perfmon_periodic.c

Go to the documentation of this file.

/*
 * perfmon_periodic.c - skeleton plug-in periodic function
 *
 * Copyright (c) <current-year> <your-organization>
 * 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 <vlib/vlib.h>
#include <vppinfra/error.h>
#include <perfmon/perfmon.h>
#include <asm/unistd.h>
#include <sys/ioctl.h>

/* "not in glibc" */
static long
perf_event_open (struct perf_event_attr *hw_event, pid_t pid, int cpu,
                 int group_fd, unsigned long flags)
{
  int ret;

  ret = syscall (__NR_perf_event_open, hw_event, pid, cpu, group_fd, flags);
  return ret;
}

static void
read_current_perf_counters (vlib_main_t * vm, u64 * c0, u64 * c1)
{
  int i;
  u64 *cc;
  perfmon_main_t *pm = &perfmon_main;
  uword my_thread_index = vm->thread_index;

  *c0 = *c1 = 0;

  for (i = 0; i < pm->n_active; i++)
    {
      cc = (i == 0) ? c0 : c1;
      if (pm->rdpmc_indices[i][my_thread_index] != ~0)
        *cc = clib_rdpmc ((int) pm->rdpmc_indices[i][my_thread_index]);
      else
        {
          u64 sw_value;
          if (read (pm->pm_fds[i][my_thread_index], &sw_value,
                    sizeof (sw_value)) != sizeof (sw_value))
            {
              clib_unix_warning
                ("counter read failed, disable collection...");
              vm->vlib_node_runtime_perf_counter_cb = 0;
              return;
            }
          *cc = sw_value;
        }
    }
}

static void
clear_counters (perfmon_main_t * pm)
{
  int i, j;
  vlib_main_t *vm = pm->vlib_main;
  vlib_main_t *stat_vm;
  vlib_node_main_t *nm;
  vlib_node_t *n;

  vlib_worker_thread_barrier_sync (vm);

  for (j = 0; j < vec_len (vlib_mains); j++)
    {
      stat_vm = vlib_mains[j];
      if (stat_vm == 0)
        continue;

      nm = &stat_vm->node_main;

      /* Clear the node runtime perfmon counters */
      for (i = 0; i < vec_len (nm->nodes); i++)
        {
          n = nm->nodes[i];
          vlib_node_sync_stats (stat_vm, n);
        }

      /* And clear the node perfmon counters */
      for (i = 0; i < vec_len (nm->nodes); i++)
        {
          n = nm->nodes[i];
          n->stats_total.perf_counter0_ticks = 0;
          n->stats_total.perf_counter1_ticks = 0;
          n->stats_total.perf_counter_vectors = 0;
          n->stats_last_clear.perf_counter0_ticks = 0;
          n->stats_last_clear.perf_counter1_ticks = 0;
          n->stats_last_clear.perf_counter_vectors = 0;
        }
    }
  vlib_worker_thread_barrier_release (vm);
}

static void
enable_current_events (perfmon_main_t * pm)
{
  struct perf_event_attr pe;
  int fd;
  struct perf_event_mmap_page *p = 0;
  perfmon_event_config_t *c;
  vlib_main_t *vm = vlib_get_main ();
  u32 my_thread_index = vm->thread_index;
  u32 index;
  int i, limit = 1;
  int cpu;

  if ((pm->current_event + 1) < vec_len (pm->single_events_to_collect))
    limit = 2;

  for (i = 0; i < limit; i++)
    {
      c = vec_elt_at_index (pm->single_events_to_collect,
                            pm->current_event + i);

      memset (&pe, 0, sizeof (struct perf_event_attr));
      pe.type = c->pe_type;
      pe.size = sizeof (struct perf_event_attr);
      pe.config = c->pe_config;
      pe.disabled = 1;
      pe.pinned = 1;
      /*
       * Note: excluding the kernel makes the
       * (software) context-switch counter read 0...
       */
      if (pe.type != PERF_TYPE_SOFTWARE)
        {
          /* Exclude kernel and hypervisor */
          pe.exclude_kernel = 1;
          pe.exclude_hv = 1;
        }

