Merge branch 'master' of https://github.com/sonic-net/sonic-platform-daemons into vdm_read_through_xcvrd

Author: mihirpat1

sonic-chassisd/scripts/chassisd

@@ -17,7 +17,7 @@ try:
     import time
     import json
     import glob
-    from datetime import datetime
+    from datetime import datetime, timezone
 
     from sonic_py_common import daemon_base, logger, device_info
     from sonic_py_common.task_base import ProcessTaskBase
@@ -136,6 +136,16 @@ def get_chassis():
         self.log_error("Failed to load chassis due to {}".format(repr(e)))
         sys.exit(CHASSIS_LOAD_ERROR)
 
+def get_formatted_time(datetimeobj=None, op_format=None):
+    """
+    Get the current time in specified format
+    :param datetimeobj: Optional - A datetime object already initialized with a specific time
+    :param op_format: Optional - Output Format for the time to be displayed
+    :returns time in string format
+    """
+    date_obj = datetimeobj if datetimeobj else datetime.now(timezone.utc)
+    return date_obj.strftime(op_format if op_format else "%a %b %d %I:%M:%S %p UTC %Y")
+
 #
 # Module Config Updater ========================================================
 #
@@ -765,28 +775,22 @@ class SmartSwitchModuleUpdater(ModuleUpdater):
 
     def update_dpu_state(self, key, state):
         """
-        Update DPU state in chassisStateDB using the given key.
+        Update specific DPU state fields in chassisStateDB using the given key.
         """
         try:
             # Connect to the CHASSIS_STATE_DB using daemon_base
             if not self.chassis_state_db:
                 self.chassis_state_db = daemon_base.db_connect("CHASSIS_STATE_DB")
 
-            # Fetch the current data for the given key and convert it to a dict
-            current_data = self._convert_to_dict(self.chassis_state_db.hgetall(key))
-
-            if current_data:
-                self.chassis_state_db.delete(key)
-
-            # Prepare the updated data
+            # Prepare the fields to update
             updates = {
                 "dpu_midplane_link_state": state,
                 "dpu_midplane_link_reason": "",
-                "dpu_midplane_link_time": datetime.now().strftime("%a %b %d %I:%M:%S %p UTC %Y"),
+                "dpu_midplane_link_time": get_formatted_time(),
             }
-            current_data.update(updates)
 
-            for field, value in current_data.items():
+            # Update each field directly
+            for field, value in updates.items():
                 self.chassis_state_db.hset(key, field, value)
 
         except Exception as e:
@@ -818,7 +822,7 @@ class SmartSwitchModuleUpdater(ModuleUpdater):
 
     def _get_current_time_str(self):
         """Returns the current time as a string in 'YYYY_MM_DD_HH_MM_SS' format."""
-        return datetime.now().strftime("%Y_%m_%d_%H_%M_%S")
+        return get_formatted_time(op_format="%Y_%m_%d_%H_%M_%S")
 
     def _get_history_path(self, module, file_name):
         """Generates the full path for history files."""
@@ -880,11 +884,9 @@ class SmartSwitchModuleUpdater(ModuleUpdater):
 
         file_path = self._get_history_path(module, file_name)
         try:
-            dt_obj = datetime.strptime(prev_reboot_time, "%Y_%m_%d_%H_%M_%S")
+            formatted_time = get_formatted_time(datetimeobj=datetime.strptime(prev_reboot_time, "%Y_%m_%d_%H_%M_%S"))
         except ValueError:
-            dt_obj = datetime.now()
-
-        formatted_time = dt_obj.strftime("%a %b %d %I:%M:%S %p UTC %Y")
+            formatted_time = get_formatted_time()
 
         reboot_cause_dict = {
             "cause": cause,
@@ -1188,7 +1190,7 @@ class DpuStateUpdater(logger.Logger):
         return True
 
     def _time_now(self):
-        return datetime.now().strftime('%Y-%m-%d %H:%M:%S')
+        return get_formatted_time()
 
     def _update_dp_dpu_state(self, state):
         self.dpu_state_table.hset(self.name, self.DP_STATE, state)

sonic-chassisd/tests/test_chassisd.py

@@ -1515,3 +1515,27 @@ def test_chassis_db_bootup_with_empty_slot():
     assert status == fvs[CHASSIS_MODULE_INFO_OPERSTATUS_FIELD]
     assert down_module_lc1_key in sup_module_updater.down_modules.keys()
 
