Fix up xcvrd to use line lanes from gearbox table if present (#689)

* Fix up xcvrd to look at line lanes from gearbox if present

- Implement new method get_gearbox_line_lanes_dict in xcvr_table_helper.py
  to return a dict of number of gearbox line lanes indexed by logical
  port.
- Building this dict and caching it in CmisTaskManager before a new
  iteration over port_dict
- Implement new method get_host_lane_count to look up the gearbox line
  lanes in this cached dict for every logical port. If found, use this
  as the host_lane_count, else use the host_lane_count from port_config
- Use this new method when setting host_lane_count for every logical
  port iteration

* Addressed nit in comment - line -> line-side

Author: aditya-nexthop

sonic-xcvrd/tests/test_xcvrd.py

@@ -2575,6 +2575,173 @@ def get_host_lane_assignment_option_side_effect(app):
         appl = common.get_cmis_application_desired(mock_xcvr_api, host_lane_count, speed)
         assert task.get_cmis_host_lanes_mask(mock_xcvr_api, appl, host_lane_count, subport) == expected
 
+    @pytest.mark.parametrize("gearbox_data, expected_dict", [
+        # Test case 1: Gearbox port with 2 line lanes
+        ({
+            "interface:0": {
+                "name": "Ethernet0",
+                "index": "0",
+                "phy_id": "1",
+                "system_lanes": "300,301,302,303",
+                "line_lanes": "304,305"
+            }
+        }, {"Ethernet0": 2}),
+        # Test case 2: Multiple gearbox ports
+        ({
+            "interface:0": {
+                "name": "Ethernet0",
+                "index": "0",
+                "phy_id": "1",
+                "system_lanes": "300,301,302,303",
+                "line_lanes": "304,305,306,307"
+            },
+            "interface:200": {
+                "name": "Ethernet200",
+                "index": "200",
+                "phy_id": "2",
+                "system_lanes": "400,401",
+                "line_lanes": "404,405"
+            }
+        }, {"Ethernet0": 4, "Ethernet200": 2}),
+        # Test case 3: Empty gearbox data
+        ({}, {}),
+        # Test case 4: Gearbox interface with empty line_lanes
+        ({
+            "interface:0": {
+                "name": "Ethernet0",
+                "index": "0",
+                "phy_id": "1",
+                "system_lanes": "300,301,302,303",
+                "line_lanes": ""
+            }
+        }, {}),
+        # Test case 5: Non-interface keys (should be ignored)
+        ({
+            "interface:0": {
+                "name": "Ethernet0",
+                "index": "0",
+                "phy_id": "1",
+                "system_lanes": "300,301,302,303",
+                "line_lanes": "304,305"
+            },
+            "phy:1": {
+                "name": "phy1",
+                "some_field": "some_value"
+            }
+        }, {"Ethernet0": 2})
+    ])
+    def test_XcvrTableHelper_get_gearbox_line_lanes_dict(self, gearbox_data, expected_dict):
+        # Mock the XcvrTableHelper and APPL_DB access
+        mock_appl_db = MagicMock()
+        mock_gearbox_table = MagicMock()
+
+        # Mock table.getKeys() to return gearbox interface keys
+        mock_gearbox_table.getKeys.return_value = list(gearbox_data.keys())
+
+        # Mock table.get() to return gearbox interface data
+        def mock_get_side_effect(key):
+            if key in gearbox_data:
+                # Convert dict to list of tuples for fvs format
+                interface_data = gearbox_data[key]
+                fvs_list = [(k, v) for k, v in interface_data.items()]
+                return (True, fvs_list)
+            return (False, [])
+
+        mock_gearbox_table.get.side_effect = mock_get_side_effect
+
+        # Mock swsscommon.Table constructor to return our mock table
+        with patch('xcvrd.xcvrd_utilities.xcvr_table_helper.swsscommon.Table', return_value=mock_gearbox_table):
+            # Mock the helper_logger to avoid logging during tests
+            with patch('xcvrd.xcvrd_utilities.xcvr_table_helper.helper_logger'):
+                helper = XcvrTableHelper(DEFAULT_NAMESPACE)
+                helper.appl_db = {0: mock_appl_db}  # Mock the appl_db dict
+
+                result = helper.get_gearbox_line_lanes_dict()
+                assert result == expected_dict
+
+    @pytest.mark.parametrize("gearbox_lanes_dict, lport, port_config_lanes, expected_count", [
+        # Test case 1: Gearbox data available, should use gearbox count
