blob: 0337bed051d6970525a8a903090c77e3129cdc05 [file] [log] [blame]
/*
* Copyright 2018-present Open Networking Foundation
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
* http://www.apache.org/licenses/LICENSE-2.0
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <iostream>
#include <memory>
#include <string>
#include "Queue.h"
#include <sstream>
#include <chrono>
#include <thread>
#include <bitset>
#include <inttypes.h>
#include <unistd.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include "device.h"
#include "core.h"
#include "indications.h"
#include "stats_collection.h"
#include "error_format.h"
#include "state.h"
#include "utils.h"
extern "C"
{
#include <bcmolt_api.h>
#include <bcmolt_host_api.h>
#include <bcmolt_api_model_supporting_enums.h>
#include <bcmolt_api_conn_mgr.h>
//CLI header files
#include <bcmcli_session.h>
#include <bcmcli.h>
#include <bcm_api_cli.h>
#include <bcmos_common.h>
#include <bcm_config.h>
// FIXME : dependency problem
// #include <bcm_common_gpon.h>
// #include <bcm_dev_log_task.h>
}
dev_log_id openolt_log_id = bcm_dev_log_id_register("OPENOLT", DEV_LOG_LEVEL_INFO, DEV_LOG_ID_TYPE_BOTH);
dev_log_id omci_log_id = bcm_dev_log_id_register("OMCI", DEV_LOG_LEVEL_INFO, DEV_LOG_ID_TYPE_BOTH);
#define BAL_RSC_MANAGER_BASE_TM_SCHED_ID 16384
#define MAX_TM_QMP_ID 16
#define TMQ_MAP_PROFILE_SIZE 8
#define MAX_TM_SCHED_ID 1023
#define MAX_SUBS_TM_SCHED_ID (MAX_SUPPORTED_PON == 16 ? MAX_TM_SCHED_ID-4-16 : MAX_TM_SCHED_ID-10-64)
#define EAP_ETHER_TYPE 34958
#define XGS_BANDWIDTH_GRANULARITY 16000
#define GPON_BANDWIDTH_GRANULARITY 32000
#define NUM_OF_PRIORITIES 8
#define NUMBER_OF_DEFAULT_INTERFACE_QUEUES 4 // <= NUM_OF_PRIORITIES
#define FILL_ARRAY(ARRAY,START,END,VALUE) for(int i=START;i<END;ARRAY[i++]=VALUE);
#define COUNT_OF(array) (sizeof(array) / sizeof(array[0]))
#define GET_FLOW_INTERFACE_TYPE(type) \
(type == BCMOLT_FLOW_INTERFACE_TYPE_PON) ? "PON" : \
(type == BCMOLT_FLOW_INTERFACE_TYPE_NNI) ? "NNI" : \
(type == BCMOLT_FLOW_INTERFACE_TYPE_HOST) ? "HOST" : "unknown"
#define GET_PKT_TAG_TYPE(type) \
(type == BCMOLT_PKT_TAG_TYPE_UNTAGGED) ? "UNTAG" : \
(type == BCMOLT_PKT_TAG_TYPE_SINGLE_TAG) ? "SINGLE_TAG" : \
(type == BCMOLT_PKT_TAG_TYPE_DOUBLE_TAG) ? "DOUBLE_TAG" : "unknown"
#define GET_ACL_ACTION_TYPE(type) \
(type == BCMOLT_ACCESS_CONTROL_FWD_ACTION_TYPE_TRAP_TO_HOST) ? "trap_to_host" : \
(type == BCMOLT_ACCESS_CONTROL_FWD_ACTION_TYPE_DROP) ? "drop" : \
(type == BCMOLT_ACCESS_CONTROL_FWD_ACTION_TYPE_REDIRECT) ? "redirction" : "unknown"
#define GET_ACL_MEMBERS_UPDATE_COMMAND(command) \
(command == BCMOLT_MEMBERS_UPDATE_COMMAND_ADD) ? "add" : \
(command == BCMOLT_MEMBERS_UPDATE_COMMAND_REMOVE) ? "remove" : \
(command == BCMOLT_MEMBERS_UPDATE_COMMAND_SET) ? "set" : "unknown"
#define GET_INTERFACE_TYPE(type) \
(type == BCMOLT_INTERFACE_TYPE_PON) ? "PON" : \
(type == BCMOLT_INTERFACE_TYPE_NNI) ? "NNI" : \
(type == BCMOLT_INTERFACE_TYPE_HOST) ? "HOST" : "unknown"
static unsigned int num_of_nni_ports = 0;
static unsigned int num_of_pon_ports = 0;
static std::string intf_technologies[MAX_SUPPORTED_PON];
static const std::string UNKNOWN_TECH("unknown");
static const std::string MIXED_TECH("mixed");
static std::string board_technology(UNKNOWN_TECH);
static std::string chip_family(UNKNOWN_TECH);
static unsigned int OPENOLT_FIELD_LEN = 200;
static std::string firmware_version = "Openolt.2019.07.01";
const uint32_t tm_upstream_sched_id_start = (MAX_SUPPORTED_PON == 16 ? \
MAX_TM_SCHED_ID-3 : MAX_TM_SCHED_ID-9);
const uint32_t tm_downstream_sched_id_start = (MAX_SUPPORTED_PON == 16 ? \
tm_upstream_sched_id_start-16 : tm_upstream_sched_id_start-64);
/* Max Queue ID supported is 7 so based on priority_q configured for GEMPORTS
in TECH PROFILE respective Queue ID from this list will be used for both
US and DS Queues*/
const uint32_t queue_id_list[8] = {0, 1, 2, 3, 4, 5, 6, 7};
const std::string upstream = "upstream";
const std::string downstream = "downstream";
const std::string multicast = "multicast";
bcmolt_oltid dev_id = 0;
/* Constants used for retrying some BAL APIs */
const uint32_t BAL_API_RETRY_TIME_IN_USECS = 1000000;
const uint32_t MAX_BAL_API_RETRY_COUNT = 5;
/* Current session */
static bcmcli_session *current_session;
static bcmcli_entry *api_parent_dir;
bcmos_bool status_bcm_cli_quit = BCMOS_FALSE;
bcmos_task bal_cli_thread;
const char *bal_cli_thread_name = "bal_cli_thread";
uint16_t flow_id_counters = 0;
State state;
static std::map<uint32_t, uint32_t> flowid_to_port; // For mapping upstream flows to logical ports
static std::map<uint32_t, uint32_t> flowid_to_gemport; // For mapping downstream flows into gemports
static std::map<uint32_t, std::set<uint32_t> > port_to_flows; // For mapping logical ports to downstream flows
/* This represents the Key to 'sched_map' map.
Represents (pon_intf_id, onu_id, uni_id, direction) */
typedef std::tuple<uint32_t, uint32_t, uint32_t, std::string> sched_map_key_tuple;
/* 'sched_map' maps sched_map_key_tuple to DBA (Upstream) or
Subscriber (Downstream) Scheduler ID */
static std::map<sched_map_key_tuple, int> sched_map;
/* Flow control is for flow_id and flow_type */
typedef std::pair<uint16_t, uint16_t> flow_pair;
static std::map<flow_pair, int32_t> flow_map;
/* This represents the Key to 'qos_type_map' map.
Represents (pon_intf_id, onu_id, uni_id) */
typedef std::tuple<uint32_t, uint32_t, uint32_t> qos_type_map_key_tuple;
/* 'qos_type_map' maps qos_type_map_key_tuple to qos_type*/
static std::map<qos_type_map_key_tuple, bcmolt_egress_qos_type> qos_type_map;
/* This represents the Key to 'sched_qmp_id_map' map.
Represents (sched_id, pon_intf_id, onu_id, uni_id) */
typedef std::tuple<uint32_t, uint32_t, uint32_t, uint32_t> sched_qmp_id_map_key_tuple;
/* 'sched_qmp_id_map' maps sched_qmp_id_map_key_tuple to TM Queue Mapping Profile ID */
static std::map<sched_qmp_id_map_key_tuple, int> sched_qmp_id_map;
/* 'qmp_id_to_qmp_map' maps TM Queue Mapping Profile ID to TM Queue Mapping Profile */
static std::map<int, std::vector < uint32_t > > qmp_id_to_qmp_map;
// Flag used to watch whether mocked alloc_cfg_compltd_key is added to alloc_cfg_compltd_map
#ifdef TEST_MODE
bool ALLOC_CFG_FLAG = false;
#endif
#define ALLOC_CFG_COMPLETE_WAIT_TIMEOUT 1000 // in milli-seconds
// Map used to track response from BAL for ITU PON Alloc Configuration.
// The key is alloc_cfg_compltd_key and value is a concurrent thread-safe queue which is
// used for pushing (from BAL) and popping (at application) the results.
std::map<alloc_cfg_compltd_key, Queue<alloc_cfg_complete_result> *> alloc_cfg_compltd_map;
// Lock to protect critical section data structure used for handling AllocObject configuration response.
bcmos_fastlock alloc_cfg_wait_lock;
/*** ACL Handling related data start ***/
static std::map<acl_classifier_key, uint16_t> acl_classifier_to_acl_id_map;
bool operator<(const acl_classifier_key& a1, const acl_classifier_key& a2)
{
return ((a1.ether_type + 2*a1.ip_proto + 3*a1.src_port + 4*a1.dst_port) <
(a2.ether_type + 2*a2.ip_proto + 3*a2.src_port + 4*a2.dst_port));
}
typedef std::tuple<uint16_t, std::string> flow_id_flow_direction;
typedef std::tuple<int16_t, uint16_t, int32_t> acl_id_gem_id_intf_id;
static std::map<flow_id_flow_direction, acl_id_gem_id_intf_id> flow_to_acl_map;
// Keeps a reference count of how many flows are referencing a given ACL ID.
// Key represents the ACL-ID and value is number of flows referencing the given ACL-ID.
// When there is at least one flow referencing the ACL-ID, the ACL should be installed.
// When there are no flows referencing the ACL-ID, the ACL should be removed.
static std::map<uint16_t, uint16_t> acl_ref_cnt;
typedef std::tuple<uint16_t, uint16_t> gem_id_intf_id; // key to gem_ref_cnt
// Keeps a reference count of how many ACL related flows are referencing a given (gem-id, pon_intf_id).
// When there is at least on flow, we should install the gem. When there are no flows
// the gem should be removed.
static std::map<gem_id_intf_id, uint16_t> gem_ref_cnt;
// Needed to keep track of how many flows for a given acl_id, intf_id and intf_type are
// installed. When there is at least on flow for this key, we should have interface registered
// for the given ACL-ID. When there are no flows, the intf should unregister itself from
// the ACL-ID.
typedef std::tuple<uint16_t, uint8_t, std::string> acl_id_intf_id_intf_type;
static std::map<acl_id_intf_id_intf_type, uint16_t> intf_acl_registration_ref_cnt;
#define MAX_ACL_ID 33
std::bitset<MAX_ACL_ID> acl_id_bitset;
/*** ACL Handling related data end ***/
std::bitset<MAX_TM_SCHED_ID> tm_sched_bitset;
std::bitset<MAX_TM_QMP_ID> tm_qmp_bitset;
static bcmos_fastlock data_lock;
#define MIN_ALLOC_ID_GPON 256
#define MIN_ALLOC_ID_XGSPON 1024
static bcmos_errno CreateSched(std::string direction, uint32_t access_intf_id, uint32_t onu_id, uint32_t uni_id, \
uint32_t port_no, uint32_t alloc_id, tech_profile::AdditionalBW additional_bw, uint32_t weight, \
uint32_t priority, tech_profile::SchedulingPolicy sched_policy,
tech_profile::TrafficShapingInfo traffic_shaping_info);
static bcmos_errno RemoveSched(int intf_id, int onu_id, int uni_id, int alloc_id, std::string direction);
static bcmos_errno CreateQueue(std::string direction, uint32_t access_intf_id, uint32_t onu_id, uint32_t uni_id, \
bcmolt_egress_qos_type qos_type, uint32_t priority, uint32_t gemport_id);
static bcmos_errno RemoveQueue(std::string direction, uint32_t access_intf_id, uint32_t onu_id, uint32_t uni_id, \
bcmolt_egress_qos_type qos_type, uint32_t priority, uint32_t gemport_id);
static bcmos_errno CreateDefaultSched(uint32_t intf_id, const std::string direction);
static bcmos_errno CreateDefaultQueue(uint32_t intf_id, const std::string direction);
uint16_t get_dev_id(void) {
return dev_id;
}
// Stubbed defntions of bcmolt_cfg_get required for unit-test
#ifdef TEST_MODE
extern bcmos_errno bcmolt_cfg_get__bal_state_stub(bcmolt_oltid olt_id, void* ptr);
extern bcmos_errno bcmolt_cfg_get__onu_state_stub(bcmolt_oltid olt_id, void* ptr);
extern bcmos_errno bcmolt_cfg_get__tm_sched_stub(bcmolt_oltid olt_id, void* ptr);
extern bcmos_errno bcmolt_cfg_get__pon_intf_stub(bcmolt_oltid olt_id, void* ptr);
extern bcmos_errno bcmolt_cfg_get__nni_intf_stub(bcmolt_oltid olt_id, void* ptr);
extern bcmos_errno bcmolt_cfg_get__olt_topology_stub(bcmolt_oltid olt_id, void* ptr);
extern bcmos_errno bcmolt_cfg_get__flow_stub(bcmolt_oltid olt_id, void* ptr);
#endif
/**
* Returns the default NNI (Upstream direction) or PON (Downstream direction) scheduler
* Every NNI port and PON port have default scheduler.
* The NNI0 default scheduler ID is 18432, and NNI1 is 18433 and so on.
* Similarly, PON0 default scheduler ID is 16384. PON1 is 16385 and so on.
*
* @param intf_id NNI or PON interface ID
* @param direction "upstream" or "downstream"
*
* @return default scheduler ID for the given interface.
*/
static inline int get_default_tm_sched_id(int intf_id, std::string direction) {
if (direction.compare(upstream) == 0) {
return tm_upstream_sched_id_start + intf_id;
} else if (direction.compare(downstream) == 0) {
return tm_downstream_sched_id_start + intf_id;
}
else {
OPENOLT_LOG(ERROR, openolt_log_id, "invalid direction - %s\n", direction.c_str());
return 0;
}
}
/**
* Gets a unique tm_sched_id for a given intf_id, onu_id, uni_id, gemport_id, direction
* The tm_sched_id is locally cached in a map, so that it can rendered when necessary.
* VOLTHA replays whole configuration on OLT reboot, so caching locally is not a problem
*
* @param intf_id NNI or PON intf ID
* @param onu_id ONU ID
* @param uni_id UNI ID
* @param gemport_id GEM Port ID
* @param direction Upstream or downstream
*
* @return tm_sched_id
*/
uint32_t get_tm_sched_id(int pon_intf_id, int onu_id, int uni_id, std::string direction) {
sched_map_key_tuple key(pon_intf_id, onu_id, uni_id, direction);
int sched_id = -1;
std::map<sched_map_key_tuple, int>::const_iterator it = sched_map.find(key);
if (it != sched_map.end()) {
sched_id = it->second;
}
if (sched_id != -1) {
return sched_id;
}
bcmos_fastlock_lock(&data_lock);
// Complexity of O(n). Is there better way that can avoid linear search?
for (sched_id = 0; sched_id < MAX_TM_SCHED_ID; sched_id++) {
if (tm_sched_bitset[sched_id] == 0) {
tm_sched_bitset[sched_id] = 1;
break;
}
}
bcmos_fastlock_unlock(&data_lock, 0);
if (sched_id < MAX_TM_SCHED_ID) {
bcmos_fastlock_lock(&data_lock);
sched_map[key] = sched_id;
bcmos_fastlock_unlock(&data_lock, 0);
return sched_id;
} else {
return -1;
}
}
/**
* Free tm_sched_id for a given intf_id, onu_id, uni_id, gemport_id, direction
*
* @param intf_id NNI or PON intf ID
* @param onu_id ONU ID
* @param uni_id UNI ID
* @param gemport_id GEM Port ID
* @param direction Upstream or downstream
*/
void free_tm_sched_id(int pon_intf_id, int onu_id, int uni_id, std::string direction) {
sched_map_key_tuple key(pon_intf_id, onu_id, uni_id, direction);
std::map<sched_map_key_tuple, int>::const_iterator it;
bcmos_fastlock_lock(&data_lock);
it = sched_map.find(key);
if (it != sched_map.end()) {
tm_sched_bitset[it->second] = 0;
sched_map.erase(it);
}
bcmos_fastlock_unlock(&data_lock, 0);
}
bool is_tm_sched_id_present(int pon_intf_id, int onu_id, int uni_id, std::string direction) {
sched_map_key_tuple key(pon_intf_id, onu_id, uni_id, direction);
std::map<sched_map_key_tuple, int>::const_iterator it = sched_map.find(key);
if (it != sched_map.end()) {
return true;
}
return false;
}
/**
* Check whether given two tm qmp profiles are equal or not
*
* @param tmq_map_profileA <vector> TM QUEUE MAPPING PROFILE
* @param tmq_map_profileB <vector> TM QUEUE MAPPING PROFILE
*
* @return boolean, true if given tmq_map_profiles are equal else false
*/
bool check_tm_qmp_equality(std::vector<uint32_t> tmq_map_profileA, std::vector<uint32_t> tmq_map_profileB) {
for (uint32_t i = 0; i < TMQ_MAP_PROFILE_SIZE; i++) {
if (tmq_map_profileA[i] != tmq_map_profileB[i]) {
return false;
}
}
return true;
}
/**
* Modifies given queues_pbit_map to parsable format
* e.g: Modifes "0b00000101" to "10100000"
*
* @param queues_pbit_map PBIT MAP configured for each GEM in TECH PROFILE
* @param size Queue count
*
* @return string queues_pbit_map
*/
std::string* get_valid_queues_pbit_map(std::string *queues_pbit_map, uint32_t size) {
for(uint32_t i=0; i < size; i++) {
/* Deletes 2 characters from index number 0 */
queues_pbit_map[i].erase(0, 2);
std::reverse(queues_pbit_map[i].begin(), queues_pbit_map[i].end());
}
return queues_pbit_map;
}
/**
* Creates TM QUEUE MAPPING PROFILE for given queues_pbit_map and queues_priority_q
*
* @param queues_pbit_map PBIT MAP configured for each GEM in TECH PROFILE
* @param queues_priority_q PRIORITY_Q configured for each GEM in TECH PROFILE
* @param size Queue count
*
* @return <vector> TM QUEUE MAPPING PROFILE
*/
std::vector<uint32_t> get_tmq_map_profile(std::string *queues_pbit_map, uint32_t *queues_priority_q, uint32_t size) {
std::vector<uint32_t> tmq_map_profile(8,0);
for(uint32_t i=0; i < size; i++) {
for (uint32_t j = 0; j < queues_pbit_map[i].size(); j++) {
if (queues_pbit_map[i][j]=='1') {
tmq_map_profile.at(j) = queue_id_list[queues_priority_q[i]];
}
}
}
return tmq_map_profile;
}
/**
* Gets corresponding tm_qmp_id for a given tmq_map_profile
*
* @param <vector> TM QUEUE MAPPING PROFILE
*
* @return tm_qmp_id
*/
int get_tm_qmp_id(std::vector<uint32_t> tmq_map_profile) {
int tm_qmp_id = -1;
std::map<int, std::vector < uint32_t > >::const_iterator it = qmp_id_to_qmp_map.begin();
while(it != qmp_id_to_qmp_map.end()) {
if(check_tm_qmp_equality(tmq_map_profile, it->second)) {
tm_qmp_id = it->first;
break;
}
it++;
}
return tm_qmp_id;
}
/**
* Updates sched_qmp_id_map with given sched_id, pon_intf_id, onu_id, uni_id, tm_qmp_id
*
* @param upstream/downstream sched_id
* @param PON intf ID
* @param onu_id ONU ID
* @param uni_id UNI ID
* @param tm_qmp_id TM QUEUE MAPPING PROFILE ID
*/
void update_sched_qmp_id_map(uint32_t sched_id,uint32_t pon_intf_id, uint32_t onu_id, \
uint32_t uni_id, int tm_qmp_id) {
bcmos_fastlock_lock(&data_lock);
sched_qmp_id_map_key_tuple key(sched_id, pon_intf_id, onu_id, uni_id);
sched_qmp_id_map.insert(make_pair(key, tm_qmp_id));
bcmos_fastlock_unlock(&data_lock, 0);
}
/**
* Gets corresponding tm_qmp_id for a given sched_id, pon_intf_id, onu_id, uni_id
*
* @param upstream/downstream sched_id
* @param PON intf ID
* @param onu_id ONU ID
* @param uni_id UNI ID
*
* @return tm_qmp_id
*/
int get_tm_qmp_id(uint32_t sched_id,uint32_t pon_intf_id, uint32_t onu_id, uint32_t uni_id) {
sched_qmp_id_map_key_tuple key(sched_id, pon_intf_id, onu_id, uni_id);
int tm_qmp_id = -1;
std::map<sched_qmp_id_map_key_tuple, int>::const_iterator it = sched_qmp_id_map.find(key);
if (it != sched_qmp_id_map.end()) {
tm_qmp_id = it->second;
}
return tm_qmp_id;
}
/**
* Gets a unique tm_qmp_id for a given tmq_map_profile
* The tm_qmp_id is locally cached in a map, so that it can be rendered when necessary.