+
+def test_smartswitch_time_format():
+    chassis = MockSmartSwitchChassis()
+    chassis_state_db = MagicMock()
+    mod_updater = SmartSwitchModuleUpdater(SYSLOG_IDENTIFIER, chassis)
+    mod_updater.chassis_state_db = chassis_state_db
+    mod_updater.chassis_state_db.hgetall = MagicMock(return_value={})
+    mod_updater.chassis_state_db.hset = MagicMock()
+    date_format = "%a %b %d %I:%M:%S %p UTC %Y"
+    def is_valid_date(date_str):
+            try:
+                datetime.strptime(date_str, date_format)
+            except ValueError:
+                # Parsing failed and we are unable to obtain the time
+                return False
+            return True
+    mod_updater.update_dpu_state("DPU1", 'up')
+    date_value = None
+    for args in (mod_updater.chassis_state_db.hset.call_args_list):
+        if args[0][0] == "DPU1" and args[0][1] == "dpu_midplane_link_time":
+            date_value = args[0][2]
+    if not date_value:
+        AssertionError("Date is not set!")
+    assert is_valid_date(date_value)

sonic-chassisd/tests/test_dpu_chassisd.py

@@ -5,6 +5,7 @@
 import signal
 import threading
 from imp import load_source
+import re
 
 from mock import MagicMock
 from sonic_py_common import daemon_base
@@ -75,14 +76,14 @@ def test_dpu_state_update_api(state, expected_state):
 
 @pytest.mark.parametrize('dpu_id, dp_state, cp_state, expected_state', [
     (0, False, False, {'DPU0': 
-        {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': '2000-01-01 00:00:00', 
-         'dpu_control_plane_state': 'down', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}),
+        {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000',
+         'dpu_control_plane_state': 'down', 'dpu_control_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000'}}),
     (0, False, True, {'DPU0': 
-        {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': '2000-01-01 00:00:00', 
-         'dpu_control_plane_state': 'up', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}),
+        {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000',
+         'dpu_control_plane_state': 'up', 'dpu_control_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000'}}),
     (0, True, True, {'DPU0': 
-        {'dpu_data_plane_state': 'up', 'dpu_data_plane_time': '2000-01-01 00:00:00', 
-         'dpu_control_plane_state': 'up', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}),
+        {'dpu_data_plane_state': 'up', 'dpu_data_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000',
+         'dpu_control_plane_state': 'up', 'dpu_control_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000'}}),
 ])
 def test_dpu_state_update(dpu_id, dp_state, cp_state, expected_state):
     chassis = MockDpuChassis()
@@ -102,7 +103,7 @@ def hset(key, field, value):
 
     with mock.patch.object(swsscommon.Table, 'hset', side_effect=hset) as hset_mock:
             dpu_updater = DpuStateUpdater(SYSLOG_IDENTIFIER, chassis)
-            dpu_updater._time_now = MagicMock(return_value='2000-01-01 00:00:00')
+            dpu_updater._time_now = MagicMock(return_value='Sat Jan 01 12:00:00 AM UTC 2000')
 
             dpu_updater.update_state()
 
@@ -112,20 +113,20 @@ def hset(key, field, value):
 
             # After the deinit we assume that the DPU state is down.
             assert chassis_state_db == {'DPU0': 
-                {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': '2000-01-01 00:00:00', 
-                 'dpu_control_plane_state': 'down', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}
+                {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000',
+                 'dpu_control_plane_state': 'down', 'dpu_control_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000'}}
 
