[Mellanox] Validate Module Presence Using Sysfs Before Accessing EEPROM in Thermal Control Daemon

platform/mellanox/mlnx-platform-api/sonic_platform/sfp.py

@@ -483,6 +483,9 @@ def get_presence(self):
         Returns:
             bool: True if device is present, False if not
         """
+        presence_sysfs = f'/sys/module/sx_core/asic0/module{self.sdk_index}/hw_present' if self.is_sw_control() else f'/sys/module/sx_core/asic0/module{self.sdk_index}/present'
+        if utils.read_int_from_file(presence_sysfs) != 1:
+            return False
         eeprom_raw = self._read_eeprom(0, 1, log_on_error=False)
         return eeprom_raw is not None
 

platform/mellanox/mlnx-platform-api/sonic_platform/thermal.py

@@ -435,6 +435,8 @@ def get_temperature(self):
             A float number of current temperature in Celsius up to nearest thousandth
             of one degree Celsius, e.g. 30.125
         """
+        if not self.sfp.get_presence():
+            return None
         value = self.sfp.get_temperature()
         return value if (value != 0.0 and value is not None) else None
 
@@ -446,6 +448,8 @@ def get_high_threshold(self):
             A float number, the high threshold temperature of thermal in Celsius
             up to nearest thousandth of one degree Celsius, e.g. 30.125
         """
+        if not self.sfp.get_presence():
+            return None
         value = self.sfp.get_temperature_warning_threshold()
         return value if (value != 0.0 and value is not None) else None
 
@@ -457,6 +461,8 @@ def get_high_critical_threshold(self):
             A float number, the high critical threshold temperature of thermal in Celsius
             up to nearest thousandth of one degree Celsius, e.g. 30.125
         """
+        if not self.sfp.get_presence():
+            return None
         value = self.sfp.get_temperature_critical_threshold()
         return value if (value != 0.0 and value is not None) else None
 

platform/mellanox/mlnx-platform-api/tests/test_sfp.py

@@ -247,15 +247,23 @@ def test_get_page_and_page_offset(self, mock_get_type_str, mock_eeprom_path, moc
         assert page == '/tmp/1/data'
         assert page_offset is 0
 
+    @mock.patch('sonic_platform.utils.read_int_from_file')
     @mock.patch('sonic_platform.sfp.SFP._read_eeprom')
-    def test_sfp_get_presence(self, mock_read):
+    def test_sfp_get_presence(self, mock_read, mock_read_int):
         sfp = SFP(0)
+
+        mock_read_int.return_value = 1
         mock_read.return_value = None
         assert not sfp.get_presence()
-
         mock_read.return_value = 0
         assert sfp.get_presence()
 
+        mock_read_int.return_value = 0
+        mock_read.return_value = None
+        assert not sfp.get_presence()
+        mock_read.return_value = 0
+        assert not sfp.get_presence()
+
     @mock.patch('sonic_platform.utils.read_int_from_file')
     def test_rj45_get_presence(self, mock_read_int):
         sfp = RJ45Port(0)

platform/mellanox/mlnx-platform-api/tests/test_thermal.py

@@ -159,12 +159,21 @@ def test_sfp_thermal(self):
         assert thermal.get_name() == rule['name'].format(start_index)
         assert thermal.get_position_in_parent() == 1
         assert thermal.is_replaceable() == False
+        sfp.get_presence = mock.MagicMock(return_value=True)
         sfp.get_temperature = mock.MagicMock(return_value=35.4)
         sfp.get_temperature_warning_threshold = mock.MagicMock(return_value=70)
         sfp.get_temperature_critical_threshold = mock.MagicMock(return_value=80)
         assert thermal.get_temperature() == 35.4
         assert thermal.get_high_threshold() == 70
         assert thermal.get_high_critical_threshold() == 80
+        sfp.get_presence = mock.MagicMock(return_value=False)
+        sfp.get_temperature = mock.MagicMock(return_value=35.4)
+        sfp.get_temperature_warning_threshold = mock.MagicMock(return_value=70)
+        sfp.get_temperature_critical_threshold = mock.MagicMock(return_value=80)
+        assert thermal.get_temperature() is None
+        assert thermal.get_high_threshold() is None
+        assert thermal.get_high_critical_threshold() is None
+        sfp.get_presence = mock.MagicMock(return_value=True)
         sfp.get_temperature = mock.MagicMock(return_value=0)
         sfp.get_temperature_warning_threshold = mock.MagicMock(return_value=0)
         sfp.get_temperature_critical_threshold = mock.MagicMock(return_value=None)