* VOLTHA replays whole configuration on OLT reboot, so caching locally is not a problem
*
* @param upstream/downstream sched_id
* @param PON intf ID
* @param onu_id ONU ID
* @param uni_id UNI ID
* @param <vector> TM QUEUE MAPPING PROFILE
*
* @return tm_qmp_id
*/
int get_tm_qmp_id(uint32_t sched_id,uint32_t pon_intf_id, uint32_t onu_id, uint32_t uni_id, \
std::vector<uint32_t> tmq_map_profile) {
int tm_qmp_id;
bcmos_fastlock_lock(&data_lock);
/* Complexity of O(n). Is there better way that can avoid linear search? */
for (tm_qmp_id = 0; tm_qmp_id < MAX_TM_QMP_ID; tm_qmp_id++) {
if (tm_qmp_bitset[tm_qmp_id] == 0) {
tm_qmp_bitset[tm_qmp_id] = 1;
break;
}
}
bcmos_fastlock_unlock(&data_lock, 0);
if (tm_qmp_id < MAX_TM_QMP_ID) {
bcmos_fastlock_lock(&data_lock);
qmp_id_to_qmp_map.insert(make_pair(tm_qmp_id, tmq_map_profile));
bcmos_fastlock_unlock(&data_lock, 0);
update_sched_qmp_id_map(sched_id, pon_intf_id, onu_id, uni_id, tm_qmp_id);
return tm_qmp_id;
} else {
return -1;
}
}
/**
* Free tm_qmp_id for a given sched_id, pon_intf_id, onu_id, uni_id
*
* @param upstream/downstream sched_id
* @param PON intf ID
* @param onu_id ONU ID
* @param uni_id UNI ID
* @param tm_qmp_id TM QUEUE MAPPING PROFILE ID
*
* @return boolean, true if no more reference for TM QMP else false
*/
bool free_tm_qmp_id(uint32_t sched_id,uint32_t pon_intf_id, uint32_t onu_id, \
uint32_t uni_id, int tm_qmp_id) {
bool result;
sched_qmp_id_map_key_tuple key(sched_id, pon_intf_id, onu_id, uni_id);
std::map<sched_qmp_id_map_key_tuple, int>::const_iterator it = sched_qmp_id_map.find(key);
bcmos_fastlock_lock(&data_lock);
if (it != sched_qmp_id_map.end()) {
sched_qmp_id_map.erase(it);
}
bcmos_fastlock_unlock(&data_lock, 0);
uint32_t tm_qmp_ref_count = 0;
std::map<sched_qmp_id_map_key_tuple, int>::const_iterator it2 = sched_qmp_id_map.begin();
while(it2 != sched_qmp_id_map.end()) {
if(it2->second == tm_qmp_id) {
tm_qmp_ref_count++;
}
it2++;
}
if (tm_qmp_ref_count == 0) {
std::map<int, std::vector < uint32_t > >::const_iterator it3 = qmp_id_to_qmp_map.find(tm_qmp_id);
if (it3 != qmp_id_to_qmp_map.end()) {
bcmos_fastlock_lock(&data_lock);
tm_qmp_bitset[tm_qmp_id] = 0;
qmp_id_to_qmp_map.erase(it3);
bcmos_fastlock_unlock(&data_lock, 0);
OPENOLT_LOG(INFO, openolt_log_id, "Reference count for tm qmp profile id %d is : %d. So clearing it\n", \
tm_qmp_id, tm_qmp_ref_count);
result = true;
}
} else {
OPENOLT_LOG(INFO, openolt_log_id, "Reference count for tm qmp profile id %d is : %d. So not clearing it\n", \
tm_qmp_id, tm_qmp_ref_count);
result = false;
}
return result;
}
// Gets free ACL ID if available, else -1
int get_acl_id() {
int acl_id;
bcmos_fastlock_lock(&data_lock);
/* Complexity of O(n). Is there better way that can avoid linear search? */
for (acl_id = 0; acl_id < MAX_ACL_ID; acl_id++) {
if (acl_id_bitset[acl_id] == 0) {
acl_id_bitset[acl_id] = 1;
break;
}
}
bcmos_fastlock_unlock(&data_lock, 0);
if (acl_id < MAX_ACL_ID) {
return acl_id ;
} else {
return -1;
}
}
// Frees up the ACL ID.
void free_acl_id (int acl_id) {
if (acl_id < MAX_ACL_ID) {
bcmos_fastlock_lock(&data_lock);
acl_id_bitset[acl_id] = 0;
bcmos_fastlock_unlock(&data_lock, 0);
}
}
/**
* Returns qos type as string
*
* @param qos_type bcmolt_egress_qos_type enum
*/
std::string get_qos_type_as_string(bcmolt_egress_qos_type qos_type) {
switch (qos_type)
{
case BCMOLT_EGRESS_QOS_TYPE_FIXED_QUEUE: return "FIXED_QUEUE";
case BCMOLT_EGRESS_QOS_TYPE_TC_TO_QUEUE: return "TC_TO_QUEUE";
case BCMOLT_EGRESS_QOS_TYPE_PBIT_TO_TC: return "PBIT_TO_TC";
case BCMOLT_EGRESS_QOS_TYPE_NONE: return "NONE";
case BCMOLT_EGRESS_QOS_TYPE_PRIORITY_TO_QUEUE: return "PRIORITY_TO_QUEUE";
default: OPENOLT_LOG(ERROR, openolt_log_id, "qos-type-not-supported %d\n", qos_type);
return "qos-type-not-supported";
}
}
/**
* Gets/Updates qos type for given pon_intf_id, onu_id, uni_id
*
* @param PON intf ID
* @param onu_id ONU ID
* @param uni_id UNI ID
* @param queue_size TrafficQueues Size
*
* @return qos_type
*/
bcmolt_egress_qos_type get_qos_type(uint32_t pon_intf_id, uint32_t onu_id, uint32_t uni_id, uint32_t queue_size = 0) {
qos_type_map_key_tuple key(pon_intf_id, onu_id, uni_id);
bcmolt_egress_qos_type egress_qos_type = BCMOLT_EGRESS_QOS_TYPE_FIXED_QUEUE;
std::string qos_string;
std::map<qos_type_map_key_tuple, bcmolt_egress_qos_type>::const_iterator it = qos_type_map.find(key);
if (it != qos_type_map.end()) {
egress_qos_type = it->second;
qos_string = get_qos_type_as_string(egress_qos_type);
OPENOLT_LOG(INFO, openolt_log_id, "Qos-type for subscriber connected to pon_intf_id %d, onu_id %d and uni_id %d is %s\n", \
pon_intf_id, onu_id, uni_id, qos_string.c_str());
}
else {
/* QOS Type has been pre-defined as Fixed Queue but it will be updated based on number of GEMPORTS
associated for a given subscriber. If GEM count = 1 for a given subscriber, qos_type will be Fixed Queue
else Priority to Queue */
egress_qos_type = (queue_size > 1) ? \
BCMOLT_EGRESS_QOS_TYPE_PRIORITY_TO_QUEUE : BCMOLT_EGRESS_QOS_TYPE_FIXED_QUEUE;
bcmos_fastlock_lock(&data_lock);
qos_type_map.insert(make_pair(key, egress_qos_type));
bcmos_fastlock_unlock(&data_lock, 0);
qos_string = get_qos_type_as_string(egress_qos_type);
OPENOLT_LOG(INFO, openolt_log_id, "Qos-type for subscriber connected to pon_intf_id %d, onu_id %d and uni_id %d is %s\n", \
pon_intf_id, onu_id, uni_id, qos_string.c_str());
}
return egress_qos_type;
}
/**
* Clears qos type for given pon_intf_id, onu_id, uni_id
*
* @param PON intf ID
* @param onu_id ONU ID
* @param uni_id UNI ID
*/
void clear_qos_type(uint32_t pon_intf_id, uint32_t onu_id, uint32_t uni_id) {
qos_type_map_key_tuple key(pon_intf_id, onu_id, uni_id);
std::map<qos_type_map_key_tuple, bcmolt_egress_qos_type>::const_iterator it = qos_type_map.find(key);
bcmos_fastlock_lock(&data_lock);
if (it != qos_type_map.end()) {
qos_type_map.erase(it);
OPENOLT_LOG(INFO, openolt_log_id, "Cleared Qos-type for subscriber connected to pon_intf_id %d, onu_id %d and uni_id %d\n", \
pon_intf_id, onu_id, uni_id);
}
bcmos_fastlock_unlock(&data_lock, 0);
}
/**
* Returns Scheduler/Queue direction as string
*
* @param direction as specified in tech_profile.proto
*/
std::string GetDirection(int direction) {
switch (direction)
{
case tech_profile::Direction::UPSTREAM: return upstream;
case tech_profile::Direction::DOWNSTREAM: return downstream;
default: OPENOLT_LOG(ERROR, openolt_log_id, "direction-not-supported %d\n", direction);
return "direction-not-supported";
}
}
inline const char *get_flow_acton_command(uint32_t command) {
char actions[200] = { };
char *s_actions_ptr = actions;
if (command & BCMOLT_ACTION_CMD_ID_ADD_OUTER_TAG) strcat(s_actions_ptr, "ADD_OUTER_TAG|");
if (command & BCMOLT_ACTION_CMD_ID_REMOVE_OUTER_TAG) strcat(s_actions_ptr, "REMOVE_OUTER_TAG|");
if (command & BCMOLT_ACTION_CMD_ID_XLATE_OUTER_TAG) strcat(s_actions_ptr, "TRANSLATE_OUTER_TAG|");
if (command & BCMOLT_ACTION_CMD_ID_ADD_INNER_TAG) strcat(s_actions_ptr, "ADD_INNTER_TAG|");
if (command & BCMOLT_ACTION_CMD_ID_REMOVE_INNER_TAG) strcat(s_actions_ptr, "REMOVE_INNER_TAG|");
if (command & BCMOLT_ACTION_CMD_ID_XLATE_INNER_TAG) strcat(s_actions_ptr, "TRANSLATE_INNER_TAG|");
if (command & BCMOLT_ACTION_CMD_ID_REMARK_OUTER_PBITS) strcat(s_actions_ptr, "REMOVE_OUTER_PBITS|");
if (command & BCMOLT_ACTION_CMD_ID_REMARK_INNER_PBITS) strcat(s_actions_ptr, "REMAKE_INNER_PBITS|");
return s_actions_ptr;
}
// This method handles waiting for AllocObject configuration.
// Returns error if the AllocObject is not in the appropriate state based on action requested.
bcmos_errno wait_for_alloc_action(uint32_t intf_id, uint32_t alloc_id, AllocCfgAction action) {
Queue<alloc_cfg_complete_result> cfg_result;
alloc_cfg_compltd_key k(intf_id, alloc_id);
alloc_cfg_compltd_map[k] = &cfg_result;
bcmos_errno err = BCM_ERR_OK;
#ifdef TEST_MODE
ALLOC_CFG_FLAG = true;
#endif
// Try to pop the result from BAL with a timeout of ALLOC_CFG_COMPLETE_WAIT_TIMEOUT ms
std::pair<alloc_cfg_complete_result, bool> result = cfg_result.pop(ALLOC_CFG_COMPLETE_WAIT_TIMEOUT);
if (result.second == false) {
OPENOLT_LOG(ERROR, openolt_log_id, "timeout waiting for alloc cfg complete indication intf_id %d, alloc_id %d\n",
intf_id, alloc_id);
// Invalidate the queue pointer.
bcmos_fastlock_lock(&alloc_cfg_wait_lock);
alloc_cfg_compltd_map[k] = NULL;
bcmos_fastlock_unlock(&alloc_cfg_wait_lock, 0);
err = BCM_ERR_INTERNAL;
#ifdef TEST_MODE
ALLOC_CFG_FLAG = false;
#endif
}
else if (result.first.status == ALLOC_CFG_STATUS_FAIL) {
OPENOLT_LOG(ERROR, openolt_log_id, "error processing alloc cfg request intf_id %d, alloc_id %d\n",
intf_id, alloc_id);
err = BCM_ERR_INTERNAL;
}
if (err == BCM_ERR_OK) {
if (action == ALLOC_OBJECT_CREATE) {
if (result.first.state != ALLOC_OBJECT_STATE_ACTIVE) {
OPENOLT_LOG(ERROR, openolt_log_id, "alloc object not in active state intf_id %d, alloc_id %d alloc_obj_state %d\n",
intf_id, alloc_id, result.first.state);
err = BCM_ERR_INTERNAL;
} else {
OPENOLT_LOG(INFO, openolt_log_id, "Create upstream bandwidth allocation success, intf_id %d, alloc_id %d\n",
intf_id, alloc_id);
}
} else { // ALLOC_OBJECT_DELETE
if (result.first.state != ALLOC_OBJECT_STATE_NOT_CONFIGURED) {
OPENOLT_LOG(ERROR, openolt_log_id, "alloc object is not reset intf_id %d, alloc_id %d alloc_obj_state %d\n",
intf_id, alloc_id, result.first.state);
err = BCM_ERR_INTERNAL;
} else {
OPENOLT_LOG(INFO, openolt_log_id, "Remove alloc object success, intf_id %d, alloc_id %d\n",
intf_id, alloc_id);
}
}
}
// Remove entry from map
bcmos_fastlock_lock(&alloc_cfg_wait_lock);
alloc_cfg_compltd_map.erase(k);
bcmos_fastlock_unlock(&alloc_cfg_wait_lock, 0);
#ifdef TEST_MODE
ALLOC_CFG_FLAG = false;
#endif
return err;
}
char* openolt_read_sysinfo(const char* field_name, char* field_val)
{
FILE *fp;
/* Prepare the command*/
char command[150];
snprintf(command, sizeof command, "bash -l -c \"onlpdump -s\" | perl -ne 'print $1 if /%s: (\\S+)/'", field_name);
/* Open the command for reading. */
fp = popen(command, "r");
if (fp == NULL) {
/*The client has to check for a Null field value in this case*/
OPENOLT_LOG(INFO, openolt_log_id, "Failed to query the %s\n", field_name);
return field_val;
}
/*Read the field value*/
if (fp) {
uint8_t ret;
ret = fread(field_val, OPENOLT_FIELD_LEN, 1, fp);
if (ret >= OPENOLT_FIELD_LEN)
OPENOLT_LOG(INFO, openolt_log_id, "Read data length %u\n", ret);
pclose(fp);
}
return field_val;
}
Status GetDeviceInfo_(openolt::DeviceInfo* device_info) {
device_info->set_vendor(VENDOR_ID);
device_info->set_model(MODEL_ID);
device_info->set_hardware_version("");
device_info->set_firmware_version(firmware_version);
device_info->set_technology(board_technology);
device_info->set_pon_ports(num_of_pon_ports);
char serial_number[OPENOLT_FIELD_LEN];
memset(serial_number, '\0', OPENOLT_FIELD_LEN);
openolt_read_sysinfo("Serial Number", serial_number);
OPENOLT_LOG(INFO, openolt_log_id, "Fetched device serial number %s\n", serial_number);
device_info->set_device_serial_number(serial_number);
char device_id[OPENOLT_FIELD_LEN];
memset(device_id, '\0', OPENOLT_FIELD_LEN);
openolt_read_sysinfo("MAC", device_id);
OPENOLT_LOG(INFO, openolt_log_id, "Fetched device mac address %s\n", device_id);
device_info->set_device_id(device_id);
// Legacy, device-wide ranges. To be deprecated when adapter
// is upgraded to support per-interface ranges
if (board_technology == "XGS-PON") {
device_info->set_onu_id_start(1);
device_info->set_onu_id_end(255);
device_info->set_alloc_id_start(MIN_ALLOC_ID_XGSPON);
device_info->set_alloc_id_end(16383);
device_info->set_gemport_id_start(1024);
device_info->set_gemport_id_end(65535);
device_info->set_flow_id_start(1);
device_info->set_flow_id_end(16383);
}
else if (board_technology == "GPON") {
device_info->set_onu_id_start(1);
device_info->set_onu_id_end(127);
device_info->set_alloc_id_start(MIN_ALLOC_ID_GPON);
device_info->set_alloc_id_end(767);
device_info->set_gemport_id_start(256);
device_info->set_gemport_id_end(4095);
device_info->set_flow_id_start(1);
device_info->set_flow_id_end(16383);
}
std::map<std::string, openolt::DeviceInfo::DeviceResourceRanges*> ranges;
for (uint32_t intf_id = 0; intf_id < num_of_pon_ports; ++intf_id) {
std::string intf_technology = intf_technologies[intf_id];
openolt::DeviceInfo::DeviceResourceRanges *range = ranges[intf_technology];
if(range == nullptr) {
range = device_info->add_ranges();
ranges[intf_technology] = range;
range->set_technology(intf_technology);
if (intf_technology == "XGS-PON") {
openolt::DeviceInfo::DeviceResourceRanges::Pool* pool;
pool = range->add_pools();
pool->set_type(openolt::DeviceInfo::DeviceResourceRanges::Pool::ONU_ID);
pool->set_sharing(openolt::DeviceInfo::DeviceResourceRanges::Pool::DEDICATED_PER_INTF);
pool->set_start(1);
pool->set_end(255);
pool = range->add_pools();
pool->set_type(openolt::DeviceInfo::DeviceResourceRanges::Pool::ALLOC_ID);
pool->set_sharing(openolt::DeviceInfo::DeviceResourceRanges::Pool::SHARED_BY_ALL_INTF_SAME_TECH);
pool->set_start(1024);
pool->set_end(16383);
pool = range->add_pools();
pool->set_type(openolt::DeviceInfo::DeviceResourceRanges::Pool::GEMPORT_ID);
pool->set_sharing(openolt::DeviceInfo::DeviceResourceRanges::Pool::SHARED_BY_ALL_INTF_ALL_TECH);
pool->set_start(1024);
pool->set_end(65535);
pool = range->add_pools();
pool->set_type(openolt::DeviceInfo::DeviceResourceRanges::Pool::FLOW_ID);
pool->set_sharing(openolt::DeviceInfo::DeviceResourceRanges::Pool::SHARED_BY_ALL_INTF_ALL_TECH);
pool->set_start(1);
pool->set_end(16383);
}
else if (intf_technology == "GPON") {
openolt::DeviceInfo::DeviceResourceRanges::Pool* pool;
pool = range->add_pools();
pool->set_type(openolt::DeviceInfo::DeviceResourceRanges::Pool::ONU_ID);
pool->set_sharing(openolt::DeviceInfo::DeviceResourceRanges::Pool::DEDICATED_PER_INTF);
pool->set_start(1);
pool->set_end(127);
pool = range->add_pools();
pool->set_type(openolt::DeviceInfo::DeviceResourceRanges::Pool::ALLOC_ID);
pool->set_sharing(openolt::DeviceInfo::DeviceResourceRanges::Pool::SHARED_BY_ALL_INTF_SAME_TECH);
pool->set_start(256);
pool->set_end(757);
pool = range->add_pools();
pool->set_type(openolt::DeviceInfo::DeviceResourceRanges::Pool::GEMPORT_ID);
pool->set_sharing(openolt::DeviceInfo::DeviceResourceRanges::Pool::SHARED_BY_ALL_INTF_ALL_TECH);
pool->set_start(256);
pool->set_end(4095);
pool = range->add_pools();
pool->set_type(openolt::DeviceInfo::DeviceResourceRanges::Pool::FLOW_ID);
pool->set_sharing(openolt::DeviceInfo::DeviceResourceRanges::Pool::SHARED_BY_ALL_INTF_ALL_TECH);
pool->set_start(1);
pool->set_end(16383);
}
}
range->add_intf_ids(intf_id);
}
// FIXME: Once dependency problem is fixed
// device_info->set_pon_ports(num_of_pon_ports);
// device_info->set_onu_id_end(XGPON_NUM_OF_ONUS - 1);
// device_info->set_alloc_id_start(1024);
// device_info->set_alloc_id_end(XGPON_NUM_OF_ALLOC_IDS * num_of_pon_ports ? - 1);
// device_info->set_gemport_id_start(XGPON_MIN_BASE_SERVICE_PORT_ID);
// device_info->set_gemport_id_end(XGPON_NUM_OF_GEM_PORT_IDS_PER_PON * num_of_pon_ports ? - 1);
// device_info->set_pon_ports(num_of_pon_ports);
return Status::OK;
}
Status pushOltOperInd(uint32_t intf_id, const char *type, const char *state)
{
openolt::Indication ind;
openolt::IntfOperIndication* intf_oper_ind = new openolt::IntfOperIndication;
intf_oper_ind->set_type(type);
intf_oper_ind->set_intf_id(intf_id);
intf_oper_ind->set_oper_state(state);
ind.set_allocated_intf_oper_ind(intf_oper_ind);
oltIndQ.push(ind);
return Status::OK;
}
#define CLI_HOST_PROMPT_FORMAT "BCM.%u> "
/* Build CLI prompt */
static void openolt_cli_get_prompt_cb(bcmcli_session *session, char *buf, uint32_t max_len)
{
snprintf(buf, max_len, CLI_HOST_PROMPT_FORMAT, dev_id);
}
static int _bal_apiend_cli_thread_handler(long data)
{
char init_string[]="\n";
bcmcli_session *sess = current_session;
bcmos_task_parm bal_cli_task_p_dummy;
/* Switch to interactive mode if not stopped in the init script */
if (!bcmcli_is_stopped(sess)) {
/* Force a CLI command prompt
* The string passed into the parse function
* must be modifiable, so a string constant like
* bcmcli_parse(current_session, "\n") will not
* work.
*/
bcmcli_parse(sess, init_string);
/* Process user input until EOF or quit command */
bcmcli_driver(sess);
}
OPENOLT_LOG(INFO, openolt_log_id, "BAL API End CLI terminated\n");
/* Cleanup */
bcmcli_session_close(current_session);
bcmcli_token_destroy(NULL);
return 0;
}
/* Init API CLI commands for the current device */
bcmos_errno bcm_openolt_api_cli_init(bcmcli_entry *parent_dir, bcmcli_session *session)
{
bcmos_errno rc;
api_parent_dir = parent_dir;
rc = bcm_api_cli_set_commands(session);
#ifdef BCM_SUBSYSTEM_HOST
/* Subscribe for device change indication */
rc = rc ? rc : bcmolt_olt_sel_ind_register(_api_cli_olt_change_ind);
#endif
return rc;
}
static bcmos_errno bcm_cli_quit(bcmcli_session *session, const bcmcli_cmd_parm parm[], uint16_t n_parms)
{
bcmcli_stop(session);
bcmcli_session_print(session, "CLI terminated by 'Quit' command\n");
status_bcm_cli_quit = BCMOS_TRUE;
return BCM_ERR_OK;
}
int get_status_bcm_cli_quit(void) {
return status_bcm_cli_quit;
}
bcmos_errno bcmolt_apiend_cli_init() {
bcmos_errno ret;
bcmos_task_parm bal_cli_task_p = {};
bcmos_task_parm bal_cli_task_p_dummy;
/** before creating the task, check if it is already created by the other half of BAL i.e. Core side */
if (BCM_ERR_OK != bcmos_task_query(&bal_cli_thread, &bal_cli_task_p_dummy)) {
/* Create BAL CLI thread */
bal_cli_task_p.name = bal_cli_thread_name;
bal_cli_task_p.handler = _bal_apiend_cli_thread_handler;
bal_cli_task_p.priority = TASK_PRIORITY_CLI;
ret = bcmos_task_create(&bal_cli_thread, &bal_cli_task_p);
if (BCM_ERR_OK != ret) {
bcmos_printf("Couldn't create BAL API end CLI thread\n");
return ret;
}
}
}
Status Enable_(int argc, char *argv[]) {
bcmos_errno err;
bcmolt_host_init_parms init_parms = {};
init_parms.transport.type = BCM_HOST_API_CONN_LOCAL;
unsigned int failed_enable_device_cnt = 0;
if (!state.is_activated()) {
vendor_init();
/* Initialize host subsystem */
err = bcmolt_host_init(&init_parms);
if (BCM_ERR_OK != err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to init OLT, err = %s\n",bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to init OLT");
}
bcmcli_session_parm mon_session_parm;
/* Create CLI session */
memset(&mon_session_parm, 0, sizeof(mon_session_parm));
mon_session_parm.get_prompt = openolt_cli_get_prompt_cb;
mon_session_parm.access_right = BCMCLI_ACCESS_ADMIN;
bcmos_errno rc = bcmcli_session_open(&mon_session_parm, &current_session);
BUG_ON(rc != BCM_ERR_OK);
/* API CLI */
bcm_openolt_api_cli_init(NULL, current_session);
/* Add quit command */
BCMCLI_MAKE_CMD_NOPARM(NULL, "quit", "Quit", bcm_cli_quit);
err = bcmolt_apiend_cli_init();
if (BCM_ERR_OK != err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to add apiend init, err = %s\n",bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to add apiend init");
}
bcmos_fastlock_init(&data_lock, 0);
bcmos_fastlock_init(&alloc_cfg_wait_lock, 0);
OPENOLT_LOG(INFO, openolt_log_id, "Enable OLT - %s-%s\n", VENDOR_ID, MODEL_ID);
//check BCM daemon is connected or not
Status status = check_connection();
if (!status.ok()) {
return status;
}
else {
Status status = SubscribeIndication();
if (!status.ok()) {
OPENOLT_LOG(ERROR, openolt_log_id, "SubscribeIndication failed - %s : %s\n",
grpc_status_code_to_string(status.error_code()).c_str(),
status.error_message().c_str());
return status;
}
//check BAL state in initial stage
status = check_bal_ready();
if (!status.ok()) {
return status;
}
}
{
bcmos_errno err;
bcmolt_odid dev;
OPENOLT_LOG(INFO, openolt_log_id, "Enabling PON %d Devices ... \n", BCM_MAX_DEVS_PER_LINE_CARD);
for (dev = 0; dev < BCM_MAX_DEVS_PER_LINE_CARD; dev++) {
bcmolt_device_cfg dev_cfg = { };
bcmolt_device_key dev_key = { };
dev_key.device_id = dev;
BCMOLT_CFG_INIT(&dev_cfg, device, dev_key);
BCMOLT_MSG_FIELD_GET(&dev_cfg, system_mode);
err = bcmolt_cfg_get(dev_id, &dev_cfg.hdr);
if (err == BCM_ERR_NOT_CONNECTED) {
bcmolt_device_key key = {.device_id = dev};
bcmolt_device_connect oper;
BCMOLT_OPER_INIT(&oper, device, connect, key);
if (MODEL_ID == "asfvolt16") {
BCMOLT_MSG_FIELD_SET(&oper, inni_config.mode, BCMOLT_INNI_MODE_ALL_10_G_XFI);
BCMOLT_MSG_FIELD_SET (&oper, system_mode, BCMOLT_SYSTEM_MODE_XGS__2_X);
} else if (MODEL_ID == "asgvolt64") {
BCMOLT_MSG_FIELD_SET(&oper, inni_config.mode, BCMOLT_INNI_MODE_ALL_10_G_XFI);
BCMOLT_MSG_FIELD_SET(&oper, inni_config.mux, BCMOLT_INNI_MUX_FOUR_TO_ONE);
BCMOLT_MSG_FIELD_SET (&oper, system_mode, BCMOLT_SYSTEM_MODE_GPON__16_X);
}
err = bcmolt_oper_submit(dev_id, &oper.hdr);
if (err) {
failed_enable_device_cnt ++;
OPENOLT_LOG(ERROR, openolt_log_id, "Enable PON device %d failed, err = %s\n", dev, bcmos_strerror(err));
if (failed_enable_device_cnt == BCM_MAX_DEVS_PER_LINE_CARD) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to enable all the pon ports, err = %s\n", bcmos_strerror(err));
return Status(grpc::StatusCode::INTERNAL, "Failed to activate all PON ports");
}
}
bcmos_usleep(200000);
}
else {
OPENOLT_LOG(WARNING, openolt_log_id, "PON deivce %d already connected\n", dev);
state.activate();
}
}
init_stats();
}
}
/* Start CLI */
OPENOLT_LOG(INFO, def_log_id, "Starting CLI\n");
//If already enabled, generate an extra indication ????
return Status::OK;
}
Status Disable_() {
//In the earlier implementation Disabling olt is done by disabling the NNI port associated with that.
