[docker-syncd-mlnx] add new mlnx-sfpd daemon to docker-syncd-mlnx (#1841)

* This new daemon will listen to the SDK for the SDK change event,
 * after got a SFP change event will publish a notification via DB.

	modified:   platform/mellanox/docker-syncd-mlnx-rpc.mk
	modified:   platform/mellanox/docker-syncd-mlnx.mk
	modified:   platform/mellanox/docker-syncd-mlnx/start.sh
	modified:   platform/mellanox/docker-syncd-mlnx/supervisord.conf
	new file:   platform/mellanox/mlnx-sfpd.mk
	new file:   platform/mellanox/mlnx-sfpd/scripts/mlnx-sfpd
	new file:   platform/mellanox/mlnx-sfpd/setup.py
	modified:   platform/mellanox/rules.mk

 signed-off-by Liu Kebo kebol@mellanox.com

platform/mellanox/docker-syncd-mlnx-rpc.mk

@@ -2,7 +2,7 @@
 
 DOCKER_SYNCD_MLNX_RPC = docker-syncd-mlnx-rpc.gz
 $(DOCKER_SYNCD_MLNX_RPC)_PATH = $(PLATFORM_PATH)/docker-syncd-mlnx-rpc
-$(DOCKER_SYNCD_MLNX_RPC)_DEPENDS += $(SYNCD_RPC) $(LIBTHRIFT)
+$(DOCKER_SYNCD_MLNX_RPC)_DEPENDS += $(SYNCD_RPC) $(LIBTHRIFT) $(MLNX_SFPD)
 $(DOCKER_SYNCD_MLNX_RPC)_LOAD_DOCKERS += $(DOCKER_SYNCD_MLNX)
 SONIC_DOCKER_IMAGES += $(DOCKER_SYNCD_MLNX_RPC)
 ifeq ($(ENABLE_SYNCD_RPC),y)

platform/mellanox/docker-syncd-mlnx.mk

@@ -2,7 +2,7 @@
 
 DOCKER_SYNCD_MLNX = docker-syncd-mlnx.gz
 $(DOCKER_SYNCD_MLNX)_PATH = $(PLATFORM_PATH)/docker-syncd-mlnx
-$(DOCKER_SYNCD_MLNX)_DEPENDS += $(SYNCD) $(PYTHON_SDK_API)
+$(DOCKER_SYNCD_MLNX)_DEPENDS += $(SYNCD) $(PYTHON_SDK_API) $(MLNX_SFPD)
 $(DOCKER_SYNCD_MLNX)_LOAD_DOCKERS += $(DOCKER_CONFIG_ENGINE)
 SONIC_DOCKER_IMAGES += $(DOCKER_SYNCD_MLNX)
 ifneq ($(ENABLE_SYNCD_RPC),y)

platform/mellanox/docker-syncd-mlnx/start.sh

@@ -5,3 +5,5 @@ rm -f /var/run/rsyslogd.pid
 supervisorctl start rsyslogd
 
 supervisorctl start syncd
+
+supervisorctl start mlnx-sfpd

platform/mellanox/docker-syncd-mlnx/supervisord.conf

@@ -26,3 +26,11 @@ autostart=false
 autorestart=false
 stdout_logfile=syslog
 stderr_logfile=syslog
+
+[program:mlnx-sfpd]
+command=/usr/bin/mlnx-sfpd
+priority=4
+autostart=false
+autorestart=false
+stdout_logfile=syslog
+stderr_logfile=syslog

platform/mellanox/mlnx-sfpd.mk

@@ -0,0 +1,5 @@
+# mlnx-sfpd (SONiC MLNX platform sfp event monitoring daemon) Debian package
+
+MLNX_SFPD = python-mlnx-sfpd_1.0-1_all.deb
+$(MLNX_SFPD)_SRC_PATH = $(PLATFORM_PATH)/mlnx-sfpd
+SONIC_PYTHON_STDEB_DEBS += $(MLNX_SFPD)

