Re: [PATCH RESEND 1/3] topology: Represent clusters of CPUs within a die

[Barry Song <21cnbao@xxxxxxxxx>]

On Wed, Oct 6, 2021 at 5:34 AM Valentin Schneider
<valentin.schneider@xxxxxxx> wrote:
>
> On 24/09/21 20:51, Barry Song wrote:
> >  void update_siblings_masks(unsigned int cpuid)
> >  {
> >       struct cpu_topology *cpu_topo, *cpuid_topo = &cpu_topology[cpuid];
> > @@ -617,6 +622,11 @@ void update_siblings_masks(unsigned int cpuid)
> >               if (cpuid_topo->package_id != cpu_topo->package_id)
> >                       continue;
> >
> > +             if (cpuid_topo->cluster_id == cpu_topo->cluster_id) {
> > +                     cpumask_set_cpu(cpu, &cpuid_topo->cluster_sibling);
> > +                     cpumask_set_cpu(cpuid, &cpu_topo->cluster_sibling);
> > +             }
> > +
>
> Hm so without cluster information (e.g. DT system), we have
> ->cluster_id=-1, we'll essentially copy the package mask into the cluster
> mask.
>
> The exposed cluster mask is still <= package mask which is sensible. Are we
> fine with that, or do we need/want the mask to be empty in the -1 case? I'm
> guessing userspace tools should check for either id!=-1 or if the exclusive
> disjucntion of cluster vs package masks is non-empty.

Hi Valentin,
Yep, this is a very good question. I'd like change the code to:
diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c
index 7cb31d959f33..fc0836f460fb 100644
--- a/drivers/base/arch_topology.c
+++ b/drivers/base/arch_topology.c
@@ -622,7 +622,8 @@ void update_siblings_masks(unsigned int cpuid)
                if (cpuid_topo->package_id != cpu_topo->package_id)
                        continue;

-               if (cpuid_topo->cluster_id == cpu_topo->cluster_id) {
+               if (cpuid_topo->cluster_id == cpu_topo->cluster_id &&
+                   cpuid_topo->cluster_id != -1) {
                        cpumask_set_cpu(cpu, &cpuid_topo->cluster_sibling);
                        cpumask_set_cpu(cpuid, &cpu_topo->cluster_sibling);
                }

This should be consistent with Tim's patch3/3 for x86 in case
id is BAD_APICID:
static bool match_l2c(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)
{
        ...
        /* Do not match if we do not have a valid APICID for cpu: */
        if (per_cpu(cpu_l2c_id, cpu1) == BAD_APICID)
                return false;
        ...
}

Thanks
Barry