Cache inclusivity?

[chriselrod]

This is probably an issue for Hwloc itself, but I figured I'd raise it here first.

Hwloc checks the inclusivity of the caches, e.g. running

hwloc-gather-topology /tmp/myhost

on the command line with create a .xml file that includes, among other output

      <object type="L3Cache" cpuset="0x000000ff" complete_cpuset="0x000000ff" nodeset="0x00000001" complete_nodeset="0x00000001" gp_index="8" cache_size="12582912" depth="3" cache_linesize="64" cache_associativity="12" cache_type="0">
        <info name="Inclusive" value="0"/>
        <object type="L2Cache" cpuset="0x00000011" complete_cpuset="0x00000011" nodeset="0x00000001" complete_nodeset="0x00000001" gp_index="7" cache_size="1310720" depth="2" cache_linesize="64" cache_associativity="20" cache_type="0">
          <info name="Inclusive" value="0"/>
          <object type="L1Cache" cpuset="0x00000011" complete_cpuset="0x00000011" nodeset="0x00000001" complete_nodeset="0x00000001" gp_index="5" cache_size="49152" depth="1" cache_linesize="64" cache_associativity="12" cache_type="1">
            <info name="Inclusive" value="0"/>
            <object type="L1iCache" cpuset="0x00000011" complete_cpuset="0x00000011" nodeset="0x00000001" complete_nodeset="0x00000001" gp_index="6" cache_size="32768" depth="1" cache_linesize="64" cache_associativity="8" cache_type="2">
              <info name="Inclusive" value="0"/>

Indicating that none of the caches are inclusive.

Hwloc uses cpuid for AMD and Intel CPUs, storing the results in a struct:

struct cacheinfo {
  hwloc_obj_cache_type_t type;
  unsigned level;
  unsigned nbthreads_sharing;
  unsigned cacheid;

  unsigned linesize;
  unsigned linepart;
  int inclusive;
  int ways;
  unsigned sets;
  unsigned long size;
};

But this seems to be an undocumented internal.
The documented hwloc_cache_attr_s that you can load with the API is missing several of these fields.

There is a documented hwloc_obj_get_info_by_name which is used to query "Inclusive", but nm -D doesn't show that symbol, and it'd take some work to figure out how to create the argument it takes (and if that object is a cacheinfo, we could just read the inclusive directly).
So perhaps it's a Hwloc issue to ask about an API for this?