//In inband scenario instead of using management interface to establish connection with adapter ,NNI interface will be used.
//Disabling NNI port on olt disable causes connection loss between adapter and agent.
//To overcome this disable is implemented by disabling all the PON ports
//associated with the device so as to support both in-band
//and out of band scenarios.
Status status;
int failedCount = 0;
for (int i = 0; i < NumPonIf_(); i++) {
status = DisablePonIf_(i);
if (!status.ok()) {
failedCount+=1;
BCM_LOG(ERROR, openolt_log_id, "Failed to disable PON interface: %d\n", i);
}
}
if (failedCount == 0) {
state.deactivate();
openolt::Indication ind;
openolt::OltIndication* olt_ind = new openolt::OltIndication;
olt_ind->set_oper_state("down");
ind.set_allocated_olt_ind(olt_ind);
BCM_LOG(INFO, openolt_log_id, "Disable OLT, add an extra indication\n");
oltIndQ.push(ind);
return Status::OK;
}
if (failedCount ==NumPonIf_()) {
return grpc::Status(grpc::StatusCode::INTERNAL, "failed to disable olt ,all the PON ports are still in enabled state");
}
return grpc::Status(grpc::StatusCode::UNKNOWN, "failed to disable olt ,few PON ports are still in enabled state");
}
Status Reenable_() {
Status status;
int failedCount = 0;
for (int i = 0; i < NumPonIf_(); i++) {
status = EnablePonIf_(i);
if (!status.ok()) {
failedCount+=1;
BCM_LOG(ERROR, openolt_log_id, "Failed to enable PON interface: %d\n", i);
}
}
if (failedCount == 0) {
state.activate();
openolt::Indication ind;
openolt::OltIndication* olt_ind = new openolt::OltIndication;
olt_ind->set_oper_state("up");
ind.set_allocated_olt_ind(olt_ind);
BCM_LOG(INFO, openolt_log_id, "Reenable OLT, add an extra indication\n");
oltIndQ.push(ind);
return Status::OK;
}
if (failedCount ==NumPonIf_()) {
return grpc::Status(grpc::StatusCode::INTERNAL, "failed to re-enable olt ,all the PON ports are still in disabled state");
}
return grpc::Status(grpc::StatusCode::UNKNOWN, "failed to re-enable olt ,few PON ports are still in disabled state");
}
bcmos_errno get_pon_interface_status(bcmolt_interface pon_ni, bcmolt_interface_state *state) {
bcmos_errno err;
bcmolt_pon_interface_key pon_key;
bcmolt_pon_interface_cfg pon_cfg;
pon_key.pon_ni = pon_ni;
BCMOLT_CFG_INIT(&pon_cfg, pon_interface, pon_key);
BCMOLT_FIELD_SET_PRESENT(&pon_cfg.data, pon_interface_cfg_data, state);
BCMOLT_FIELD_SET_PRESENT(&pon_cfg.data, pon_interface_cfg_data, itu);
#ifdef TEST_MODE
// It is impossible to mock the setting of pon_cfg.data.state because
// the actual bcmolt_cfg_get passes the address of pon_cfg.hdr and we cannot
// set the pon_cfg.data.state. So a new stub function is created and address
// of pon_cfg is passed. This is one-of case where we need to add test specific
// code in production code.
err = bcmolt_cfg_get__pon_intf_stub(dev_id, &pon_cfg);
#else
err = bcmolt_cfg_get(dev_id, &pon_cfg.hdr);
#endif
*state = pon_cfg.data.state;
return err;
}
inline uint64_t get_flow_status(uint16_t flow_id, uint16_t flow_type, uint16_t data_id) {
bcmos_errno err;
bcmolt_flow_key flow_key;
bcmolt_flow_cfg flow_cfg;
flow_key.flow_id = flow_id;
flow_key.flow_type = (bcmolt_flow_type)flow_type;
BCMOLT_CFG_INIT(&flow_cfg, flow, flow_key);
switch (data_id) {
case ONU_ID: //onu_id
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, onu_id);
#ifdef TEST_MODE
// It is impossible to mock the setting of flow_cfg.data.state because
// the actual bcmolt_cfg_get passes the address of flow_cfg.hdr and we cannot
// set the flow_cfg.data. So a new stub function is created and address
// of flow_cfg is passed. This is one-of case where we need to add test specific
// code in production code.
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get onu_id, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.onu_id;
case FLOW_TYPE:
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get flow_type, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.key.flow_type;
case SVC_PORT_ID: //svc_port_id
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, svc_port_id);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get svc_port_id, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.svc_port_id;
case PRIORITY:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, priority);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get priority, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.priority;
case COOKIE: //cookie
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, cookie);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get cookie, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.cookie;
case INGRESS_INTF_TYPE: //ingress intf_type
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, ingress_intf);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get ingress intf_type, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.ingress_intf.intf_type;
case EGRESS_INTF_TYPE: //egress intf_type
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, egress_intf);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get egress intf_type, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.egress_intf.intf_type;
case INGRESS_INTF_ID: //ingress intf_id
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, ingress_intf);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get ingress intf_id, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.ingress_intf.intf_id;
case EGRESS_INTF_ID: //egress intf_id
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, egress_intf);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get egress intf_id, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.egress_intf.intf_id;
case CLASSIFIER_O_VID:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, classifier);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get classifier o_vid, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.classifier.o_vid;
case CLASSIFIER_O_PBITS:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, classifier);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get classifier o_pbits, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.classifier.o_pbits;
case CLASSIFIER_I_VID:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, classifier);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get classifier i_vid, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.classifier.i_vid;
case CLASSIFIER_I_PBITS:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, classifier);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get classifier i_pbits, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.classifier.i_pbits;
case CLASSIFIER_ETHER_TYPE:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, classifier);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get classifier ether_type, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.classifier.ether_type;
case CLASSIFIER_IP_PROTO:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, classifier);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get classifier ip_proto, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.classifier.ip_proto;
case CLASSIFIER_SRC_PORT:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, classifier);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get classifier src_port, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.classifier.src_port;
case CLASSIFIER_DST_PORT:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, classifier);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get classifier dst_port, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.classifier.dst_port;
case CLASSIFIER_PKT_TAG_TYPE:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, classifier);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get classifier pkt_tag_type, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.classifier.pkt_tag_type;
case EGRESS_QOS_TYPE:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, egress_qos);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get egress_qos type, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.egress_qos.type;
case EGRESS_QOS_QUEUE_ID:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, egress_qos);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get egress_qos queue_id, err = %s\n",bcmos_strerror(err));
return err;
}
switch (flow_cfg.data.egress_qos.type) {
case BCMOLT_EGRESS_QOS_TYPE_FIXED_QUEUE:
return flow_cfg.data.egress_qos.u.fixed_queue.queue_id;
case BCMOLT_EGRESS_QOS_TYPE_TC_TO_QUEUE:
return flow_cfg.data.egress_qos.u.tc_to_queue.tc_to_queue_id;
case BCMOLT_EGRESS_QOS_TYPE_PBIT_TO_TC:
return flow_cfg.data.egress_qos.u.pbit_to_tc.tc_to_queue_id;
case BCMOLT_EGRESS_QOS_TYPE_PRIORITY_TO_QUEUE:
return flow_cfg.data.egress_qos.u.priority_to_queue.tm_q_set_id;
case BCMOLT_EGRESS_QOS_TYPE_NONE:
default:
return -1;
}
case EGRESS_QOS_TM_SCHED_ID:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, egress_qos);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get egress_qos tm_sched_id, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.egress_qos.tm_sched.id;
case ACTION_CMDS_BITMASK:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, action);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get action cmds_bitmask, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.action.cmds_bitmask;
case ACTION_O_VID:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, action);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get action o_vid, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.action.o_vid;
case ACTION_O_PBITS:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, action);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get action o_pbits, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.action.o_pbits;
case ACTION_I_VID:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, action);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get action i_vid, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.action.i_vid;
case ACTION_I_PBITS:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, action);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get action i_pbits, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.action.i_pbits;
case STATE:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, state);
#ifdef TEST_MODE
err = bcmolt_cfg_get__flow_stub(dev_id, &flow_cfg);
#else
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get state, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.state;
case GROUP_ID:
BCMOLT_FIELD_SET_PRESENT(&flow_cfg.data, flow_cfg_data, group_id);
err = bcmolt_cfg_get(dev_id, &flow_cfg.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get group_id, err = %s\n",bcmos_strerror(err));
return err;
}
return flow_cfg.data.group_id;
default:
return BCM_ERR_INTERNAL;
}
return err;
}
Status EnablePonIf_(uint32_t intf_id) {
bcmos_errno err = BCM_ERR_OK;
bcmolt_pon_interface_cfg interface_obj;
bcmolt_pon_interface_key intf_key = {.pon_ni = (bcmolt_interface)intf_id};
bcmolt_pon_interface_set_pon_interface_state pon_interface_set_state;
bcmolt_interface_state state;
err = get_pon_interface_status((bcmolt_interface)intf_id, &state);
if (err == BCM_ERR_OK) {
if (state == BCMOLT_INTERFACE_STATE_ACTIVE_WORKING) {
OPENOLT_LOG(WARNING, openolt_log_id, "PON interface: %d already enabled\n", intf_id);
return Status::OK;
}
}
BCMOLT_CFG_INIT(&interface_obj, pon_interface, intf_key);
BCMOLT_OPER_INIT(&pon_interface_set_state, pon_interface, set_pon_interface_state, intf_key);
BCMOLT_MSG_FIELD_SET(&interface_obj, discovery.control, BCMOLT_CONTROL_STATE_ENABLE);
BCMOLT_MSG_FIELD_SET(&interface_obj, discovery.interval, 5000);
BCMOLT_MSG_FIELD_SET(&interface_obj, discovery.onu_post_discovery_mode,
BCMOLT_ONU_POST_DISCOVERY_MODE_ACTIVATE);
BCMOLT_MSG_FIELD_SET(&interface_obj, itu.automatic_onu_deactivation.los, true);
BCMOLT_MSG_FIELD_SET(&interface_obj, itu.automatic_onu_deactivation.onu_alarms, true);
BCMOLT_MSG_FIELD_SET(&interface_obj, itu.automatic_onu_deactivation.tiwi, true);
BCMOLT_MSG_FIELD_SET(&interface_obj, itu.automatic_onu_deactivation.ack_timeout, true);
BCMOLT_MSG_FIELD_SET(&interface_obj, itu.automatic_onu_deactivation.sfi, true);
BCMOLT_MSG_FIELD_SET(&interface_obj, itu.automatic_onu_deactivation.loki, true);
BCMOLT_FIELD_SET(&pon_interface_set_state.data, pon_interface_set_pon_interface_state_data,
operation, BCMOLT_INTERFACE_OPERATION_ACTIVE_WORKING);
err = bcmolt_cfg_set(dev_id, &interface_obj.hdr);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to enable discovery onu, PON interface %d, err = %s\n", intf_id, bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to enable discovery onu");
}
err = bcmolt_oper_submit(dev_id, &pon_interface_set_state.hdr);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to enable PON interface: %d, err = %s\n", intf_id, bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to enable PON interface");
}
else {
OPENOLT_LOG(INFO, openolt_log_id, "Successfully enabled PON interface: %d\n", intf_id);
OPENOLT_LOG(INFO, openolt_log_id, "Initializing tm sched creation for PON interface: %d\n", intf_id);
CreateDefaultSched(intf_id, downstream);
CreateDefaultQueue(intf_id, downstream);
}
return Status::OK;
}
/* Same as bcmolt_cfg_get but with added logic of retrying the API
in case of some specific failures like timeout or object not yet ready
*/
bcmos_errno bcmolt_cfg_get_mult_retry(bcmolt_oltid olt, bcmolt_cfg *cfg) {
bcmos_errno err;
uint32_t current_try = 0;
while (current_try < MAX_BAL_API_RETRY_COUNT) {
err = bcmolt_cfg_get(olt, cfg);
current_try++;
if (err == BCM_ERR_STATE || err == BCM_ERR_TIMEOUT) {
OPENOLT_LOG(WARNING, openolt_log_id, "bcmolt_cfg_get: err = %s\n", bcmos_strerror(err));
bcmos_usleep(BAL_API_RETRY_TIME_IN_USECS);
continue;
}
else {
break;
}
}
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "bcmolt_cfg_get tried (%d) times with retry time(%d usecs) err = %s\n",
current_try,
BAL_API_RETRY_TIME_IN_USECS,
bcmos_strerror(err));
}
return err;
}
Status ProbeDeviceCapabilities_() {
bcmos_errno err;
bcmolt_device_cfg dev_cfg = { };
bcmolt_device_key dev_key = { };
bcmolt_olt_cfg olt_cfg = { };
bcmolt_olt_key olt_key = { };
bcmolt_topology_map topo_map[BCM_MAX_PONS_PER_OLT] = { };
bcmolt_topology topo = { };
topo.topology_maps.len = BCM_MAX_PONS_PER_OLT;
topo.topology_maps.arr = &topo_map[0];
BCMOLT_CFG_INIT(&olt_cfg, olt, olt_key);
BCMOLT_MSG_FIELD_GET(&olt_cfg, bal_state);
BCMOLT_FIELD_SET_PRESENT(&olt_cfg.data, olt_cfg_data, topology);
BCMOLT_CFG_LIST_BUF_SET(&olt_cfg, olt, topo.topology_maps.arr,
sizeof(bcmolt_topology_map) * topo.topology_maps.len);
#ifdef TEST_MODE
// It is impossible to mock the setting of olt_cfg.data.bal_state because
// the actual bcmolt_cfg_get passes the address of olt_cfg.hdr and we cannot
// set the olt_cfg.data.topology. So a new stub function is created and address
// of olt_cfg is passed. This is one-of case where we need to test add specific
// code in production code.
err = bcmolt_cfg_get__olt_topology_stub(dev_id, &olt_cfg);
#else
err = bcmolt_cfg_get_mult_retry(dev_id, &olt_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "cfg: Failed to query OLT topology, err = %s\n", bcmos_strerror(err));
return bcm_to_grpc_err(err, "cfg: Failed to query OLT topology");
}
num_of_nni_ports = olt_cfg.data.topology.num_switch_ports;
num_of_pon_ports = olt_cfg.data.topology.topology_maps.len;
OPENOLT_LOG(INFO, openolt_log_id, "OLT capabilitites, oper_state: %s\n",
olt_cfg.data.bal_state == BCMOLT_BAL_STATE_BAL_AND_SWITCH_READY
? "up" : "down");
OPENOLT_LOG(INFO, openolt_log_id, "topology nni: %d pon: %d dev: %d\n",
num_of_nni_ports,
num_of_pon_ports,
BCM_MAX_DEVS_PER_LINE_CARD);
uint32_t num_failed_cfg_gets = 0;
static std::string openolt_version = firmware_version;
for (int devid = 0; devid < BCM_MAX_DEVS_PER_LINE_CARD; devid++) {
dev_key.device_id = devid;
BCMOLT_CFG_INIT(&dev_cfg, device, dev_key);
BCMOLT_MSG_FIELD_GET(&dev_cfg, firmware_sw_version);
BCMOLT_MSG_FIELD_GET(&dev_cfg, chip_family);
BCMOLT_MSG_FIELD_GET(&dev_cfg, system_mode);
err = bcmolt_cfg_get_mult_retry(dev_id, &dev_cfg.hdr);
if (err) {
OPENOLT_LOG(WARNING, openolt_log_id,"Failed to query PON MAC Device %d (errno = %s). Skipping the device.\n", devid, bcmos_strerror(err));
num_failed_cfg_gets++;
continue;
}
std::string bal_version;
bal_version += std::to_string(dev_cfg.data.firmware_sw_version.major)
+ "." + std::to_string(dev_cfg.data.firmware_sw_version.minor)
+ "." + std::to_string(dev_cfg.data.firmware_sw_version.revision);
firmware_version = "BAL." + bal_version + "__" + openolt_version;
switch(dev_cfg.data.system_mode) {
case 10: board_technology = "GPON"; FILL_ARRAY(intf_technologies,devid*4,(devid+1)*4,"GPON"); break;
case 11: board_technology = "GPON"; FILL_ARRAY(intf_technologies,devid*8,(devid+1)*8,"GPON"); break;
case 12: board_technology = "GPON"; FILL_ARRAY(intf_technologies,devid*16,(devid+1)*16,"GPON"); break;
case 13: board_technology = "XGPON"; FILL_ARRAY(intf_technologies,devid*2,(devid+1)*2,"XGPON"); break;
case 14: board_technology = "XGPON"; FILL_ARRAY(intf_technologies,devid*4,(devid+1)*4,"XGPON"); break;
case 15: board_technology = "XGPON"; FILL_ARRAY(intf_technologies,devid*8,(devid+1)*8,"XGPON"); break;
case 16: board_technology = "XGPON"; FILL_ARRAY(intf_technologies,devid*16,(devid+1)*16,"XGPON"); break;
case 18: board_technology = "XGS-PON"; FILL_ARRAY(intf_technologies,devid*2,(devid+1)*2,"XGS-PON"); break;
case 19: board_technology = "XGS-PON"; FILL_ARRAY(intf_technologies,devid*16,(devid+1)*16,"XGS-PON"); break;
case 20: board_technology = MIXED_TECH; FILL_ARRAY(intf_technologies,devid*2,(devid+1)*2,MIXED_TECH); break;
}
switch(dev_cfg.data.chip_family) {
case BCMOLT_CHIP_FAMILY_CHIP_FAMILY_6862_X: chip_family = "Maple"; break;
case BCMOLT_CHIP_FAMILY_CHIP_FAMILY_6865_X: chip_family = "Aspen"; break;
}
OPENOLT_LOG(INFO, openolt_log_id, "device %d, pon: %d, version %s, family: %s, board_technology: %s\n",
devid, BCM_MAX_PONS_PER_DEV, bal_version.c_str(), chip_family.c_str(), board_technology.c_str());
bcmos_usleep(500000);
}
/* If all the devices returned errors then we tell the caller that this is an error else we work with
only the devices that retured success*/
if (num_failed_cfg_gets == BCM_MAX_DEVS_PER_LINE_CARD) {
OPENOLT_LOG(ERROR, openolt_log_id, "device: Query of all the devices failed\n");
return bcm_to_grpc_err(err, "device: All devices failed query");
}
return Status::OK;
}
#if 0
Status ProbePonIfTechnology_() {
// Probe maximum extent possible as configured into BAL driver to determine
// which are active in the current BAL topology. And for those
// that are active, determine each port's access technology, i.e. "gpon" or "xgspon".
for (uint32_t intf_id = 0; intf_id < num_of_pon_ports; ++intf_id) {
bcmolt_pon_interface_cfg interface_obj;
bcmolt_pon_interface_key interface_key;
interface_key.pon_ni = intf_id;
BCMOLT_CFG_INIT(&interface_obj, pon_interface, interface_key);
if (board_technology == "XGS-PON"
BCMOLT_MSG_FIELD_GET(&interface_obj, xgs_ngpon2_trx);
else if (board_technology == "GPON")
BCMOLT_MSG_FIELD_GET(&interface_obj, gpon_trx);
bcmos_errno err = bcmolt_cfg_get(dev_id, &interface_obj.hdr);
if (err != BCM_ERR_OK) {
intf_technologies[intf_id] = UNKNOWN_TECH;
if(err != BCM_ERR_RANGE) OPENOLT_LOG(ERROR, openolt_log_id, "Failed to get PON config: %d err %d\n", intf_id, err);
}
else {
if (board_technology == "XGS-PON") {
switch(interface_obj.data.xgpon_trx.transceiver_type) {
case BCMOLT_XGPON_TRX_TYPE_LTH_7222_PC:
case BCMOLT_XGPON_TRX_TYPE_WTD_RTXM266_702:
case BCMOLT_XGPON_TRX_TYPE_LTH_7222_BC_PLUS:
case BCMOLT_XGPON_TRX_TYPE_LTH_7226_PC:
case BCMOLT_XGPON_TRX_TYPE_LTH_5302_PC:
case BCMOLT_XGPON_TRX_TYPE_LTH_7226_A_PC_PLUS:
case BCMOLT_XGPON_TRX_TYPE_D272RR_SSCB_DM:
intf_technologies[intf_id] = "XGS-PON";
break;
}
} else if (board_technology == "GPON") {
switch(interface_obj.data.gpon_trx.transceiver_type) {
case BCMOLT_TRX_TYPE_SPS_43_48_H_HP_CDE_SD_2013:
case BCMOLT_TRX_TYPE_LTE_3680_M:
case BCMOLT_TRX_TYPE_SOURCE_PHOTONICS:
case BCMOLT_TRX_TYPE_LTE_3680_P_TYPE_C_PLUS:
case BCMOLT_TRX_TYPE_LTE_3680_P_BC:
intf_technologies[intf_id] = "GPON";
break;
}
}
if (board_technology != UNKNOWN_TECH) {
board_technology = intf_technologies[intf_id];
} else if (board_technology != MIXED_TECH && board_technology != intf_technologies[intf_id]) {
intf_technologies[intf_id] = MIXED_TECH;
}
}
}
return Status::OK;
}
#endif
unsigned NumNniIf_() {return num_of_nni_ports;}
unsigned NumPonIf_() {return num_of_pon_ports;}
bcmos_errno get_nni_interface_status(bcmolt_interface id, bcmolt_interface_state *state) {
bcmos_errno err;
bcmolt_nni_interface_key nni_key;
bcmolt_nni_interface_cfg nni_cfg;
nni_key.id = id;
BCMOLT_CFG_INIT(&nni_cfg, nni_interface, nni_key);
BCMOLT_FIELD_SET_PRESENT(&nni_cfg.data, nni_interface_cfg_data, state);
#ifdef TEST_MODE
// It is impossible to mock the setting of nni_cfg.data.state because
// the actual bcmolt_cfg_get passes the address of nni_cfg.hdr and we cannot
// set the nni_cfg.data.state. So a new stub function is created and address
// of nni_cfg is passed. This is one-of case where we need to add test specific
// code in production code.
err = bcmolt_cfg_get__nni_intf_stub(dev_id, &nni_cfg);
#else
err = bcmolt_cfg_get(dev_id, &nni_cfg.hdr);
#endif
*state = nni_cfg.data.state;
return err;
}
Status SetStateUplinkIf_(uint32_t intf_id, bool set_state) {
bcmos_errno err = BCM_ERR_OK;
bcmolt_nni_interface_key intf_key = {.id = (bcmolt_interface)intf_id};
bcmolt_nni_interface_set_nni_state nni_interface_set_state;
bcmolt_interface_state state;
err = get_nni_interface_status((bcmolt_interface)intf_id, &state);
if (err == BCM_ERR_OK) {
if (set_state && state == BCMOLT_INTERFACE_STATE_ACTIVE_WORKING) {
OPENOLT_LOG(WARNING, openolt_log_id, "NNI interface: %d already enabled\n", intf_id);
OPENOLT_LOG(INFO, openolt_log_id, "Initializing tm sched creation for NNI interface: %d\n", intf_id);
CreateDefaultSched(intf_id, upstream);
CreateDefaultQueue(intf_id, upstream);
return Status::OK;
} else if (!set_state && state == BCMOLT_INTERFACE_STATE_INACTIVE) {
OPENOLT_LOG(INFO, openolt_log_id, "NNI interface: %d already disabled\n", intf_id);
return Status::OK;
}
}
BCMOLT_OPER_INIT(&nni_interface_set_state, nni_interface, set_nni_state, intf_key);
if (set_state) {
BCMOLT_FIELD_SET(&nni_interface_set_state.data, nni_interface_set_nni_state_data,
nni_state, BCMOLT_INTERFACE_OPERATION_ACTIVE_WORKING);
} else {
BCMOLT_FIELD_SET(&nni_interface_set_state.data, nni_interface_set_nni_state_data,
nni_state, BCMOLT_INTERFACE_OPERATION_INACTIVE);
}
err = bcmolt_oper_submit(dev_id, &nni_interface_set_state.hdr);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to %s NNI interface: %d, err = %s\n",
(set_state)?"enable":"disable", intf_id, bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to enable NNI interface");
}
else {
OPENOLT_LOG(INFO, openolt_log_id, "Successfully %s NNI interface: %d\n", (set_state)?"enable":"disable", intf_id);
if (set_state) {
OPENOLT_LOG(INFO, openolt_log_id, "Initializing tm sched creation for NNI interface: %d\n", intf_id);
CreateDefaultSched(intf_id, upstream);
CreateDefaultQueue(intf_id, upstream);
}
}
return Status::OK;
}
Status DisablePonIf_(uint32_t intf_id) {
bcmos_errno err;
bcmolt_pon_interface_cfg interface_obj;
bcmolt_pon_interface_key intf_key = {.pon_ni = (bcmolt_interface)intf_id};
bcmolt_pon_interface_set_pon_interface_state pon_interface_set_state;
BCMOLT_CFG_INIT(&interface_obj, pon_interface, intf_key);
BCMOLT_OPER_INIT(&pon_interface_set_state, pon_interface, set_pon_interface_state, intf_key);
BCMOLT_MSG_FIELD_SET(&interface_obj, discovery.control, BCMOLT_CONTROL_STATE_DISABLE);
err = bcmolt_cfg_set(dev_id, &interface_obj.hdr);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to disable discovery of onu, PON interface %d, err %d\n", intf_id, err);
return bcm_to_grpc_err(err, "Failed to disable discovery of onu");
}
BCMOLT_FIELD_SET(&pon_interface_set_state.data, pon_interface_set_pon_interface_state_data,
operation, BCMOLT_INTERFACE_OPERATION_INACTIVE);
err = bcmolt_oper_submit(dev_id, &pon_interface_set_state.hdr);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to disable PON interface: %d\n , err %d\n", intf_id, err);
return bcm_to_grpc_err(err, "Failed to disable PON interface");
}
OPENOLT_LOG(INFO, openolt_log_id, "Successfully disabled PON interface: %d\n", intf_id);
return Status::OK;
}
Status ActivateOnu_(uint32_t intf_id, uint32_t onu_id,
const char *vendor_id, const char *vendor_specific, uint32_t pir) {
bcmos_errno err = BCM_ERR_OK;
bcmolt_onu_cfg onu_cfg;
bcmolt_onu_key onu_key;
bcmolt_serial_number serial_number; /**< ONU serial number */
bcmolt_bin_str_36 registration_id; /**< ONU registration ID */
onu_key.onu_id = onu_id;
onu_key.pon_ni = intf_id;
BCMOLT_CFG_INIT(&onu_cfg, onu, onu_key);
BCMOLT_FIELD_SET_PRESENT(&onu_cfg.data, onu_cfg_data, onu_state);
#ifdef TEST_MODE
// It is impossible to mock the setting of onu_cfg.data.onu_state because
// the actual bcmolt_cfg_get passes the address of onu_cfg.hdr and we cannot
// set the onu_cfg.data.onu_state. So a new stub function is created and address
// of onu_cfg is passed. This is one-of case where we need to add test specific
// code in production code.
err = bcmolt_cfg_get__onu_state_stub(dev_id, &onu_cfg);
#else
err = bcmolt_cfg_get(dev_id, &onu_cfg.hdr);
#endif
if (err == BCM_ERR_OK) {
if ((onu_cfg.data.onu_state == BCMOLT_ONU_STATE_PROCESSING ||
onu_cfg.data.onu_state == BCMOLT_ONU_STATE_ACTIVE) ||
(onu_cfg.data.onu_state == BCMOLT_ONU_STATE_INACTIVE &&
onu_cfg.data.onu_old_state == BCMOLT_ONU_STATE_NOT_CONFIGURED))
return Status::OK;
}
OPENOLT_LOG(INFO, openolt_log_id, "Enabling ONU %d on PON %d : vendor id %s, \
vendor specific %s, pir %d\n", onu_id, intf_id, vendor_id,
vendor_specific_to_str(vendor_specific).c_str(), pir);
memcpy(serial_number.vendor_id.arr, vendor_id, 4);
memcpy(serial_number.vendor_specific.arr, vendor_specific, 4);
BCMOLT_CFG_INIT(&onu_cfg, onu, onu_key);
BCMOLT_MSG_FIELD_SET(&onu_cfg, itu.serial_number, serial_number);
BCMOLT_MSG_FIELD_SET(&onu_cfg, itu.auto_learning, BCMOS_TRUE);
/*set burst and data profiles to fec disabled*/
if (board_technology == "XGS-PON") {
BCMOLT_MSG_FIELD_SET(&onu_cfg, itu.xgpon.ranging_burst_profile, 2);
BCMOLT_MSG_FIELD_SET(&onu_cfg, itu.xgpon.data_burst_profile, 1);
} else if (board_technology == "GPON") {
BCMOLT_MSG_FIELD_SET(&onu_cfg, itu.gpon.ds_ber_reporting_interval, 1000000);
BCMOLT_MSG_FIELD_SET(&onu_cfg, itu.gpon.omci_port_id, onu_id);
}
err = bcmolt_cfg_set(dev_id, &onu_cfg.hdr);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to set activate ONU %d on PON %d, err = %s\n", onu_id, intf_id, bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to activate ONU");
}
return Status::OK;
}
Status DeactivateOnu_(uint32_t intf_id, uint32_t onu_id,
const char *vendor_id, const char *vendor_specific) {
bcmos_errno err = BCM_ERR_OK;
bcmolt_onu_set_onu_state onu_oper; /* declare main API struct */
bcmolt_onu_cfg onu_cfg;
bcmolt_onu_key onu_key; /**< Object key. */
bcmolt_onu_state onu_state;
onu_key.onu_id = onu_id;
onu_key.pon_ni = intf_id;
BCMOLT_CFG_INIT(&onu_cfg, onu, onu_key);
BCMOLT_FIELD_SET_PRESENT(&onu_cfg.data, onu_cfg_data, onu_state);
#ifdef TEST_MODE
// It is impossible to mock the setting of onu_cfg.data.onu_state because
// the actual bcmolt_cfg_get passes the address of onu_cfg.hdr and we cannot
// set the onu_cfg.data.onu_state. So a new stub function is created and address
// of onu_cfg is passed. This is one-of case where we need to add test specific
// code in production code.
err = bcmolt_cfg_get__onu_state_stub(dev_id, &onu_cfg);
onu_state = onu_cfg.data.onu_state;
#else
err = bcmolt_cfg_get(dev_id, &onu_cfg.hdr);
#endif
if (err == BCM_ERR_OK) {
switch (onu_state) {
case BCMOLT_ONU_STATE_ACTIVE:
BCMOLT_OPER_INIT(&onu_oper, onu, set_onu_state, onu_key);
BCMOLT_FIELD_SET(&onu_oper.data, onu_set_onu_state_data,
onu_state, BCMOLT_ONU_OPERATION_INACTIVE);
err = bcmolt_oper_submit(dev_id, &onu_oper.hdr);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to deactivate ONU %d on PON %d, err = %s\n", onu_id, intf_id, bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to deactivate ONU");
}
break;
}
}
return Status::OK;
}
Status DeleteOnu_(uint32_t intf_id, uint32_t onu_id,
const char *vendor_id, const char *vendor_specific) {
OPENOLT_LOG(INFO, openolt_log_id, "DeleteOnu ONU %d on PON %d : vendor id %s, vendor specific %s\n",
onu_id, intf_id, vendor_id, vendor_specific_to_str(vendor_specific).c_str());
// Need to deactivate before removing it (BAL rules)
DeactivateOnu_(intf_id, onu_id, vendor_id, vendor_specific);
// Sleep to allow the state to propagate
// We need the subscriber terminal object to be admin down before removal
// Without sleep the race condition is lost by ~ 20 ms
std::this_thread::sleep_for(std::chrono::milliseconds(100));
// TODO: Delete the schedulers and queues.
bcmolt_onu_cfg cfg_obj;
bcmolt_onu_key key;
//OPENOLT_LOG(INFO, openolt_log_id, "Processing subscriber terminal cfg clear for sub_term_id %d and intf_id %d\n",
// onu_id, intf_id);
OPENOLT_LOG(INFO, openolt_log_id, "Processing onu cfg clear for onu_id %d and intf_id %d\n",
onu_id, intf_id);
key.onu_id = onu_id;
key.pon_ni = intf_id;
BCMOLT_CFG_INIT(&cfg_obj, onu, key);
bcmos_errno err = bcmolt_cfg_clear(dev_id, &cfg_obj.hdr);
if (err != BCM_ERR_OK)
{
//OPENOLT_LOG(ERROR, openolt_log_id, "Failed to clear information for BAL subscriber_terminal_id %d, Interface ID %d, err = %s\n", onu_id, intf_id, bcmos_strerror(err));
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to clear information for BAL onu_id %d, Interface ID %d, err = %s\n", onu_id, intf_id, bcmos_strerror(err));
return Status(grpc::StatusCode::INTERNAL, "Failed to delete ONU");
}
return Status::OK;
}
#define MAX_CHAR_LENGTH 20
#define MAX_OMCI_MSG_LENGTH 44
Status OmciMsgOut_(uint32_t intf_id, uint32_t onu_id, const std::string pkt) {
bcmolt_bin_str buf = {};
bcmolt_onu_cpu_packets omci_cpu_packets;
bcmolt_onu_key key;
key.pon_ni = intf_id;
key.onu_id = onu_id;
BCMOLT_OPER_INIT(&omci_cpu_packets, onu, cpu_packets, key);
BCMOLT_MSG_FIELD_SET(&omci_cpu_packets, packet_type, BCMOLT_PACKET_TYPE_OMCI);
BCMOLT_MSG_FIELD_SET(&omci_cpu_packets, calc_crc, BCMOS_TRUE);
// ???
if ((pkt.size()/2) > MAX_OMCI_MSG_LENGTH) {
buf.len = MAX_OMCI_MSG_LENGTH;
} else {
buf.len = pkt.size()/2;
}
/* Send the OMCI packet using the BAL remote proxy API */
uint16_t idx1 = 0;
uint16_t idx2 = 0;
uint8_t arraySend[buf.len];
char str1[MAX_CHAR_LENGTH];
char str2[MAX_CHAR_LENGTH];
memset(&arraySend, 0, buf.len);
for (idx1=0,idx2=0; idx1<((buf.len)*2); idx1++,idx2++) {
sprintf(str1,"%c", pkt[idx1]);
sprintf(str2,"%c", pkt[++idx1]);
strcat(str1,str2);
arraySend[idx2] = strtol(str1, NULL, 16);
}
buf.arr = (uint8_t *)malloc((buf.len)*sizeof(uint8_t));
memcpy(buf.arr, (uint8_t *)arraySend, buf.len);
BCMOLT_MSG_FIELD_SET(&omci_cpu_packets, number_of_packets, 1);
BCMOLT_MSG_FIELD_SET(&omci_cpu_packets, packet_size, buf.len);
BCMOLT_MSG_FIELD_SET(&omci_cpu_packets, buffer, buf);
bcmos_errno err = bcmolt_oper_submit(dev_id, &omci_cpu_packets.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Error sending OMCI message to ONU %d on PON %d, err = %s\n", onu_id, intf_id, bcmos_strerror(err));
return bcm_to_grpc_err(err, "send OMCI failed");
} else {
OPENOLT_LOG(DEBUG, omci_log_id, "OMCI request msg of length %d sent to ONU %d on PON %d : %s\n",
buf.len, onu_id, intf_id, pkt.c_str());
}
free(buf.arr);
return Status::OK;
}
Status OnuPacketOut_(uint32_t intf_id, uint32_t onu_id, uint32_t port_no, uint32_t gemport_id, const std::string pkt) {
bcmolt_pon_interface_cpu_packets pon_interface_cpu_packets; /**< declare main API struct */
bcmolt_pon_interface_key key = {.pon_ni = (bcmolt_interface)intf_id}; /**< declare key */
bcmolt_bin_str buf = {};
bcmolt_gem_port_id gem_port_id_array[1];
bcmolt_gem_port_id_list_u8_max_16 gem_port_list = {};
if (port_no > 0) {
bool found = false;
if (gemport_id == 0) {
bcmos_fastlock_lock(&data_lock);
// Map the port_no to one of the flows that owns it to find a gemport_id for that flow.
// Pick any flow that is mapped with the same port_no.
std::map<uint32_t, std::set<uint32_t> >::const_iterator it = port_to_flows.find(port_no);
if (it != port_to_flows.end() && !it->second.empty()) {
uint32_t flow_id = *(it->second.begin()); // Pick any flow_id out of the bag set
std::map<uint32_t, uint32_t>::const_iterator fit = flowid_to_gemport.find(flow_id);
if (fit != flowid_to_gemport.end()) {
found = true;
gemport_id = fit->second;
}
}
bcmos_fastlock_unlock(&data_lock, 0);
if (!found) {
OPENOLT_LOG(ERROR, openolt_log_id, "Packet out failed to find destination for ONU %d port_no %u on PON %d\n",
onu_id, port_no, intf_id);
return grpc::Status(grpc::StatusCode::NOT_FOUND, "no flow for port_no");
}
OPENOLT_LOG(INFO, openolt_log_id, "Gem port %u found for ONU %d port_no %u on PON %d\n",
gemport_id, onu_id, port_no, intf_id);
}
gem_port_id_array[0] = gemport_id;
gem_port_list.len = 1;
gem_port_list.arr = gem_port_id_array;
buf.len = pkt.size();
buf.arr = (uint8_t *)malloc((buf.len)*sizeof(uint8_t));
memcpy(buf.arr, (uint8_t *)pkt.data(), buf.len);
/* init the API struct */
BCMOLT_OPER_INIT(&pon_interface_cpu_packets, pon_interface, cpu_packets, key);
BCMOLT_MSG_FIELD_SET(&pon_interface_cpu_packets, packet_type, BCMOLT_PACKET_TYPE_ETH);
BCMOLT_MSG_FIELD_SET(&pon_interface_cpu_packets, calc_crc, BCMOS_TRUE);
BCMOLT_MSG_FIELD_SET(&pon_interface_cpu_packets, gem_port_list, gem_port_list);
BCMOLT_MSG_FIELD_SET(&pon_interface_cpu_packets, buffer, buf);
OPENOLT_LOG(INFO, openolt_log_id, "Packet out of length %d sent to gemport %d on pon %d port_no %u\n",
(uint8_t)pkt.size(), gemport_id, intf_id, port_no);
/* call API */
bcmolt_oper_submit(dev_id, &pon_interface_cpu_packets.hdr);
}
else {
//TODO: Port No is 0, it is coming sender requirement.
OPENOLT_LOG(INFO, openolt_log_id, "port_no %d onu %d on pon %d\n",
port_no, onu_id, intf_id);
}
free(buf.arr);
return Status::OK;
}
Status UplinkPacketOut_(uint32_t intf_id, const std::string pkt) {
bcmolt_flow_key key = {}; /* declare key */
bcmolt_bin_str buffer = {};
bcmolt_flow_send_eth_packet oper; /* declare main API struct */
// TODO: flow_id is currently not passed in UplinkPacket message from voltha.
bcmolt_flow_id flow_id = 0;
//validate flow_id and find flow_id/flow type: upstream/ingress type: PON/egress type: NNI
if (get_flow_status(flow_id, BCMOLT_FLOW_TYPE_UPSTREAM, FLOW_TYPE) == BCMOLT_FLOW_TYPE_UPSTREAM && \
get_flow_status(flow_id, BCMOLT_FLOW_TYPE_UPSTREAM, INGRESS_INTF_TYPE) == BCMOLT_FLOW_INTERFACE_TYPE_PON && \
get_flow_status(flow_id, BCMOLT_FLOW_TYPE_UPSTREAM, EGRESS_INTF_TYPE) == BCMOLT_FLOW_INTERFACE_TYPE_NNI)
key.flow_id = flow_id;
else {
if (flow_id_counters) {
std::map<flow_pair, int>::iterator it;
for(it = flow_map.begin(); it != flow_map.end(); it++) {
int flow_index = it->first.first;
if (get_flow_status(flow_index, BCMOLT_FLOW_TYPE_UPSTREAM, FLOW_TYPE) == BCMOLT_FLOW_TYPE_UPSTREAM && \
get_flow_status(flow_index, BCMOLT_FLOW_TYPE_UPSTREAM, INGRESS_INTF_TYPE) == BCMOLT_FLOW_INTERFACE_TYPE_PON && \
get_flow_status(flow_index, BCMOLT_FLOW_TYPE_UPSTREAM, EGRESS_INTF_TYPE) == BCMOLT_FLOW_INTERFACE_TYPE_NNI) {
key.flow_id = flow_index;
break;
}
}
}
else {
OPENOLT_LOG(ERROR, openolt_log_id, "no flow id found for uplink packetout\n");
return grpc::Status(grpc::StatusCode::NOT_FOUND, "no flow id found");
}
}
key.flow_type = BCMOLT_FLOW_TYPE_UPSTREAM; /* send from uplink direction */
/* Initialize the API struct. */
BCMOLT_OPER_INIT(&oper, flow, send_eth_packet, key);
buffer.len = pkt.size();
buffer.arr = (uint8_t *)malloc((buffer.len)*sizeof(uint8_t));
memcpy(buffer.arr, (uint8_t *)pkt.data(), buffer.len);
if (buffer.arr == NULL) {
OPENOLT_LOG(ERROR, openolt_log_id, "allocate packet buffer failed\n");
return bcm_to_grpc_err(BCM_ERR_PARM, "allocate packet buffer failed");
}
BCMOLT_FIELD_SET(&oper.data, flow_send_eth_packet_data, buffer, buffer);
bcmos_errno err = bcmolt_oper_submit(dev_id, &oper.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Error sending packets via nni port %d, flow_id %d, err = %s\n", intf_id, key.flow_id, bcmos_strerror(err));
return bcm_to_grpc_err(BCM_ERR_SYSCALL_ERR, "Error sending packets via nni port");
} else {
OPENOLT_LOG(INFO, openolt_log_id, "sent packets to port %d in upstream direction, flow_id %d \n", intf_id, key.flow_id);
}
return Status::OK;
}
uint32_t GetPortNum_(uint32_t flow_id) {
bcmos_fastlock_lock(&data_lock);
uint32_t port_no = 0;
std::map<uint32_t, uint32_t >::const_iterator it = flowid_to_port.find(flow_id);
if (it != flowid_to_port.end()) {
port_no = it->second;
}
bcmos_fastlock_unlock(&data_lock, 0);
return port_no;
}
#define ACL_LOG(level,msg,err) \
do { \
OPENOLT_LOG(level, openolt_log_id, "--------> %s (acl_id %d) err: %d <--------\n", msg, key.id, err); \
OPENOLT_LOG(level, openolt_log_id, "action_type %s\n", \
GET_ACL_ACTION_TYPE(action_type)); \
OPENOLT_LOG(level, openolt_log_id, "classifier(ether type %d), ip_proto %d, src_port %d, dst_port %d\n", \
acl_key.ether_type, acl_key.ip_proto, acl_key.src_port, acl_key.dst_port); \
} while(0)
#define FLOW_LOG(level,msg,err) \
do { \
OPENOLT_LOG(level, openolt_log_id, "--------> %s (flow_id %d) err: %d <--------\n", msg, key.flow_id, err); \
OPENOLT_LOG(level, openolt_log_id, "intf_id %d, onu_id %d, uni_id %d, port_no %u, cookie %"PRIu64"\n", \
access_intf_id, onu_id, uni_id, port_no, cookie); \
OPENOLT_LOG(level, openolt_log_id, "flow_type %s, queue_id %d, sched_id %d\n", flow_type.c_str(), \
cfg.data.egress_qos.u.fixed_queue.queue_id, cfg.data.egress_qos.tm_sched.id); \
OPENOLT_LOG(level, openolt_log_id, "Ingress(intfd_type %s, intf_id %d), Egress(intf_type %s, intf_id %d)\n", \
GET_FLOW_INTERFACE_TYPE(cfg.data.ingress_intf.intf_type), cfg.data.ingress_intf.intf_id, \
GET_FLOW_INTERFACE_TYPE(cfg.data.egress_intf.intf_type), cfg.data.egress_intf.intf_id); \
OPENOLT_LOG(level, openolt_log_id, "classifier(o_vid %d, o_pbits %d, i_vid %d, i_pbits %d, ether type 0x%x)\n", \
c_val.o_vid, c_val.o_pbits, c_val.i_vid, c_val.i_pbits, classifier.eth_type()); \
OPENOLT_LOG(level, openolt_log_id, "classifier(ip_proto 0x%x, gemport_id %d, src_port %d, dst_port %d, pkt_tag_type %s)\n", \
c_val.ip_proto, gemport_id, c_val.src_port, c_val.dst_port, GET_PKT_TAG_TYPE(c_val.pkt_tag_type)); \
OPENOLT_LOG(level, openolt_log_id, "action(cmds_bitmask %s, o_vid %d, o_pbits %d, i_vid %d, i_pbits %d)\n\n", \
get_flow_acton_command(a_val.cmds_bitmask), a_val.o_vid, a_val.o_pbits, a_val.i_vid, a_val.i_pbits); \
} while(0)
#define FLOW_PARAM_LOG() \
do { \
OPENOLT_LOG(INFO, openolt_log_id, "--------> flow comparison (now before) <--------\n"); \
OPENOLT_LOG(INFO, openolt_log_id, "flow_id (%d %d)\n", \
key.flow_id, it->first.first); \
OPENOLT_LOG(INFO, openolt_log_id, "onu_id (%d %lu)\n", \
cfg.data.onu_id , get_flow_status(it->first.first, it->first.second, ONU_ID)); \
OPENOLT_LOG(INFO, openolt_log_id, "type (%d %lu)\n", \
key.flow_type, get_flow_status(it->first.first, it->first.second, FLOW_TYPE)); \
OPENOLT_LOG(INFO, openolt_log_id, "svc_port_id (%d %lu)\n", \
cfg.data.svc_port_id, get_flow_status(it->first.first, it->first.second, SVC_PORT_ID)); \
OPENOLT_LOG(INFO, openolt_log_id, "priority (%d %lu)\n", \
cfg.data.priority, get_flow_status(it->first.first, it->first.second, PRIORITY)); \
OPENOLT_LOG(INFO, openolt_log_id, "cookie (%lu %lu)\n", \
cfg.data.cookie, get_flow_status(it->first.first, it->first.second, COOKIE)); \
OPENOLT_LOG(INFO, openolt_log_id, "ingress intf_type (%s %s)\n", \
GET_FLOW_INTERFACE_TYPE(cfg.data.ingress_intf.intf_type), \
GET_FLOW_INTERFACE_TYPE(get_flow_status(it->first.first, it->first.second, INGRESS_INTF_TYPE))); \
OPENOLT_LOG(INFO, openolt_log_id, "ingress intf id (%d %lu)\n", \
cfg.data.ingress_intf.intf_id , get_flow_status(it->first.first, it->first.second, INGRESS_INTF_ID)); \
OPENOLT_LOG(INFO, openolt_log_id, "egress intf_type (%d %lu)\n", \
cfg.data.egress_intf.intf_type , get_flow_status(it->first.first, it->first.second, EGRESS_INTF_TYPE)); \
OPENOLT_LOG(INFO, openolt_log_id, "egress intf_id (%d %lu)\n", \
cfg.data.egress_intf.intf_id , get_flow_status(it->first.first, it->first.second, EGRESS_INTF_ID)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier o_vid (%d %lu)\n", \
c_val.o_vid , get_flow_status(it->first.first, it->first.second, CLASSIFIER_O_VID)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier o_pbits (%d %lu)\n", \
c_val.o_pbits , get_flow_status(it->first.first, it->first.second, CLASSIFIER_O_PBITS)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier i_vid (%d %lu)\n", \
c_val.i_vid , get_flow_status(it->first.first, it->first.second, CLASSIFIER_I_VID)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier i_pbits (%d %lu)\n", \
c_val.i_pbits , get_flow_status(it->first.first, it->first.second, CLASSIFIER_I_PBITS)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier ether_type (0x%x 0x%lx)\n", \
c_val.ether_type , get_flow_status(it->first.first, it->first.second, CLASSIFIER_ETHER_TYPE)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier ip_proto (%d %lu)\n", \
c_val.ip_proto , get_flow_status(it->first.first, it->first.second, CLASSIFIER_IP_PROTO)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier src_port (%d %lu)\n", \
