[action] [PR:19473] [Mellanox] implement state machine for always firmware control ports (#19473)

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

@@ -126,6 +126,9 @@
                 "comex_amb": False,
                 "pch_temp": True
             }
+        },
+        'sfp': {
+            'fw_control_ports': [64, 65]  # 0 based sfp index list
         }
     },
     'x86_64-nvidia_sn5600-r0': {
@@ -134,6 +137,9 @@
                 "comex_amb": False,
                 "pch_temp": True
             }
+        },
+        'sfp': {
+            'fw_control_ports': [64]  # 0 based sfp index list
         }
     }
 }
@@ -299,3 +305,16 @@ def get_watchdog_max_period(cls):
             return DEFAULT_WD_PERIOD
 
         return watchdog_data.get('max_period', None)
+
+    @classmethod
+    @utils.read_only_cache()
+    def get_always_fw_control_ports(cls):
+        platform_data = DEVICE_DATA.get(cls.get_platform_name())
+        if not platform_data:
+            return None
+
+        sfp_data = platform_data.get('sfp')
+        if not sfp_data:
+            return None
+
+        return sfp_data.get('fw_control_ports')

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

@@ -253,6 +253,14 @@
 ACTION_ON_FW_CONTROL = 'On Firmware Control'
 ACTION_ON_POWER_LIMIT_ERROR = 'On Power Limit Error'
 ACTION_ON_CANCEL_WAIT = 'On Cancel Wait'
+
+# States/actions for always firmware control ports
+STATE_FCP_DOWN = 'Down(Firmware Control)'
+STATE_FCP_INIT = 'Initializing(Firmware Control)'
+STATE_FCP_NOT_PRESENT = 'Not Present(Firmware Control)'
+STATE_FCP_PRESENT = 'Present(Firmware Control)'
+
+ACTION_FCP_ON_START = 'On Start(Firmware Control)'
 # Module host management definitions end
 
 # SFP EEPROM limited bytes
@@ -463,7 +471,11 @@ def __init__(self, sfp_index, sfp_type=None, slot_id=0, linecard_port_count=0, l
         self.slot_id = slot_id
         self._sfp_type_str = None
         # SFP state, only applicable for module host management
-        self.state = STATE_DOWN
+        fw_control_ports = DeviceDataManager.get_always_fw_control_ports()
+        if not fw_control_ports or self.sdk_index not in fw_control_ports:
+            self.state = STATE_DOWN
+        else:
+            self.state = STATE_FCP_DOWN
 
     def __str__(self):
         return f'SFP {self.sdk_index}'
@@ -1431,7 +1443,7 @@ def get_state_machine(cls):
             sm.add_state(STATE_POWER_LIMIT_ERROR).set_entry_action(ACTION_ON_POWER_LIMIT_ERROR) \
               .add_transition(EVENT_POWER_GOOD, STATE_POWERED_ON) \
               .add_transition(EVENT_NOT_PRESENT, STATE_NOT_PRESENT)
-
+                          
             cls.action_table = {}
             cls.action_table[ACTION_ON_START] = cls.action_on_start
             cls.action_table[ACTION_ON_RESET] = cls.action_on_reset
@@ -1441,6 +1453,16 @@ def get_state_machine(cls):
             cls.action_table[ACTION_ON_CANCEL_WAIT] = cls.action_on_cancel_wait
             cls.action_table[ACTION_ON_POWER_LIMIT_ERROR] = cls.action_on_power_limit_error
 
+            # For always firewire control ports
+            sm.add_state(STATE_FCP_DOWN).add_transition(EVENT_START, STATE_FCP_INIT)
+            sm.add_state(STATE_FCP_INIT).set_entry_action(ACTION_FCP_ON_START) \
+              .add_transition(EVENT_NOT_PRESENT, STATE_FCP_NOT_PRESENT) \
+              .add_transition(EVENT_PRESENT, STATE_FCP_PRESENT)
+            sm.add_state(STATE_FCP_NOT_PRESENT).add_transition(EVENT_PRESENT, STATE_FCP_PRESENT)
+            sm.add_state(STATE_FCP_PRESENT).add_transition(EVENT_NOT_PRESENT, STATE_FCP_NOT_PRESENT)
+
+            cls.action_table[ACTION_FCP_ON_START] = cls.action_fcp_on_start
+
             cls.sm = sm
 
         return cls.sm
@@ -1469,6 +1491,14 @@ def action_on_start(cls, sfp):
                 sfp.on_event(EVENT_RESET)
             else:
                 sfp.on_event(EVENT_POWER_ON)
+
+    @classmethod
+    def action_fcp_on_start(cls, sfp):
+        present = utils.read_int_from_file(f'/sys/module/sx_core/asic0/module{sfp.sdk_index}/present')
+        if present:
+            sfp.on_event(EVENT_PRESENT)
+        else:
+            sfp.on_event(EVENT_NOT_PRESENT)
 
