examples/vm_power_mgr: add policy to channels
authorRory Sexton <rory.sexton@intel.com>
Wed, 11 Oct 2017 16:18:51 +0000 (17:18 +0100)
committerFerruh Yigit <ferruh.yigit@intel.com>
Wed, 11 Oct 2017 23:45:18 +0000 (00:45 +0100)
Signed-off-by: Nemanja Marjanovic <nemanja.marjanovic@intel.com>
Signed-off-by: Rory Sexton <rory.sexton@intel.com>
Signed-off-by: David Hunt <david.hunt@intel.com>
Reviewed-by: Santosh Shukla <santosh.shukla@caviumnetworks.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
examples/vm_power_manager/Makefile
examples/vm_power_manager/channel_monitor.c
examples/vm_power_manager/channel_monitor.h

index 59a9641..9cf20a2 100644 (file)
@@ -54,6 +54,22 @@ CFLAGS += $(WERROR_FLAGS)
 
 LDLIBS += -lvirt
 
+ifeq ($(CONFIG_RTE_BUILD_SHARED_LIB),y)
+
+ifeq ($(CONFIG_RTE_LIBRTE_IXGBE_PMD),y)
+LDLIBS += -lrte_pmd_ixgbe
+endif
+
+ifeq ($(CONFIG_RTE_LIBRTE_I40E_PMD),y)
+LDLIBS += -lrte_pmd_i40e
+endif
+
+ifeq ($(CONFIG_RTE_LIBRTE_BNXT_PMD),y)
+LDLIBS += -lrte_pmd_bnxt
+endif
+
+endif
+
 # workaround for a gcc bug with noreturn attribute
 # http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12603
 ifeq ($(CONFIG_RTE_TOOLCHAIN_GCC),y)
index ac40dac..f16358d 100644 (file)
 #include <sys/types.h>
 #include <sys/epoll.h>
 #include <sys/queue.h>
+#include <sys/time.h>
 
 #include <rte_log.h>
 #include <rte_memory.h>
 #include <rte_malloc.h>
 #include <rte_atomic.h>
+#include <rte_cycles.h>
+#include <rte_ethdev.h>
+#include <rte_pmd_i40e.h>
 
-
+#include <libvirt/libvirt.h>
 #include "channel_monitor.h"
 #include "channel_commands.h"
 #include "channel_manager.h"
 
 #define MAX_EVENTS 256
 
+uint64_t vsi_pkt_count_prev[384];
+uint64_t rdtsc_prev[384];
 
+double time_period_s = 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];
 
 void channel_monitor_exit(void)
 {
@@ -68,6 +77,286 @@ void channel_monitor_exit(void)
        rte_free(global_events_list);
 }
 