+        ({"Ethernet0": 2}, "Ethernet0", "25,26,27,28", 2),
+        # Test case 2: Gearbox data available with 4 lanes
+        ({"Ethernet0": 4}, "Ethernet0", "29,30", 4),
+        # Test case 3: No gearbox data for this port, should use port config
+        ({"Ethernet4": 2}, "Ethernet0", "33,34,35,36", 4),
+        # Test case 4: Empty gearbox dict, should use port config
+        ({}, "Ethernet0", "37,38", 2),
+        # Test case 5: Multiple ports in gearbox dict
+        ({"Ethernet0": 2, "Ethernet4": 4}, "Ethernet0", "25,26,27,28", 2),
+        # Test case 6: Port not in gearbox dict
+        ({"Ethernet4": 4}, "Ethernet8", "41,42,43", 3)
+    ])
+    def test_CmisManagerTask_get_host_lane_count(self, gearbox_lanes_dict, lport, port_config_lanes, expected_count):
+        port_mapping = PortMapping()
+        stop_event = threading.Event()
+        task = CmisManagerTask(DEFAULT_NAMESPACE, port_mapping, stop_event)
+
+        result = task.get_host_lane_count(lport, port_config_lanes, gearbox_lanes_dict)
+        assert result == expected_count
+
+    def test_CmisManagerTask_gearbox_integration_end_to_end(self):
+        """Test end-to-end integration of gearbox line lanes with CMIS application selection"""
+        port_mapping = PortMapping()
+        stop_event = threading.Event()
+        task = CmisManagerTask(DEFAULT_NAMESPACE, port_mapping, stop_event)
+
+        # Mock gearbox lanes dictionary - port has 4 system lanes but only 2 line lanes
+        gearbox_lanes_dict = {"Ethernet0": 2}  # 2 line lanes from gearbox
+
+        # Mock port config - would normally give 4 lanes
+        port_config_lanes = "25,26,27,28"  # 4 lanes from port config
+
+        # Mock CMIS API with application advertisement
+        mock_xcvr_api = MagicMock()
+        mock_xcvr_api.get_application_advertisement.return_value = {
+            1: {
+                'host_electrical_interface_id': '100GAUI-2 C2M (Annex 135G)',
+                'module_media_interface_id': '100G-FR/100GBASE-FR1 (Cl 140)',
+                'media_lane_count': 1,
+                'host_lane_count': 2,  # Matches our gearbox line lanes
+                'host_lane_assignment_options': 85
+            },
+            2: {
+                'host_electrical_interface_id': 'CAUI-4 C2M (Annex 83E)',
+                'module_media_interface_id': 'Active Cable assembly',
+                'media_lane_count': 4,
+                'host_lane_count': 4,  # Would match port config lanes
+                'host_lane_assignment_options': 17
+            }
+        }
+
+        # Test the integration: should use gearbox line lanes (2) not port config lanes (4)
+        host_lane_count = task.get_host_lane_count("Ethernet0", port_config_lanes, gearbox_lanes_dict)
+        assert host_lane_count == 2  # Should use gearbox line lanes, not port config
+
+        # Test that this leads to correct CMIS application selection
+        with patch('xcvrd.xcvrd_utilities.common.is_cmis_api', return_value=True):
+            appl = common.get_cmis_application_desired(mock_xcvr_api, host_lane_count, 100000)
+            assert appl == 1  # Should select application 1 (2 lanes) not application 2 (4 lanes)
+
+    def test_CmisManagerTask_gearbox_caching_integration(self):
+        """Test that gearbox lanes dictionary is properly cached and used in task worker"""
+        port_mapping = PortMapping()
+        stop_event = threading.Event()
+        task = CmisManagerTask(DEFAULT_NAMESPACE, port_mapping, stop_event)
+
+        # Mock the XcvrTableHelper to return a gearbox lanes dictionary
+        mock_gearbox_lanes_dict = {"Ethernet0": 2, "Ethernet4": 4}
+        task.xcvr_table_helper = MagicMock()
+        task.xcvr_table_helper.get_gearbox_line_lanes_dict.return_value = mock_gearbox_lanes_dict
+
+        # Test that get_host_lane_count uses the cached dictionary correctly
+        result1 = task.get_host_lane_count("Ethernet0", "25,26,27,28", mock_gearbox_lanes_dict)
+        assert result1 == 2  # Should use gearbox count
+
+        result2 = task.get_host_lane_count("Ethernet4", "29,30", mock_gearbox_lanes_dict)
+        assert result2 == 4  # Should use gearbox count
+
+        result3 = task.get_host_lane_count("Ethernet8", "33,34,35", mock_gearbox_lanes_dict)
+        assert result3 == 3  # Should fall back to port config count
+
     @patch('swsscommon.swsscommon.FieldValuePairs')
     def test_CmisManagerTask_post_port_active_apsel_to_db_error_cases(self, mock_field_value_pairs):
         mock_xcvr_api = MagicMock()