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

@@ -0,0 +1,199 @@
+#!/usr/bin/env python
+'''
+This code is for a mlnx platform specific daemon, mlnx-sfpd.
+Which listen to the SDK for the SFP change event and post the event to DB.
+'''
+
+from __future__ import print_function
+import sys, errno
+import os
+import time
+import syslog
+import signal
+import json
+from python_sdk_api.sx_api import *
+from swsssdk import SonicV2Connector
+
+VERSION = '1.0'
+
+SYSLOG_IDENTIFIER = "mlnx-sfpd"
+
+REDIS_HOSTIP = "127.0.0.1"
+
+SDK_SFP_STATE_IN = 0x1
+SDK_SFP_STATE_OUT = 0x2
+STATUS_PLUGIN = '1'
+STATUS_PLUGOUT = '0'
+STATUS_UNKNOWN = '2'
+
+sfp_value_status_dict = {SDK_SFP_STATE_IN:STATUS_PLUGIN, SDK_SFP_STATE_OUT:STATUS_PLUGOUT}
+
+#========================== Syslog wrappers ==========================
+
+def log_info(msg, also_print_to_console=False):
+    syslog.openlog(SYSLOG_IDENTIFIER)
+    syslog.syslog(syslog.LOG_INFO, msg)
+    syslog.closelog()
+
+    if also_print_to_console:
+        print(msg)
+
+def log_warning(msg, also_print_to_console=False):
+    syslog.openlog(SYSLOG_IDENTIFIER)
+    syslog.syslog(syslog.LOG_WARNING, msg)
+    syslog.closelog()
+
+    if also_print_to_console:
+        print(msg)
+
+def log_error(msg, also_print_to_console=False):
+    syslog.openlog(SYSLOG_IDENTIFIER)
+    syslog.syslog(syslog.LOG_ERR, msg)
+    syslog.closelog()
+
+    if also_print_to_console:
+        print(msg)
+
+#========================== Signal Handling ==========================
+
+def signal_handler(sig, frame):
+    if sig == signal.SIGHUP:
+        log_info("Caught SIGHUP - ignoring...")
+        return
+    elif sig == signal.SIGINT:
+        log_info("Caught SIGINT - exiting...")
+        sys.exit(128 + sig)
+    elif sig == signal.SIGTERM:
+        log_info("Caught SIGTERM - exiting...")
+        sys.exit(128 + sig)
+    else:
+        log_warning("Caught unhandled signal '" + sig + "'")
+
+
+def sx_recv(fd_p, handle):
+    # recv parameters
+    pkt_size = 2000
+    pkt_size_p = new_uint32_t_p()
+    uint32_t_p_assign(pkt_size_p, pkt_size)
+    pkt = new_uint8_t_arr(pkt_size)
+    recv_info_p = new_sx_receive_info_t_p()
+    pmpe_t = sx_event_pmpe_t()
+    logical_port_list = new_sx_port_log_id_t_arr(4)
+    port_attributes_list = new_sx_port_attributes_t_arr(64)
+    port_cnt_p = new_uint32_t_p()
+    uint32_t_p_assign(port_cnt_p,64)
+    label_port_list = []
+
+    rc = sx_lib_host_ifc_recv(fd_p, pkt, pkt_size_p, recv_info_p)
+    if rc != 0:
+        log_error("event receive exit with error, rc %d" % rc)
+        exit(rc)
+
+    pmpe_t = recv_info_p.event_info.pmpe
+    port_list_size = pmpe_t.list_size
+    logical_port_list = pmpe_t.log_port_list
+    module_state = pmpe_t.module_state
+
+    for i in range(0, port_list_size):
+        logical_port = sx_port_log_id_t_arr_getitem(logical_port_list, i)
+        rc = sx_api_port_device_get(handle, 1 , 0, port_attributes_list,  port_cnt_p)
+        port_cnt = uint32_t_p_value(port_cnt_p)
+
+        for i in range(0,port_cnt):
+            port_attributes = sx_port_attributes_t_arr_getitem(port_attributes_list,i)
+            if port_attributes.log_port == logical_port:
+                lable_port = port_attributes.port_mapping.module_port
+                break
+        label_port_list.append(lable_port)
+
+    return label_port_list, module_state
+
+def send_sfp_notification(db, interface, state):
+    sfp_notify = [interface,state]
+    msg = json.dumps(sfp_notify,separators=(',',':'))
+    db.publish('STATE_DB','TRANSCEIVER_NOTIFY', msg)
+    return
+
+# main start
+def main():
+    # Register our signal handlers
+    signal.signal(signal.SIGHUP, signal_handler)
+    signal.signal(signal.SIGINT, signal_handler)
+    signal.signal(signal.SIGTERM, signal_handler)
+
+    #open SDK handler
+    log_info("starting mlnx-sfpd...")
+    rc, handle = sx_api_open(None)
+    retry_time  = 1
+    while rc != SX_STATUS_SUCCESS:
+        time.sleep(2**retry_time)
+        retry_time += 1
+        rc, handle = sx_api_open(None)
+        if retry_time > 20:
+            log_error("Failed to open api handle.\nPlease check that SDK is running.")
+            sys.exit(errno.EACCES)
+
+    #open recv fd
+    rx_fd_p  =  new_sx_fd_t_p()
+    rc = sx_api_host_ifc_open(handle, rx_fd_p)
+    if rc != 0:
+        log_error("sx_api_host_ifc_open exit with error, rc %d" % rc)
+        exit(rc)
+
+    # set up general host ifc parameters
+    swid = 0
+    cmd = SX_ACCESS_CMD_REGISTER
+    uc_p = new_sx_user_channel_t_p()
+    uc_p.type = SX_USER_CHANNEL_TYPE_FD
+    uc_p.channel.fd = rx_fd_p
+    trap_id = SX_TRAP_ID_PMPE
+
+    rc = sx_api_host_ifc_trap_id_register_set(handle, cmd, swid, trap_id, uc_p)
+    if rc != 0:
+        log_error("sx_api_host_ifc_trap_id_register_set exit with error, rc %d" % rc)
+        exit(rc)
+
+    #connect to state db for notification sending
+    state_db = SonicV2Connector(host=REDIS_HOSTIP)
+    state_db.connect(state_db.STATE_DB)
+
+    #main loop for sfp event listening
+    log_info("mlnx-sfpd started")
+    while True:
+        state = STATUS_UNKNOWN
+        port_list, module_state = sx_recv(rx_fd_p, handle)
+        if module_state in sfp_value_status_dict: state = sfp_value_status_dict[module_state]
+
+        if state != STATUS_UNKNOWN:
+            for port in port_list:
+                log_info("SFP on port %d state %s" % (port, state))
+                send_sfp_notification(state_db, str(port), state)
+
+            log_info("sfp change event handling done")
+
+    '''
+    # TODO: clean open handlers before exit, need find out which errors can be raised by SDK in this case.
+    # unregister trap id
+    cmd = SX_ACCESS_CMD_DEREGISTER
+    rc = sx_api_host_ifc_trap_id_register_set(handle, cmd, swid, trap_id, uc_p)
+    if rc != 0:
+        log_error("sx_api_host_ifc_trap_id_register_set exit with error, rc %d" % rc)
+        exit(rc)
+
+    # close read fp
+    rc = sx_api_host_ifc_close(handle, rx_fd_p)
+    if rc != 0:
+        log_error("sx_api_host_ifc_close exit with error, rc %d" % rc)
+        exit(rc)
+
+    # close sdk handler
+    rc = sx_api_close(handle)
+    if rc != 0:
+        log_error("exit with error, rc %d" % rc)
+        exit(rc)
+
+    log_info("mlnx-sfpd exited")
+    '''
+
+if __name__ == '__main__':
+    main()