c_val.src_port , get_flow_status(it->first.first, it->first.second, CLASSIFIER_SRC_PORT)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier dst_port (%d %lu)\n", \
c_val.dst_port , get_flow_status(it->first.first, it->first.second, CLASSIFIER_DST_PORT)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier pkt_tag_type (%s %s)\n", \
GET_PKT_TAG_TYPE(c_val.pkt_tag_type), \
GET_PKT_TAG_TYPE(get_flow_status(it->first.first, it->first.second, CLASSIFIER_PKT_TAG_TYPE))); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier egress_qos type (%d %lu)\n", \
cfg.data.egress_qos.type , get_flow_status(it->first.first, it->first.second, EGRESS_QOS_TYPE)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier egress_qos queue_id (%d %lu)\n", \
cfg.data.egress_qos.u.fixed_queue.queue_id, \
get_flow_status(it->first.first, it->first.second, EGRESS_QOS_QUEUE_ID)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier egress_qos sched_id (%d %lu)\n", \
cfg.data.egress_qos.tm_sched.id, \
get_flow_status(it->first.first, it->first.second, EGRESS_QOS_TM_SCHED_ID)); \
OPENOLT_LOG(INFO, openolt_log_id, "classifier cmds_bitmask (%s %s)\n", \
get_flow_acton_command(a_val.cmds_bitmask), \
get_flow_acton_command(get_flow_status(it->first.first, it->first.second, ACTION_CMDS_BITMASK))); \
OPENOLT_LOG(INFO, openolt_log_id, "action o_vid (%d %lu)\n", \
a_val.o_vid , get_flow_status(it->first.first, it->first.second, ACTION_O_VID)); \
OPENOLT_LOG(INFO, openolt_log_id, "action i_vid (%d %lu)\n", \
a_val.i_vid , get_flow_status(it->first.first, it->first.second, ACTION_I_VID)); \
OPENOLT_LOG(INFO, openolt_log_id, "action o_pbits (%d %lu)\n", \
a_val.o_pbits , get_flow_status(it->first.first, it->first.second, ACTION_O_PBITS)); \
OPENOLT_LOG(INFO, openolt_log_id, "action i_pbits (%d %lu)\n\n", \
a_val.i_pbits, get_flow_status(it->first.first, it->first.second, ACTION_I_PBITS)); \
OPENOLT_LOG(INFO, openolt_log_id, "group_id (%d %lu)\n\n", \
a_val.group_id, get_flow_status(it->first.first, it->first.second, GROUP_ID)); \
} while(0)
#define FLOW_CHECKER
//#define SHOW_FLOW_PARAM
Status install_gem_port(int32_t intf_id, int32_t onu_id, int32_t gemport_id) {
bcmos_errno err;
bcmolt_itupon_gem_cfg cfg; /* declare main API struct */
bcmolt_itupon_gem_key key = {}; /* declare key */
bcmolt_gem_port_configuration configuration = {};
key.pon_ni = intf_id;
key.gem_port_id = gemport_id;
BCMOLT_CFG_INIT(&cfg, itupon_gem, key);
bcmolt_gem_port_direction configuration_direction;
configuration_direction = BCMOLT_GEM_PORT_DIRECTION_BIDIRECTIONAL;
BCMOLT_FIELD_SET(&configuration, gem_port_configuration, direction, configuration_direction);
bcmolt_gem_port_type configuration_type;
configuration_type = BCMOLT_GEM_PORT_TYPE_UNICAST;
BCMOLT_FIELD_SET(&configuration, gem_port_configuration, type, configuration_type);
BCMOLT_FIELD_SET(&cfg.data, itupon_gem_cfg_data, configuration, configuration);
BCMOLT_FIELD_SET(&cfg.data, itupon_gem_cfg_data, onu_id, onu_id);
bcmolt_control_state encryption_mode;
encryption_mode = BCMOLT_CONTROL_STATE_DISABLE;
BCMOLT_FIELD_SET(&cfg.data, itupon_gem_cfg_data, encryption_mode, encryption_mode);
bcmolt_us_gem_port_destination upstream_destination_queue;
upstream_destination_queue = BCMOLT_US_GEM_PORT_DESTINATION_DATA;
BCMOLT_FIELD_SET(&cfg.data, itupon_gem_cfg_data, upstream_destination_queue, upstream_destination_queue);
bcmolt_control_state control;
control = BCMOLT_CONTROL_STATE_ENABLE;
BCMOLT_FIELD_SET(&cfg.data, itupon_gem_cfg_data, control, control);
err = bcmolt_cfg_set(dev_id, &cfg.hdr);
if(err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "failed to install gem_port = %d\n", gemport_id);
return bcm_to_grpc_err(err, "Access_Control set ITU PON Gem port failed");
}
OPENOLT_LOG(INFO, openolt_log_id, "gem port installed successfully = %d\n", gemport_id);
return Status::OK;
}
Status remove_gem_port(int32_t intf_id, int32_t gemport_id) {
bcmolt_itupon_gem_cfg gem_cfg;
bcmolt_itupon_gem_key key = {
.pon_ni = (bcmolt_interface)intf_id,
.gem_port_id = (bcmolt_gem_port_id)gemport_id
};
bcmos_errno err;
BCMOLT_CFG_INIT(&gem_cfg, itupon_gem, key);
err = bcmolt_cfg_clear(dev_id, &gem_cfg.hdr);
if (err != BCM_ERR_OK)
{
OPENOLT_LOG(ERROR, openolt_log_id, "failed to remove gem_port = %d err=%s\n", gemport_id, gem_cfg.hdr.hdr.err_text);
return bcm_to_grpc_err(err, "Access_Control clear ITU PON Gem port failed");
}
OPENOLT_LOG(INFO, openolt_log_id, "gem port removed successfully = %d\n", gemport_id);
return Status::OK;
}
Status update_acl_interface(int32_t intf_id, bcmolt_interface_type intf_type, uint32_t access_control_id,
bcmolt_members_update_command acl_cmd) {
bcmos_errno err;
bcmolt_access_control_interfaces_update oper; /* declare main API struct */
bcmolt_access_control_key acl_key = {}; /* declare key */
bcmolt_intf_ref interface_ref_list_elem = {};
bcmolt_interface_type interface_ref_list_elem_intf_type;
bcmolt_interface_id interface_ref_list_elem_intf_id;
bcmolt_intf_ref_list_u8 interface_ref_list = {};
if (acl_cmd != BCMOLT_MEMBERS_UPDATE_COMMAND_ADD && acl_cmd != BCMOLT_MEMBERS_UPDATE_COMMAND_REMOVE) {
OPENOLT_LOG(ERROR, openolt_log_id, "acl cmd = %d not supported currently\n", acl_cmd);
return bcm_to_grpc_err(BCM_ERR_PARM, "unsupported acl cmd");
}
interface_ref_list.arr = (bcmolt_intf_ref*)bcmos_calloc(sizeof(bcmolt_intf_ref)*1);
if (interface_ref_list.arr == NULL)
return bcm_to_grpc_err(BCM_ERR_PARM, "allocate interface_ref_list failed");
OPENOLT_LOG(INFO, openolt_log_id, "update acl interface received for intf_id = %d, intf_type = %s, acl_id = %d, acl_cmd = %s\n",
intf_id, intf_type == BCMOLT_INTERFACE_TYPE_PON? "pon": "nni", access_control_id,
acl_cmd == BCMOLT_MEMBERS_UPDATE_COMMAND_ADD? "add": "remove");
acl_key.id = access_control_id;
/* Initialize the API struct. */
BCMOLT_OPER_INIT(&oper, access_control, interfaces_update, acl_key);
bcmolt_members_update_command command;
command = acl_cmd;
BCMOLT_FIELD_SET(&oper.data, access_control_interfaces_update_data, command, command);
interface_ref_list_elem_intf_type = intf_type;
BCMOLT_FIELD_SET(&interface_ref_list_elem, intf_ref, intf_type, interface_ref_list_elem_intf_type);
interface_ref_list_elem_intf_id = intf_id;
BCMOLT_FIELD_SET(&interface_ref_list_elem, intf_ref, intf_id, interface_ref_list_elem_intf_id);
interface_ref_list.len = 1;
BCMOLT_ARRAY_ELEM_SET(&interface_ref_list, 0, interface_ref_list_elem);
BCMOLT_FIELD_SET(&oper.data, access_control_interfaces_update_data, interface_ref_list, interface_ref_list);
err = bcmolt_oper_submit(dev_id, &oper.hdr);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "update acl interface fail for intf_id = %d, intf_type = %s, acl_id = %d, acl_cmd = %s\n",
intf_id, intf_type == BCMOLT_INTERFACE_TYPE_PON? "pon": "nni", access_control_id,
acl_cmd == BCMOLT_MEMBERS_UPDATE_COMMAND_ADD? "add": "remove");
return bcm_to_grpc_err(err, "Access_Control submit interface failed");
}
bcmos_free(interface_ref_list.arr);
OPENOLT_LOG(INFO, openolt_log_id, "update acl interface success for intf_id = %d, intf_type = %s, acl_id = %d, acl_cmd = %s\n",
intf_id, intf_type == BCMOLT_INTERFACE_TYPE_PON? "pon": "nni", access_control_id,
acl_cmd == BCMOLT_MEMBERS_UPDATE_COMMAND_ADD? "add": "remove");
return Status::OK;
}
Status install_acl(const acl_classifier_key acl_key) {
bcmos_errno err;
bcmolt_access_control_cfg cfg;
bcmolt_access_control_key key = { };
bcmolt_classifier c_val = { };
// hardcode the action for now.
bcmolt_access_control_fwd_action_type action_type = BCMOLT_ACCESS_CONTROL_FWD_ACTION_TYPE_TRAP_TO_HOST;
int acl_id = get_acl_id();
if (acl_id < 0) {
OPENOLT_LOG(ERROR, openolt_log_id, "exhausted acl_id for eth_type = %d, ip_proto = %d, src_port = %d, dst_port = %d\n",
acl_key.ether_type, acl_key.ip_proto, acl_key.src_port, acl_key.dst_port);
bcmos_fastlock_unlock(&data_lock, 0);
return bcm_to_grpc_err(BCM_ERR_INTERNAL, "exhausted acl id");
}
key.id = acl_id;
/* config access control instance */
BCMOLT_CFG_INIT(&cfg, access_control, key);
if (acl_key.ether_type > 0) {
OPENOLT_LOG(DEBUG, openolt_log_id, "Access_Control classify ether_type 0x%04x\n", acl_key.ether_type);
BCMOLT_FIELD_SET(&c_val, classifier, ether_type, acl_key.ether_type);
}
if (acl_key.ip_proto > 0) {
OPENOLT_LOG(DEBUG, openolt_log_id, "Access_Control classify ip_proto %d\n", acl_key.ip_proto);
BCMOLT_FIELD_SET(&c_val, classifier, ip_proto, acl_key.ip_proto);
}
if (acl_key.dst_port > 0) {
OPENOLT_LOG(DEBUG, openolt_log_id, "Access_Control classify dst_port %d\n", acl_key.dst_port);
BCMOLT_FIELD_SET(&c_val, classifier, dst_port, acl_key.dst_port);
}
if (acl_key.src_port > 0) {
OPENOLT_LOG(DEBUG, openolt_log_id, "Access_Control classify src_port %d\n", acl_key.src_port);
BCMOLT_FIELD_SET(&c_val, classifier, src_port, acl_key.src_port);
}
BCMOLT_MSG_FIELD_SET(&cfg, classifier, c_val);
BCMOLT_MSG_FIELD_SET(&cfg, priority, 10000);
BCMOLT_MSG_FIELD_SET(&cfg, statistics_control, BCMOLT_CONTROL_STATE_ENABLE);
BCMOLT_MSG_FIELD_SET(&cfg, forwarding_action.action, action_type);
err = bcmolt_cfg_set(dev_id, &cfg.hdr);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Access_Control set configuration failed, Error %d\n", err);
// Free the acl_id
free_acl_id(acl_id);
return bcm_to_grpc_err(err, "Access_Control set configuration failed");
}
ACL_LOG(INFO, "ACL add ok", err);
// Update the map that we have installed an acl for the given classfier.
acl_classifier_to_acl_id_map[acl_key] = acl_id;
return Status::OK;
}
Status remove_acl(int acl_id) {
bcmos_errno err;
bcmolt_access_control_cfg cfg; /* declare main API struct */
bcmolt_access_control_key key = {}; /* declare key */
key.id = acl_id;
/* Initialize the API struct. */
BCMOLT_CFG_INIT(&cfg, access_control, key);
BCMOLT_FIELD_SET_PRESENT(&cfg.data, access_control_cfg_data, state);
err = bcmolt_cfg_get(dev_id, &cfg.hdr);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Access_Control get state failed\n");
return bcm_to_grpc_err(err, "Access_Control get state failed");
}
if (cfg.data.state == BCMOLT_CONFIG_STATE_CONFIGURED) {
key.id = acl_id;
/* Initialize the API struct. */
BCMOLT_CFG_INIT(&cfg, access_control, key);
err = bcmolt_cfg_clear(dev_id, &cfg.hdr);
if (err != BCM_ERR_OK) {
// Should we free acl_id here ? We should ideally never land here..
OPENOLT_LOG(ERROR, openolt_log_id, "Error %d while removing Access_Control rule ID %d\n",
err, acl_id);
return Status(grpc::StatusCode::INTERNAL, "Failed to remove Access_Control");
}
}
// Free up acl_id
free_acl_id(acl_id);
OPENOLT_LOG(INFO, openolt_log_id, "acl removed successfully %d\n", acl_id);
return Status::OK;
}
// Formulates ACL Classifier Key based on the following fields
// a. ether_type b. ip_proto c. src_port d. dst_port
// If any of the field is not available it is populated as -1.
void formulate_acl_classifier_key(acl_classifier_key *key, const ::openolt::Classifier& classifier) {
// TODO: Is 0 a valid value for any of the following classifiers?
// because in the that case, the 'if' check would fail and -1 would be filled as value.
//
if (classifier.eth_type()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "classify ether_type 0x%04x\n", classifier.eth_type());
key->ether_type = classifier.eth_type();
} else key->ether_type = -1;
if (classifier.ip_proto()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "classify ip_proto %d\n", classifier.ip_proto());
key->ip_proto = classifier.ip_proto();
} else key->ip_proto = -1;
if (classifier.src_port()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "classify src_port %d\n", classifier.src_port());
key->src_port = classifier.src_port();
} else key->src_port = -1;
if (classifier.dst_port()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "classify dst_port %d\n", classifier.dst_port());
key->dst_port = classifier.dst_port();
} else key->dst_port = -1;
}
Status handle_acl_rule_install(int32_t onu_id, uint32_t flow_id,
const std::string flow_type, int32_t access_intf_id,
int32_t network_intf_id, int32_t gemport_id,
const ::openolt::Classifier& classifier) {
int acl_id;
int32_t intf_id = flow_type.compare(upstream) == 0? access_intf_id: network_intf_id;
const std::string intf_type = flow_type.compare(upstream) == 0? "pon": "nni";
bcmolt_interface_type olt_if_type = intf_type == "pon"? BCMOLT_INTERFACE_TYPE_PON: BCMOLT_INTERFACE_TYPE_NNI;
Status resp;
// few map keys we are going to use later.
flow_id_flow_direction fl_id_fl_dir(flow_id, flow_type);
gem_id_intf_id gem_intf(gemport_id, access_intf_id);
acl_classifier_key acl_key;
formulate_acl_classifier_key(&acl_key, classifier);
const acl_classifier_key acl_key_const = {.ether_type=acl_key.ether_type, .ip_proto=acl_key.ip_proto,
.src_port=acl_key.src_port, .dst_port=acl_key.dst_port};
bcmos_fastlock_lock(&data_lock);
// Check if the acl is already installed
if (acl_classifier_to_acl_id_map.count(acl_key_const) > 0) {
// retreive the acl_id
acl_id = acl_classifier_to_acl_id_map[acl_key_const];
acl_id_gem_id_intf_id ac_id_gm_id_if_id(acl_id, gemport_id, intf_id);
if (flow_to_acl_map.count(fl_id_fl_dir)) {
// coult happen if same trap flow is received again
OPENOLT_LOG(INFO, openolt_log_id, "flow and related acl already handled, nothing more to do\n");
bcmos_fastlock_unlock(&data_lock, 0);
return Status::OK;
}
OPENOLT_LOG(INFO, openolt_log_id, "Acl for flow_id=%u with eth_type = %d, ip_proto = %d, src_port = %d, dst_port = %d already installed with acl id = %u\n",
flow_id, acl_key.ether_type, acl_key.ip_proto, acl_key.src_port, acl_key.dst_port, acl_id);
// The acl_ref_cnt is needed to know how many flows refer an ACL.
// When the flow is removed, we decrement the reference count.
// When the reference count becomes 0, we remove the ACL.
if (acl_ref_cnt.count(acl_id) > 0) {
acl_ref_cnt[acl_id] ++;
} else {
// We should ideally not land here. The acl_ref_cnt should have been
// initialized the first time acl was installed.
acl_ref_cnt[acl_id] = 1;
}
} else {
resp = install_acl(acl_key_const);
if (!resp.ok()) {
OPENOLT_LOG(ERROR, openolt_log_id, "Acl for flow_id=%u with eth_type = %d, ip_proto = %d, src_port = %d, dst_port = %d failed\n",
flow_id, acl_key_const.ether_type, acl_key_const.ip_proto, acl_key_const.src_port, acl_key_const.dst_port);
bcmos_fastlock_unlock(&data_lock, 0);
return resp;
}
acl_id = acl_classifier_to_acl_id_map[acl_key_const];
// Initialize the acl reference count
acl_ref_cnt[acl_id] = 1;
OPENOLT_LOG(INFO, openolt_log_id, "acl add success for flow_id=%u with acl_id=%d\n", flow_id, acl_id);
}
// Register the interface for the given acl
acl_id_intf_id_intf_type ac_id_inf_id_inf_type(acl_id, intf_id, intf_type);
// This is needed to keep a track of which interface (pon/nni) has registered for an ACL.
// If it is registered, how many flows refer to it.
if (intf_acl_registration_ref_cnt.count(ac_id_inf_id_inf_type) > 0) {
intf_acl_registration_ref_cnt[ac_id_inf_id_inf_type]++;
} else {
// The given interface is not registered for the ACL. We need to do it now.
resp = update_acl_interface(intf_id, olt_if_type, acl_id, BCMOLT_MEMBERS_UPDATE_COMMAND_ADD);
if (!resp.ok()){
OPENOLT_LOG(ERROR, openolt_log_id, "failed to update acl interfaces intf_id=%d, intf_type=%s, acl_id=%d", intf_id, intf_type.c_str(), acl_id);
// TODO: Ideally we should return error from hear and clean up other other stateful
// counters we creaed earlier. Will leave it out for now.
}
intf_acl_registration_ref_cnt[ac_id_inf_id_inf_type] = 1;
}
// Install the gem port if needed.
if (gemport_id > 0 && access_intf_id >= 0) {
if (gem_ref_cnt.count(gem_intf) > 0) {
// The gem port is already installed
// Increment the ref counter indicating number of flows referencing this gem port
gem_ref_cnt[gem_intf]++;
OPENOLT_LOG(DEBUG, openolt_log_id, "increment gem_ref_cnt in acl handler, ref_cnt=%d\n", gem_ref_cnt[gem_intf]);
} else {
// We should ideally never land here. The gem port should have been created the
// first time ACL was installed.
// Install the gem port
Status resp = install_gem_port(access_intf_id, onu_id, gemport_id);
if (!resp.ok()) {
// TODO: We might need to reverse all previous data, but leave it out for now.
OPENOLT_LOG(ERROR, openolt_log_id, "failed to install the gemport=%d for acl_id=%d, intf_id=%d\n", gemport_id, acl_id, access_intf_id);
bcmos_fastlock_unlock(&data_lock, 0);
return resp;
}
// Initialize the refence count for the gemport.
gem_ref_cnt[gem_intf] = 1;
OPENOLT_LOG(DEBUG, openolt_log_id, "intialized gem ref count in acl handler\n");
}
} else {
OPENOLT_LOG(DEBUG, openolt_log_id, "not incrementing gem_ref_cnt in acl handler flow_id=%d, gemport_id=%d, intf_id=%d\n", flow_id, gemport_id, access_intf_id);
}
// Update the flow_to_acl_map
// This info is needed during flow remove. We need to which ACL ID and GEM PORT ID
// the flow was referring to.
// After retrieving the ACL ID and GEM PORT ID, we decrement the corresponding
// reference counters for those ACL ID and GEMPORT ID.
acl_id_gem_id_intf_id ac_id_gm_id_if_id(acl_id, gemport_id, intf_id);
flow_to_acl_map[fl_id_fl_dir] = ac_id_gm_id_if_id;
bcmos_fastlock_unlock(&data_lock, 0);
return Status::OK;
}
Status FlowAdd_(int32_t access_intf_id, int32_t onu_id, int32_t uni_id, uint32_t port_no,
uint32_t flow_id, const std::string flow_type,
int32_t alloc_id, int32_t network_intf_id,
int32_t gemport_id, const ::openolt::Classifier& classifier,
const ::openolt::Action& action, int32_t priority_value, uint64_t cookie,
int32_t group_id) {
bcmolt_flow_cfg cfg;
bcmolt_flow_key key = { }; /**< Object key. */
int32_t o_vid = -1;
bool single_tag = false;
uint32_t ether_type = 0;
bcmolt_classifier c_val = { };
bcmolt_action a_val = { };
bcmolt_tm_queue_ref tm_val = { };
int tm_qmp_id, tm_q_set_id;
bcmolt_egress_qos_type qos_type;
OPENOLT_LOG(INFO, openolt_log_id, "flow add received for flow_id=%u, flow_type=%s\n", flow_id, flow_type.c_str());
key.flow_id = flow_id;
if (flow_type.compare(upstream) == 0 ) {
key.flow_type = BCMOLT_FLOW_TYPE_UPSTREAM;
} else if (flow_type.compare(downstream) == 0) {
key.flow_type = BCMOLT_FLOW_TYPE_DOWNSTREAM;
} else if (flow_type.compare(multicast) == 0) {
key.flow_type = BCMOLT_FLOW_TYPE_MULTICAST;
} else {
OPENOLT_LOG(ERROR, openolt_log_id, "Invalid flow type %s\n", flow_type.c_str());
return bcm_to_grpc_err(BCM_ERR_PARM, "Invalid flow type");
}
BCMOLT_CFG_INIT(&cfg, flow, key);
BCMOLT_MSG_FIELD_SET(&cfg, cookie, cookie);
if (action.cmd().trap_to_host()) {
Status resp = handle_acl_rule_install(onu_id, flow_id, flow_type, access_intf_id,
network_intf_id, gemport_id, classifier);
return resp;
}
if (key.flow_type != BCMOLT_FLOW_TYPE_MULTICAST) {
if (access_intf_id >= 0 && network_intf_id >= 0) {
if (key.flow_type == BCMOLT_FLOW_TYPE_UPSTREAM) { //upstream
BCMOLT_MSG_FIELD_SET(&cfg, ingress_intf.intf_type, BCMOLT_FLOW_INTERFACE_TYPE_PON);
BCMOLT_MSG_FIELD_SET(&cfg, ingress_intf.intf_id, access_intf_id);
BCMOLT_MSG_FIELD_SET(&cfg, egress_intf.intf_type, BCMOLT_FLOW_INTERFACE_TYPE_NNI);
BCMOLT_MSG_FIELD_SET(&cfg, egress_intf.intf_id, network_intf_id);
} else if (key.flow_type == BCMOLT_FLOW_TYPE_DOWNSTREAM) { //downstream
BCMOLT_MSG_FIELD_SET(&cfg, ingress_intf.intf_type, BCMOLT_FLOW_INTERFACE_TYPE_NNI);
BCMOLT_MSG_FIELD_SET(&cfg, ingress_intf.intf_id, network_intf_id);
BCMOLT_MSG_FIELD_SET(&cfg, egress_intf.intf_type, BCMOLT_FLOW_INTERFACE_TYPE_PON);
BCMOLT_MSG_FIELD_SET(&cfg, egress_intf.intf_id, access_intf_id);
}
} else {
OPENOLT_LOG(ERROR, openolt_log_id, "flow network setting invalid\n");
return bcm_to_grpc_err(BCM_ERR_PARM, "flow network setting invalid");
}
if (onu_id >= 0) {
BCMOLT_MSG_FIELD_SET(&cfg, onu_id, onu_id);
}
if (gemport_id >= 0) {
BCMOLT_MSG_FIELD_SET(&cfg, svc_port_id, gemport_id);
}
if (gemport_id >= 0 && port_no != 0) {
bcmos_fastlock_lock(&data_lock);
if (key.flow_type == BCMOLT_FLOW_TYPE_DOWNSTREAM) {
port_to_flows[port_no].insert(key.flow_id);
flowid_to_gemport[key.flow_id] = gemport_id;
}
else
{
flowid_to_port[key.flow_id] = port_no;
}
bcmos_fastlock_unlock(&data_lock, 0);
}
if (gemport_id >= 0 && access_intf_id >= 0) {
// Update the flow_to_acl_map. Note that since this is a datapath flow, acl_id is -1
// This info is needed during flow remove where we need to retrieve the gemport_id
// and access_intf id for the given flow id and flow direction.