sonic-xcvrd/xcvrd/xcvrd.py

@@ -465,6 +465,29 @@ def get_cmis_module_power_up_duration_secs(self, api):
     def get_cmis_module_power_down_duration_secs(self, api):
         return api.get_module_pwr_down_duration()/1000
 
+    def get_host_lane_count(self, lport, port_config_lanes, gearbox_lanes_dict):
+        """
+        Get host lane count from gearbox configuration if available, otherwise from port config
+
+        Args:
+            lport: logical port name (e.g., "Ethernet0")
+            port_config_lanes: lanes string from port config (e.g., "25,26,27,28")
+            gearbox_lanes_dict: dictionary of gearbox line lanes counts
+
+        Returns:
+            Integer: number of host lanes
+        """
+        # First try to get from gearbox configuration
+        gearbox_host_lane_count = gearbox_lanes_dict.get(lport, 0)
+        if gearbox_host_lane_count > 0:
+            self.log_debug("{}: Using gearbox line lanes count: {}".format(lport, gearbox_host_lane_count))
+            return gearbox_host_lane_count
+
+        # Fallback to port config lanes
+        host_lane_count = len(port_config_lanes.split(','))
+        self.log_debug("{}: Using port config lanes count: {}".format(lport, host_lane_count))
+        return host_lane_count
+
     def get_cmis_host_lanes_mask(self, api, appl, host_lane_count, subport):
         """
         Retrieves mask of active host lanes based on appl, host lane count and subport
@@ -972,6 +995,9 @@ def task_worker(self):
             # Handle port change event from main thread
             port_change_observer.handle_port_update_event()
 
+            # Cache gearbox line lanes dictionary for this set of iterations over the port_dict
+            gearbox_lanes_dict = self.xcvr_table_helper.get_gearbox_line_lanes_dict()
+
             for lport, info in self.port_dict.items():
                 if self.task_stopping_event.is_set():
                     break
@@ -1003,7 +1029,7 @@ def task_worker(self):
 
                 # Desired port speed on the host side
                 host_speed = speed
-                host_lane_count = len(lanes.split(','))
+                host_lane_count = self.get_host_lane_count(lport, lanes, gearbox_lanes_dict)
 