     @classmethod
     def action_on_reset(cls, sfp):
@@ -1565,10 +1595,12 @@ def in_stable_state(self):
         Returns:
             bool: True if the module is in a stable state
         """
-        return self.state in (STATE_NOT_PRESENT, STATE_SW_CONTROL, STATE_FW_CONTROL, STATE_POWER_BAD, STATE_POWER_LIMIT_ERROR)
+        return self.state in (STATE_NOT_PRESENT, STATE_SW_CONTROL, STATE_FW_CONTROL, 
+                              STATE_POWER_BAD, STATE_POWER_LIMIT_ERROR, STATE_FCP_NOT_PRESENT,
+                              STATE_FCP_PRESENT)
 
     def get_fds_for_poling(self):            
-        if self.state == STATE_FW_CONTROL:
+        if self.state == STATE_FW_CONTROL or self.state == STATE_FCP_NOT_PRESENT or self.state == STATE_FCP_PRESENT:
             return {
                 'present': self.get_fd('present')
             } 
@@ -1584,11 +1616,9 @@ def fill_change_event(self, port_dict):
         Args:
             port_dict (dict): {<sfp_index>:<sfp_state>}
         """
-        if self.state == STATE_NOT_PRESENT:
+        if self.state == STATE_NOT_PRESENT or self.state == STATE_FCP_NOT_PRESENT:
             port_dict[self.sdk_index + 1] = SFP_STATUS_REMOVED
-        elif self.state == STATE_SW_CONTROL:
-            port_dict[self.sdk_index + 1] = SFP_STATUS_INSERTED
-        elif self.state == STATE_FW_CONTROL:
+        elif self.state == STATE_SW_CONTROL or self.state == STATE_FW_CONTROL or self.state == STATE_FCP_PRESENT:
             port_dict[self.sdk_index + 1] = SFP_STATUS_INSERTED
         elif self.state == STATE_POWER_BAD or self.state == STATE_POWER_LIMIT_ERROR:
             sfp_state = SFP.SFP_ERROR_BIT_POWER_BUDGET_EXCEEDED | SFP.SFP_STATUS_BIT_INSERTED
@@ -1607,7 +1637,7 @@ def refresh_poll_obj(self, poll_obj, all_registered_fds):
         # find fds registered by this SFP
         current_registered_fds = {item[2]: (fileno, item[1]) for fileno, item in all_registered_fds.items() if item[0] == self.sdk_index}
         logger.log_debug(f'SFP {self.sdk_index} registered fds are: {current_registered_fds}')
-        if self.state == STATE_FW_CONTROL:
+        if self.state == STATE_FW_CONTROL or self.state == STATE_FCP_NOT_PRESENT or self.state == STATE_FCP_PRESENT:
             target_poll_types = ['present']
         else:
             target_poll_types = ['hw_present', 'power_good']
@@ -1643,9 +1673,10 @@ def is_dummy_event(self, fd_type, fd_value):
         """
         if fd_type == 'hw_present' or fd_type == 'present':
             if fd_value == int(SFP_STATUS_INSERTED):
-                return self.state in (STATE_SW_CONTROL, STATE_FW_CONTROL, STATE_POWER_BAD, STATE_POWER_LIMIT_ERROR)
+                return self.state in (STATE_SW_CONTROL, STATE_FW_CONTROL, STATE_POWER_BAD,
+                                      STATE_POWER_LIMIT_ERROR, STATE_FCP_PRESENT)
             elif fd_value == int(SFP_STATUS_REMOVED):
-                return self.state == STATE_NOT_PRESENT
+                return self.state in (STATE_NOT_PRESENT, STATE_FCP_NOT_PRESENT)
         elif fd_type == 'power_good':
             if fd_value == 1:
                 return self.state in (STATE_SW_CONTROL, STATE_NOT_PRESENT, STATE_RESETTING)

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

@@ -168,3 +168,16 @@ def test_sw_control(self):
         s.disable_tx_for_sff_optics = mock.MagicMock()
         s.on_event(sfp.EVENT_START)
         assert s.get_state() == sfp.STATE_SW_CONTROL
+
+    @mock.patch('sonic_platform.device_data.DeviceDataManager.get_always_fw_control_ports', mock.MagicMock(return_value=[0]))
+    def test_fcp_state(self):
+        self.mock_value('present', 1)
+        s = sfp.SFP(0)
+        s.on_event(sfp.EVENT_START)
+        assert s.get_state() == sfp.STATE_FCP_PRESENT
+
+        self.mock_value('present', 0)
+        s = sfp.SFP(0)
+        s.on_event(sfp.EVENT_START)
+        assert s.get_state() == sfp.STATE_FCP_NOT_PRESENT
+