+static void
+core_share(int pNo, int z, int x, int t)
+{
+       if (policies[pNo].core_share[z].pcpu == lvm_info[x].pcpus[t]) {
+               if (strcmp(policies[pNo].pkt.vm_name,
+                               lvm_info[x].vm_name) != 0) {
+                       policies[pNo].core_share[z].status = 1;
+                       power_manager_scale_core_max(
+                                       policies[pNo].core_share[z].pcpu);
+               }
+       }
+}
+
+static void
+core_share_status(int pNo)
+{
+
+       int noVms, noVcpus, z, x, t;
+
+       get_all_vm(&noVms, &noVcpus);
+
+       /* Reset Core Share Status. */
+       for (z = 0; z < noVcpus; z++)
+               policies[pNo].core_share[z].status = 0;
+
+       /* Foreach vcpu in a policy. */
+       for (z = 0; z < policies[pNo].pkt.num_vcpu; z++) {
+               /* Foreach VM on the platform. */
+               for (x = 0; x < noVms; x++) {
+                       /* Foreach vcpu of VMs on platform. */
+                       for (t = 0; t < lvm_info[x].num_cpus; t++)
+                               core_share(pNo, z, x, t);
+               }
+       }
+}
+
+static void
+get_pcpu_to_control(struct policy *pol)
+{
+
+       /* Convert vcpu to pcpu. */
+       struct vm_info info;
+       int pcpu, count;
+       uint64_t mask_u64b;
+
+       RTE_LOG(INFO, CHANNEL_MONITOR, "Looking for pcpu for %s\n",
+                       pol->pkt.vm_name);
+       get_info_vm(pol->pkt.vm_name, &info);
+
+       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;
+               }
+       }
+}
+
+static int
+get_pfid(struct policy *pol)
+{
+
+       int i, x, ret = 0, nb_ports;
+
+       nb_ports = rte_eth_dev_count();
+       for (i = 0; i < pol->pkt.nb_mac_to_monitor; i++) {
+
+               for (x = 0; x < nb_ports; x++) {
+                       ret = rte_pmd_i40e_query_vfid_by_mac(x,
+                               (struct ether_addr *)&(pol->pkt.vfid[i]));
+                       if (ret != -EINVAL) {
+                               pol->port[i] = x;
+                               break;
+                       }
+               }
+               if (ret == -EINVAL || ret == -ENOTSUP || ret == ENODEV) {
+                       RTE_LOG(INFO, CHANNEL_MONITOR,
+                               "Error with Policy. MAC not found on "
+                               "attached ports ");
+                       pol->enabled = 0;
+                       return ret;
+               }
+               pol->pfid[i] = ret;
+       }
+       return 1;
+}
+
+static int
+update_policy(struct channel_packet *pkt)
+{
+
+       unsigned int updated = 0;
+
+       for (int i = 0; i < MAX_VMS; i++) {
+               if (strcmp(policies[i].pkt.vm_name, pkt->vm_name) == 0) {
+                       policies[i].pkt = *pkt;
+                       get_pcpu_to_control(&policies[i]);
+                       if (get_pfid(&policies[i]) == -1) {
+                               updated = 1;
+                               break;
+                       }
+                       core_share_status(i);
+                       policies[i].enabled = 1;
+                       updated = 1;
+               }
+       }
+       if (!updated) {
+               for (int i = 0; i < MAX_VMS; i++) {
+                       if (policies[i].enabled == 0) {
+                               policies[i].pkt = *pkt;
+                               get_pcpu_to_control(&policies[i]);
+                               if (get_pfid(&policies[i]) == -1)
+                                       break;
+                               core_share_status(i);
+                               policies[i].enabled = 1;
+                               break;
+                       }
+               }
+       }
+       return 0;
+}
+
+static uint64_t
+get_pkt_diff(struct policy *pol)
+{
+
+       uint64_t vsi_pkt_count,
+               vsi_pkt_total = 0,
+               vsi_pkt_count_prev_total = 0;
+       double rdtsc_curr, rdtsc_diff, diff;
+       int x;
+       struct rte_eth_stats vf_stats;
+
+       for (x = 0; x < pol->pkt.nb_mac_to_monitor; x++) {
+
+               /*Read vsi stats*/
+               if (rte_pmd_i40e_get_vf_stats(x, pol->pfid[x], &vf_stats) == 0)
+                       vsi_pkt_count = vf_stats.ipackets;
+               else
+                       vsi_pkt_count = -1;
+
+               vsi_pkt_total += vsi_pkt_count;
+
+               vsi_pkt_count_prev_total += vsi_pkt_count_prev[pol->pfid[x]];
+               vsi_pkt_count_prev[pol->pfid[x]] = vsi_pkt_count;
+       }
+
+       rdtsc_curr = rte_rdtsc_precise();
+       rdtsc_diff = rdtsc_curr - rdtsc_prev[pol->pfid[x-1]];
+       rdtsc_prev[pol->pfid[x-1]] = rdtsc_curr;
+
+       diff = (vsi_pkt_total - vsi_pkt_count_prev_total) *
+                       ((double)rte_get_tsc_hz() / rdtsc_diff);
+
+       return diff;
+}
+
+static void
+apply_traffic_profile(struct policy *pol)
+{
+
+       int count;
+       uint64_t diff = 0;
+
+       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)
+                               power_manager_scale_core_max(
+                                               pol->core_share[count].pcpu);
+               }
+       } else if (diff >= (pol->pkt.traffic_policy.avg_max_packet_thresh)) {
+               for (count = 0; count < pol->pkt.num_vcpu; count++) {
+                       if (pol->core_share[count].status != 1)
+                               power_manager_scale_core_med(
+                                               pol->core_share[count].pcpu);
+               }
+       } else if (diff < (pol->pkt.traffic_policy.avg_max_packet_thresh)) {
+               for (count = 0; count < pol->pkt.num_vcpu; count++) {
+                       if (pol->core_share[count].status != 1)
+                               power_manager_scale_core_min(
+                                               pol->core_share[count].pcpu);
+               }
+       }
+}
+
+static void
+apply_time_profile(struct policy *pol)
+{
+
+       int count, x;
+       struct timeval tv;
+       struct tm *ptm;
+       char time_string[40];
+
+       /* Obtain the time of day, and convert it to a tm struct. */
+       gettimeofday(&tv, NULL);
+       ptm = localtime(&tv.tv_sec);
+       /* Format the date and time, down to a single second. */
+       strftime(time_string, sizeof(time_string), "%Y-%m-%d %H:%M:%S", ptm);
+
+       for (x = 0; x < HOURS; x++) {
+
+               if (ptm->tm_hour == pol->pkt.timer_policy.busy_hours[x]) {
+                       for (count = 0; count < pol->pkt.num_vcpu; count++) {
+                               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;
+               } else if (ptm->tm_hour ==
+                               pol->pkt.timer_policy.quiet_hours[x]) {
+                       for (count = 0; count < pol->pkt.num_vcpu; count++) {
+                               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;
+               } else if (ptm->tm_hour ==
+                       pol->pkt.timer_policy.hours_to_use_traffic_profile[x]) {
+                       apply_traffic_profile(pol);
+                       break;
+               }
+       }
+}
+
+static void
+apply_workload_profile(struct policy *pol)
+{
+
+       int count;
+
+       if (pol->pkt.workload == HIGH) {
+               for (count = 0; count < pol->pkt.num_vcpu; count++) {
+                       if (pol->core_share[count].status != 1)
+                               power_manager_scale_core_max(
+                                               pol->core_share[count].pcpu);
+               }
+       } else if (pol->pkt.workload == MEDIUM) {
+               for (count = 0; count < pol->pkt.num_vcpu; count++) {
+                       if (pol->core_share[count].status != 1)
+                               power_manager_scale_core_med(
+                                               pol->core_share[count].pcpu);
+               }
+       } else if (pol->pkt.workload == LOW) {
+               for (count = 0; count < pol->pkt.num_vcpu; count++) {
+                       if (pol->core_share[count].status != 1)
+                               power_manager_scale_core_min(
+                                               pol->core_share[count].pcpu);
+               }
+       }
+}
+
+static void
+apply_policy(struct policy *pol)
+{
+
+       struct channel_packet *pkt = &pol->pkt;
+
+       /*Check policy to use*/
+       if (pkt->policy_to_use == TRAFFIC)
+               apply_traffic_profile(pol);
+       else if (pkt->policy_to_use == TIME)
+               apply_time_profile(pol);
+       else if (pkt->policy_to_use == WORKLOAD)
+               apply_workload_profile(pol);
+}
+
+
 static int
 process_request(struct channel_packet *pkt, struct channel_info *chan_info)
 {
@@ -140,6 +429,13 @@ process_request(struct channel_packet *pkt, struct channel_info *chan_info)
 
