src/vppinfra/atomics.h

/*
 * Copyright (c) 2018 Cisco and/or its affiliates.
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this
 * 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 <vcl/vcl_private.h>

static pthread_key_t vcl_worker_stop_key;

static const char *
vppcom_app_state_str (app_state_t state)
{
  char *st;

  switch (state)
    {
    case STATE_APP_START:
      st = "STATE_APP_START";
      break;

    case STATE_APP_CONN_VPP:
      st = "STATE_APP_CONN_VPP";
      break;

    case STATE_APP_ENABLED:
      st = "STATE_APP_ENABLED";
      break;

    case STATE_APP_ATTACHED:
      st = "STATE_APP_ATTACHED";
      break;

    default:
      st = "UNKNOWN_APP_STATE";
      break;
    }

  return st;
}

int
vcl_wait_for_app_state_change (app_state_t app_state)
{
  vcl_worker_t *wrk = vcl_worker_get_current ();
  f64 timeout = clib_time_now (&wrk->clib_time) + vcm->cfg.app_timeout;

  while (clib_time_now (&wrk->clib_time) < timeout)
    {
      if (vcm->app_state == app_state)
	return VPPCOM_OK;
      if (vcm->app_state == STATE_APP_FAILED)
	return VPPCOM_ECONNABORTED;
    }
  VDBG (0, "VCL<%d>: timeout waiting for state %s (%d)", getpid (),
	vppcom_app_state_str (app_state), app_state);
  vcl_evt (VCL_EVT_SESSION_TIMEOUT, vcm, app_state);

  return VPPCOM_ETIMEDOUT;
}

vcl_cut_through_registration_t *
vcl_ct_registration_lock_and_alloc (vcl_worker_t * wrk)
{
  vcl_cut_through_registration_t *cr;
  clib_spinlock_lock (&wrk->ct_registration_lock);
  pool_get (wrk->cut_through_registrations, cr);
  memset (cr, 0, sizeof (*cr));
  cr->epoll_evt_conn_index = -1;
  return cr;
}

u32
vcl_ct_registration_index (vcl_worker_t * wrk,
			   vcl_cut_through_registration_t * ctr)
{
  return (ctr - wrk->cut_through_registrations);
}

void
vcl_ct_registration_lock (vcl_worker_t * wrk)
{
  clib_spinlock_lock (&wrk->ct_registration_lock);
}

void
vcl_ct_registration_unlock (vcl_worker_t * wrk)
{
  clib_spinlock_unlock (&wrk->ct_registration_lock);
}

vcl_cut_through_registration_t *
vcl_ct_registration_get (vcl_worker_t * wrk, u32 ctr_index)
{
  if (pool_is_free_index (wrk->cut_through_registrations, ctr_index))
    return 0;
  return pool_elt_at_index (wrk->cut_through_registrations, ctr_index);
}

vcl_cut_through_registration_t *
vcl_ct_registration_lock_and_lookup (vcl_worker_t * wrk, uword mq_addr)
{
  uword *p;
  clib_spinlock_lock (&wrk->ct_registration_lock);
  p = hash_get (wrk->ct_registration_by_mq, mq_addr);
  if (!p)
    return 0;
  return vcl_ct_registration_get (wrk, p[0]);
}

void
vcl_ct_registration_lookup_add (vcl_worker_t * wrk, uword mq_addr,
				u32 ctr_index)
{
  hash_set (wrk->ct_registration_by_mq, mq_addr, ctr_index);
}

void
vcl_ct_registration_lookup_del (vcl_worker_t * wrk, uword mq_addr)
{
  hash_unset (wrk->ct_registration_by_mq, mq_addr);
}

void
vcl_ct_registration_del (vcl_worker_t * wrk,
			 vcl_cut_through_registration_t * ctr)
{
  pool_put (wrk->cut_through_registrations, ctr);
}

vcl_mq_evt_conn_t *
vcl_mq_evt_conn_alloc (vcl_worker_t * wrk)
{
  vcl_mq_evt_conn_t *mqc;
  pool_get (wrk->mq_evt_conns, mqc);
  memset (mqc, 0, sizeof (*mqc));
  return mqc;
}

u32
vcl_mq_evt_conn_index (vcl_worker_t * wrk, vcl_mq_evt_conn_t * mqc)
{
  return (mqc - wrk->mq_evt_conns);
}

vcl_mq_evt_conn_t *
vcl_mq_evt_conn_get (vcl_worker_t * wrk, u32 mq_conn_idx)
{
  return pool_elt_at_index (wrk->mq_evt_conns, mq_conn_idx);
}

int
vcl_mq_epoll_add_evfd (vcl_worker_t * wrk, svm_msg_q_t * mq)
{
  struct epoll_event e = { 0 };
  vcl_mq_evt_conn_t *mqc;
  u32 mqc_index;
  int mq_fd;

  mq_fd = svm_msg_q_get_consumer_eventfd (mq);

  if (wrk->mqs_epfd < 0 || mq_fd == -1)
    return -1;