      cpu = vm->cpu_id;

      fd = perf_event_open (&pe, 0, cpu, -1, 0);
      if (fd == -1)
        {
          clib_unix_warning ("event open: type %d config %d", c->pe_type,
                             c->pe_config);
          return;
        }

      if (pe.type != PERF_TYPE_SOFTWARE)
        {
          p = mmap (0, pm->page_size, PROT_READ, MAP_SHARED, fd, 0);
          if (p == MAP_FAILED)
            {
              clib_unix_warning ("mmap");
              close (fd);
              return;
            }
        }
      else
        p = 0;

      if (ioctl (fd, PERF_EVENT_IOC_RESET, 0) < 0)
        clib_unix_warning ("reset ioctl");

      if (ioctl (fd, PERF_EVENT_IOC_ENABLE, 0) < 0)
        clib_unix_warning ("enable ioctl");

      pm->perf_event_pages[i][my_thread_index] = (void *) p;
      pm->pm_fds[i][my_thread_index] = fd;
    }

  /*
   * Hardware events must be all opened and enabled before aquiring
   * pmc indices, otherwise the pmc indices might be out-dated.
   */
  for (i = 0; i < limit; i++)
    {
      p =
        (struct perf_event_mmap_page *)
        pm->perf_event_pages[i][my_thread_index];

      /*
       * Software event counters - and others not capable of being
       * read via the "rdpmc" instruction - will be read
       * by system calls.
       */
      if (p == 0 || p->cap_user_rdpmc == 0)
        index = ~0;
      else
        index = p->index - 1;

      pm->rdpmc_indices[i][my_thread_index] = index;
    }

  pm->n_active = i;
  /* Enable the main loop counter snapshot mechanism */
  vm->vlib_node_runtime_perf_counter_cb = read_current_perf_counters;
}

static void
disable_events (perfmon_main_t * pm)
{
  vlib_main_t *vm = vlib_get_main ();
  u32 my_thread_index = vm->thread_index;
  int i;

  /* Stop main loop collection */
  vm->vlib_node_runtime_perf_counter_cb = 0;

  for (i = 0; i < pm->n_active; i++)
    {
      if (pm->pm_fds[i][my_thread_index] == 0)
        continue;

      if (ioctl (pm->pm_fds[i][my_thread_index], PERF_EVENT_IOC_DISABLE, 0) <
          0)
        clib_unix_warning ("disable ioctl");

      if (pm->perf_event_pages[i][my_thread_index])
        if (munmap (pm->perf_event_pages[i][my_thread_index],
                    pm->page_size) < 0)
          clib_unix_warning ("munmap");

      (void) close (pm->pm_fds[i][my_thread_index]);
      pm->pm_fds[i][my_thread_index] = 0;
    }
}

static void
worker_thread_start_event (vlib_main_t * vm)
{
  perfmon_main_t *pm = &perfmon_main;

  enable_current_events (pm);
  vm->worker_thread_main_loop_callback = 0;
}

static void
worker_thread_stop_event (vlib_main_t * vm)
{
  perfmon_main_t *pm = &perfmon_main;
  disable_events (pm);
  vm->worker_thread_main_loop_callback = 0;
}

static void
start_event (perfmon_main_t * pm, f64 now, uword event_data)
{
  int i;
  int last_set;
  int all = 0;
  pm->current_event = 0;

  if (vec_len (pm->single_events_to_collect) == 0)
    {
      pm->state = PERFMON_STATE_OFF;
      return;
    }

  last_set = clib_bitmap_last_set (pm->thread_bitmap);
  all = (last_set == ~0);

  pm->state = PERFMON_STATE_RUNNING;
  clear_counters (pm);

  /* Start collection on thread 0? */
  if (all || clib_bitmap_get (pm->thread_bitmap, 0))
    {
      /* Start collection on this thread */
      enable_current_events (pm);
    }