// After retrieving the ACL ID and GEM PORT ID, we decrement the corresponding
// reference counters for those ACL ID and GEMPORT ID.
acl_id_gem_id_intf_id ac_id_gm_id_if_id(-1, gemport_id, access_intf_id);
flow_id_flow_direction fl_id_fl_dir(flow_id, flow_type);
bcmos_fastlock_lock(&data_lock);
flow_to_acl_map[fl_id_fl_dir] = ac_id_gm_id_if_id;
bcmos_fastlock_unlock(&data_lock, 0);
}
if (priority_value >= 0) {
BCMOLT_MSG_FIELD_SET(&cfg, priority, priority_value);
}
} else { // MULTICAST FLOW
if (group_id >= 0) {
BCMOLT_MSG_FIELD_SET(&cfg, group_id, group_id);
}
BCMOLT_MSG_FIELD_SET(&cfg, ingress_intf.intf_type, BCMOLT_FLOW_INTERFACE_TYPE_NNI);
BCMOLT_MSG_FIELD_SET(&cfg, ingress_intf.intf_id, network_intf_id);
}
{
/* removed by BAL v3.0
if (classifier.o_tpid()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "classify o_tpid 0x%04x\n", classifier.o_tpid());
BCMBAL_ATTRIBUTE_PROP_SET(&val, classifier, o_tpid, classifier.o_tpid());
}
*/
/* removed by BAL v3.0
if (classifier.i_tpid()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "classify i_tpid 0x%04x\n", classifier.i_tpid());
BCMBAL_ATTRIBUTE_PROP_SET(&val, classifier, i_tpid, classifier.i_tpid());
}
*/
if (classifier.eth_type()) {
ether_type = classifier.eth_type();
OPENOLT_LOG(DEBUG, openolt_log_id, "classify ether_type 0x%04x\n", classifier.eth_type());
BCMOLT_FIELD_SET(&c_val, classifier, ether_type, classifier.eth_type());
}
if (classifier.dst_mac().size() > 0) {
bcmos_mac_address d_mac = {};
bcmos_mac_address_init(&d_mac);
memcpy(d_mac.u8, classifier.dst_mac().data(), sizeof(d_mac.u8));
OPENOLT_LOG(DEBUG, openolt_log_id, "classify dst_mac %02x:%02x:%02x:%02x:%02x:%02x\n", d_mac.u8[0],
d_mac.u8[1], d_mac.u8[2], d_mac.u8[3], d_mac.u8[4], d_mac.u8[5]);
BCMOLT_FIELD_SET(&c_val, classifier, dst_mac, d_mac);
}
/*
if (classifier.src_mac()) {
BCMBAL_ATTRIBUTE_PROP_SET(&val, classifier, src_mac, classifier.src_mac());
}
*/
if (classifier.ip_proto()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "classify ip_proto %d\n", classifier.ip_proto());
BCMOLT_FIELD_SET(&c_val, classifier, ip_proto, classifier.ip_proto());
}
/*
if (classifier.dst_ip()) {
BCMBAL_ATTRIBUTE_PROP_SET(&val, classifier, dst_ip, classifier.dst_ip());
}
if (classifier.src_ip()) {
BCMBAL_ATTRIBUTE_PROP_SET(&val, classifier, src_ip, classifier.src_ip());
}
*/
if (classifier.src_port()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "classify src_port %d\n", classifier.src_port());
BCMOLT_FIELD_SET(&c_val, classifier, src_port, classifier.src_port());
}
if (classifier.dst_port()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "classify dst_port %d\n", classifier.dst_port());
BCMOLT_FIELD_SET(&c_val, classifier, dst_port, classifier.dst_port());
}
if (!classifier.pkt_tag_type().empty()) {
if (classifier.o_vid()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "classify o_vid %d\n", classifier.o_vid());
BCMOLT_FIELD_SET(&c_val, classifier, o_vid, classifier.o_vid());
}
if (classifier.i_vid()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "classify i_vid %d\n", classifier.i_vid());
BCMOLT_FIELD_SET(&c_val, classifier, i_vid, classifier.i_vid());
}
OPENOLT_LOG(DEBUG, openolt_log_id, "classify tag_type %s\n", classifier.pkt_tag_type().c_str());
if (classifier.pkt_tag_type().compare("untagged") == 0) {
BCMOLT_FIELD_SET(&c_val, classifier, pkt_tag_type, BCMOLT_PKT_TAG_TYPE_UNTAGGED);
} else if (classifier.pkt_tag_type().compare("single_tag") == 0) {
BCMOLT_FIELD_SET(&c_val, classifier, pkt_tag_type, BCMOLT_PKT_TAG_TYPE_SINGLE_TAG);
single_tag = true;
OPENOLT_LOG(DEBUG, openolt_log_id, "classify o_pbits 0x%x\n", classifier.o_pbits());
//According to makeOpenOltClassifierField in voltha-openolt-adapter, o_pbits 0xFF means PCP value 0.
if(0xFF == classifier.o_pbits()){
BCMOLT_FIELD_SET(&c_val, classifier, o_pbits, 0);
}
else{
BCMOLT_FIELD_SET(&c_val, classifier, o_pbits, classifier.o_pbits());
}
} else if (classifier.pkt_tag_type().compare("double_tag") == 0) {
BCMOLT_FIELD_SET(&c_val, classifier, pkt_tag_type, BCMOLT_PKT_TAG_TYPE_DOUBLE_TAG);
OPENOLT_LOG(DEBUG, openolt_log_id, "classify o_pbits 0x%x\n", classifier.o_pbits());
if(0xFF == classifier.o_pbits()){
BCMOLT_FIELD_SET(&c_val, classifier, o_pbits, 0);
}
else{
BCMOLT_FIELD_SET(&c_val, classifier, o_pbits, classifier.o_pbits());
}
}
}
BCMOLT_MSG_FIELD_SET(&cfg, classifier, c_val);
}
const ::openolt::ActionCmd& cmd = action.cmd();
if (cmd.add_outer_tag()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "action add o_tag\n");
BCMOLT_FIELD_SET(&a_val, action, cmds_bitmask, BCMOLT_ACTION_CMD_ID_ADD_OUTER_TAG);
}
if (cmd.remove_outer_tag()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "action pop o_tag\n");
BCMOLT_FIELD_SET(&a_val, action, cmds_bitmask, BCMOLT_ACTION_CMD_ID_REMOVE_OUTER_TAG);
}
/* removed by BAL v3.0
if (cmd.trap_to_host()) {
OPENOLT_LOG(INFO, openolt_log_id, "action trap-to-host\n");
BCMBAL_ATTRIBUTE_PROP_SET(&val, action, cmds_bitmask, BCMBAL_ACTION_CMD_ID_TRAP_TO_HOST);
}
*/
if (action.o_vid()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "action o_vid=%d\n", action.o_vid());
o_vid = action.o_vid();
BCMOLT_FIELD_SET(&a_val, action, o_vid, action.o_vid());
}
if (action.o_pbits()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "action o_pbits=0x%x\n", action.o_pbits());
BCMOLT_FIELD_SET(&a_val, action, o_pbits, action.o_pbits());
}
/* removed by BAL v3.0
if (action.o_tpid()) {
OPENOLT_LOG(INFO, openolt_log_id, "action o_tpid=0x%04x\n", action.o_tpid());
BCMBAL_ATTRIBUTE_PROP_SET(&val, action, o_tpid, action.o_tpid());
}
*/
if (action.i_vid()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "action i_vid=%d\n", action.i_vid());
BCMOLT_FIELD_SET(&a_val, action, i_vid, action.i_vid());
}
if (action.i_pbits()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "action i_pbits=0x%x\n", action.i_pbits());
BCMOLT_FIELD_SET(&a_val, action, i_pbits, action.i_pbits());
}
/* removed by BAL v3.0
if (action.i_tpid()) {
OPENOLT_LOG(DEBUG, openolt_log_id, "action i_tpid=0x%04x\n", action.i_tpid());
BCMBAL_ATTRIBUTE_PROP_SET(&val, action, i_tpid, action.i_tpid());
}
*/
BCMOLT_MSG_FIELD_SET(&cfg, action, a_val);
if ((access_intf_id >= 0) && (onu_id >= 0)) {
qos_type = get_qos_type(access_intf_id, onu_id, uni_id);
if (key.flow_type == BCMOLT_FLOW_TYPE_DOWNSTREAM) {
tm_val.sched_id = get_tm_sched_id(access_intf_id, onu_id, uni_id, downstream);
if (qos_type == BCMOLT_EGRESS_QOS_TYPE_FIXED_QUEUE) {
// Queue 0 on DS subscriber scheduler
tm_val.queue_id = 0;
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.type, qos_type);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.tm_sched.id, tm_val.sched_id);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.u.fixed_queue.queue_id, tm_val.queue_id);
OPENOLT_LOG(DEBUG, openolt_log_id, "direction = %s, queue_id = %d, sched_id = %d, intf_type %s\n", \
downstream.c_str(), tm_val.queue_id, tm_val.sched_id, \
GET_FLOW_INTERFACE_TYPE(cfg.data.ingress_intf.intf_type));
} else if (qos_type == BCMOLT_EGRESS_QOS_TYPE_PRIORITY_TO_QUEUE) {
/* Fetch TM QMP ID mapped to DS subscriber scheduler */
tm_qmp_id = tm_q_set_id = get_tm_qmp_id(tm_val.sched_id, access_intf_id, onu_id, uni_id);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.type, qos_type);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.tm_sched.id, tm_val.sched_id);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.u.priority_to_queue.tm_qmp_id, tm_qmp_id);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.u.priority_to_queue.tm_q_set_id, tm_q_set_id);
OPENOLT_LOG(DEBUG, openolt_log_id, "direction = %s, q_set_id = %d, sched_id = %d, intf_type %s\n", \
downstream.c_str(), tm_q_set_id, tm_val.sched_id, \
GET_FLOW_INTERFACE_TYPE(cfg.data.ingress_intf.intf_type));
}
} else if (key.flow_type == BCMOLT_FLOW_TYPE_UPSTREAM) {
// NNI Scheduler ID
tm_val.sched_id = get_default_tm_sched_id(network_intf_id, upstream);
if (qos_type == BCMOLT_EGRESS_QOS_TYPE_FIXED_QUEUE) {
// Queue 0 on NNI scheduler
tm_val.queue_id = 0;
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.type, qos_type);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.tm_sched.id, tm_val.sched_id);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.u.fixed_queue.queue_id, tm_val.queue_id);
OPENOLT_LOG(DEBUG, openolt_log_id, "direction = %s, queue_id = %d, sched_id = %d, intf_type %s\n", \
upstream.c_str(), tm_val.queue_id, tm_val.sched_id, \
GET_FLOW_INTERFACE_TYPE(cfg.data.ingress_intf.intf_type));
} else if (qos_type == BCMOLT_EGRESS_QOS_TYPE_PRIORITY_TO_QUEUE) {
/* Fetch TM QMP ID mapped to US NNI scheduler */
tm_qmp_id = tm_q_set_id = get_tm_qmp_id(tm_val.sched_id, access_intf_id, onu_id, uni_id);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.type, qos_type);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.tm_sched.id, tm_val.sched_id);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.u.priority_to_queue.tm_qmp_id, tm_qmp_id);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.u.priority_to_queue.tm_q_set_id, tm_q_set_id);
OPENOLT_LOG(DEBUG, openolt_log_id, "direction = %s, q_set_id = %d, sched_id = %d, intf_type %s\n", \
upstream.c_str(), tm_q_set_id, tm_val.sched_id, \
GET_FLOW_INTERFACE_TYPE(cfg.data.ingress_intf.intf_type));
}
}
} else {
tm_val.sched_id = get_default_tm_sched_id(network_intf_id, upstream);
tm_val.queue_id = 0;
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.type, BCMOLT_EGRESS_QOS_TYPE_FIXED_QUEUE);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.tm_sched.id, tm_val.sched_id);
BCMOLT_MSG_FIELD_SET(&cfg , egress_qos.u.fixed_queue.queue_id, tm_val.queue_id);
OPENOLT_LOG(DEBUG, openolt_log_id, "direction = %s, queue_id = %d, sched_id = %d, intf_type %s\n", \
flow_type.c_str(), tm_val.queue_id, tm_val.sched_id, \
GET_FLOW_INTERFACE_TYPE(cfg.data.ingress_intf.intf_type));
}
BCMOLT_MSG_FIELD_SET(&cfg, state, BCMOLT_FLOW_STATE_ENABLE);
// BAL 3.1 supports statistics only for unicast flows.
if (key.flow_type != BCMOLT_FLOW_TYPE_MULTICAST) {
BCMOLT_MSG_FIELD_SET(&cfg, statistics, BCMOLT_CONTROL_STATE_ENABLE);
}
#ifdef FLOW_CHECKER
//Flow Checker, To avoid duplicate flow.
if (flow_id_counters != 0) {
bool b_duplicate_flow = false;
std::map<flow_pair, int>::iterator it;
for(it = flow_map.begin(); it != flow_map.end(); it++) {
b_duplicate_flow = (cfg.data.onu_id == get_flow_status(it->first.first, it->first.second, ONU_ID)) && \
(key.flow_type == it->first.second) && \
(cfg.data.svc_port_id == get_flow_status(it->first.first, it->first.second, SVC_PORT_ID)) && \
(cfg.data.priority == get_flow_status(it->first.first, it->first.second, PRIORITY)) && \
(cfg.data.cookie == get_flow_status(it->first.first, it->first.second, COOKIE)) && \
(cfg.data.ingress_intf.intf_type == get_flow_status(it->first.first, it->first.second, INGRESS_INTF_TYPE)) && \
(cfg.data.ingress_intf.intf_id == get_flow_status(it->first.first, it->first.second, INGRESS_INTF_ID)) && \
(cfg.data.egress_intf.intf_type == get_flow_status(it->first.first, it->first.second, EGRESS_INTF_TYPE)) && \
(cfg.data.egress_intf.intf_id == get_flow_status(it->first.first, it->first.second, EGRESS_INTF_ID)) && \
(c_val.o_vid == get_flow_status(it->first.first, it->first.second, CLASSIFIER_O_VID)) && \
(c_val.o_pbits == get_flow_status(it->first.first, it->first.second, CLASSIFIER_O_PBITS)) && \
(c_val.i_vid == get_flow_status(it->first.first, it->first.second, CLASSIFIER_I_VID)) && \
(c_val.i_pbits == get_flow_status(it->first.first, it->first.second, CLASSIFIER_I_PBITS)) && \
(c_val.ether_type == get_flow_status(it->first.first, it->first.second, CLASSIFIER_ETHER_TYPE)) && \
(c_val.ip_proto == get_flow_status(it->first.first, it->first.second, CLASSIFIER_IP_PROTO)) && \
(c_val.src_port == get_flow_status(it->first.first, it->first.second, CLASSIFIER_SRC_PORT)) && \
(c_val.dst_port == get_flow_status(it->first.first, it->first.second, CLASSIFIER_DST_PORT)) && \
(c_val.pkt_tag_type == get_flow_status(it->first.first, it->first.second, CLASSIFIER_PKT_TAG_TYPE)) && \
(cfg.data.egress_qos.type == get_flow_status(it->first.first, it->first.second, EGRESS_QOS_TYPE)) && \
(cfg.data.egress_qos.u.fixed_queue.queue_id == get_flow_status(it->first.first, it->first.second, EGRESS_QOS_QUEUE_ID)) && \
(cfg.data.egress_qos.tm_sched.id == get_flow_status(it->first.first, it->first.second, EGRESS_QOS_TM_SCHED_ID)) && \
(a_val.cmds_bitmask == get_flow_status(it->first.first, it->first.second, ACTION_CMDS_BITMASK)) && \
(a_val.o_vid == get_flow_status(it->first.first, it->first.second, ACTION_O_VID)) && \
(a_val.i_vid == get_flow_status(it->first.first, it->first.second, ACTION_I_VID)) && \
(a_val.o_pbits == get_flow_status(it->first.first, it->first.second, ACTION_O_PBITS)) && \
(a_val.i_pbits == get_flow_status(it->first.first, it->first.second, ACTION_I_PBITS)) && \
(cfg.data.state == get_flow_status(it->first.first, it->first.second, STATE)) && \
(cfg.data.group_id == get_flow_status(it->first.first, it->first.second, GROUP_ID));
#ifdef SHOW_FLOW_PARAM
// Flow Parameter
FLOW_PARAM_LOG();
#endif
if (b_duplicate_flow) {
FLOW_LOG(WARNING, "Flow duplicate", 0);
return bcm_to_grpc_err(BCM_ERR_ALREADY, "flow exists");
}
}
}
#endif
bcmos_errno err = bcmolt_cfg_set(dev_id, &cfg.hdr);
if (err) {
FLOW_LOG(ERROR, "Flow add failed", err);
return bcm_to_grpc_err(err, "flow add failed");
} else {
FLOW_LOG(INFO, "Flow add ok", err);
bcmos_fastlock_lock(&data_lock);
flow_map[std::pair<int, int>(key.flow_id,key.flow_type)] = flow_map.size();
flow_id_counters = flow_map.size();
if (gemport_id > 0 && access_intf_id >= 0) {
gem_id_intf_id gem_intf(gemport_id, access_intf_id);
if (gem_ref_cnt.count(gem_intf) > 0) {
// The gem port is already installed
// Increment the ref counter indicating number of flows referencing this gem port
gem_ref_cnt[gem_intf]++;
OPENOLT_LOG(DEBUG, openolt_log_id, "incremented gem_ref_cnt, gem_ref_cnt=%d\n", gem_ref_cnt[gem_intf]);
} else {
// Initialize the refence count for the gemport.
gem_ref_cnt[gem_intf] = 1;
OPENOLT_LOG(DEBUG, openolt_log_id, "initialized gem_ref_cnt\n");
}
} else {
OPENOLT_LOG(DEBUG, openolt_log_id, "not incrementing gem_ref_cnt flow_id=%d gemport_id=%d access_intf_id=%d\n", flow_id, gemport_id, access_intf_id);
}
bcmos_fastlock_unlock(&data_lock, 0);
}
return Status::OK;
}
void clear_gem_port(int gemport_id, int access_intf_id) {
gem_id_intf_id gem_intf(gemport_id, access_intf_id);
if (gemport_id > 0 && access_intf_id >= 0 && gem_ref_cnt.count(gem_intf) > 0) {
OPENOLT_LOG(DEBUG, openolt_log_id, "decrementing gem_ref_cnt gemport_id=%d access_intf_id=%d\n", gemport_id, access_intf_id);
gem_ref_cnt[gem_intf]--;
if (gem_ref_cnt[gem_intf] == 0) {
// For datapath flow this may not be necessary (to be verified)
remove_gem_port(access_intf_id, gemport_id);
gem_ref_cnt.erase(gem_intf);
OPENOLT_LOG(DEBUG, openolt_log_id, "removing gem_ref_cnt entry gemport_id=%d access_intf_id=%d\n", gemport_id, access_intf_id);
} else {
OPENOLT_LOG(DEBUG, openolt_log_id, "gem_ref_cnt not zero yet gemport_id=%d access_intf_id=%d\n", gemport_id, access_intf_id);
}
} else {
OPENOLT_LOG(DEBUG, openolt_log_id, "not decrementing gem_ref_cnt gemport_id=%d access_intf_id=%d\n", gemport_id, access_intf_id);
}
}
Status handle_acl_rule_cleanup(int16_t acl_id, int32_t gemport_id, int32_t intf_id, const std::string flow_type) {
const std::string intf_type= flow_type.compare(upstream) == 0 ? "pon": "nni";
acl_id_intf_id_intf_type ac_id_inf_id_inf_type(acl_id, intf_id, intf_type);
intf_acl_registration_ref_cnt[ac_id_inf_id_inf_type]--;
if (intf_acl_registration_ref_cnt[ac_id_inf_id_inf_type] == 0) {
bcmolt_interface_type olt_if_type = intf_type == "pon"? BCMOLT_INTERFACE_TYPE_PON: BCMOLT_INTERFACE_TYPE_NNI;
Status resp = update_acl_interface(intf_id, olt_if_type, acl_id, BCMOLT_MEMBERS_UPDATE_COMMAND_REMOVE);
if (!resp.ok()){
OPENOLT_LOG(ERROR, openolt_log_id, "failed to update acl interfaces intf_id=%d, intf_type=%s, acl_id=%d", intf_id, intf_type.c_str(), acl_id);
}
intf_acl_registration_ref_cnt.erase(ac_id_inf_id_inf_type);
}
acl_ref_cnt[acl_id]--;
if (acl_ref_cnt[acl_id] == 0) {
remove_acl(acl_id);
acl_ref_cnt.erase(acl_id);
// Iterate acl_classifier_to_acl_id_map and delete classifier the key corresponding to acl_id
std::map<acl_classifier_key, uint16_t>::iterator it;
for (it=acl_classifier_to_acl_id_map.begin(); it!=acl_classifier_to_acl_id_map.end(); ++it) {
if (it->second == acl_id) {
OPENOLT_LOG(INFO, openolt_log_id, "cleared classifier key corresponding to acl_id = %d\n", acl_id);
acl_classifier_to_acl_id_map.erase(it->first);
break;
}
}
}
clear_gem_port(gemport_id, intf_id);
return Status::OK;
}
Status FlowRemove_(uint32_t flow_id, const std::string flow_type) {
bcmolt_flow_cfg cfg;
bcmolt_flow_key key = { };
key.flow_id = (bcmolt_flow_id) flow_id;
key.flow_id = flow_id;
if (flow_type.compare(upstream) == 0 ) {
key.flow_type = BCMOLT_FLOW_TYPE_UPSTREAM;
} else if (flow_type.compare(downstream) == 0) {
key.flow_type = BCMOLT_FLOW_TYPE_DOWNSTREAM;
} else if(flow_type.compare(multicast) == 0) {
key.flow_type = BCMOLT_FLOW_TYPE_MULTICAST;
} else {
OPENOLT_LOG(WARNING, openolt_log_id, "Invalid flow type %s\n", flow_type.c_str());
return bcm_to_grpc_err(BCM_ERR_PARM, "Invalid flow type");
}
OPENOLT_LOG(INFO, openolt_log_id, "flow remove received for flow_id=%u, flow_type=%s\n",
flow_id, flow_type.c_str());
bcmos_fastlock_lock(&data_lock);
flow_id_flow_direction fl_id_fl_dir(flow_id, flow_type);
int32_t gemport_id = -1;
int32_t intf_id = -1;
int16_t acl_id = -1;
if (flow_to_acl_map.count(fl_id_fl_dir) > 0) {
acl_id_gem_id_intf_id ac_id_gm_id_if_id = flow_to_acl_map[fl_id_fl_dir];
acl_id = std::get<0>(ac_id_gm_id_if_id);
gemport_id = std::get<1>(ac_id_gm_id_if_id);
intf_id = std::get<2>(ac_id_gm_id_if_id);
// cleanup acl only if it is a valid acl. If not valid acl, it may be datapath flow.
if (acl_id >= 0) {
Status resp = handle_acl_rule_cleanup(acl_id, gemport_id, intf_id, flow_type);
bcmos_fastlock_unlock(&data_lock, 0);
if (resp.ok()) {
OPENOLT_LOG(INFO, openolt_log_id, "acl removed ok for flow_id = %u with acl_id = %d\n", flow_id, acl_id);
flow_to_acl_map.erase(fl_id_fl_dir);
} else {
OPENOLT_LOG(ERROR, openolt_log_id, "acl remove error for flow_id = %u with acl_id = %d\n", flow_id, acl_id);
}
return resp;
}
}
uint32_t port_no = flowid_to_port[key.flow_id];
if (key.flow_type == BCMOLT_FLOW_TYPE_DOWNSTREAM) {
flowid_to_gemport.erase(key.flow_id);
port_to_flows[port_no].erase(key.flow_id);
if (port_to_flows[port_no].empty()) port_to_flows.erase(port_no);
}
else
{
flowid_to_port.erase(key.flow_id);
}
bcmos_fastlock_unlock(&data_lock, 0);
BCMOLT_CFG_INIT(&cfg, flow, key);
bcmos_errno err = bcmolt_cfg_clear(dev_id, &cfg.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Error while removing %s flow, flow_id=%d, err = %s\n", flow_type.c_str(), flow_id, bcmos_strerror(err));
return Status(grpc::StatusCode::INTERNAL, "Failed to remove flow");
}
bcmos_fastlock_lock(&data_lock);
if (flow_id_counters != 0) {
std::map<flow_pair, int>::iterator it;
for(it = flow_map.begin(); it != flow_map.end(); it++) {
if (it->first.first == flow_id && it->first.second == key.flow_type) {
flow_id_counters -= 1;
flow_map.erase(it);
}
}
}
OPENOLT_LOG(INFO, openolt_log_id, "Flow %d, %s removed\n", flow_id, flow_type.c_str());
clear_gem_port(gemport_id, intf_id);
flow_to_acl_map.erase(fl_id_fl_dir);
bcmos_fastlock_unlock(&data_lock, 0);
return Status::OK;
}
bcmos_errno CreateDefaultSched(uint32_t intf_id, const std::string direction) {
bcmos_errno err;
bcmolt_tm_sched_cfg tm_sched_cfg;
bcmolt_tm_sched_key tm_sched_key = {.id = 1};
tm_sched_key.id = get_default_tm_sched_id(intf_id, direction);
//check TM scheduler has configured or not
BCMOLT_CFG_INIT(&tm_sched_cfg, tm_sched, tm_sched_key);
BCMOLT_MSG_FIELD_GET(&tm_sched_cfg, state);
#ifdef TEST_MODE
// It is impossible to mock the setting of tm_sched_cfg.data.state because
// the actual bcmolt_cfg_get passes the address of tm_sched_cfg.hdr and we cannot
// set the tm_sched_cfg.data.state. So a new stub function is created and address
// of tm_sched_cfg is passed. This is one-of case where we need to add test specific
// code in production code.
err = bcmolt_cfg_get__tm_sched_stub(dev_id, &tm_sched_cfg);
#else
err = bcmolt_cfg_get(dev_id, &tm_sched_cfg.hdr);
#endif
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "cfg: Failed to query TM scheduler, err = %s\n",bcmos_strerror(err));
return err;
}
else if (tm_sched_cfg.data.state == BCMOLT_CONFIG_STATE_CONFIGURED) {
OPENOLT_LOG(WARNING, openolt_log_id, "tm scheduler default config has already with id %d\n", tm_sched_key.id);
return BCM_ERR_OK;
}
// bcmbal_tm_sched_owner
BCMOLT_CFG_INIT(&tm_sched_cfg, tm_sched, tm_sched_key);
/**< The output of the tm_sched object instance */
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, attachment_point.type, BCMOLT_TM_SCHED_OUTPUT_TYPE_INTERFACE);
if (direction.compare(upstream) == 0) {
// In upstream it is NNI scheduler
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, attachment_point.u.interface.interface_ref.intf_type, BCMOLT_INTERFACE_TYPE_NNI);
} else if (direction.compare(downstream) == 0) {
// In downstream it is PON scheduler
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, attachment_point.u.interface.interface_ref.intf_type, BCMOLT_INTERFACE_TYPE_PON);
}
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, attachment_point.u.interface.interface_ref.intf_id, intf_id);
// bcmbal_tm_sched_type
// set the deafult policy to strict priority
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, sched_type, BCMOLT_TM_SCHED_TYPE_SP);
// num_priorities: Max number of strict priority scheduling elements
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, num_priorities, NUM_OF_PRIORITIES);
err = bcmolt_cfg_set(dev_id, &tm_sched_cfg.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to create %s scheduler, id %d, intf_id %d, err = %s\n",
direction.c_str(), tm_sched_key.id, intf_id, bcmos_strerror(err));
return err;
}
OPENOLT_LOG(INFO, openolt_log_id, "Create %s scheduler success, id %d, intf_id %d\n", \
direction.c_str(), tm_sched_key.id, intf_id);
return BCM_ERR_OK;
}
bcmos_errno CreateSched(std::string direction, uint32_t intf_id, uint32_t onu_id, uint32_t uni_id, uint32_t port_no,
uint32_t alloc_id, tech_profile::AdditionalBW additional_bw, uint32_t weight, uint32_t priority,
tech_profile::SchedulingPolicy sched_policy, tech_profile::TrafficShapingInfo tf_sh_info) {
bcmos_errno err;
if (direction == downstream) {
bcmolt_tm_sched_cfg tm_sched_cfg;
bcmolt_tm_sched_key tm_sched_key = {.id = 1};
tm_sched_key.id = get_tm_sched_id(intf_id, onu_id, uni_id, direction);
// bcmbal_tm_sched_owner
// In downstream it is sub_term scheduler
BCMOLT_CFG_INIT(&tm_sched_cfg, tm_sched, tm_sched_key);
/**< The output of the tm_sched object instance */
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, attachment_point.type, BCMOLT_TM_SCHED_OUTPUT_TYPE_TM_SCHED);
// bcmbal_tm_sched_parent
// The parent for the sub_term scheduler is the PON scheduler in the downstream
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, attachment_point.u.tm_sched.tm_sched_id, get_default_tm_sched_id(intf_id, direction));
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, attachment_point.u.tm_sched.tm_sched_param.u.priority.priority, priority);
/* removed by BAL v3.0, N/A - No direct attachment point of type ONU, same functionality may
be achieved using the' virtual' type of attachment.
tm_sched_owner.u.sub_term.intf_id = intf_id;
tm_sched_owner.u.sub_term.sub_term_id = onu_id;
*/
// bcmbal_tm_sched_type
// set the deafult policy to strict priority
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, sched_type, BCMOLT_TM_SCHED_TYPE_SP);
// num_priorities: Max number of strict priority scheduling elements
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, num_priorities, 8);
// bcmbal_tm_shaping
if (tf_sh_info.cir() >= 0 && tf_sh_info.pir() > 0) {
uint32_t cir = tf_sh_info.cir();
uint32_t pir = tf_sh_info.pir();
uint32_t burst = tf_sh_info.pbs();
OPENOLT_LOG(INFO, openolt_log_id, "applying traffic shaping in DL cir=%u, pir=%u, burst=%u\n",
cir, pir, burst);
BCMOLT_FIELD_SET_PRESENT(&tm_sched_cfg.data.rate, tm_shaping, pir);
BCMOLT_FIELD_SET_PRESENT(&tm_sched_cfg.data.rate, tm_shaping, burst);
// FIXME: Setting CIR, results in BAL throwing error 'tm_sched minimum rate is not supported yet'
//BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, rate.cir, cir);
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, rate.pir, pir);
BCMOLT_MSG_FIELD_SET(&tm_sched_cfg, rate.burst, burst);
}
err = bcmolt_cfg_set(dev_id, &tm_sched_cfg.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to create downstream subscriber scheduler, id %d, \
intf_id %d, onu_id %d, uni_id %d, port_no %u, err = %s\n", tm_sched_key.id, intf_id, onu_id, uni_id, \
port_no, bcmos_strerror(err));
return err;
}
OPENOLT_LOG(INFO, openolt_log_id, "Create downstream subscriber sched, id %d, intf_id %d, onu_id %d, \
uni_id %d, port_no %u\n", tm_sched_key.id, intf_id, onu_id, uni_id, port_no);
} else { //upstream
bcmolt_itupon_alloc_cfg cfg;
bcmolt_itupon_alloc_key key = { };
key.pon_ni = intf_id;
key.alloc_id = alloc_id;
int bw_granularity = (board_technology == "XGS-PON")?XGS_BANDWIDTH_GRANULARITY:GPON_BANDWIDTH_GRANULARITY;
int pir_bw = tf_sh_info.pir()*125; // conversion from kbps to bytes/sec
int cir_bw = tf_sh_info.cir()*125; // conversion from kbps to bytes/sec
//offset to match bandwidth granularity
int offset_pir_bw = pir_bw%bw_granularity;
int offset_cir_bw = cir_bw%bw_granularity;
pir_bw = pir_bw - offset_pir_bw;
cir_bw = cir_bw - offset_cir_bw;
BCMOLT_CFG_INIT(&cfg, itupon_alloc, key);
switch (additional_bw) {
case 2: //AdditionalBW_BestEffort
if (pir_bw == 0) {
OPENOLT_LOG(ERROR, openolt_log_id, "Maximum bandwidth was set to 0, must be at least \
%d bytes/sec\n", (board_technology == "XGS-PON")?XGS_BANDWIDTH_GRANULARITY:GPON_BANDWIDTH_GRANULARITY);
return BCM_ERR_PARM;
} else if (pir_bw < cir_bw) {
OPENOLT_LOG(ERROR, openolt_log_id, "Maximum bandwidth (%d) can't be less than Guaranteed \
bandwidth (%d)\n", pir_bw, cir_bw);
return BCM_ERR_PARM;
} else if (pir_bw == cir_bw) {
OPENOLT_LOG(ERROR, openolt_log_id, "Maximum bandwidth must be greater than Guaranteed \
bandwidth for additional bandwidth eligibility of type best_effort\n");
return BCM_ERR_PARM;
}
BCMOLT_MSG_FIELD_SET(&cfg, sla.additional_bw_eligibility, BCMOLT_ADDITIONAL_BW_ELIGIBILITY_BEST_EFFORT);
break;
case 1: //AdditionalBW_NA
if (pir_bw == 0) {
OPENOLT_LOG(ERROR, openolt_log_id, "Maximum bandwidth was set to 0, must be at least \
%d bytes/sec\n", (board_technology == "XGS-PON")?XGS_BANDWIDTH_GRANULARITY:GPON_BANDWIDTH_GRANULARITY);
return BCM_ERR_PARM;
} else if (cir_bw == 0) {
OPENOLT_LOG(ERROR, openolt_log_id, "Guaranteed bandwidth must be greater than zero for \
additional bandwidth eligibility of type Non-Assured (NA)\n");
return BCM_ERR_PARM;
} else if (pir_bw < cir_bw) {
OPENOLT_LOG(ERROR, openolt_log_id, "Maximum bandwidth (%d) can't be less than Guaranteed \
bandwidth (%d)\n", pir_bw, cir_bw);
return BCM_ERR_PARM;
} else if (pir_bw == cir_bw) {
OPENOLT_LOG(ERROR, openolt_log_id, "Maximum bandwidth must be greater than Guaranteed \
bandwidth for additional bandwidth eligibility of type non_assured\n");
return BCM_ERR_PARM;
}
BCMOLT_MSG_FIELD_SET(&cfg, sla.additional_bw_eligibility, BCMOLT_ADDITIONAL_BW_ELIGIBILITY_NON_ASSURED);
break;
case 0: //AdditionalBW_None
if (pir_bw == 0) {
OPENOLT_LOG(ERROR, openolt_log_id, "Maximum bandwidth was set to 0, must be at least \
16000 bytes/sec\n");
return BCM_ERR_PARM;
} else if (cir_bw == 0) {
OPENOLT_LOG(ERROR, openolt_log_id, "Maximum bandwidth must be equal to Guaranteed bandwidth \
for additional bandwidth eligibility of type None\n");
return BCM_ERR_PARM;
} else if (pir_bw > cir_bw) {
OPENOLT_LOG(ERROR, openolt_log_id, "Maximum bandwidth must be equal to Guaranteed bandwidth \
for additional bandwidth eligibility of type None\n");
OPENOLT_LOG(ERROR, openolt_log_id, "Setting Maximum bandwidth (%d) to Guaranteed \
bandwidth in None eligibility\n", pir_bw);
cir_bw = pir_bw;
} else if (pir_bw < cir_bw) {
OPENOLT_LOG(ERROR, openolt_log_id, "Maximum bandwidth (%d) can't be less than Guaranteed \
bandwidth (%d)\n", pir_bw, cir_bw);
OPENOLT_LOG(ERROR, openolt_log_id, "Setting Maximum bandwidth (%d) to Guaranteed \
bandwidth in None eligibility\n", pir_bw);
cir_bw = pir_bw;
}
BCMOLT_MSG_FIELD_SET(&cfg, sla.additional_bw_eligibility, BCMOLT_ADDITIONAL_BW_ELIGIBILITY_NONE);
break;
default:
return BCM_ERR_PARM;
}
/* CBR Real Time Bandwidth which require shaping of the bandwidth allocations
in a fine granularity. */
BCMOLT_MSG_FIELD_SET(&cfg, sla.cbr_rt_bw, 0);
/* Fixed Bandwidth with no critical requirement of shaping */
BCMOLT_MSG_FIELD_SET(&cfg, sla.cbr_nrt_bw, 0);
/* Dynamic bandwidth which the OLT is committed to allocate upon demand */
BCMOLT_MSG_FIELD_SET(&cfg, sla.guaranteed_bw, cir_bw);
/* Maximum allocated bandwidth allowed for this alloc ID */
BCMOLT_MSG_FIELD_SET(&cfg, sla.maximum_bw, pir_bw);
BCMOLT_MSG_FIELD_SET(&cfg, sla.alloc_type, BCMOLT_ALLOC_TYPE_NSR);
/* Set to True for AllocID with CBR RT Bandwidth that requires compensation
for skipped allocations during quiet window */
BCMOLT_MSG_FIELD_SET(&cfg, sla.cbr_rt_compensation, BCMOS_FALSE);
/**< Allocation Profile index for CBR non-RT Bandwidth */
BCMOLT_MSG_FIELD_SET(&cfg, sla.cbr_nrt_ap_index, 0);
/**< Allocation Profile index for CBR RT Bandwidth */
BCMOLT_MSG_FIELD_SET(&cfg, sla.cbr_rt_ap_index, 0);
/**< Alloc ID Weight used in case of Extended DBA mode */
BCMOLT_MSG_FIELD_SET(&cfg, sla.weight, 0);
/**< Alloc ID Priority used in case of Extended DBA mode */
BCMOLT_MSG_FIELD_SET(&cfg, sla.priority, 0);
BCMOLT_MSG_FIELD_SET(&cfg, onu_id, onu_id);
err = bcmolt_cfg_set(dev_id, &cfg.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to create upstream bandwidth allocation, intf_id %d, onu_id %d, uni_id %d,\
port_no %u, alloc_id %d, err = %s\n", intf_id, onu_id,uni_id,port_no,alloc_id, bcmos_strerror(err));
return err;
}
err = wait_for_alloc_action(intf_id, alloc_id, ALLOC_OBJECT_CREATE);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to create upstream bandwidth allocation, intf_id %d, onu_id %d, uni_id %d,\
port_no %u, alloc_id %d, err = %s\n", intf_id, onu_id,uni_id,port_no,alloc_id, bcmos_strerror(err));
return err;
}
OPENOLT_LOG(INFO, openolt_log_id, "create upstream bandwidth allocation success, intf_id %d, onu_id %d, uni_id %d,\
port_no %u, alloc_id %d\n", intf_id, onu_id,uni_id,port_no,alloc_id);
}
return BCM_ERR_OK;
}
Status CreateTrafficSchedulers_(const tech_profile::TrafficSchedulers *traffic_scheds) {
uint32_t intf_id = traffic_scheds->intf_id();
uint32_t onu_id = traffic_scheds->onu_id();
uint32_t uni_id = traffic_scheds->uni_id();
uint32_t port_no = traffic_scheds->port_no();
std::string direction;
unsigned int alloc_id;
tech_profile::SchedulerConfig sched_config;
tech_profile::AdditionalBW additional_bw;
uint32_t priority;
uint32_t weight;
tech_profile::SchedulingPolicy sched_policy;
tech_profile::TrafficShapingInfo traffic_shaping_info;
bcmos_errno err;
for (int i = 0; i < traffic_scheds->traffic_scheds_size(); i++) {
tech_profile::TrafficScheduler traffic_sched = traffic_scheds->traffic_scheds(i);
direction = GetDirection(traffic_sched.direction());
if (direction.compare("direction-not-supported") == 0)
return bcm_to_grpc_err(BCM_ERR_PARM, "direction-not-supported");
alloc_id = traffic_sched.alloc_id();
sched_config = traffic_sched.scheduler();
additional_bw = sched_config.additional_bw();
priority = sched_config.priority();
weight = sched_config.weight();
sched_policy = sched_config.sched_policy();
traffic_shaping_info = traffic_sched.traffic_shaping_info();
err = CreateSched(direction, intf_id, onu_id, uni_id, port_no, alloc_id, additional_bw, weight, priority,
sched_policy, traffic_shaping_info);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to create scheduler, err = %s\n", bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to create scheduler");
}
}
return Status::OK;
}
bcmos_errno RemoveSched(int intf_id, int onu_id, int uni_id, int alloc_id, std::string direction) {
bcmos_errno err;
uint16_t sched_id;
if (direction == upstream) {
bcmolt_itupon_alloc_cfg cfg;
bcmolt_itupon_alloc_key key = { };
key.pon_ni = intf_id;
key.alloc_id = alloc_id;
sched_id = alloc_id;
BCMOLT_CFG_INIT(&cfg, itupon_alloc, key);
err = bcmolt_cfg_clear(dev_id, &cfg.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to remove scheduler, direction = %s, intf_id %d, alloc_id %d, err = %s\n",
direction.c_str(), intf_id, alloc_id, bcmos_strerror(err));
return err;
}
err = wait_for_alloc_action(intf_id, alloc_id, ALLOC_OBJECT_DELETE);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to remove scheduler, direction = %s, intf_id %d, alloc_id %d, err = %s\n",
direction.c_str(), intf_id, alloc_id, bcmos_strerror(err));
return err;
}
} else if (direction == downstream) {
bcmolt_tm_sched_cfg cfg;
bcmolt_tm_sched_key key = { };
if (is_tm_sched_id_present(intf_id, onu_id, uni_id, direction)) {
key.id = get_tm_sched_id(intf_id, onu_id, uni_id, direction);
sched_id = key.id;
} else {
OPENOLT_LOG(INFO, openolt_log_id, "schduler not present in %s, err %d\n", direction.c_str(), err);
return BCM_ERR_OK;
}
BCMOLT_CFG_INIT(&cfg, tm_sched, key);
err = bcmolt_cfg_clear(dev_id, &(cfg.hdr));
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to remove scheduler, direction = %s, sched_id %d, \
intf_id %d, onu_id %d, err = %s\n", direction.c_str(), key.id, intf_id, onu_id, bcmos_strerror(err));
return err;
}
}
OPENOLT_LOG(INFO, openolt_log_id, "Removed sched, direction = %s, id %d, intf_id %d, onu_id %d\n",
direction.c_str(), sched_id, intf_id, onu_id);
free_tm_sched_id(intf_id, onu_id, uni_id, direction);
return BCM_ERR_OK;
}
Status RemoveTrafficSchedulers_(const tech_profile::TrafficSchedulers *traffic_scheds) {
uint32_t intf_id = traffic_scheds->intf_id();
uint32_t onu_id = traffic_scheds->onu_id();
uint32_t uni_id = traffic_scheds->uni_id();
std::string direction;
bcmos_errno err;
for (int i = 0; i < traffic_scheds->traffic_scheds_size(); i++) {
tech_profile::TrafficScheduler traffic_sched = traffic_scheds->traffic_scheds(i);
direction = GetDirection(traffic_sched.direction());
if (direction.compare("direction-not-supported") == 0)
return bcm_to_grpc_err(BCM_ERR_PARM, "direction-not-supported");
int alloc_id = traffic_sched.alloc_id();
err = RemoveSched(intf_id, onu_id, uni_id, alloc_id, direction);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Error-removing-traffic-scheduler, err = %s\n",bcmos_strerror(err));
return bcm_to_grpc_err(err, "error-removing-traffic-scheduler");
}
}
return Status::OK;
}
bcmos_errno CreateTrafficQueueMappingProfile(uint32_t sched_id, uint32_t intf_id, uint32_t onu_id, uint32_t uni_id, \
std::string direction, std::vector<uint32_t> tmq_map_profile) {
bcmos_errno err;
bcmolt_tm_qmp_cfg tm_qmp_cfg;
bcmolt_tm_qmp_key tm_qmp_key;
bcmolt_arr_u8_8 pbits_to_tmq_id = {0};
int tm_qmp_id = get_tm_qmp_id(sched_id, intf_id, onu_id, uni_id, tmq_map_profile);
if (tm_qmp_id == -1) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to create tm queue mapping profile. Max allowed tm queue mapping profile count is 16.\n");
return BCM_ERR_RANGE;
}
tm_qmp_key.id = tm_qmp_id;
for (uint32_t priority=0; priority<tmq_map_profile.size(); priority++) {
pbits_to_tmq_id.arr[priority] = tmq_map_profile[priority];
}
BCMOLT_CFG_INIT(&tm_qmp_cfg, tm_qmp, tm_qmp_key);
BCMOLT_MSG_FIELD_SET(&tm_qmp_cfg, type, BCMOLT_TM_QMP_TYPE_PBITS);
BCMOLT_MSG_FIELD_SET(&tm_qmp_cfg, pbits_to_tmq_id, pbits_to_tmq_id);
//BCMOLT_MSG_FIELD_SET(&tm_qmp_cfg, ref_count, 0);
//BCMOLT_MSG_FIELD_SET(&tm_qmp_cfg, state, BCMOLT_CONFIG_STATE_CONFIGURED);
err = bcmolt_cfg_set(dev_id, &tm_qmp_cfg.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to create tm queue mapping profile, tm_qmp_id %d, err = %s\n",
tm_qmp_key.id, bcmos_strerror(err));
return err;
}
OPENOLT_LOG(INFO, openolt_log_id, "Create tm queue mapping profile success, id %d\n", \
tm_qmp_key.id);
return BCM_ERR_OK;
}
bcmos_errno RemoveTrafficQueueMappingProfile(uint32_t tm_qmp_id) {
bcmos_errno err;
bcmolt_tm_qmp_cfg tm_qmp_cfg;
bcmolt_tm_qmp_key tm_qmp_key;
tm_qmp_key.id = tm_qmp_id;
BCMOLT_CFG_INIT(&tm_qmp_cfg, tm_qmp, tm_qmp_key);
err = bcmolt_cfg_clear(dev_id, &tm_qmp_cfg.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to remove tm queue mapping profile, tm_qmp_id %d, err = %s\n",
tm_qmp_key.id, bcmos_strerror(err));
return err;
}
OPENOLT_LOG(INFO, openolt_log_id, "Remove tm queue mapping profile success, id %d\n", \
tm_qmp_key.id);
return BCM_ERR_OK;
}
bcmos_errno CreateDefaultQueue(uint32_t intf_id, const std::string direction) {
bcmos_errno err;
/* Create default queues on the given PON/NNI scheduler */
for (int queue_id = 0; queue_id < NUMBER_OF_DEFAULT_INTERFACE_QUEUES; queue_id++) {
bcmolt_tm_queue_cfg tm_queue_cfg;
bcmolt_tm_queue_key tm_queue_key = {};
tm_queue_key.sched_id = get_default_tm_sched_id(intf_id, direction);
tm_queue_key.id = queue_id;
/* DefaultQueues on PON/NNI schedulers are created with egress_qos_type as
BCMOLT_EGRESS_QOS_TYPE_FIXED_QUEUE - with tm_q_set_id 32768 */
tm_queue_key.tm_q_set_id = BCMOLT_TM_QUEUE_SET_ID_QSET_NOT_USE;
BCMOLT_CFG_INIT(&tm_queue_cfg, tm_queue, tm_queue_key);
BCMOLT_MSG_FIELD_SET(&tm_queue_cfg, tm_sched_param.type, BCMOLT_TM_SCHED_PARAM_TYPE_PRIORITY);
BCMOLT_MSG_FIELD_SET(&tm_queue_cfg, tm_sched_param.u.priority.priority, queue_id);
err = bcmolt_cfg_set(dev_id, &tm_queue_cfg.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to create %s tm queue, id %d, sched_id %d, tm_q_set_id %d, err = %s\n", \
direction.c_str(), tm_queue_key.id, tm_queue_key.sched_id, tm_queue_key.tm_q_set_id, bcmos_strerror(err));
return err;
}
OPENOLT_LOG(INFO, openolt_log_id, "Create %s tm_queue success, id %d, sched_id %d, tm_q_set_id %d\n", \
direction.c_str(), tm_queue_key.id, tm_queue_key.sched_id, tm_queue_key.tm_q_set_id);
}
return BCM_ERR_OK;
}
bcmos_errno CreateQueue(std::string direction, uint32_t access_intf_id, uint32_t onu_id, uint32_t uni_id,
bcmolt_egress_qos_type qos_type, uint32_t priority, uint32_t gemport_id) {
bcmos_errno err;
bcmolt_tm_queue_cfg cfg;
bcmolt_tm_queue_key key = { };
OPENOLT_LOG(INFO, openolt_log_id, "creating %s queue. access_intf_id = %d, onu_id = %d, uni_id = %d \
gemport_id = %d\n", direction.c_str(), access_intf_id, onu_id, uni_id, gemport_id);
key.sched_id = (direction.compare(upstream) == 0) ? get_default_tm_sched_id(nni_intf_id, direction) : \
get_tm_sched_id(access_intf_id, onu_id, uni_id, direction);
if (priority > 7) {
return BCM_ERR_RANGE;
}
/* FIXME: The upstream queues have to be created once only.