 
 @pytest.mark.parametrize('dpu_id, dp_state, cp_state, expected_state', [
     (0, False, False, {'DPU0':
-        {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': '2000-01-01 00:00:00',
-         'dpu_control_plane_state': 'down', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}),
+        {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000',
+         'dpu_control_plane_state': 'down', 'dpu_control_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000'}}),
     (0, False, True, {'DPU0':
-        {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': '2000-01-01 00:00:00',
-         'dpu_control_plane_state': 'up', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}),
+        {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000',
+         'dpu_control_plane_state': 'up', 'dpu_control_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000'}}),
     (0, True, True, {'DPU0':
-        {'dpu_data_plane_state': 'up', 'dpu_data_plane_time': '2000-01-01 00:00:00',
-         'dpu_control_plane_state': 'up', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}),
+        {'dpu_data_plane_state': 'up', 'dpu_data_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000',
+         'dpu_control_plane_state': 'up', 'dpu_control_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000'}}),
 ])
 def test_dpu_state_manager(dpu_id, dp_state, cp_state, expected_state):
     chassis = MockDpuChassis()
@@ -146,7 +147,7 @@ def hset(key, field, value):
     with mock.patch.object(swsscommon.Table, 'hset', side_effect=hset):
         with mock.patch.object(swsscommon.Select, 'select', side_effect=((swsscommon.Select.OBJECT, None), (swsscommon.Select.OBJECT, None), KeyboardInterrupt)):
             dpu_updater = DpuStateUpdater(SYSLOG_IDENTIFIER, chassis)
-            dpu_updater._time_now = MagicMock(return_value='2000-01-01 00:00:00')
+            dpu_updater._time_now = MagicMock(return_value='Sat Jan 01 12:00:00 AM UTC 2000')
 
             dpu_state_mng = DpuStateManagerTask(SYSLOG_IDENTIFIER, dpu_updater)
 
@@ -158,8 +159,8 @@ def hset(key, field, value):
 
             # After the deinit we assume that the DPU state is down.
             assert chassis_state_db == {'DPU0':
-                {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': '2000-01-01 00:00:00',
-                 'dpu_control_plane_state': 'down', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}
+                {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000',
+                 'dpu_control_plane_state': 'down', 'dpu_control_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000'}}
 
 
 def test_dpu_chassis_daemon():
@@ -180,7 +181,7 @@ def hset(key, field, value):
         chassis_state_db[key][field] = value
 
     with mock.patch.object(swsscommon.Table, 'hset', side_effect=hset) as hset_mock:
-            with mock.patch.object(DpuStateUpdater, '_time_now', side_effect=lambda: '2000-01-01 00:00:00') as mock_time_now:
+            with mock.patch.object(DpuStateUpdater, '_time_now', side_effect=lambda: 'Sat Jan 01 12:00:00 AM UTC 2000') as mock_time_now:
 
                 daemon_chassisd = DpuChassisdDaemon(SYSLOG_IDENTIFIER, chassis)
                 daemon_chassisd.CHASSIS_INFO_UPDATE_PERIOD_SECS = MagicMock(return_value=1)
@@ -195,14 +196,40 @@ def hset(key, field, value):
                 time.sleep(3)
 
                 assert chassis_state_db == {'DPU1':
-                    {'dpu_data_plane_state': 'up', 'dpu_data_plane_time': '2000-01-01 00:00:00',
-                    'dpu_control_plane_state': 'up', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}
+                    {'dpu_data_plane_state': 'up', 'dpu_data_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000',
+                    'dpu_control_plane_state': 'up', 'dpu_control_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000'}}
 
                 daemon_chassisd.signal_handler(signal.SIGINT, None)
                 daemon_chassisd.stop.wait.return_value = True
 
                 thread.join()
 
                 assert chassis_state_db == {'DPU1':
-                    {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': '2000-01-01 00:00:00',
-                    'dpu_control_plane_state': 'down', 'dpu_control_plane_time': '2000-01-01 00:00:00'}}
+                    {'dpu_data_plane_state': 'down', 'dpu_data_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000',
+                    'dpu_control_plane_state': 'down', 'dpu_control_plane_time': 'Sat Jan 01 12:00:00 AM UTC 2000'}}
+    with mock.patch.object(swsscommon.Table, 'hset', side_effect=hset):
+        daemon_chassisd = DpuChassisdDaemon(SYSLOG_IDENTIFIER, chassis)
+        daemon_chassisd.CHASSIS_INFO_UPDATE_PERIOD_SECS = MagicMock(return_value=1)
+
+        daemon_chassisd.stop = MagicMock()
+        daemon_chassisd.stop.wait.return_value = False
+
+        thread = threading.Thread(target=daemon_chassisd.run)
+        thread.start()
+        # Wait for thread to start and update DB
+        time.sleep(3)
+        date_format = "%a %b %d %I:%M:%S %p UTC %Y"
+
+        def is_valid_date(date_str):
+            try:
+                datetime.strptime(date_str, date_format)
+            except ValueError:
+                # Parsing failed and we are unable to obtain the time
+                return False
+            return True
+        assert is_valid_date(chassis_state_db['DPU1']['dpu_data_plane_time'])
+        assert is_valid_date(chassis_state_db['DPU1']['dpu_control_plane_time'])
+        daemon_chassisd.signal_handler(signal.SIGINT, None)
+        daemon_chassisd.stop.wait.return_value = True
+
+        thread.join()