  mqc = vcl_mq_evt_conn_alloc (wrk);
  mqc_index = vcl_mq_evt_conn_index (wrk, mqc);
  mqc->mq_fd = mq_fd;
  mqc->mq = mq;

  e.events = EPOLLIN;
  e.data.u32 = mqc_index;
  if (epoll_ctl (wrk->mqs_epfd, EPOLL_CTL_ADD, mq_fd, &e) < 0)
    {
      clib_warning ("failed to add mq eventfd to mq epoll fd");
      return -1;
    }

  return mqc_index;
}

int
vcl_mq_epoll_del_evfd (vcl_worker_t * wrk, u32 mqc_index)
{
  vcl_mq_evt_conn_t *mqc;

  if (wrk->mqs_epfd || mqc_index == ~0)
    return -1;

  mqc = vcl_mq_evt_conn_get (wrk, mqc_index);
  if (epoll_ctl (wrk->mqs_epfd, EPOLL_CTL_DEL, mqc->mq_fd, 0) < 0)
    {
      clib_warning ("failed to del mq eventfd to mq epoll fd");
      return -1;
    }
  return 0;
}

static vcl_worker_t *
vcl_worker_alloc (void)
{
  vcl_worker_t *wrk;
  pool_get (vcm->workers, wrk);
  memset (wrk, 0, sizeof (*wrk));
  wrk->wrk_index = wrk - vcm->workers;
  return wrk;
}

static void
vcl_worker_free (vcl_worker_t * wrk)
{
  pool_put (vcm->workers, wrk);
}

void
vcl_worker_cleanup (void)
{
  vcl_worker_t *wrk = vcl_worker_get_current ();

  clib_spinlock_lock (&vcm->workers_lock);
  vcl_send_app_worker_add_del (0 /* is_add */ );
  close (wrk->mqs_epfd);
  hash_free (wrk->session_index_by_vpp_handles);
  hash_free (wrk->ct_registration_by_mq);
  clib_spinlock_free (&wrk->ct_registration_lock);
  vec_free (wrk->mq_events);
  vec_free (wrk->mq_msg_vector);
  vcl_set_worker_index (~0);
  vcl_worker_free (wrk);
  clib_spinlock_unlock (&vcm->workers_lock);
  VDBG (0, "cleaned up worker %u", wrk->wrk_index);
}

static void
vcl_worker_cleanup_cb (void *arg)
{
  vcl_worker_cleanup ();
}

vcl_worker_t *
vcl_worker_alloc_and_init ()
{
  vcl_worker_t *wrk;

  /* This was initialized already */
  if (vcl_get_worker_index () != ~0)
    return 0;

  if (pool_elts (vcm->workers) == vcm->cfg.max_workers)
    {
      VDBG (0, "max-workers %u limit reached", vcm->cfg.max_workers);
      return 0;
    }

  clib_spinlock_lock (&vcm->workers_lock);
  wrk = vcl_worker_alloc ();
  vcl_set_worker_index (wrk->wrk_index);
  wrk->thread_id = pthread_self ();
  wrk->current_pid = getpid ();

  wrk->mqs_epfd = -1;
  if (vcm->cfg.use_mq_eventfd)
    {
      wrk->mqs_epfd = epoll_create (1);
      if (wrk->mqs_epfd < 0)
	{
	  clib_unix_warning ("epoll_create() returned");
	  goto done;
	}
    }

  wrk->session_index_by_vpp_handles = hash_create (0, sizeof (uword));
  wrk->ct_registration_by_mq = hash_create (0, sizeof (uword));
  clib_spinlock_init (&wrk->ct_registration_lock);
  clib_time_init (&wrk->clib_time);
  vec_validate (wrk->mq_events, 64);
  vec_validate (wrk->mq_msg_vector, 128);
  vec_reset_length (wrk->mq_msg_vector);
  vec_validate (wrk->unhandled_evts_vector, 128);
  vec_reset_length (wrk->unhandled_evts_vector);
  clib_spinlock_unlock (&vcm->workers_lock);

done:
  return wrk;
}

int
vcl_worker_register_with_vpp (void)
{
  vcl_worker_t *wrk = vcl_worker_get_current ();

  clib_spinlock_lock (&vcm->workers_lock);

  vcm->app_state = STATE_APP_ADDING_WORKER;
  vcl_send_app_worker_add_del (1 /* is_add */ );
  if (vcl_wait_for_app_state_change (STATE_APP_READY))
    {
      clib_warning ("failed to add worker to vpp");
      return -1;
    }

  if (pthread_key_create (&vcl_worker_stop_key, vcl_worker_cleanup_cb))
    clib_warning ("failed to add pthread cleanup function");
  if (pthread_setspecific (vcl_worker_stop_key, &wrk->thread_id))
    clib_warning ("failed to setup key value");

  clib_spinlock_unlock (&vcm->workers_lock);