platform/mellanox/mlnx-sfpd/setup.py

@@ -0,0 +1,28 @@
+from setuptools import setup
+
+setup(
+    name='mlnx-sfpd',
+    version='1.0',
+    description='SFP event mmonitoring daemon for SONiC on mellanox platform',
+    license='Apache 2.0',
+    author='SONiC Community',
+    url='https://github.com/Azure/sonic-buildimage/',
+    maintainer='Kebo Liu',
+    maintainer_email='kebol@mellanox.com',
+    scripts=[
+        'scripts/mlnx-sfpd',
+    ],
+    classifiers=[
+        'Development Status :: 4 - Beta',
+        'Environment :: No Input/Output (Daemon)',
+        'Intended Audience :: Developers',
+        'Intended Audience :: Information Technology',
+        'Intended Audience :: System Administrators',
+        'License :: OSI Approved :: Apache Software License',
+        'Natural Language :: English',
+        'Operating System :: POSIX :: Linux',
+        'Programming Language :: Python :: 2.7',
+        'Topic :: System :: Hardware',
+    ],
+    keywords='sonic SONiC SFP sfp MELLANOX mellanox daemon SFPD sfpd',
+)

platform/mellanox/rules.mk

@@ -10,6 +10,7 @@ include $(PLATFORM_PATH)/one-image.mk
 include $(PLATFORM_PATH)/libsaithrift-dev.mk
 include $(PLATFORM_PATH)/python-saithrift.mk
 include $(PLATFORM_PATH)/docker-ptf-mlnx.mk
+include $(PLATFORM_PATH)/mlnx-sfpd.mk
 
 SONIC_ALL += $(SONIC_ONE_IMAGE) \
              $(DOCKER_FPM)