                 # double-check the HW presence before moving forward
                 sfp = platform_chassis.get_sfp(pport)

sonic-xcvrd/xcvrd/xcvrd_utilities/xcvr_table_helper.py

@@ -56,6 +56,7 @@ def __init__(self, namespaces):
         self.int_tbl, self.dom_tbl, self.dom_threshold_tbl, self.status_tbl, self.app_port_tbl, \
 		self.cfg_port_tbl, self.state_port_tbl, self.pm_tbl, self.firmware_info_tbl = {}, {}, {}, {}, {}, {}, {}, {}, {}
         self.state_db = {}
+        self.appl_db = {}
         self.cfg_db = {}
         self.dom_flag_tbl = {}
         self.dom_flag_change_count_tbl = {}
@@ -92,8 +93,8 @@ def __init__(self, namespaces):
             self.pm_tbl[asic_id] = swsscommon.Table(self.state_db[asic_id], TRANSCEIVER_PM_TABLE)
             self.firmware_info_tbl[asic_id] = swsscommon.Table(self.state_db[asic_id], TRANSCEIVER_FIRMWARE_INFO_TABLE)
             self.state_port_tbl[asic_id] = swsscommon.Table(self.state_db[asic_id], swsscommon.STATE_PORT_TABLE_NAME)
-            appl_db = daemon_base.db_connect("APPL_DB", namespace)
-            self.app_port_tbl[asic_id] = swsscommon.ProducerStateTable(appl_db, swsscommon.APP_PORT_TABLE_NAME)
+            self.appl_db[asic_id] = daemon_base.db_connect("APPL_DB", namespace)
+            self.app_port_tbl[asic_id] = swsscommon.ProducerStateTable(self.appl_db[asic_id], swsscommon.APP_PORT_TABLE_NAME)
             self.cfg_db[asic_id] = daemon_base.db_connect("CONFIG_DB", namespace)
             self.cfg_port_tbl[asic_id] = swsscommon.Table(self.cfg_db[asic_id], swsscommon.CFG_PORT_TABLE_NAME)
             self.vdm_real_value_tbl[asic_id] = swsscommon.Table(self.state_db[asic_id], TRANSCEIVER_VDM_REAL_VALUE_TABLE)
@@ -238,3 +239,52 @@ def is_npu_si_settings_update_required(self, lport, port_mapping):
 
         # If npu_si_settings_sync_val is None, it can also mean that the key is not present in the table
         return npu_si_settings_sync_val is None or npu_si_settings_sync_val == NPU_SI_SETTINGS_DEFAULT_VALUE
+
+    def get_gearbox_line_lanes_dict(self):
+        """
+        Retrieves the gearbox line lanes dictionary from APPL_DB
+
+        This method scans all ASICs for gearbox interface configurations and extracts
+        the line_lanes count for each logical port. The line_lanes represent the
+        number of lanes on the line side (towards the optical module) which is the
+        correct count to use for CMIS host lane configuration.
+
+        Returns:
+            dict: A dictionary where:
+                - key (str): logical port name (e.g., "Ethernet0")
+                - value (int): number of line-side lanes for that port
+
+        Example:
+            {"Ethernet0": 2, "Ethernet200": 4}
+
+        Note:
+            - Returns empty dict if no gearbox configuration is found
+            - Silently skips invalid or malformed entries
+            - Only processes keys that start with "interface:"
+        """
+        gearbox_line_lanes_dict = {}
+        try:
+            for asic_id in self.appl_db:
+                appl_db = self.appl_db[asic_id]
+                gearbox_table = swsscommon.Table(appl_db, "_GEARBOX_TABLE")
+                interface_keys = gearbox_table.getKeys()
+                for key in interface_keys:
+                    if key.startswith("interface:"):
+                        (found, fvs) = gearbox_table.get(key)
+                        if found:
+                            fvs_dict = dict(fvs)
+                            interface_name = fvs_dict.get('name', '')
+                            line_lanes_str = fvs_dict.get('line_lanes', '')
+                            if interface_name and line_lanes_str:
+                                line_lanes_count = len(line_lanes_str.split(','))
+                                gearbox_line_lanes_dict[interface_name] = line_lanes_count
+                            else:
+                                if not interface_name:
+                                    helper_logger.log_warning("get_gearbox_line_lanes_dict: ASIC {}: Interface {} missing 'name' field".format(asic_id, key))
+                                if not line_lanes_str:
+                                    helper_logger.log_debug("get_gearbox_line_lanes_dict: ASIC {}: Interface {} has empty 'line_lanes' field".format(asic_id, interface_name))
+        except Exception as e:
+            helper_logger.log_error("Error in get_gearbox_line_lanes_dict: {}".format(str(e)))
+            return gearbox_line_lanes_dict
+
+        return gearbox_line_lanes_dict