The upstream queues on the NNI scheduler are shared by all subscribers.
When the first scheduler comes in, the queues get created, and are re-used by all others.
Also, these queues should be present until the last subscriber exits the system.
One solution is to have these queues always, i.e., create it as soon as OLT is enabled.
There is one queue per gem port and Queue ID is fetched based on priority_q configuration
for each GEM in TECH PROFILE */
key.id = queue_id_list[priority];
if (qos_type == BCMOLT_EGRESS_QOS_TYPE_FIXED_QUEUE) {
// Reset the Queue ID to 0, if it is fixed queue, i.e., there is only one queue for subscriber.
key.id = 0;
key.tm_q_set_id = BCMOLT_TM_QUEUE_SET_ID_QSET_NOT_USE;
}
else if (qos_type == BCMOLT_EGRESS_QOS_TYPE_PRIORITY_TO_QUEUE) {
key.tm_q_set_id = get_tm_qmp_id(key.sched_id, access_intf_id, onu_id, uni_id);
}
else {
key.tm_q_set_id = BCMOLT_TM_QUEUE_KEY_TM_Q_SET_ID_DEFAULT;
}
OPENOLT_LOG(INFO, openolt_log_id, "queue assigned queue_id = %d\n", key.id);
BCMOLT_CFG_INIT(&cfg, tm_queue, key);
BCMOLT_MSG_FIELD_SET(&cfg, tm_sched_param.u.priority.priority, priority);
err = bcmolt_cfg_set(dev_id, &cfg.hdr);
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to create subscriber tm queue, direction = %s, queue_id %d, \
sched_id %d, tm_q_set_id %d, intf_id %d, onu_id %d, uni_id %d, err = %s\n", \
direction.c_str(), key.id, key.sched_id, key.tm_q_set_id, access_intf_id, onu_id, uni_id, bcmos_strerror(err));
return err;
}
OPENOLT_LOG(INFO, openolt_log_id, "Created tm_queue, direction %s, id %d, sched_id %d, tm_q_set_id %d, \
intf_id %d, onu_id %d, uni_id %d\n", direction.c_str(), key.id, key.sched_id, key.tm_q_set_id, access_intf_id, onu_id, uni_id);
return BCM_ERR_OK;
}
Status CreateTrafficQueues_(const tech_profile::TrafficQueues *traffic_queues) {
uint32_t intf_id = traffic_queues->intf_id();
uint32_t onu_id = traffic_queues->onu_id();
uint32_t uni_id = traffic_queues->uni_id();
uint32_t sched_id;
std::string direction;
bcmos_errno err;
bcmolt_egress_qos_type qos_type = get_qos_type(intf_id, onu_id, uni_id, traffic_queues->traffic_queues_size());
if (qos_type == BCMOLT_EGRESS_QOS_TYPE_PRIORITY_TO_QUEUE) {
uint32_t queues_priority_q[traffic_queues->traffic_queues_size()] = {0};
std::string queues_pbit_map[traffic_queues->traffic_queues_size()];
for (int i = 0; i < traffic_queues->traffic_queues_size(); i++) {
tech_profile::TrafficQueue traffic_queue = traffic_queues->traffic_queues(i);
direction = GetDirection(traffic_queue.direction());
if (direction.compare("direction-not-supported") == 0)
return bcm_to_grpc_err(BCM_ERR_PARM, "direction-not-supported");
queues_priority_q[i] = traffic_queue.priority();
queues_pbit_map[i] = traffic_queue.pbit_map();
}
std::vector<uint32_t> tmq_map_profile(8, 0);
tmq_map_profile = get_tmq_map_profile(get_valid_queues_pbit_map(queues_pbit_map, COUNT_OF(queues_pbit_map)), \
queues_priority_q, COUNT_OF(queues_priority_q));
sched_id = (direction.compare(upstream) == 0) ? get_default_tm_sched_id(nni_intf_id, direction) : \
get_tm_sched_id(intf_id, onu_id, uni_id, direction);
int tm_qmp_id = get_tm_qmp_id(tmq_map_profile);
if (tm_qmp_id == -1) {
err = CreateTrafficQueueMappingProfile(sched_id, intf_id, onu_id, uni_id, direction, tmq_map_profile);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to create tm queue mapping profile, err = %s\n", bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to create tm queue mapping profile");
}
} else if (tm_qmp_id != -1 && get_tm_qmp_id(sched_id, intf_id, onu_id, uni_id) == -1) {
OPENOLT_LOG(INFO, openolt_log_id, "tm queue mapping profile present already with id %d\n", tm_qmp_id);
update_sched_qmp_id_map(sched_id, intf_id, onu_id, uni_id, tm_qmp_id);
}
}
for (int i = 0; i < traffic_queues->traffic_queues_size(); i++) {
tech_profile::TrafficQueue traffic_queue = traffic_queues->traffic_queues(i);
direction = GetDirection(traffic_queue.direction());
if (direction.compare("direction-not-supported") == 0)
return bcm_to_grpc_err(BCM_ERR_PARM, "direction-not-supported");
err = CreateQueue(direction, intf_id, onu_id, uni_id, qos_type, traffic_queue.priority(), traffic_queue.gemport_id());
// If the queue exists already, lets not return failure and break the loop.
if (err && err != BCM_ERR_ALREADY) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to create queue, err = %s\n",bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to create queue");
}
}
return Status::OK;
}
bcmos_errno RemoveQueue(std::string direction, uint32_t access_intf_id, uint32_t onu_id, uint32_t uni_id,
bcmolt_egress_qos_type qos_type, uint32_t priority, uint32_t gemport_id) {
bcmolt_tm_queue_cfg cfg;
bcmolt_tm_queue_key key = { };
bcmos_errno err;
if (direction == downstream) {
if (is_tm_sched_id_present(access_intf_id, onu_id, uni_id, direction)) {
key.sched_id = get_tm_sched_id(access_intf_id, onu_id, uni_id, direction);
key.id = queue_id_list[priority];
} else {
OPENOLT_LOG(INFO, openolt_log_id, "queue not present in DS. Not clearing, access_intf_id %d, onu_id %d, uni_id %d, gemport_id %d, direction %s\n", access_intf_id, onu_id, uni_id, gemport_id, direction.c_str());
return BCM_ERR_OK;
}
} else {
/* In the upstream we use pre-created queues on the NNI scheduler that are used by all subscribers.
They should not be removed. So, lets return OK. */
return BCM_ERR_OK;
}
if (qos_type == BCMOLT_EGRESS_QOS_TYPE_FIXED_QUEUE) {
key.tm_q_set_id = BCMOLT_TM_QUEUE_SET_ID_QSET_NOT_USE;
// Reset the queue id to 0 when using fixed queue.
key.id = 0;
}
else if (qos_type == BCMOLT_EGRESS_QOS_TYPE_PRIORITY_TO_QUEUE) {
key.tm_q_set_id = get_tm_qmp_id(key.sched_id, access_intf_id, onu_id, uni_id);
}
else {
key.tm_q_set_id = BCMOLT_TM_QUEUE_KEY_TM_Q_SET_ID_DEFAULT;
}
BCMOLT_CFG_INIT(&cfg, tm_queue, key);
err = bcmolt_cfg_clear(dev_id, &(cfg.hdr));
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to remove queue, direction = %s, queue_id %d, sched_id %d, \
tm_q_set_id %d, intf_id %d, onu_id %d, uni_id %d, err = %s\n",
direction.c_str(), key.id, key.sched_id, key.tm_q_set_id, access_intf_id, onu_id, uni_id, bcmos_strerror(err));
return err;
}
OPENOLT_LOG(INFO, openolt_log_id, "Removed tm_queue, direction %s, id %d, sched_id %d, tm_q_set_id %d, \
intf_id %d, onu_id %d, uni_id %d\n", direction.c_str(), key.id, key.sched_id, key.tm_q_set_id, access_intf_id, onu_id, uni_id);
return BCM_ERR_OK;
}
Status RemoveTrafficQueues_(const tech_profile::TrafficQueues *traffic_queues) {
uint32_t intf_id = traffic_queues->intf_id();
uint32_t onu_id = traffic_queues->onu_id();
uint32_t uni_id = traffic_queues->uni_id();
uint32_t port_no = traffic_queues->port_no();
uint32_t sched_id;
std::string direction;
bcmos_errno err;
bcmolt_egress_qos_type qos_type = get_qos_type(intf_id, onu_id, uni_id, traffic_queues->traffic_queues_size());
for (int i = 0; i < traffic_queues->traffic_queues_size(); i++) {
tech_profile::TrafficQueue traffic_queue = traffic_queues->traffic_queues(i);
direction = GetDirection(traffic_queue.direction());
if (direction.compare("direction-not-supported") == 0)
return bcm_to_grpc_err(BCM_ERR_PARM, "direction-not-supported");
err = RemoveQueue(direction, intf_id, onu_id, uni_id, qos_type, traffic_queue.priority(), traffic_queue.gemport_id());
if (err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to remove queue, err = %s\n",bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to remove queue");
}
}
if (qos_type == BCMOLT_EGRESS_QOS_TYPE_PRIORITY_TO_QUEUE && (direction.compare(upstream) == 0 || direction.compare(downstream) == 0 && is_tm_sched_id_present(intf_id, onu_id, uni_id, direction))) {
sched_id = (direction.compare(upstream) == 0) ? get_default_tm_sched_id(nni_intf_id, direction) : \
get_tm_sched_id(intf_id, onu_id, uni_id, direction);
int tm_qmp_id = get_tm_qmp_id(sched_id, intf_id, onu_id, uni_id);
if (free_tm_qmp_id(sched_id, intf_id, onu_id, uni_id, tm_qmp_id)) {
err = RemoveTrafficQueueMappingProfile(tm_qmp_id);
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to remove tm queue mapping profile, err = %s\n", bcmos_strerror(err));
return bcm_to_grpc_err(err, "Failed to remove tm queue mapping profile");
}
}
}
clear_qos_type(intf_id, onu_id, uni_id);
return Status::OK;
}
Status check_connection() {
int maxTrials = 60;
while (!bcmolt_api_conn_mgr_is_connected(dev_id)) {
sleep(1);
if (--maxTrials == 0)
return grpc::Status(grpc::StatusCode::UNAVAILABLE, "check connection failed");
else
OPENOLT_LOG(INFO, openolt_log_id, "waiting for daemon connection ...\n");
}
OPENOLT_LOG(INFO, openolt_log_id, "daemon is connected\n");
return Status::OK;
}
Status check_bal_ready() {
bcmos_errno err;
int maxTrials = 30;
bcmolt_olt_cfg olt_cfg = { };
bcmolt_olt_key olt_key = { };
BCMOLT_CFG_INIT(&olt_cfg, olt, olt_key);
BCMOLT_MSG_FIELD_GET(&olt_cfg, bal_state);
while (olt_cfg.data.bal_state != BCMOLT_BAL_STATE_BAL_AND_SWITCH_READY) {
if (--maxTrials == 0)
return grpc::Status(grpc::StatusCode::UNAVAILABLE, "check bal ready failed");
sleep(5);
#ifdef TEST_MODE
// It is impossible to mock the setting of olt_cfg.data.bal_state because
// the actual bcmolt_cfg_get passes the address of olt_cfg.hdr and we cannot
// set the olt_cfg.data.bal_state. So a new stub function is created and address
// of olt_cfg is passed. This is one-of case where we need to add test specific
// code in production code.
if (bcmolt_cfg_get__bal_state_stub(dev_id, &olt_cfg)) {
#else
if (bcmolt_cfg_get(dev_id, &olt_cfg.hdr)) {
#endif
continue;
}
else
OPENOLT_LOG(INFO, openolt_log_id, "waiting for BAL ready ...\n");
}
OPENOLT_LOG(INFO, openolt_log_id, "BAL is ready\n");
return Status::OK;
}
Status PerformGroupOperation_(const openolt::Group *group_cfg) {
bcmos_errno err;
bcmolt_group_key key = {};
bcmolt_group_cfg grp_cfg_obj;
bcmolt_group_members_update grp_mem_upd;
bcmolt_members_update_command grp_mem_upd_cmd;
bcmolt_group_member_info member_info = {};
bcmolt_group_member_info_list_u8 members = {};
bcmolt_intf_ref interface_ref = {};
bcmolt_egress_qos egress_qos = {};
bcmolt_tm_sched_ref tm_sched_ref = {};
bcmolt_action a_val = {};
uint32_t group_id = group_cfg->group_id();
OPENOLT_LOG(INFO, openolt_log_id, "PerformGroupOperation request received for Group %d\n", group_id);
if (group_id >= 0) {
key.id = group_id;
}
else {
OPENOLT_LOG(ERROR, openolt_log_id, "Invalid group id %d.\n", group_id);
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "Invalid group id");
}
BCMOLT_CFG_INIT(&grp_cfg_obj, group, key);
BCMOLT_FIELD_SET_PRESENT(&grp_cfg_obj.data, group_cfg_data, state);
OPENOLT_LOG(INFO, openolt_log_id, "Checking if Group %d exists...\n",group_id);
err = bcmolt_cfg_get(dev_id, &(grp_cfg_obj.hdr));
if (err != BCM_ERR_OK) {
OPENOLT_LOG(ERROR, openolt_log_id, "Error in querying Group %d, err = %s\n", group_id, bcmos_strerror(err));
return bcm_to_grpc_err(err, "Error in querying group");
}
members.len = group_cfg->members_size();
// IMPORTANT: A member cannot be added to a group if the group type is not determined.
// Group type is determined after a flow is assigned to it.
// Therefore, a group must be created first, then a flow (with multicast type) must be assigned to it.
// Only then we can add members to the group.
// if group does not exist, create it and return.
if (grp_cfg_obj.data.state == BCMOLT_GROUP_STATE_NOT_CONFIGURED) {
if (members.len != 0) {
OPENOLT_LOG(ERROR, openolt_log_id, "Member list is not empty for non-existent Group %d. Members can be added only after a flow is assigned to this newly-created group.\n", group_id);
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "Non-empty member list given for non-existent group");
} else {
BCMOLT_CFG_INIT(&grp_cfg_obj, group, key);
BCMOLT_MSG_FIELD_SET(&grp_cfg_obj, cookie, key.id);
/* Setting group actions and action parameters, if any.
Only remove_outer_tag and translate_inner_tag actions and i_vid action parameter
are supported for multicast groups in BAL 3.1.
*/
const ::openolt::Action& action = group_cfg->action();
const ::openolt::ActionCmd &cmd = action.cmd();
bcmolt_action_cmd_id cmd_bmask = BCMOLT_ACTION_CMD_ID_NONE;
if (cmd.remove_outer_tag()) {
OPENOLT_LOG(INFO, openolt_log_id, "Action remove_outer_tag applied to Group %d.\n", group_id);
cmd_bmask = (bcmolt_action_cmd_id) (cmd_bmask | BCMOLT_ACTION_CMD_ID_REMOVE_OUTER_TAG);
}
if (cmd.translate_inner_tag()) {
OPENOLT_LOG(INFO, openolt_log_id, "Action translate_inner_tag applied to Group %d.\n", group_id);
cmd_bmask = (bcmolt_action_cmd_id) (cmd_bmask | BCMOLT_ACTION_CMD_ID_XLATE_INNER_TAG);
}
BCMOLT_FIELD_SET(&a_val, action, cmds_bitmask, cmd_bmask);
if (action.i_vid()) {
OPENOLT_LOG(INFO, openolt_log_id, "Setting action parameter i_vid=%d for Group %d.\n", action.i_vid(), group_id);
BCMOLT_FIELD_SET(&a_val, action, i_vid, action.i_vid());
}
BCMOLT_MSG_FIELD_SET(&grp_cfg_obj, action, a_val);
// Create group
err = bcmolt_cfg_set(dev_id, &(grp_cfg_obj.hdr));
if (BCM_ERR_OK != err) {
BCM_LOG(ERROR, openolt_log_id, "Failed to create Group %d, err = %s (%d)\n", key.id, bcmos_strerror(err), err);
return bcm_to_grpc_err(err, "Error in creating group");
}
BCM_LOG(INFO, openolt_log_id, "Group %d has been created and configured with empty member list.\n", key.id);
return Status::OK;
}
}
// The group already exists. Continue configuring it according to the update member command.
OPENOLT_LOG(INFO, openolt_log_id, "Configuring existing Group %d.\n",group_id);
// MEMBER LIST CONSTRUCTION
// Note that members.len can be 0 here. if the group already exists and the command is SET then sending
// empty list to the group is a legit operation and this actually empties the member list.
members.arr = (bcmolt_group_member_info*)bcmos_calloc((members.len)*sizeof(bcmolt_group_member_info));
if (!members.arr) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to allocate memory for group member list.\n");
return grpc::Status(grpc::StatusCode::RESOURCE_EXHAUSTED, "Memory exhausted during member list creation");
}
/* SET GROUP MEMBERS UPDATE COMMAND */
openolt::Group::GroupMembersCommand command = group_cfg->command();
switch(command) {
case openolt::Group::SET_MEMBERS :
grp_mem_upd_cmd = BCMOLT_MEMBERS_UPDATE_COMMAND_SET;
OPENOLT_LOG(INFO, openolt_log_id, "Setting %d members for Group %d.\n", members.len, group_id);
break;
case openolt::Group::ADD_MEMBERS :
grp_mem_upd_cmd = BCMOLT_MEMBERS_UPDATE_COMMAND_ADD;
OPENOLT_LOG(INFO, openolt_log_id, "Adding %d members to Group %d.\n", members.len, group_id);
break;
case openolt::Group::REMOVE_MEMBERS :
grp_mem_upd_cmd = BCMOLT_MEMBERS_UPDATE_COMMAND_REMOVE;
OPENOLT_LOG(INFO, openolt_log_id, "Removing %d members from Group %d.\n", members.len, group_id);
break;
default :
OPENOLT_LOG(ERROR, openolt_log_id, "Invalid value %d for group member command.\n", command);
bcmos_free(members.arr);
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "Invalid group member command");
}
// SET MEMBERS LIST
for (int i = 0; i < members.len; i++) {
if (command == openolt::Group::REMOVE_MEMBERS) {
OPENOLT_LOG(INFO, openolt_log_id, "Removing group member %d from group %d\n",i,key.id);
} else {
OPENOLT_LOG(INFO, openolt_log_id, "Adding group member %d to group %d\n",i,key.id);
}
openolt::GroupMember *member = (openolt::GroupMember *) &group_cfg->members()[i];
// Set member interface type
openolt::GroupMember::InterfaceType if_type = member->interface_type();
switch(if_type){
case openolt::GroupMember::PON :
BCMOLT_FIELD_SET(&interface_ref, intf_ref, intf_type, BCMOLT_INTERFACE_TYPE_PON);
OPENOLT_LOG(INFO, openolt_log_id, "Interface type PON is assigned to GroupMember %d\n",i);
break;
case openolt::GroupMember::EPON_1G_PATH :
BCMOLT_FIELD_SET(&interface_ref, intf_ref, intf_type, BCMOLT_INTERFACE_TYPE_EPON_1_G);
OPENOLT_LOG(INFO, openolt_log_id, "Interface type EPON_1G is assigned to GroupMember %d\n",i);
break;
case openolt::GroupMember::EPON_10G_PATH :
BCMOLT_FIELD_SET(&interface_ref, intf_ref, intf_type, BCMOLT_INTERFACE_TYPE_EPON_10_G);
OPENOLT_LOG(INFO, openolt_log_id, "Interface type EPON_10G is assigned to GroupMember %d\n",i);
break;
default :
OPENOLT_LOG(ERROR, openolt_log_id, "Invalid interface type value %d for GroupMember %d.\n",if_type,i);
bcmos_free(members.arr);
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "Invalid interface type for a group member");
}
// Set member interface id
if (member->interface_id() >= 0) {
BCMOLT_FIELD_SET(&interface_ref, intf_ref, intf_id, member->interface_id());
OPENOLT_LOG(INFO, openolt_log_id, "Interface %d is assigned to GroupMember %d\n", member->interface_id(), i);
} else {
bcmos_free(members.arr);
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "Invalid interface id for a group member");
}
// Set member interface_ref
BCMOLT_FIELD_SET(&member_info, group_member_info, intf, interface_ref);
// Set member gem_port_id. This must be a multicast gemport.
if (member->gem_port_id() >= 0) {
BCMOLT_FIELD_SET(&member_info, group_member_info, svc_port_id, member->gem_port_id());
OPENOLT_LOG(INFO, openolt_log_id, "GEM Port %d is assigned to GroupMember %d\n", member->gem_port_id(), i);
} else {
bcmos_free(members.arr);
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "Invalid gem port id for a group member");
}
// Set member scheduler id and queue_id
uint32_t tm_sched_id = get_default_tm_sched_id(member->interface_id(), downstream);
OPENOLT_LOG(INFO, openolt_log_id, "Scheduler %d is assigned to GroupMember %d\n", tm_sched_id, i);
BCMOLT_FIELD_SET(&tm_sched_ref, tm_sched_ref, id, tm_sched_id);
BCMOLT_FIELD_SET(&egress_qos, egress_qos, tm_sched, tm_sched_ref);
// We assume that all multicast traffic destined to a PON port is using the same fixed queue.
uint32_t tm_queue_id;
if (member->priority() >= 0 && member->priority() < NUMBER_OF_DEFAULT_INTERFACE_QUEUES) {
tm_queue_id = queue_id_list[member->priority()];
OPENOLT_LOG(INFO, openolt_log_id, "Queue %d is assigned to GroupMember %d\n", tm_queue_id, i);
BCMOLT_FIELD_SET(&egress_qos, egress_qos, type, BCMOLT_EGRESS_QOS_TYPE_FIXED_QUEUE);
BCMOLT_FIELD_SET(&egress_qos.u.fixed_queue, egress_qos_fixed_queue, queue_id, tm_queue_id);
} else {
OPENOLT_LOG(ERROR, openolt_log_id, "Invalid fixed queue priority/ID %d for GroupMember %d\n", member->priority(), i);
bcmos_free(members.arr);
return grpc::Status(grpc::StatusCode::INVALID_ARGUMENT, "Invalid queue priority for a group member");
}
BCMOLT_FIELD_SET(&member_info, group_member_info, egress_qos, egress_qos);
BCMOLT_ARRAY_ELEM_SET(&(members), i, member_info);
}
BCMOLT_OPER_INIT(&grp_mem_upd, group, members_update, key);
BCMOLT_MSG_FIELD_SET(&grp_mem_upd, members_cmd.members, members);
BCMOLT_MSG_FIELD_SET(&grp_mem_upd, members_cmd.command, grp_mem_upd_cmd);
err = bcmolt_oper_submit(dev_id, &(grp_mem_upd.hdr));
bcmos_free(members.arr);
if (BCM_ERR_OK != err) {
OPENOLT_LOG(ERROR, openolt_log_id, "Failed to submit members update operation for Group %d err = %s (%d)\n", key.id, bcmos_strerror(err), err);
return bcm_to_grpc_err(err, "Failed to submit members update operation for the group");
}
OPENOLT_LOG(INFO, openolt_log_id, "Successfully submitted members update operation for Group %d\n", key.id);
return Status::OK;
}