sonic-pcied/scripts/pcied

@@ -27,6 +27,10 @@ SYSLOG_IDENTIFIER = "pcied"
 PCIE_RESULT_REGEX = "PCIe Device Checking All Test"
 PCIE_DEVICE_TABLE_NAME = "PCIE_DEVICE"
 PCIE_STATUS_TABLE_NAME = "PCIE_DEVICES"
+PCIE_DETACH_INFO_TABLE = "PCIE_DETACH_INFO"
+
+PCIE_DETACH_BUS_INFO_FIELD = "bus_info"
+PCIE_DETACH_DPU_STATE_FIELD = "dpu_state"
 
 PCIED_MAIN_THREAD_SLEEP_SECS = 60
 
@@ -92,6 +96,7 @@ class DaemonPcied(daemon_base.DaemonBase):
         self.state_db = daemon_base.db_connect("STATE_DB")
         self.device_table = swsscommon.Table(self.state_db, PCIE_DEVICE_TABLE_NAME)
         self.status_table = swsscommon.Table(self.state_db, PCIE_STATUS_TABLE_NAME)
+        self.detach_info = swsscommon.Table(self.state_db, PCIE_DETACH_INFO_TABLE)
 
     def __del__(self):
         if self.device_table:
@@ -102,6 +107,10 @@ class DaemonPcied(daemon_base.DaemonBase):
             stable_keys = self.status_table.getKeys()
             for stk in stable_keys:
                 self.status_table._del(stk)
+        if self.detach_info:
+            detach_info_keys = self.detach_info.getKeys()
+            for dk in detach_info_keys:
+                self.detach_info._del(dk)
 
     # load aer-fields into statedb
     def update_aer_to_statedb(self):
@@ -151,6 +160,28 @@ class DaemonPcied(daemon_base.DaemonBase):
 
         self.status_table.set("status", fvs)
 
+    # Check if any PCI interface is in detaching mode by querying the state_db
+    def is_dpu_in_detaching_mode(self, pcie_dev):
+        # Ensure detach_info is not None
+        if self.detach_info is None:
+            self.log_debug("detach_info is None")
+            return False
+
+        # Query the state_db for the device detaching status
+        detach_info_keys = list(self.detach_info.getKeys())
+        if not detach_info_keys:
+            return False
+
+        for key in detach_info_keys:
+            dpu_info = self.detach_info.get(key)
+            if dpu_info:
+                bus_info = dpu_info.get(PCIE_DETACH_BUS_INFO_FIELD)
+                dpu_state = dpu_info.get(PCIE_DETACH_DPU_STATE_FIELD)
+                if bus_info == pcie_dev and dpu_state == "detaching":
+                    return True
+
+        return False
+
     # Check the PCIe devices
     def check_pcie_devices(self):
         self.resultInfo = platform_pcieutil.get_pcie_check()
@@ -160,6 +191,14 @@ class DaemonPcied(daemon_base.DaemonBase):
 
         for result in self.resultInfo:
             if result["result"] == "Failed":
+                # Convert bus, device, and function to a bus_info format like "0000:03:00.0"
+                pcie_dev = "0000:{int(result['bus'], 16):02x}:{int(result['dev'], 16):02x}.{int(result['fn'], 16)}"
+
+                # Check if the device is in detaching mode
+                if device_info.is_smartswitch() and self.is_dpu_in_detaching_mode(pcie_dev):
+                    self.log_debug("PCIe Device: {} is in detaching mode, skipping warning.".format(pcie_dev))
+                    continue
+
                 self.log_warning("PCIe Device: " + result["name"] + " Not Found")
                 err += 1
             else: