered invalid): "
> > -                    "CPU topology parameters must be greater than zero");
> > +        error_setg(errp, "Invalid CPU topology: "
> > +                   "CPU topology parameters must be greater than zero");
> > +        return;
> >      }
> 
> unsigned maxcpus = config->has_maxcpus ? config->maxcpus : 0;

This indicates the default maxcpus is initialized as 0 if user doesn't
specifies it.

For this case - no user configuration - maxcpus will be re-calculated
as:

    maxcpus = maxcpus > 0 ? maxcpus : drawers * books * sockets * dies *
                                      clusters * cores * threads; (*)

>  ...
>  if (config->has_maxcpus && config->maxcpus == 0)

This check only wants to identify the case that user sets the 0.

> 
> * The check (has_maxcpus && maxcpus == 0) seems to be repeating above,
> maybe we could check if (maxcpus == 0) error_setg().

If the default maxcpus is initialized as 0, then (maxcpus == 0) will
fail if user doesn't set maxcpus.

However, we could initialize maxcpus as other default value, e.g., 

    maxcpus = config->has_maxcpus ? config->maxcpus : 1.

But it is still necessary to distinguish whether maxcpus is user-set or
auto-initialized.

If it is user-set, -smp should fail is there's invalid maxcpus/invalid
topology.

Otherwise, if it is auto-initialized, its value should be adjusted based
on other topology components as the above calculation in (*).

> And same for
> other topology parameters?

Other parameters also have the similar needs to distinguish if they're
set by user. So the check needs to also cover has_* fields.

> * Also a check to ensure cpus <= maxcpus is required I think.
>

Yes, the valid topology needs this. This code block already covers this
case ;-):

    if (maxcpus < cpus) {
        g_autofree char *topo_msg = cpu_hierarchy_to_string(ms);
        error_setg(errp, "Invalid CPU topology: "
                   "maxcpus must be equal to or greater than smp: "
                   "%s == maxcpus (%u) < smp_cpus (%u)",
                   topo_msg, maxcpus, cpus);
        return;
    }

Thanks,
Zhao