From da480b4589b198a3d29f87054daa1985ddb2df66 Mon Sep 17 00:00:00 2001
From: "Bill(Shuzhou) Liu" <shuzhou.liu@amd.com>
Date: Tue, 30 Mar 2021 10:44:01 -0400
Subject: [PATCH] Add support for the HBM temperature

The rsmi_dev_temp_metric_get() can also support the HBM
temperatures which is retrieved from gpu_metrics.

Change-Id: I96b979296e90cf881523627b41b1a02849676416
---
 include/rocm_smi/rocm_smi.h                 |  7 +++--
 src/rocm_smi.cc                             | 34 +++++++++++++++++++++
 tests/rocm_smi_test/functional/temp_read.cc |  4 +++
 3 files changed, 43 insertions(+), 2 deletions(-)

diff --git a/include/rocm_smi/rocm_smi.h b/include/rocm_smi/rocm_smi.h
index 894ba534f2..24ec69e5d2 100755
--- a/include/rocm_smi/rocm_smi.h
+++ b/include/rocm_smi/rocm_smi.h
@@ -407,8 +407,11 @@ typedef enum {
   RSMI_TEMP_TYPE_JUNCTION,                     //!< Junction/hotspot
                                                //!< temperature
   RSMI_TEMP_TYPE_MEMORY,                       //!< VRAM temperature
-
-  RSMI_TEMP_TYPE_LAST = RSMI_TEMP_TYPE_MEMORY,
+  RSMI_TEMP_TYPE_HBM_0,                        //!< HBM temperature instance 0
+  RSMI_TEMP_TYPE_HBM_1,                        //!< HBM temperature instance 1
+  RSMI_TEMP_TYPE_HBM_2,                        //!< HBM temperature instance 2
+  RSMI_TEMP_TYPE_HBM_3,                        //!< HBM temperature instance 3
+  RSMI_TEMP_TYPE_LAST = RSMI_TEMP_TYPE_HBM_3,
   RSMI_TEMP_TYPE_INVALID = 0xFFFFFFFF          //!< Invalid type
 } rsmi_temperature_type_t;
 
diff --git a/src/rocm_smi.cc b/src/rocm_smi.cc
index eebd59bc28..525311a568 100755
--- a/src/rocm_smi.cc
+++ b/src/rocm_smi.cc
@@ -2024,6 +2024,40 @@ rsmi_dev_temp_metric_get(uint32_t dv_ind, uint32_t sensor_type,
       mon_type = amd::smi::kMonInvalid;
   }
 
+  // The HBM temperature is retreived from the gpu_metrics
+  if (sensor_type == RSMI_TEMP_TYPE_HBM_0
+     || sensor_type == RSMI_TEMP_TYPE_HBM_1
+     || sensor_type == RSMI_TEMP_TYPE_HBM_2
+     || sensor_type == RSMI_TEMP_TYPE_HBM_3) {
+       if (metric != RSMI_TEMP_CURRENT) {   // only support RSMI_TEMP_CURRENT
+          return RSMI_STATUS_NOT_SUPPORTED;
+       }
+
+       rsmi_gpu_metrics_t gpu_metrics;
+       ret = rsmi_dev_gpu_metrics_info_get(dv_ind, &gpu_metrics);
+       if (ret != RSMI_STATUS_SUCCESS) {
+         return ret;
+       }
+
+       if (temperature == nullptr) {
+          return RSMI_STATUS_INVALID_ARGS;
+       }
+
+       if (sensor_type == RSMI_TEMP_TYPE_HBM_0) {
+         *temperature = gpu_metrics.temperature_hbm[0];
+       } else if (sensor_type == RSMI_TEMP_TYPE_HBM_1) {
+         *temperature = gpu_metrics.temperature_hbm[1];
+       } else if (sensor_type == RSMI_TEMP_TYPE_HBM_2) {
+         *temperature = gpu_metrics.temperature_hbm[2];
+       } else if (sensor_type == RSMI_TEMP_TYPE_HBM_3) {
+         *temperature = gpu_metrics.temperature_hbm[3];
+       } else {
+         return RSMI_STATUS_NOT_SUPPORTED;
+       }
+
+       return RSMI_STATUS_SUCCESS;
+  }  // end HBM temperature
+
   DEVICE_MUTEX
 
   GET_DEV_FROM_INDX
diff --git a/tests/rocm_smi_test/functional/temp_read.cc b/tests/rocm_smi_test/functional/temp_read.cc
index 938ddc3777..859d50399a 100755
--- a/tests/rocm_smi_test/functional/temp_read.cc
+++ b/tests/rocm_smi_test/functional/temp_read.cc
@@ -60,6 +60,10 @@ static const std::map<uint32_t, std::string> kTempSensorNameMap = {
     {RSMI_TEMP_TYPE_MEMORY, "Memory"},
     {RSMI_TEMP_TYPE_JUNCTION, "Junction"},
     {RSMI_TEMP_TYPE_EDGE, "Edge"},
+    {RSMI_TEMP_TYPE_HBM_0, "HBM_0"},
+    {RSMI_TEMP_TYPE_HBM_1, "HBM_1"},
+    {RSMI_TEMP_TYPE_HBM_2, "HBM_2"},
+    {RSMI_TEMP_TYPE_HBM_3, "HBM_3"},
 };
 TestTempRead::TestTempRead() : TestBase() {
   set_title("RSMI Temp Read Test");