[Mellanox] implement sfp.reset for CMIS management (#16862)

- Why I did it
For CMIS host management module, we need a different implementation for sfp.reset. This PR is to implement it

- How I did it
For SW control modules, do reset from hw_reset
For FW control modules, do reset as the original way

- How to verify it
Manual test
sonic-mgmt platform test

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

@@ -26,6 +26,7 @@ try:
     import ctypes
     import subprocess
     import os
+    import threading
     from sonic_py_common.logger import Logger
     from sonic_py_common.general import check_output_pipe
     from . import utils
@@ -219,6 +220,9 @@ class SdkHandleContext(object):
         deinitialize_sdk_handle(self.sdk_handle)
 
 class NvidiaSFPCommon(SfpOptoeBase):
+    sfp_index_to_logical_port_dict = {}
+    sfp_index_to_logical_lock = threading.Lock()
+    
     def __init__(self, sfp_index):
         super(NvidiaSFPCommon, self).__init__()
         self.index = sfp_index + 1
@@ -248,6 +252,30 @@ class NvidiaSFPCommon(SfpOptoeBase):
 
         return oper_state, error_type
     
+    @classmethod
+    def get_sfp_index_to_logical_port(cls, force=False):
+        if not cls.sfp_index_to_logical_port_dict or force:
+            config_db = utils.DbUtils.get_db_instance('CONFIG_DB')
+            port_data = config_db.get_table('PORT')
+            for key, data in port_data.items():
+                if data['index'] not in cls.sfp_index_to_logical_port_dict:
+                    cls.sfp_index_to_logical_port_dict[int(data['index']) - 1] = key
+    
+    @classmethod
+    def get_logical_port_by_sfp_index(cls, sfp_index):
+        with cls.sfp_index_to_logical_lock:
+            cls.get_sfp_index_to_logical_port()
+            logical_port_name = cls.sfp_index_to_logical_port_dict.get(sfp_index)
+            if not logical_port_name:
+                cls.get_sfp_index_to_logical_port(force=True)
+            else:
+                config_db = utils.DbUtils.get_db_instance('CONFIG_DB')
+                current_index = int(config_db.get('CONFIG_DB', f'PORT|{logical_port_name}', 'index'))
+                if current_index != sfp_index:
+                    cls.get_sfp_index_to_logical_port(force=True)
+                    logical_port_name = cls.sfp_index_to_logical_port_dict.get(sfp_index)
+            return logical_port_name
+  
 
 class SFP(NvidiaSFPCommon):
     """Platform-specific SFP class"""
@@ -299,6 +327,17 @@ class SFP(NvidiaSFPCommon):
         Returns:
             bool: True if device is present, False if not
         """
+        if DeviceDataManager.is_independent_mode():
+            if utils.read_int_from_file(f'/sys/module/sx_core/asic0/module{self.sdk_index}/control') != 0:
+                if not utils.read_int_from_file(f'/sys/module/sx_core/asic0/module{self.sdk_index}/hw_present'):
+                    return False
+                if not utils.read_int_from_file(f'/sys/module/sx_core/asic0/module{self.sdk_index}/power_good'):
+                    return False
+                if not utils.read_int_from_file(f'/sys/module/sx_core/asic0/module{self.sdk_index}/power_on'):
+                    return False
+                if utils.read_int_from_file(f'/sys/module/sx_core/asic0/module{self.sdk_index}/hw_reset') == 1:
+                    return False
+                
         eeprom_raw = self._read_eeprom(0, 1, log_on_error=False)
         return eeprom_raw is not None
 
@@ -455,8 +494,17 @@ class SFP(NvidiaSFPCommon):
 
         refer plugins/sfpreset.py
         """
+        try:
+            if not self.is_sw_control():
                 file_path = SFP_SDK_MODULE_SYSFS_ROOT_TEMPLATE.format(self.sdk_index) + SFP_SYSFS_RESET
                 return utils.write_file(file_path, '1')
+            else:
+                file_path = SFP_SDK_MODULE_SYSFS_ROOT_TEMPLATE.format(self.sdk_index) + SFP_SYSFS_HWRESET
+                return utils.write_file(file_path, '0') and utils.write_file(file_path, '1')
+        except Exception as e:
+            print(f'Failed to reset module - {e}')
+            logger.log_error(f'Failed to reset module - {e}')
+            return False
 
 
     @classmethod
@@ -918,15 +966,15 @@ class SFP(NvidiaSFPCommon):
             return False
 
         db = utils.DbUtils.get_db_instance('STATE_DB')
-        control_type = db.get('STATE_DB', f'TRANSCEIVER_MODULES_MGMT|{self.sdk_index}', 'control_type')
+        logical_port = NvidiaSFPCommon.get_logical_port_by_sfp_index(self.sdk_index)
-        control_file_value = utils.read_int_from_file(f'/sys/module/sx_core/asic0/module{self.sdk_index}/control')
+        if not logical_port:
+            raise Exception(f'Module {self.sdk_index} is not present or in initialization')
         
-        if control_type == 'SW_CONTROL' and control_file_value == 1:
+        initialized = db.exists('STATE_DB', f'TRANSCEIVER_STATUS|{logical_port}')
-            return True
+        if not initialized:
-        elif control_type == 'FW_CONTROL' and control_file_value == 0:
+            raise Exception(f'Module {self.sdk_index} is not present or in initialization')
-            return False
+        
-        else:
+        return utils.read_int_from_file(f'/sys/module/sx_core/asic0/module{self.sdk_index}/control') == 1
-            raise Exception(f'Module {self.sdk_index} is in initialization, please retry later')
 
 
 class RJ45Port(NvidiaSFPCommon):

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

@@ -381,9 +381,9 @@ class DbUtils:
                         cls.db_instances.data = {}
 
             if db_name not in cls.db_instances.data:
-                from swsscommon.swsscommon import SonicV2Connector
+                from swsscommon.swsscommon import ConfigDBConnector
-                db = SonicV2Connector(use_unix_socket_path=True)
+                db = ConfigDBConnector(use_unix_socket_path=True)
-                db.connect(db_name)
+                db.db_connect(db_name)
                 cls.db_instances.data[db_name] = db
             return cls.db_instances.data[db_name]
         except Exception as e:

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

@@ -266,9 +266,14 @@ class TestSfp:
     @mock.patch('sonic_platform.utils.write_file')
     def test_reset(self, mock_write):
         sfp = SFP(0)
+        sfp.is_sw_control = mock.MagicMock(return_value=False)
         mock_write.return_value = True
         assert sfp.reset()
         mock_write.assert_called_with('/sys/module/sx_core/asic0/module0/reset', '1')
+        sfp.is_sw_control.return_value = True
+        assert sfp.reset()
+        sfp.is_sw_control.side_effect = Exception('')
+        assert not sfp.reset()
 
     @mock.patch('sonic_platform.sfp.SFP.read_eeprom')
     def test_get_xcvr_api(self, mock_read):
@@ -332,31 +337,33 @@ class TestSfp:
         assert sfp.get_temperature_warning_threashold() == 75.0
         assert sfp.get_temperature_critical_threashold() == 85.0
 
+    @mock.patch('sonic_platform.sfp.NvidiaSFPCommon.get_logical_port_by_sfp_index')
     @mock.patch('sonic_platform.utils.read_int_from_file')
     @mock.patch('sonic_platform.device_data.DeviceDataManager.is_independent_mode')
     @mock.patch('sonic_platform.utils.DbUtils.get_db_instance')
-    def test_is_sw_control(self, mock_get_db, mock_mode, mock_read):
+    def test_is_sw_control(self, mock_get_db, mock_mode, mock_read, mock_get_logical):
         sfp = SFP(0)
         mock_mode.return_value = False
         assert not sfp.is_sw_control()
         mock_mode.return_value = True
         
-        mock_db = mock.MagicMock()
+        mock_get_logical.return_value = None
-        mock_get_db.return_value = mock_db
-        mock_db.get = mock.MagicMock(return_value=None)
         with pytest.raises(Exception):
             sfp.is_sw_control()
 
-        mock_read.return_value = 0
+        mock_get_logical.return_value = 'Ethernet0'
-        mock_db.get.return_value = 'FW_CONTROL'
+        mock_db = mock.MagicMock()
-        assert not sfp.is_sw_control()
+        mock_get_db.return_value = mock_db
-        mock_read.return_value = 1
+        mock_db.exists = mock.MagicMock(return_value=False)
-        mock_db.get.return_value = 'SW_CONTROL'
-        assert sfp.is_sw_control()
-        mock_read.return_value = 0
         with pytest.raises(Exception):
             sfp.is_sw_control()
 
+        mock_db.exists.return_value = True
+        mock_read.return_value = 0
+        assert not sfp.is_sw_control()
+        mock_read.return_value = 1
+        assert sfp.is_sw_control()
+
     @mock.patch('sonic_platform.device_data.DeviceDataManager.is_independent_mode', mock.MagicMock(return_value=False))
     @mock.patch('sonic_platform.sfp.SFP.is_sw_control', mock.MagicMock(return_value=False))
     @mock.patch('sonic_platform.utils.is_host', mock.MagicMock(side_effect = [True, True, False, False]))