[Mellanox] do not use system clock to avoid issues caused by system time change (sonic-net#21446)

Junchao-Mellanox · BYGX-wcr · commit ed9f50dc8475 · 2025-01-21T18:43:32.000Z
- Why I did it
Mellanox platform API uses standard python time function time.time() in many places. time.time() gets time from system clock which could be changed by NTP or user. Adjusting system clock will affect the code logical and causes bugs. For example, in platform/mellanox/mlnx-platform-api/sonic_platform/utils.py there is a Timer class, the timer will trigger event with unexpected interval if user/NTP changes the system clock. This PR changes time.time() to time.monotonic to avoid such issue.

- How I did it
Use time.monotonic() instead of time.time .

- How to verify it
Manual test.
Unit test.
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/chassis.py b/platform/mellanox/mlnx-platform-api/sonic_platform/chassis.py
@@ -1,6 +1,6 @@
 #
 # SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES
-# Copyright (c) 2019-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# Copyright (c) 2019-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -447,7 +447,7 @@ def get_change_event_for_module_host_management_mode(self, timeout):
         timeout = 1000.0 if timeout >= 1000 else float(timeout)
         port_dict = {}
         error_dict = {}
-        begin = time.time()
+        begin = time.monotonic()
         wait_ready_task = sfp.SFP.get_wait_ready_task()
         
         while True:        
@@ -524,7 +524,7 @@ def get_change_event_for_module_host_management_mode(self, timeout):
                 }
             else:
                 if not wait_forever:
-                    elapse = time.time() - begin
+                    elapse = time.monotonic() - begin
                     if elapse * 1000 >= timeout:
                         return True, {'sfp': {}}
 
@@ -569,7 +569,7 @@ def get_change_event_legacy(self, timeout):
         timeout = 1000.0 if timeout >= 1000 else float(timeout)
         port_dict = {}
         error_dict = {}
-        begin = time.time()
+        begin = time.monotonic()
         
         while True:
             fds_events = self.poll_obj.poll(timeout)
@@ -619,7 +619,7 @@ def get_change_event_legacy(self, timeout):
                 }
             else:
                 if not wait_forever:
-                    elapse = time.time() - begin
+                    elapse = time.monotonic() - begin
                     if elapse * 1000 >= timeout:
                         return True, {'sfp': {}}
 
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/dpuctlplat.py b/platform/mellanox/mlnx-platform-api/sonic_platform/dpuctlplat.py
@@ -1,6 +1,6 @@
 #
 # SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES
-# Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -206,8 +206,8 @@ def wait_for_pci(self):
                     return True
                 poll_obj = poll()
                 poll_obj.register(dir_fd, POLLIN)
-                start = time.time()
-                while (time.time() - start) < WAIT_FOR_PCI_DEV:
+                start = time.monotonic()
+                while (time.monotonic() - start) < WAIT_FOR_PCI_DEV:
                     events = poll_obj.poll(WAIT_FOR_PCI_DEV * 1000)
                     if events:
                         if os.path.exists(os.path.dirname(self.get_pci_dev_path())):
@@ -473,9 +473,9 @@ def boot_prog_context(self):
     @contextmanager
     def time_check_context(self, msg):
         if self.verbosity:
-            start_time = time.time()
+            start_time = time.monotonic()
             yield
-            end_time = time.time()
+            end_time = time.monotonic()
             self.log_info(f"Total time taken = {end_time - start_time} for {msg}")
             return
         yield
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/sfp.py b/platform/mellanox/mlnx-platform-api/sonic_platform/sfp.py
@@ -1,6 +1,6 @@
 #
 # SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES
-# Copyright (c) 2019-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# Copyright (c) 2019-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -1475,14 +1475,14 @@ def initialize_sfp_modules(cls, sfp_list):
             # Resetting SFP requires a reloading of module firmware, it takes up to 3 seconds
             # according to standard
             max_wait_time = 3.5
-            begin = time.time()
+            begin = time.monotonic()
             while True:
                 ready_sfp_set = wait_ready_task.get_ready_set()
                 for sfp_index in ready_sfp_set:
                     s = sfp_list[sfp_index]
                     logger.log_debug(f'SFP {sfp_index} is recovered from resetting state')
                     s.on_event(EVENT_RESET_DONE)
-                elapse = time.time() - begin
+                elapse = time.monotonic() - begin
                 if elapse < max_wait_time:
                     time.sleep(0.5)
                 else:
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/utils.py b/platform/mellanox/mlnx-platform-api/sonic_platform/utils.py
@@ -1,6 +1,6 @@
 #
 # SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES
-# Copyright (c) 2020-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# Copyright (c) 2020-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -340,7 +340,7 @@ def schedule(self, interval, cb, repeat=True, run_now=True):
         self.add_timer_event(timer_event, run_now)
 
     def add_timer_event(self, timer_event, run_now=True):
-        timestamp = time.time()
+        timestamp = time.monotonic()
         if not run_now:
             timestamp += timer_event.interval
 
