gpu_sysman: show also mem health + allocatable amount at init

Sysman may return zero for physical memory size, but allocatable size
should always be valid.  This can be useful in identifying different
GPU types (on same machine) from each other.

Skip showing of invalid memory information.

Signed-off-by: Eero Tamminen <eero.t.tamminen@intel.com>

diff --git a/src/gpu_sysman.c b/src/gpu_sysman.c
index f2ff0f04786a33d0b95d3b81e16e3f48d55565f8..1ecf01d627c9c651fa6cfbe42fa540a24e7d1a83 100644 (file)
--- a/src/gpu_sysman.c
+++ b/src/gpu_sysman.c
@@ -447,6 +447,25 @@ static const char *map_mem_type(zes_mem_type_t type) {
   }
 }
 
+/* returns NULL if health is unknown, to avoid adding useless labels */
+static const char *map_mem_health(zes_mem_health_t value) {
+  if (value == ZES_MEM_HEALTH_UNKNOWN) {
+    return NULL;
+  }
+  switch (value) {
+  case ZES_MEM_HEALTH_OK:
+    return "ok";
+  case ZES_MEM_HEALTH_DEGRADED:
+    return "degraded";
+  case ZES_MEM_HEALTH_CRITICAL:
+    return "critical";
+  case ZES_MEM_HEALTH_REPLACE:
+    return "replace";
+  default:
+    return "unknown";
+  }
+}
+
 /* If GPU info setting is enabled, log Sysman API provided info for
  * given GPU, and set PCI device ID to 'pci_dev'.  On success, return
  * true and set GPU PCI address to 'pci_bdf' as string in BDF notation:
@@ -576,21 +595,44 @@ static bool gpu_info(zes_device_handle_t dev, char **pci_bdf, char **pci_dev) {
   /* HW info for all memories */
   for (i = 0; i < mem_count; i++) {
     zes_mem_properties_t props = {.pNext = NULL};
-    /* properties call here would mess unit-testing for metric values */
-#ifndef SYSMAN_UNIT_TEST_BUILD
+#ifndef SYSMAN_UNIT_TEST_BUILD /* call here would mess unit-testing */
     ret = zesMemoryGetProperties(mems[i], &props);
 #endif
     if (ret != ZE_RESULT_SUCCESS) {
-      WARNING(PLUGIN_NAME ": failed to get memory %d properties => 0x%x", i,
-              ret);
+      WARNING(PLUGIN_NAME ": mem %d props query fail => 0x%x", i, ret);
       continue;
     }
+
     const char *location = map_mem_location(props.location);
     const char *type = map_mem_type(props.type);
-    INFO("Memory[%i] - %s / %s:", i, location, type);
-    INFO("- size:       %lu MiB", props.physicalSize / (1024 * 1024));
-    INFO("- channels:   %u ", props.numChannels);
-    INFO("- bus width:  %u", props.busWidth);
+    INFO("Memory-%i:", i);
+    INFO("- type:      %s", type);
+    INFO("- location:  %s", location);
+    if (props.onSubdevice) {
+      INFO("- subdevice: %u", props.subdeviceId);
+    }
+    if (props.busWidth > 0 && props.numChannels) {
+      INFO("- bus width: %u", props.busWidth);
+      INFO("- channels:  %u", props.numChannels);
+    }
+    if (props.physicalSize > 0) {
+      INFO("- size:      %lu MiB", props.physicalSize / (1024 * 1024));
+    }
+
+    zes_mem_state_t state = {0};
+#ifndef SYSMAN_UNIT_TEST_BUILD /* call here would mess unit-testing */
+    ret = zesMemoryGetState(mems[i], &state);
+#endif
+    if (ret != ZE_RESULT_SUCCESS) {
+      WARNING(PLUGIN_NAME ": mem %d state query fail => 0x%x", i, ret);
+      continue;
+    }
+    INFO("Memory state:");
+    INFO("- allocatable: %lu MiB", state.size / (1024 * 1024));
+    const char *health = map_mem_health(state.health);
+    if (health) {
+      INFO("- status:      %s", health);
+    }
   }
   free(mems);
   return true;
@@ -1181,25 +1223,8 @@ static bool gpu_mems(gpu_device_t *gpu, unsigned int cache_idx) {
       break;
     }
     /* get health status from last i.e. zeroeth sample */
-    zes_mem_health_t value = gpu->memory[0][i].health;
-    if (value != ZES_MEM_HEALTH_UNKNOWN) {
-      const char *health;
-      switch (value) {
-      case ZES_MEM_HEALTH_OK:
-        health = "ok";
-        break;
-      case ZES_MEM_HEALTH_DEGRADED:
-        health = "degraded";
-        break;
-      case ZES_MEM_HEALTH_CRITICAL:
-        health = "critical";
-        break;
-      case ZES_MEM_HEALTH_REPLACE:
-        health = "replace";
-        break;
-      default:
-        health = "unknown";
-      }
+    const char *health = map_mem_health(gpu->memory[0][i].health);
+    if (health) {
       metric_label_set(&metric, "health", health);
     }
     double mem_used;