linux/cpukinds/freq: use acpi_cppc/nominal_freq when available

cpufreq/base_frequency is only available on Intel so far, and works well.

acpi_cppc/nominal_freq is already available on AMD (and ARM or soon),
so it's likely good for the future.
However it reports incorrect values on Intel SPR and MTL at least.

Hence try cpufreq/base_frequency first,
then fallback to acpi_cppc/nominal_freq.

Refs #690

Signed-off-by: Brice Goglin <[email protected]>

Author: bgoglin

doc/hwloc.doxy

@@ -2466,8 +2466,9 @@ These CPU kinds may be annotated with the following native attributes:
 as reported by <tt>cpufreq</tt> drivers on Linux.
 </dd>
 <dt>FrequencyBaseMHz (Linux)</dt>
-<dd>The base operating frequency of the core,
-as reported by some <tt>cpufreq</tt> drivers on Linux (e.g. <tt>intel_pstate</tt>).
+<dd>The base/nominal operating frequency of the core,
+as reported by some <tt>cpufreq</tt> or ACPI drivers on Linux
+(e.g. <tt>cpufreq_cppc</tt> or <tt>intel_pstate</tt>).
 </dd>
 <dt>CoreType (x86)</dt>
 <dd>A string describing the kind of core,

hwloc/topology-linux.c

@@ -4480,8 +4480,8 @@ look_sysfsnode(struct hwloc_topology *topology,
 
 struct hwloc_linux_cpukinds_by_pu {
   unsigned pu;
-  unsigned long max_freq;
-  unsigned long base_freq;
+  unsigned long max_freq; /* kHz */
+  unsigned long base_freq; /* kHz */
   unsigned long capacity;
   int done; /* temporary bit to identify PU that were processed by the current algorithm
              * (only hwloc_linux_cpukinds_adjust_maxfreqs() for now)
@@ -4724,6 +4724,7 @@ look_sysfscpukinds(struct hwloc_topology *topology,
   char *env;
   hwloc_bitmap_t atom_pmu_set, core_pmu_set;
   int maxfreq_enabled = -1; /* -1 means adjust (default), 0 means ignore, 1 means enforce */
+  int use_cppc_nominal_freq = -1; /* -1 means try, 0 no, 1 yes */
   unsigned adjust_max = 10;
   int force_homogeneous;
   const char *info;
@@ -4763,10 +4764,26 @@ look_sysfscpukinds(struct hwloc_topology *topology,
     sprintf(str, "/sys/devices/system/cpu/cpu%d/cpufreq/cpuinfo_max_freq", i);
     if (hwloc_read_path_as_uint(str, &maxfreq, data->root_fd) >= 0)
       by_pu[i].max_freq = maxfreq;
-    /* base_frequency is intel_pstate specific */
+    /* base_frequency is in intel_pstate and works fine */
     sprintf(str, "/sys/devices/system/cpu/cpu%d/cpufreq/base_frequency", i);
-    if (hwloc_read_path_as_uint(str, &basefreq, data->root_fd) >= 0)
+    if (hwloc_read_path_as_uint(str, &basefreq, data->root_fd) >= 0) {
       by_pu[i].base_freq = basefreq;
+      use_cppc_nominal_freq = 0;
+    }
+    /* try acpi_cppc/nominal_freq only if cpufreq/base_frequency failed
+     * acpi_cppc/nominal_freq is widely available, but it returns 0 on some Intel SPR,
+     * same freq for all cores on RPL,
+     * maxfreq for E-cores and LP-E-cores but basefreq for P-cores on MTL.
+     */
+    if (use_cppc_nominal_freq != 0) {
+      sprintf(str, "/sys/devices/system/cpu/cpu%d/acpi_cppc/nominal_freq", i);
+      if (hwloc_read_path_as_uint(str, &basefreq, data->root_fd) >= 0 && basefreq > 0) {
+        by_pu[i].base_freq = basefreq * 1000; /* nominal_freq is already in MHz */
+        use_cppc_nominal_freq = 1;
+      } else {
+        use_cppc_nominal_freq = 0;
+      }
+    }
     if (maxfreq && !basefreq)
       max_without_basefreq = 1;
     /* capacity */