@@ -356,7 +356,7 @@ def stop(self):
 
     def run(self):
         while not self._stop_event.is_set():
-            now = time.time()
+            now = time.monotonic()
             item = self._timestamp_queue.get()
             self._min_timestamp = item[0]
             if self._min_timestamp > now:
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/wait_sfp_ready_task.py b/platform/mellanox/mlnx-platform-api/sonic_platform/wait_sfp_ready_task.py
@@ -1,5 +1,6 @@
 #
-# Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES
+# Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -65,7 +66,7 @@ def schedule_wait(self, sfp_index):
             is_empty = len(self._wait_dict) == 0
   
             # The item will be expired in 3 seconds
-            self._wait_dict[sfp_index] = time.time() + self.WAIT_TIME
+            self._wait_dict[sfp_index] = time.monotonic() + self.WAIT_TIME
 
         if is_empty:
             logger.log_debug('An item arrives, wake up WaitSfpReadyTask')
@@ -120,7 +121,7 @@ def run(self):
                 self.event.wait()
                 self.event.clear()
 
-            now = time.time()
+            now = time.monotonic()
             with self.lock:
                 logger.log_debug(f'Processing wait SFP dict: {self._wait_dict}, now={now}')
                 for sfp_index, expire_time in self._wait_dict.items():
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/watchdog.py b/platform/mellanox/mlnx-platform-api/sonic_platform/watchdog.py
@@ -1,5 +1,6 @@
 #
-# Copyright (c) 2019-2024 NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES
+# Copyright (c) 2019-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -229,7 +230,7 @@ def arm(self, seconds):
         # Save the watchdog arm timestamp
         # requiered for get_remaining_time()
         os.makedirs('/tmp/nvidia', exist_ok=True)
-        utils.write_file(self.TIMESTAMP_FILE, str(time.time()))
+        utils.write_file(self.TIMESTAMP_FILE, str(time.monotonic()))
 
         return ret
 
@@ -244,7 +245,7 @@ def get_remaining_time(self):
 
         if self.is_armed():
             arm_timestamp = utils.read_float_from_file(self.TIMESTAMP_FILE)
-            timeleft = int(self.timeout - (time.time() - arm_timestamp))
+            timeleft = int(self.timeout - (time.monotonic() - arm_timestamp))
 
         return timeleft
 
diff --git a/platform/mellanox/mlnx-platform-api/tests/test_change_event.py b/platform/mellanox/mlnx-platform-api/tests/test_change_event.py
@@ -1,5 +1,6 @@
 #
-# Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES
+# Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -34,7 +35,7 @@
 class TestChangeEvent:
     @mock.patch('sonic_platform.sfp.SFP.get_fd_for_polling_legacy')
     @mock.patch('select.poll')
-    @mock.patch('time.time')
+    @mock.patch('time.monotonic')
     @mock.patch('sonic_platform.device_data.DeviceDataManager.is_module_host_management_mode', mock.MagicMock(return_value=False))
     @mock.patch('sonic_platform.device_data.DeviceDataManager.get_sfp_count', mock.MagicMock(return_value=1))
     @mock.patch('sonic_platform.chassis.extract_RJ45_ports_index', mock.MagicMock(return_value=[]))
@@ -87,7 +88,7 @@ def test_get_change_event_legacy(self, mock_status, mock_time, mock_create_poll,
     @mock.patch('sonic_platform.wait_sfp_ready_task.WaitSfpReadyTask.get_ready_set')    
     @mock.patch('sonic_platform.sfp.SFP.get_fd')
     @mock.patch('select.poll')
-    @mock.patch('time.time')
+    @mock.patch('time.monotonic')
     @mock.patch('sonic_platform.device_data.DeviceDataManager.is_module_host_management_mode', mock.MagicMock(return_value=True))
     @mock.patch('sonic_platform.device_data.DeviceDataManager.get_sfp_count', mock.MagicMock(return_value=1))
     @mock.patch('sonic_platform.chassis.extract_RJ45_ports_index', mock.MagicMock(return_value=[]))
diff --git a/platform/mellanox/mlnx-platform-api/tests/test_dpuctlplat.py b/platform/mellanox/mlnx-platform-api/tests/test_dpuctlplat.py
@@ -1,6 +1,6 @@
 #
 # SPDX-FileCopyrightText: NVIDIA CORPORATION & AFFILIATES
-# Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # Apache-2.0
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -461,7 +461,7 @@ def mock_time_diff():
             mock_time_diff.counter += 1
             return mock_time_diff.counter * timeout_val
         mock_time_diff.counter = 0
-        with patch("time.time", wraps=mock_time_diff):
+        with patch("time.monotonic", wraps=mock_time_diff):
             # PCI Device is not recognized
             assert not dpuctl_obj.wait_for_pci()
             pci_parent_path = os.path.dirname(pci_dev_path)