                }
        }
+
+       if (pkt->command == PKT_POLICY) {
+               RTE_LOG(INFO, CHANNEL_MONITOR, "\nProcessing Policy request from Guest\n");
+               update_policy(pkt);
+               policy_is_set = 1;
+       }
+
        /* Return is not checked as channel status may have been set to DISABLED
         * from management thread
         */
@@ -209,9 +505,10 @@ run_channel_monitor(void)
                        struct channel_info *chan_info = (struct channel_info *)
                                        global_events_list[i].data.ptr;
                        if ((global_events_list[i].events & EPOLLERR) ||
-                                       (global_events_list[i].events & EPOLLHUP)) {
+                               (global_events_list[i].events & EPOLLHUP)) {
                                RTE_LOG(DEBUG, CHANNEL_MONITOR, "Remote closed connection for "
-                                               "channel '%s'\n", chan_info->channel_path);
+                                               "channel '%s'\n",
+                                               chan_info->channel_path);
                                remove_channel(&chan_info);
                                continue;
                        }
@@ -223,14 +520,17 @@ run_channel_monitor(void)
                                int buffer_len = sizeof(pkt);
 
                                while (buffer_len > 0) {
-                                       n_bytes = read(chan_info->fd, buffer, buffer_len);
+                                       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));
+                                               RTE_LOG(DEBUG, CHANNEL_MONITOR,
+                                                       "Received error on "
+                                                       "channel '%s' read: %s\n",
+                                                       chan_info->channel_path,
+                                                       strerror(err));
                                                remove_channel(&chan_info);
                                                break;
                                        }
@@ -241,5 +541,12 @@ run_channel_monitor(void)
                                        process_request(&pkt, chan_info);
                        }
                }
+               rte_delay_us(time_period_s*1000000);
+               if (policy_is_set) {
+                       for (int j = 0; j < MAX_VMS; j++) {
+                               if (policies[j].enabled == 1)
+                                       apply_policy(&policies[j]);
+                       }
+               }
        }
 }
index c138607..b52c1fc 100644 (file)
 #define CHANNEL_MONITOR_H_
 
 #include "channel_manager.h"
+#include "channel_commands.h"
+
+struct core_share {
+       unsigned int pcpu;
+       /*
+        * 1 CORE SHARE
+        * 0 NOT SHARED
+        */
+       int status;
+};
+
+struct policy {
+       struct channel_packet pkt;
+       uint32_t pfid[MAX_VFS];
+       uint32_t port[MAX_VFS];
+       unsigned int enabled;
+       struct core_share core_share[MAX_VCPU_PER_VM];
+};
 
 #ifdef __cplusplus
 extern "C" {