  /* And also on worker threads */
  for (i = 1; i < vec_len (vlib_mains); i++)
    {
      if (vlib_mains[i] == 0)
        continue;

      if (all || clib_bitmap_get (pm->thread_bitmap, i))
        vlib_mains[i]->worker_thread_main_loop_callback = (void *)
          worker_thread_start_event;
    }
}

void
scrape_and_clear_counters (perfmon_main_t * pm)
{
  int i, j, k;
  vlib_main_t *vm = pm->vlib_main;
  vlib_main_t *stat_vm;
  vlib_node_main_t *nm;
  vlib_node_t ***node_dups = 0;
  vlib_node_t **nodes;
  vlib_node_t *n;
  perfmon_capture_t *c;
  perfmon_event_config_t *current_event;
  uword *p;
  u8 *counter_name;
  u64 vectors_this_counter;

  /* snapshoot the nodes, including pm counters */
  vlib_worker_thread_barrier_sync (vm);

  for (j = 0; j < vec_len (vlib_mains); j++)
    {
      stat_vm = vlib_mains[j];
      if (stat_vm == 0)
        continue;

      nm = &stat_vm->node_main;

      for (i = 0; i < vec_len (nm->nodes); i++)
        {
          n = nm->nodes[i];
          vlib_node_sync_stats (stat_vm, n);
        }

      nodes = 0;
      vec_validate (nodes, vec_len (nm->nodes) - 1);
      vec_add1 (node_dups, nodes);

      /* Snapshoot and clear the per-node perfmon counters */
      for (i = 0; i < vec_len (nm->nodes); i++)
        {
          n = nm->nodes[i];
          nodes[i] = clib_mem_alloc (sizeof (*n));
          clib_memcpy_fast (nodes[i], n, sizeof (*n));
          n->stats_total.perf_counter0_ticks = 0;
          n->stats_total.perf_counter1_ticks = 0;
          n->stats_total.perf_counter_vectors = 0;
          n->stats_last_clear.perf_counter0_ticks = 0;
          n->stats_last_clear.perf_counter1_ticks = 0;
          n->stats_last_clear.perf_counter_vectors = 0;
        }
    }

  vlib_worker_thread_barrier_release (vm);

  for (j = 0; j < vec_len (vlib_mains); j++)
    {
      stat_vm = vlib_mains[j];
      if (stat_vm == 0)
        continue;

      nodes = node_dups[j];

      for (i = 0; i < vec_len (nodes); i++)
        {
          u8 *capture_name;

          n = nodes[i];

          if (n->stats_total.perf_counter0_ticks == 0 &&
              n->stats_total.perf_counter1_ticks == 0)
            goto skip_this_node;

          for (k = 0; k < 2; k++)
            {
              u64 counter_value, counter_last_clear;

              /*
               * We collect 2 counters at once, except for the
               * last counter when the user asks for an odd number of
               * counters
               */
              if ((pm->current_event + k)
                  >= vec_len (pm->single_events_to_collect))
                break;

              if (k == 0)
                {
                  counter_value = n->stats_total.perf_counter0_ticks;
                  counter_last_clear =
                    n->stats_last_clear.perf_counter0_ticks;
                }
              else
                {
                  counter_value = n->stats_total.perf_counter1_ticks;
                  counter_last_clear =
                    n->stats_last_clear.perf_counter1_ticks;
                }

              capture_name = format (0, "t%d-%v%c", j, n->name, 0);

              p = hash_get_mem (pm->capture_by_thread_and_node_name,
                                capture_name);

              if (p == 0)
                {
                  pool_get (pm->capture_pool, c);
                  memset (c, 0, sizeof (*c));
                  c->thread_and_node_name = capture_name;
                  hash_set_mem (pm->capture_by_thread_and_node_name,
                                capture_name, c - pm->capture_pool);
                }
              else
                {
                  c = pool_elt_at_index (pm->capture_pool, p[0]);
                  vec_free (capture_name);
                }

