fix sfp eeprom unreadable after switching from SW to FW control mode

Signed-off-by: Yuanzhe, Liu <yualiu@nvidia.com>
yuazhe · Jul 12, 2024 · b1c5927 · b1c5927
1 parent 11730d5
commit b1c5927
Show file tree

Hide file tree

Showing 2 changed files with 17 additions and 3 deletions.
diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/chassis.py b/platform/mellanox/mlnx-platform-api/sonic_platform/chassis.py
@@ -499,16 +499,20 @@ def get_change_event_for_module_host_management_mode(self, timeout):
                     s.on_event(event)
 
                 if s.in_stable_state():
+                    if s.get_control_type() == SFP_FW_CONTROL and not s.is_eeprom_ready(2):
+                            logger.log_error(f'SFP {sfp_index} eeprom is not readable')
                     s.fill_change_event(port_dict)
                     s.refresh_poll_obj(self.poll_obj, self.registered_fds)
                 else:
                     logger.log_debug(f'SFP {sfp_index} does not reach stable state, state={s.state}')
-                                        
+
             ready_sfp_set = wait_ready_task.get_ready_set()
             for sfp_index in ready_sfp_set:
                 s = self._sfp_list[sfp_index]
                 s.on_event(sfp.EVENT_RESET_DONE)
                 if s.in_stable_state():
+                    if s.get_control_type() == SFP_FW_CONTROL and not s.is_eeprom_ready(2):
+                        logger.log_error(f'SFP {sfp_index} eeprom is not readable')                  
                     s.fill_change_event(port_dict)
                     s.refresh_poll_obj(self.poll_obj, self.registered_fds)
                 else:

diff --git a/platform/mellanox/mlnx-platform-api/sonic_platform/sfp.py b/platform/mellanox/mlnx-platform-api/sonic_platform/sfp.py
@@ -488,6 +488,12 @@ def get_presence(self):
             return False
         eeprom_raw = self._read_eeprom(0, 1, log_on_error=False)
         return eeprom_raw is not None
+
+    def is_eeprom_ready(self, wait_time):
+        while wait_time > 0 and self._read_eeprom(0, 1, False) == None:
+            sleep(0.1)
+            wait_time -= 0.1
+        return self._read_eeprom(0, 1, False) == None:
 
     # read eeprom specfic bytes beginning from offset with size as num_bytes
     def read_eeprom(self, offset, num_bytes):
@@ -1691,8 +1697,12 @@ def initialize_sfp_modules(cls, sfp_list):
             if not s.in_stable_state():
                 logger.log_error(f'SFP {index} is not in stable state after initializing, state={s.state}')
             logger.log_notice(f'SFP {index} is in state {s.state} after module initialization')
-
-
+
+        # If the module is FW control, verify that it's ready
+        for s in sfp_list:
+            if not s.is_eeprom_ready(2):
+                logger.log_error(f'SFP {s.sdk_index} eeprom is not readable')
+
 class RJ45Port(NvidiaSFPCommon):
     """class derived from SFP, representing RJ45 ports"""