  VDBG (0, "added worker %u", wrk->wrk_index);
  return 0;
}

int
vcl_worker_set_bapi (void)
{
  vcl_worker_t *wrk = vcl_worker_get_current ();
  int i;

  /* Find the first worker with the same pid */
  for (i = 0; i < vec_len (vcm->workers); i++)
    {
      if (i == wrk->wrk_index)
	continue;
      if (vcm->workers[i].current_pid == wrk->current_pid)
	{
	  wrk->vl_input_queue = vcm->workers[i].vl_input_queue;
	  wrk->my_client_index = vcm->workers[i].my_client_index;
	  return 0;
	}
    }
  return -1;
}

vcl_shared_session_t *
vcl_shared_session_alloc (void)
{
  vcl_shared_session_t *ss;
  pool_get (vcm->shared_sessions, ss);
  memset (ss, 0, sizeof (*ss));
  ss->ss_index = ss - vcm->shared_sessions;
  return ss;
}

vcl_shared_session_t *
vcl_shared_session_get (u32 ss_index)
{
  if (pool_is_free_index (vcm->shared_sessions, ss_index))
    return 0;
  return pool_elt_at_index (vcm->shared_sessions, ss_index);
}

void
vcl_shared_session_free (vcl_shared_session_t * ss)
{
  pool_put (vcm->shared_sessions, ss);
}

void
vcl_worker_share_session (vcl_worker_t * parent, vcl_worker_t * wrk,
			  vcl_session_t * new_s)
{
  vcl_shared_session_t *ss;
  vcl_session_t *s;

  s = vcl_session_get (parent, new_s->session_index);
  if (s->shared_index == ~0)
    {
      ss = vcl_shared_session_alloc ();
      vec_add1 (ss->workers, parent->wrk_index);
      s->shared_index = ss->ss_index;
    }
  else
    {
      ss = vcl_shared_session_get (s->shared_index);
    }
  new_s->shared_index = ss->ss_index;
  vec_add1 (ss->workers, wrk->wrk_index);
}

int
vcl_worker_unshare_session (vcl_worker_t * wrk, vcl_session_t * s)
{
  vcl_shared_session_t *ss;
  int i;

  ss = vcl_shared_session_get (s->shared_index);
  for (i = 0; i < vec_len (ss->workers); i++)
    {
      if (ss->workers[i] == wrk->wrk_index)
	{
	  vec_del1 (ss->workers, i);
	  break;
	}
    }

  if (vec_len (ss->workers) == 0)
    {
      vcl_shared_session_free (ss);
      return 1;
    }

  return 0;
}

void
vcl_worker_share_sessions (u32 parent_wrk_index)
{
  vcl_worker_t *parent_wrk, *wrk;
  vcl_session_t *new_s;

  parent_wrk = vcl_worker_get (parent_wrk_index);
  if (!parent_wrk->sessions)
    return;

  wrk = vcl_worker_get_current ();
  wrk->sessions = pool_dup (parent_wrk->sessions);
  wrk->session_index_by_vpp_handles =
    hash_dup (parent_wrk->session_index_by_vpp_handles);

  /* *INDENT-OFF* */
  pool_foreach (new_s, wrk->sessions, ({
    vcl_worker_share_session (parent_wrk, wrk, new_s);
  }));
  /* *INDENT-ON* */
}

int
vcl_session_get_refcnt (vcl_session_t * s)
{
  vcl_shared_session_t *ss;
  ss = vcl_shared_session_get (s->shared_index);
  if (ss)
    return vec_len (ss->workers);
  return 0;
}

/*
 * fd.io coding-style-patch-verification: ON
 *
 * Local Variables:
 * eval: (c-set-style "gnu")
 * End:
 */
Age	Commit message (Expand)	Author	Files	Lines
2021-10-12	ipsec: Performance improvement of ipsec4_output_node using flow cache	Govindarajan Mohandoss	1	-0/+2
2021-02-08	virtio: add atomic call for kicking	Mohsin Kazmi	1	-0/+3
2020-12-23	svm: remove fifo segment heap	Florin Coras	1	-0/+3
2020-11-19	svm: move chunk locks to linked list	Florin Coras	1	-0/+3
2019-08-01	vppinfra: refactor clib_spinlock_t to use compare and swap	jaszha03	1	-0/+3
2019-06-05	Switch atomic release API from __sync to __atomic builtin.	Sirshak Das	1	-1/+1
2019-06-05	Switch atomic test and set API from __sync to __atomic builtin	Sirshak Das	1	-1/+1
2019-04-16	svm_fifo rework to avoid contention on cursize	Sirshak Das	1	-0/+3
2019-03-22	svm/atomics: add clib_atomic_swap_rel_n	Florin Coras	1	-0/+1
2018-11-28	Use acquire/release ordering when accessing svm_fifo shared variable cursize	Sirshak Das	1	-0/+4
2018-11-05	Enable atomic swap and store macro with acquire and release ordering	Sirshak Das	1	-2/+3
2018-10-19	vppinfra: add atomic macros for __sync builtins	Sirshak Das	1	-0/+45