              /* Snapshoot counters, etc. into the capture */
              current_event = pm->single_events_to_collect
                + pm->current_event + k;
              counter_name = (u8 *) current_event->name;
              vectors_this_counter = n->stats_total.perf_counter_vectors -
                n->stats_last_clear.perf_counter_vectors;

              vec_add1 (c->counter_names, counter_name);
              vec_add1 (c->counter_values,
                        counter_value - counter_last_clear);
              vec_add1 (c->vectors_this_counter, vectors_this_counter);
            }
        skip_this_node:
          clib_mem_free (n);
        }
      vec_free (nodes);
    }
  vec_free (node_dups);
}

static void
handle_timeout (vlib_main_t * vm, perfmon_main_t * pm, f64 now)
{
  int i;
  int last_set, all;

  last_set = clib_bitmap_last_set (pm->thread_bitmap);
  all = (last_set == ~0);

  if (all || clib_bitmap_get (pm->thread_bitmap, 0))
    disable_events (pm);

  /* And also on worker threads */
  for (i = 1; i < vec_len (vlib_mains); i++)
    {
      if (vlib_mains[i] == 0)
        continue;
      if (all || clib_bitmap_get (pm->thread_bitmap, i))
        vlib_mains[i]->worker_thread_main_loop_callback = (void *)
          worker_thread_stop_event;
    }

  /* Make sure workers have stopped collection */
  if (i > 1)
    {
      f64 deadman = vlib_time_now (vm) + 1.0;

      for (i = 1; i < vec_len (vlib_mains); i++)
        {
          /* Has the worker actually stopped collecting data? */
          while (vlib_mains[i]->worker_thread_main_loop_callback)
            {
              if (vlib_time_now (vm) > deadman)
                {
                  clib_warning ("Thread %d deadman timeout!", i);
                  break;
                }
              vlib_process_suspend (pm->vlib_main, 1e-3);
            }
        }
    }
  scrape_and_clear_counters (pm);
  pm->current_event += pm->n_active;
  if (pm->current_event >= vec_len (pm->single_events_to_collect))
    {
      pm->current_event = 0;
      pm->state = PERFMON_STATE_OFF;
      return;
    }

  if (all || clib_bitmap_get (pm->thread_bitmap, 0))
    enable_current_events (pm);

  /* And also on worker threads */
  for (i = 1; i < vec_len (vlib_mains); i++)
    {
      if (vlib_mains[i] == 0)
        continue;
      if (all || clib_bitmap_get (pm->thread_bitmap, i))
        vlib_mains[i]->worker_thread_main_loop_callback = (void *)
          worker_thread_start_event;
    }
}

static uword
perfmon_periodic_process (vlib_main_t * vm,
                          vlib_node_runtime_t * rt, vlib_frame_t * f)
{
  perfmon_main_t *pm = &perfmon_main;
  f64 now;
  uword *event_data = 0;
  uword event_type;
  int i;

  while (1)
    {
      if (pm->state == PERFMON_STATE_RUNNING)
        vlib_process_wait_for_event_or_clock (vm, pm->timeout_interval);
      else
        vlib_process_wait_for_event (vm);

      now = vlib_time_now (vm);

      event_type = vlib_process_get_events (vm, (uword **) & event_data);

      switch (event_type)
        {
        case PERFMON_START:
          for (i = 0; i < vec_len (event_data); i++)
            start_event (pm, now, event_data[i]);
          break;

          /* Handle timeout */
        case ~0:
          handle_timeout (vm, pm, now);
          break;

        default:
          clib_warning ("Unexpected event %d", event_type);
          break;
        }
      vec_reset_length (event_data);
    }
  return 0;                     /* or not */
}

/* *INDENT-OFF* */
VLIB_REGISTER_NODE (perfmon_periodic_node) =
{
  .function = perfmon_periodic_process,
  .type = VLIB_NODE_TYPE_PROCESS,
  .name = "perfmon-periodic-process",
};
/* *INDENT-ON* */

/*
 * fd.io coding-style-patch-verification: ON
 *
 * Local Variables:
 * eval: (c-set-style "gnu")
 * End:
 */