[Mellanox] mlnx-sfpd init flow enhancement (#3294)

* fix sfpd initialize issue
* fix review comments
* rephrase the output log
* fix retry counter
* change the retry time to 10, means set max waiting time 1024s
* fix mlnx-sfpd init flow with new solution
* [mlnx-sfpd] address comments
1. wait for 5 seconds * 30 times, 150 seconds totally. use constant wait time for each retry.
2. use try/except structure so that error can be handled in a graceful way
* [mlnx-sfpd] wait 5 seconds after SDK_DAEMON_READY_FILE exists to make sure SDK is fully up.
* [mlnx-sfpd]simplify initialization by using deinitialize on initializing failure

platform/mellanox/mlnx-sfpd/scripts/mlnx-sfpd

@@ -30,6 +30,8 @@ STATUS_UNKNOWN = '2'
 
 SFPD_LIVENESS_EXPIRE_SECS = 30
 
+SDK_DAEMON_READY_FILE = '/tmp/sdk_ready'
+
 sfp_value_status_dict = {
         SDK_SFP_STATE_IN:  STATUS_PLUGIN,
         SDK_SFP_STATE_OUT: STATUS_PLUGOUT,
@@ -64,7 +66,8 @@ def log_error(msg, also_print_to_console=False):
 class MlnxSfpd:
     ''' Listen to plugin/plugout cable events '''
 
-    SX_OPEN_RETRIES = 20
+    SX_OPEN_RETRIES = 30
+    SX_OPEN_TIMEOUT = 5
     SELECT_TIMEOUT = 1
 
     def __init__(self):
@@ -75,7 +78,6 @@ class MlnxSfpd:
         # Allocate SDK fd and user channel structures
         self.rx_fd_p = new_sx_fd_t_p()
         self.user_channel_p = new_sx_user_channel_t_p()
-
         self.state_db = SonicV2Connector(host=REDIS_HOSTIP)
 
         # Register our signal handlers
@@ -98,37 +100,78 @@ class MlnxSfpd:
     def initialize(self):
         self.state_db.connect("STATE_DB")
 
-        # open SDK API handle
+        swid_cnt_p = None
-        # retry at most SX_OPEN_RETRIES times to wait
+
-        # until SDK is started during system startup
+        try:
-        retry = 1
+            # Wait for SDK daemon to be started with detect the sdk_ready file
-        while True:
+            retry = 0
+            while not os.path.exists(SDK_DAEMON_READY_FILE):  
+                if retry >= self.SX_OPEN_RETRIES:
+                    raise RuntimeError("SDK daemon failed to start after {} retries and {} seconds waiting, exiting..."
+                        .format(retry, self.SX_OPEN_TIMEOUT * self.SX_OPEN_RETRIES))
+                else:
+                    log_info("SDK daemon not started yet, retry {} times".format(retry))
+                    retry = retry + 1
+                    time.sleep(self.SX_OPEN_TIMEOUT)
+
+            # to make sure SDK daemon has started
+            time.sleep(self.SX_OPEN_TIMEOUT)
+
+            # After SDK daemon started, sx_api_open and sx_api_host_ifc_open is ready for call
             rc, self.handle = sx_api_open(None)
-            if rc == SX_STATUS_SUCCESS:
+            if rc != SX_STATUS_SUCCESS:
-                break
+                raise RuntimeError("failed to call sx_api_open with rc {}, exiting...".format(rc))
 
-            log_warning("failed to open SDK API handle... retrying {}".format(retry))
+            rc = sx_api_host_ifc_open(self.handle, self.rx_fd_p)
+            if rc != SX_STATUS_SUCCESS:
+                raise RuntimeError("failed to call sx_api_host_ifc_open with rc {}, exiting...".format(rc))
 
-            time.sleep(2 ** retry)
+            self.user_channel_p.type = SX_USER_CHANNEL_TYPE_FD
-            retry += 1
+            self.user_channel_p.channel.fd = self.rx_fd_p
 
-            if retry > self.SX_OPEN_RETRIES:
+            # Wait for switch to be created and initialized inside SDK
-                raise RuntimeError("failed to open SDK API handle after {} retries".format(retry))
+            retry = 0
+            swid_cnt_p = new_uint32_t_p()
+            uint32_t_p_assign(swid_cnt_p, 0)
+            swid_cnt = 0
+            while True:
+                if retry >= self.SX_OPEN_RETRIES:
+                    raise RuntimeError("switch not created after {} retries and {} seconds waiting, exiting..."
+                        .format(retry, self.SX_OPEN_RETRIES * self.SX_OPEN_TIMEOUT))
+                else:
+                    rc = sx_api_port_swid_list_get(self.handle, None, swid_cnt_p)
+                    if rc == SX_STATUS_SUCCESS:
+                        swid_cnt = uint32_t_p_value(swid_cnt_p)
+                        if swid_cnt > 0:
+                            delete_uint32_t_p(swid_cnt_p)
+                            swid_cnt_p = None
+                            break
+                        else:
+                            log_info("switch not created yet, swid_cnt {}, retry {} times and wait for {} seconds"
+                                .format(swid_cnt, retry, self.SX_OPEN_TIMEOUT * retry))
+                    else:
+                        raise RuntimeError("sx_api_port_swid_list_get fail with rc {}, retry {} times and wait for {} seconds".
+                            format(rc, retry, self.SX_OPEN_TIMEOUT * retry))
 
-        rc = sx_api_host_ifc_open(self.handle, self.rx_fd_p)
+                    retry = retry + 1
-        if rc != SX_STATUS_SUCCESS:
+                    time.sleep(self.SX_OPEN_TIMEOUT)
-            raise RuntimeError("sx_api_host_ifc_open exited with error, rc {}".format(rc))
 
-        self.user_channel_p.type = SX_USER_CHANNEL_TYPE_FD
+            # After switch was created inside SDK, sx_api_host_ifc_trap_id_register_set is ready to call
-        self.user_channel_p.channel.fd = self.rx_fd_p
+            rc = sx_api_host_ifc_trap_id_register_set(self.handle,
+                                                    SX_ACCESS_CMD_REGISTER,
+                                                    self.swid,
+                                                    SX_TRAP_ID_PMPE,
+                                                    self.user_channel_p)
 
-        rc = sx_api_host_ifc_trap_id_register_set(self.handle,
+            if rc != SX_STATUS_SUCCESS:
-                                                  SX_ACCESS_CMD_REGISTER,
+                raise RuntimeError("sx_api_host_ifc_trap_id_register_set failed with rc {}, exiting...".format(rc))
-                                                  self.swid,
+
-                                                  SX_TRAP_ID_PMPE,
+            self.running = True
-                                                  self.user_channel_p)
+        except Exception as e:
-        if rc != SX_STATUS_SUCCESS:
+            log_error("mlnx-sfpd initialization failed due to {}, exiting...".format(repr(e)))
-            raise RuntimeError("sx_api_host_ifc_trap_id_register_set exited with error, rc {}".format(c))
+            if swid_cnt_p is not None:
+                delete_uint32_t_p(swid_cnt_p)
+            self.deinitialize()
 
     def deinitialize(self):
         # remove mlnx-sfpd liveness key in DB if not expired yet
@@ -156,7 +199,6 @@ class MlnxSfpd:
             log_error("sx_api_close exited with error, rc {}".format(rc))
 
     def run(self):
-        self.running = True
 
         while self.running:
             try: