net: add rte prefix to ether defines
[dpdk.git] / examples / vm_power_manager / channel_monitor.c
index 73bddd9..aab19ba 100644 (file)
@@ -9,11 +9,19 @@
 #include <signal.h>
 #include <errno.h>
 #include <string.h>
+#include <fcntl.h>
 #include <sys/types.h>
 #include <sys/epoll.h>
 #include <sys/queue.h>
 #include <sys/time.h>
-
+#include <sys/socket.h>
+#include <sys/select.h>
+#ifdef USE_JANSSON
+#include <jansson.h>
+#else
+#pragma message "Jansson dev libs unavailable, not including JSON parsing"
+#endif
+#include <rte_string_fns.h>
 #include <rte_log.h>
 #include <rte_memory.h>
 #include <rte_malloc.h>
@@ -27,6 +35,7 @@
 #include "channel_commands.h"
 #include "channel_manager.h"
 #include "power_manager.h"
+#include "oob_monitor.h"
 
 #define RTE_LOGTYPE_CHANNEL_MONITOR RTE_LOGTYPE_USER1
 
 
 uint64_t vsi_pkt_count_prev[384];
 uint64_t rdtsc_prev[384];
+#define MAX_JSON_STRING_LEN 1024
+char json_data[MAX_JSON_STRING_LEN];
 
 double time_period_ms = 1;
 static volatile unsigned run_loop = 1;
 static int global_event_fd;
 static unsigned int policy_is_set;
 static struct epoll_event *global_events_list;
-static struct policy policies[MAX_VMS];
+static struct policy policies[MAX_CLIENTS];
+
+#ifdef USE_JANSSON
+
+union PFID {
+       struct rte_ether_addr addr;
+       uint64_t pfid;
+};
+
+static int
+str_to_ether_addr(const char *a, struct rte_ether_addr *ether_addr)
+{
+       int i;
+       char *end;
+       unsigned long o[RTE_ETHER_ADDR_LEN];
+
+       i = 0;
+       do {
+               errno = 0;
+               o[i] = strtoul(a, &end, 16);
+               if (errno != 0 || end == a || (end[0] != ':' && end[0] != 0))
+                       return -1;
+               a = end + 1;
+       } while (++i != RTE_DIM(o) / sizeof(o[0]) && end[0] != 0);
+
+       /* Junk at the end of line */
+       if (end[0] != 0)
+               return -1;
+
+       /* Support the format XX:XX:XX:XX:XX:XX */
+       if (i == RTE_ETHER_ADDR_LEN) {
+               while (i-- != 0) {
+                       if (o[i] > UINT8_MAX)
+                               return -1;
+                       ether_addr->addr_bytes[i] = (uint8_t)o[i];
+               }
+       /* Support the format XXXX:XXXX:XXXX */
+       } else if (i == RTE_ETHER_ADDR_LEN / 2) {
+               while (i-- != 0) {
+                       if (o[i] > UINT16_MAX)
+                               return -1;
+                       ether_addr->addr_bytes[i * 2] =
+                                       (uint8_t)(o[i] >> 8);
+                       ether_addr->addr_bytes[i * 2 + 1] =
+                                       (uint8_t)(o[i] & 0xff);
+               }
+       /* unknown format */
+       } else
+               return -1;
+
+       return 0;
+}
+
+static int
+set_policy_mac(struct channel_packet *pkt, int idx, char *mac)
+{
+       union PFID pfid;
+       int ret;
+
+       /* Use port MAC address as the vfid */
+       ret = str_to_ether_addr(mac, &pfid.addr);
+
+       if (ret != 0) {
+               RTE_LOG(ERR, CHANNEL_MONITOR,
+                       "Invalid mac address received in JSON\n");
+               pkt->vfid[idx] = 0;
+               return -1;
+       }
+
+       printf("Received MAC Address: %02" PRIx8 ":%02" PRIx8 ":%02" PRIx8 ":"
+                       "%02" PRIx8 ":%02" PRIx8 ":%02" PRIx8 "\n",
+                       pfid.addr.addr_bytes[0], pfid.addr.addr_bytes[1],
+                       pfid.addr.addr_bytes[2], pfid.addr.addr_bytes[3],
+                       pfid.addr.addr_bytes[4], pfid.addr.addr_bytes[5]);
+
+       pkt->vfid[idx] = pfid.pfid;
+       return 0;
+}
+
+
+static int
+parse_json_to_pkt(json_t *element, struct channel_packet *pkt)
+{
+       const char *key;
+       json_t *value;
+       int ret;
+
+       memset(pkt, 0, sizeof(struct channel_packet));
+
+       pkt->nb_mac_to_monitor = 0;
+       pkt->t_boost_status.tbEnabled = false;
+       pkt->workload = LOW;
+       pkt->policy_to_use = TIME;
+       pkt->command = PKT_POLICY;
+       pkt->core_type = CORE_TYPE_PHYSICAL;
+
+       json_object_foreach(element, key, value) {
+               if (!strcmp(key, "policy")) {
+                       /* Recurse in to get the contents of profile */
+                       ret = parse_json_to_pkt(value, pkt);
+                       if (ret)
+                               return ret;
+               } else if (!strcmp(key, "instruction")) {
+                       /* Recurse in to get the contents of instruction */
+                       ret = parse_json_to_pkt(value, pkt);
+                       if (ret)
+                               return ret;
+               } else if (!strcmp(key, "name")) {
+                       strlcpy(pkt->vm_name, json_string_value(value),
+                                       sizeof(pkt->vm_name));
+               } else if (!strcmp(key, "command")) {
+                       char command[32];
+                       strlcpy(command, json_string_value(value), 32);
+                       if (!strcmp(command, "power")) {
+                               pkt->command = CPU_POWER;
+                       } else if (!strcmp(command, "create")) {
+                               pkt->command = PKT_POLICY;
+                       } else if (!strcmp(command, "destroy")) {
+                               pkt->command = PKT_POLICY_REMOVE;
+                       } else {
+                               RTE_LOG(ERR, CHANNEL_MONITOR,
+                                       "Invalid command received in JSON\n");
+                               return -1;
+                       }
+               } else if (!strcmp(key, "policy_type")) {
+                       char command[32];
+                       strlcpy(command, json_string_value(value), 32);
+                       if (!strcmp(command, "TIME")) {
+                               pkt->policy_to_use = TIME;
+                       } else if (!strcmp(command, "TRAFFIC")) {
+                               pkt->policy_to_use = TRAFFIC;
+                       } else if (!strcmp(command, "WORKLOAD")) {
+                               pkt->policy_to_use = WORKLOAD;
+                       } else if (!strcmp(command, "BRANCH_RATIO")) {
+                               pkt->policy_to_use = BRANCH_RATIO;
+                       } else {
+                               RTE_LOG(ERR, CHANNEL_MONITOR,
+                                       "Wrong policy_type received in JSON\n");
+                               return -1;
+                       }
+               } else if (!strcmp(key, "workload")) {
+                       char command[32];
+                       strlcpy(command, json_string_value(value), 32);
+                       if (!strcmp(command, "HIGH")) {
+                               pkt->workload = HIGH;
+                       } else if (!strcmp(command, "MEDIUM")) {
+                               pkt->workload = MEDIUM;
+                       } else if (!strcmp(command, "LOW")) {
+                               pkt->workload = LOW;
+                       } else {
+                               RTE_LOG(ERR, CHANNEL_MONITOR,
+                                       "Wrong workload received in JSON\n");
+                               return -1;
+                       }
+               } else if (!strcmp(key, "busy_hours")) {
+                       unsigned int i;
+                       size_t size = json_array_size(value);
+
+                       for (i = 0; i < size; i++) {
+                               int hour = (int)json_integer_value(
+                                               json_array_get(value, i));
+                               pkt->timer_policy.busy_hours[i] = hour;
+                       }
+               } else if (!strcmp(key, "quiet_hours")) {
+                       unsigned int i;
+                       size_t size = json_array_size(value);
+
+                       for (i = 0; i < size; i++) {
+                               int hour = (int)json_integer_value(
+                                               json_array_get(value, i));
+                               pkt->timer_policy.quiet_hours[i] = hour;
+                       }
+               } else if (!strcmp(key, "core_list")) {
+                       unsigned int i;
+                       size_t size = json_array_size(value);
+
+                       for (i = 0; i < size; i++) {
+                               int core = (int)json_integer_value(
+                                               json_array_get(value, i));
+                               pkt->vcpu_to_control[i] = core;
+                       }
+                       pkt->num_vcpu = size;
+               } else if (!strcmp(key, "mac_list")) {
+                       unsigned int i;
+                       size_t size = json_array_size(value);
+
+                       for (i = 0; i < size; i++) {
+                               char mac[32];
+                               strlcpy(mac,
+                                       json_string_value(json_array_get(value, i)),
+                                       32);
+                               set_policy_mac(pkt, i, mac);
+                       }
+                       pkt->nb_mac_to_monitor = size;
+               } else if (!strcmp(key, "avg_packet_thresh")) {
+                       pkt->traffic_policy.avg_max_packet_thresh =
+                                       (uint32_t)json_integer_value(value);
+               } else if (!strcmp(key, "max_packet_thresh")) {
+                       pkt->traffic_policy.max_max_packet_thresh =
+                                       (uint32_t)json_integer_value(value);
+               } else if (!strcmp(key, "unit")) {
+                       char unit[32];
+                       strlcpy(unit, json_string_value(value), 32);
+                       if (!strcmp(unit, "SCALE_UP")) {
+                               pkt->unit = CPU_POWER_SCALE_UP;
+                       } else if (!strcmp(unit, "SCALE_DOWN")) {
+                               pkt->unit = CPU_POWER_SCALE_DOWN;
+                       } else if (!strcmp(unit, "SCALE_MAX")) {
+                               pkt->unit = CPU_POWER_SCALE_MAX;
+                       } else if (!strcmp(unit, "SCALE_MIN")) {
+                               pkt->unit = CPU_POWER_SCALE_MIN;
+                       } else if (!strcmp(unit, "ENABLE_TURBO")) {
+                               pkt->unit = CPU_POWER_ENABLE_TURBO;
+                       } else if (!strcmp(unit, "DISABLE_TURBO")) {
+                               pkt->unit = CPU_POWER_DISABLE_TURBO;
+                       } else {
+                               RTE_LOG(ERR, CHANNEL_MONITOR,
+                                       "Invalid command received in JSON\n");
+                               return -1;
+                       }
+               } else if (!strcmp(key, "resource_id")) {
+                       pkt->resource_id = (uint32_t)json_integer_value(value);
+               } else {
+                       RTE_LOG(ERR, CHANNEL_MONITOR,
+                               "Unknown key received in JSON string: %s\n",
+                               key);
+               }
+       }
+       return 0;
+}
+#endif
 
 void channel_monitor_exit(void)
 {
@@ -65,7 +306,7 @@ static void
 core_share_status(int pNo)
 {
 
-       int noVms, noVcpus, z, x, t;
+       int noVms = 0, noVcpus = 0, z, x, t;
 
        get_all_vm(&noVms, &noVcpus);
 
@@ -84,6 +325,33 @@ core_share_status(int pNo)
        }
 }
 
+
+static int
+pcpu_monitor(struct policy *pol, struct core_info *ci, int pcpu, int count)
+{
+       int ret = 0;
+
+       if (pol->pkt.policy_to_use == BRANCH_RATIO) {
+               ci->cd[pcpu].oob_enabled = 1;
+               ret = add_core_to_monitor(pcpu);
+               if (ret == 0)
+                       RTE_LOG(INFO, CHANNEL_MONITOR,
+                                       "Monitoring pcpu %d OOB for %s\n",
+                                       pcpu, pol->pkt.vm_name);
+               else
+                       RTE_LOG(ERR, CHANNEL_MONITOR,
+                                       "Error monitoring pcpu %d OOB for %s\n",
+                                       pcpu, pol->pkt.vm_name);
+
+       } else {
+               pol->core_share[count].pcpu = pcpu;
+               RTE_LOG(INFO, CHANNEL_MONITOR,
+                               "Monitoring pcpu %d for %s\n",
+                               pcpu, pol->pkt.vm_name);
+       }
+       return ret;
+}
+
 static void
 get_pcpu_to_control(struct policy *pol)
 {
@@ -91,17 +359,37 @@ get_pcpu_to_control(struct policy *pol)
        /* Convert vcpu to pcpu. */
        struct vm_info info;
        int pcpu, count;
-       uint64_t mask_u64b;
+       struct core_info *ci;
+
+       ci = get_core_info();
 
-       RTE_LOG(INFO, CHANNEL_MONITOR, "Looking for pcpu for %s\n",
-                       pol->pkt.vm_name);
-       get_info_vm(pol->pkt.vm_name, &info);
+       RTE_LOG(DEBUG, CHANNEL_MONITOR,
+                       "Looking for pcpu for %s\n", pol->pkt.vm_name);
 
-       for (count = 0; count < pol->pkt.num_vcpu; count++) {
-               mask_u64b = info.pcpu_mask[pol->pkt.vcpu_to_control[count]];
-               for (pcpu = 0; mask_u64b; mask_u64b &= ~(1ULL << pcpu++)) {
-                       if ((mask_u64b >> pcpu) & 1)
-                               pol->core_share[count].pcpu = pcpu;
+       /*
+        * So now that we're handling virtual and physical cores, we need to
+        * differenciate between them when adding them to the branch monitor.
+        * Virtual cores need to be converted to physical cores.
+        */
+       if (pol->pkt.core_type == CORE_TYPE_VIRTUAL) {
+               /*
+                * If the cores in the policy are virtual, we need to map them
+                * to physical core. We look up the vm info and use that for
+                * the mapping.
+                */
+               get_info_vm(pol->pkt.vm_name, &info);
+               for (count = 0; count < pol->pkt.num_vcpu; count++) {
+                       pcpu = info.pcpu_map[pol->pkt.vcpu_to_control[count]];
+                       pcpu_monitor(pol, ci, pcpu, count);
+               }
+       } else {
+               /*
+                * If the cores in the policy are physical, we just use
+                * those core id's directly.
+                */
+               for (count = 0; count < pol->pkt.num_vcpu; count++) {
+                       pcpu = pol->pkt.vcpu_to_control[count];
+                       pcpu_monitor(pol, ci, pcpu, count);
                }
        }
 }
@@ -116,7 +404,7 @@ get_pfid(struct policy *pol)
 
                RTE_ETH_FOREACH_DEV(x) {
                        ret = rte_pmd_i40e_query_vfid_by_mac(x,
-                               (struct ether_addr *)&(pol->pkt.vfid[i]));
+                               (struct rte_ether_addr *)&(pol->pkt.vfid[i]));
                        if (ret != -EINVAL) {
                                pol->port[i] = x;
                                break;
@@ -141,13 +429,21 @@ update_policy(struct channel_packet *pkt)
        unsigned int updated = 0;
        int i;
 
-       for (i = 0; i < MAX_VMS; i++) {
+
+       RTE_LOG(INFO, CHANNEL_MONITOR,
+                       "Applying policy for %s\n", pkt->vm_name);
+
+       for (i = 0; i < MAX_CLIENTS; i++) {
                if (strcmp(policies[i].pkt.vm_name, pkt->vm_name) == 0) {
+                       /* Copy the contents of *pkt into the policy.pkt */
                        policies[i].pkt = *pkt;
                        get_pcpu_to_control(&policies[i]);
-                       if (get_pfid(&policies[i]) == -1) {
-                               updated = 1;
-                               break;
+                       /* Check Eth dev only for Traffic policy */
+                       if (policies[i].pkt.policy_to_use == TRAFFIC) {
+                               if (get_pfid(&policies[i]) < 0) {
+                                       updated = 1;
+                                       break;
+                               }
                        }
                        core_share_status(i);
                        policies[i].enabled = 1;
@@ -155,12 +451,17 @@ update_policy(struct channel_packet *pkt)
                }
        }
        if (!updated) {
-               for (i = 0; i < MAX_VMS; i++) {
+               for (i = 0; i < MAX_CLIENTS; i++) {
                        if (policies[i].enabled == 0) {
                                policies[i].pkt = *pkt;
                                get_pcpu_to_control(&policies[i]);
-                               if (get_pfid(&policies[i]) == -1)
-                                       break;
+                               /* Check Eth dev only for Traffic policy */
+                               if (policies[i].pkt.policy_to_use == TRAFFIC) {
+                                       if (get_pfid(&policies[i]) < 0) {
+                                               updated = 1;
+                                               break;
+                                       }
+                               }
                                core_share_status(i);
                                policies[i].enabled = 1;
                                break;
@@ -170,6 +471,24 @@ update_policy(struct channel_packet *pkt)
        return 0;
 }
 
+static int
+remove_policy(struct channel_packet *pkt __rte_unused)
+{
+       int i;
+
+       /*
+        * Disabling the policy is simply a case of setting
+        * enabled to 0
+        */
+       for (i = 0; i < MAX_CLIENTS; i++) {
+               if (strcmp(policies[i].pkt.vm_name, pkt->vm_name) == 0) {
+                       policies[i].enabled = 0;
+                       return 0;
+               }
+       }
+       return -1;
+}
+
 static uint64_t
 get_pkt_diff(struct policy *pol)
 {
@@ -214,8 +533,6 @@ apply_traffic_profile(struct policy *pol)
 
        diff = get_pkt_diff(pol);
 
-       RTE_LOG(INFO, CHANNEL_MONITOR, "Applying traffic profile\n");
-
        if (diff >= (pol->pkt.traffic_policy.max_max_packet_thresh)) {
                for (count = 0; count < pol->pkt.num_vcpu; count++) {
                        if (pol->core_share[count].status != 1)
@@ -259,9 +576,6 @@ apply_time_profile(struct policy *pol)
                                if (pol->core_share[count].status != 1) {
                                        power_manager_scale_core_max(
                                                pol->core_share[count].pcpu);
-                               RTE_LOG(INFO, CHANNEL_MONITOR,
-                                       "Scaling up core %d to max\n",
-                                       pol->core_share[count].pcpu);
                                }
                        }
                        break;
@@ -271,9 +585,6 @@ apply_time_profile(struct policy *pol)
                                if (pol->core_share[count].status != 1) {
                                        power_manager_scale_core_min(
                                                pol->core_share[count].pcpu);
-                               RTE_LOG(INFO, CHANNEL_MONITOR,
-                                       "Scaling down core %d to min\n",
-                                       pol->core_share[count].pcpu);
                        }
                }
                        break;
@@ -327,11 +638,10 @@ apply_policy(struct policy *pol)
                apply_workload_profile(pol);
 }
 
-
 static int
 process_request(struct channel_packet *pkt, struct channel_info *chan_info)
 {
-       uint64_t core_mask;
+       int ret;
 
        if (chan_info == NULL)
                return -1;
@@ -341,73 +651,59 @@ process_request(struct channel_packet *pkt, struct channel_info *chan_info)
                return -1;
 
        if (pkt->command == CPU_POWER) {
-               core_mask = get_pcpus_mask(chan_info, pkt->resource_id);
-               if (core_mask == 0) {
-                       RTE_LOG(ERR, CHANNEL_MONITOR, "Error get physical CPU mask for "
-                               "channel '%s' using vCPU(%u)\n", chan_info->channel_path,
-                               (unsigned)pkt->unit);
-                       return -1;
-               }
-               if (__builtin_popcountll(core_mask) == 1) {
+               unsigned int core_num;
 
-                       unsigned core_num = __builtin_ffsll(core_mask) - 1;
+               if (pkt->core_type == CORE_TYPE_VIRTUAL)
+                       core_num = get_pcpu(chan_info, pkt->resource_id);
+               else
+                       core_num = pkt->resource_id;
 
-                       switch (pkt->unit) {
-                       case(CPU_POWER_SCALE_MIN):
-                                       power_manager_scale_core_min(core_num);
-                       break;
-                       case(CPU_POWER_SCALE_MAX):
-                                       power_manager_scale_core_max(core_num);
-                       break;
-                       case(CPU_POWER_SCALE_DOWN):
-                                       power_manager_scale_core_down(core_num);
-                       break;
-                       case(CPU_POWER_SCALE_UP):
-                                       power_manager_scale_core_up(core_num);
-                       break;
-                       case(CPU_POWER_ENABLE_TURBO):
-                               power_manager_enable_turbo_core(core_num);
-                       break;
-                       case(CPU_POWER_DISABLE_TURBO):
-                               power_manager_disable_turbo_core(core_num);
+               RTE_LOG(DEBUG, CHANNEL_MONITOR, "Processing requested cmd for cpu:%d\n",
+                       core_num);
+
+               switch (pkt->unit) {
+               case(CPU_POWER_SCALE_MIN):
+                       power_manager_scale_core_min(core_num);
                        break;
-                       default:
-                               break;
-                       }
-               } else {
-                       switch (pkt->unit) {
-                       case(CPU_POWER_SCALE_MIN):
-                                       power_manager_scale_mask_min(core_mask);
+               case(CPU_POWER_SCALE_MAX):
+                       power_manager_scale_core_max(core_num);
                        break;
-                       case(CPU_POWER_SCALE_MAX):
-                                       power_manager_scale_mask_max(core_mask);
+               case(CPU_POWER_SCALE_DOWN):
+                       power_manager_scale_core_down(core_num);
                        break;
-                       case(CPU_POWER_SCALE_DOWN):
-                                       power_manager_scale_mask_down(core_mask);
+               case(CPU_POWER_SCALE_UP):
+                       power_manager_scale_core_up(core_num);
                        break;
-                       case(CPU_POWER_SCALE_UP):
-                                       power_manager_scale_mask_up(core_mask);
+               case(CPU_POWER_ENABLE_TURBO):
+                       power_manager_enable_turbo_core(core_num);
                        break;
-                       case(CPU_POWER_ENABLE_TURBO):
-                               power_manager_enable_turbo_mask(core_mask);
+               case(CPU_POWER_DISABLE_TURBO):
+                       power_manager_disable_turbo_core(core_num);
                        break;
-                       case(CPU_POWER_DISABLE_TURBO):
-                               power_manager_disable_turbo_mask(core_mask);
+               default:
                        break;
-                       default:
-                               break;
-                       }
-
                }
        }
 
        if (pkt->command == PKT_POLICY) {
-               RTE_LOG(INFO, CHANNEL_MONITOR, "\nProcessing Policy request from Guest\n");
+               RTE_LOG(INFO, CHANNEL_MONITOR, "Processing policy request %s\n",
+                               pkt->vm_name);
                update_policy(pkt);
                policy_is_set = 1;
        }
 
-       /* Return is not checked as channel status may have been set to DISABLED
+       if (pkt->command == PKT_POLICY_REMOVE) {
+               ret = remove_policy(pkt);
+               if (ret == 0)
+                       RTE_LOG(INFO, CHANNEL_MONITOR,
+                                "Removed policy %s\n", pkt->vm_name);
+               else
+                       RTE_LOG(INFO, CHANNEL_MONITOR,
+                                "Policy %s does not exist\n", pkt->vm_name);
+       }
+
+       /*
+        * Return is not checked as channel status may have been set to DISABLED
         * from management thread
         */
        rte_atomic32_cmpset(&(chan_info->status), CHANNEL_MGR_CHANNEL_PROCESSING,
@@ -429,13 +725,16 @@ add_channel_to_monitor(struct channel_info **chan_info)
                                "to epoll\n", info->channel_path);
                return -1;
        }
+       RTE_LOG(ERR, CHANNEL_MONITOR, "Added channel '%s' "
+                       "to monitor\n", info->channel_path);
        return 0;
 }
 
 int
 remove_channel_from_monitor(struct channel_info *chan_info)
 {
-       if (epoll_ctl(global_event_fd, EPOLL_CTL_DEL, chan_info->fd, NULL) < 0) {
+       if (epoll_ctl(global_event_fd, EPOLL_CTL_DEL,
+                       chan_info->fd, NULL) < 0) {
                RTE_LOG(ERR, CHANNEL_MONITOR, "Unable to remove channel '%s' "
                                "from epoll\n", chan_info->channel_path);
                return -1;
@@ -448,11 +747,13 @@ channel_monitor_init(void)
 {
        global_event_fd = epoll_create1(0);
        if (global_event_fd == 0) {
-               RTE_LOG(ERR, CHANNEL_MONITOR, "Error creating epoll context with "
-                               "error %s\n", strerror(errno));
+               RTE_LOG(ERR, CHANNEL_MONITOR,
+                               "Error creating epoll context with error %s\n",
+                               strerror(errno));
                return -1;
        }
-       global_events_list = rte_malloc("epoll_events", sizeof(*global_events_list)
+       global_events_list = rte_malloc("epoll_events",
+                       sizeof(*global_events_list)
                        * MAX_EVENTS, RTE_CACHE_LINE_SIZE);
        if (global_events_list == NULL) {
                RTE_LOG(ERR, CHANNEL_MONITOR, "Unable to rte_malloc for "
@@ -462,6 +763,98 @@ channel_monitor_init(void)
        return 0;
 }
 
+static void
+read_binary_packet(struct channel_info *chan_info)
+{
+       struct channel_packet pkt;
+       void *buffer = &pkt;
+       int buffer_len = sizeof(pkt);
+       int n_bytes, err = 0;
+
+       while (buffer_len > 0) {
+               n_bytes = read(chan_info->fd,
+                               buffer, buffer_len);
+               if (n_bytes == buffer_len)
+                       break;
+               if (n_bytes < 0) {
+                       err = errno;
+                       RTE_LOG(DEBUG, CHANNEL_MONITOR,
+                               "Received error on "
+                               "channel '%s' read: %s\n",
+                               chan_info->channel_path,
+                               strerror(err));
+                       remove_channel(&chan_info);
+                       break;
+               }
+               buffer = (char *)buffer + n_bytes;
+               buffer_len -= n_bytes;
+       }
+       if (!err)
+               process_request(&pkt, chan_info);
+}
+
+#ifdef USE_JANSSON
+static void
+read_json_packet(struct channel_info *chan_info)
+{
+       struct channel_packet pkt;
+       int n_bytes, ret;
+       json_t *root;
+       json_error_t error;
+
+       /* read opening brace to closing brace */
+       do {
+               int idx = 0;
+               int indent = 0;
+               do {
+                       n_bytes = read(chan_info->fd, &json_data[idx], 1);
+                       if (n_bytes == 0)
+                               break;
+                       if (json_data[idx] == '{')
+                               indent++;
+                       if (json_data[idx] == '}')
+                               indent--;
+                       if ((indent > 0) || (idx > 0))
+                               idx++;
+                       if (indent <= 0)
+                               json_data[idx] = 0;
+                       if (idx >= MAX_JSON_STRING_LEN-1)
+                               break;
+               } while (indent > 0);
+
+               json_data[idx] = '\0';
+
+               if (strlen(json_data) == 0)
+                       continue;
+
+               printf("got [%s]\n", json_data);
+
+               root = json_loads(json_data, 0, &error);
+
+               if (root) {
+                       /*
+                        * Because our data is now in the json
+                        * object, we can overwrite the pkt
+                        * with a channel_packet struct, using
+                        * parse_json_to_pkt()
+                        */
+                       ret = parse_json_to_pkt(root, &pkt);
+                       json_decref(root);
+                       if (ret) {
+                               RTE_LOG(ERR, CHANNEL_MONITOR,
+                                       "Error validating JSON profile data\n");
+                               break;
+                       }
+                       process_request(&pkt, chan_info);
+               } else {
+                       RTE_LOG(ERR, CHANNEL_MONITOR,
+                                       "JSON error on line %d: %s\n",
+                                       error.line, error.text);
+               }
+       } while (n_bytes > 0);
+}
+#endif
+
 void
 run_channel_monitor(void)
 {
@@ -477,7 +870,8 @@ run_channel_monitor(void)
                                        global_events_list[i].data.ptr;
                        if ((global_events_list[i].events & EPOLLERR) ||
                                (global_events_list[i].events & EPOLLHUP)) {
-                               RTE_LOG(DEBUG, CHANNEL_MONITOR, "Remote closed connection for "
+                               RTE_LOG(INFO, CHANNEL_MONITOR,
+                                               "Remote closed connection for "
                                                "channel '%s'\n",
                                                chan_info->channel_path);
                                remove_channel(&chan_info);
@@ -485,38 +879,25 @@ run_channel_monitor(void)
                        }
                        if (global_events_list[i].events & EPOLLIN) {
 
-                               int n_bytes, err = 0;
-                               struct channel_packet pkt;
-                               void *buffer = &pkt;
-                               int buffer_len = sizeof(pkt);
-
-                               while (buffer_len > 0) {
-                                       n_bytes = read(chan_info->fd,
-                                                       buffer, buffer_len);
-                                       if (n_bytes == buffer_len)
-                                               break;
-                                       if (n_bytes == -1) {
-                                               err = errno;
-                                               RTE_LOG(DEBUG, CHANNEL_MONITOR,
-                                                       "Received error on "
-                                                       "channel '%s' read: %s\n",
-                                                       chan_info->channel_path,
-                                                       strerror(err));
-                                               remove_channel(&chan_info);
-                                               break;
-                                       }
-                                       buffer = (char *)buffer + n_bytes;
-                                       buffer_len -= n_bytes;
+                               switch (chan_info->type) {
+                               case CHANNEL_TYPE_BINARY:
+                                       read_binary_packet(chan_info);
+                                       break;
+#ifdef USE_JANSSON
+                               case CHANNEL_TYPE_JSON:
+                                       read_json_packet(chan_info);
+                                       break;
+#endif
+                               default:
+                                       break;
                                }
-                               if (!err)
-                                       process_request(&pkt, chan_info);
                        }
                }
                rte_delay_us(time_period_ms*1000);
                if (policy_is_set) {
                        int j;
 
-                       for (j = 0; j < MAX_VMS; j++) {
+                       for (j = 0; j < MAX_CLIENTS; j++) {
                                if (policies[j].enabled == 1)
                                        apply_policy(&policies[j]);
                        }