/* * Copyright (c) 2015 Cisco and/or its affiliates. * 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. */ /* * main.c: main vector processing loop * * Copyright (c) 2008 Eliot Dresselhaus * * Permission is hereby granted, free of charge, to any person obtaining * a copy of this software and associated documentation files (the * "Software"), to deal in the Software without restriction, including * without limitation the rights to use, copy, modify, merge, publish, * distribute, sublicense, and/or sell copies of the Software, and to * permit persons to whom the Software is furnished to do so, subject to * the following conditions: * * The above copyright notice and this permission notice shall be * included in all copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ #include #include #include #include #include #include #include CJ_GLOBAL_LOG_PROTOTYPE; /* Actually allocate a few extra slots of vector data to support speculative vector enqueues which overflow vector data in next frame. */ #define VLIB_FRAME_SIZE_ALLOC (VLIB_FRAME_SIZE + 4) u32 wraps; always_inline u32 vlib_frame_bytes (u32 n_scalar_bytes, u32 n_vector_bytes) { u32 n_bytes; /* Make room for vlib_frame_t plus scalar arguments. */ n_bytes = vlib_frame_vector_byte_offset (n_scalar_bytes); /* Make room for vector arguments. Allocate a few extra slots of vector data to support speculative vector enqueues which overflow vector data in next frame. */ #define VLIB_FRAME_SIZE_EXTRA 4 n_bytes += (VLIB_FRAME_SIZE + VLIB_FRAME_SIZE_EXTRA) * n_vector_bytes; /* Magic number is first 32bit number after vector data. Used to make sure that vector data is never overrun. */ #define VLIB_FRAME_MAGIC (0xabadc0ed) n_bytes += sizeof (u32); /* Pad to cache line. */ n_bytes = round_pow2 (n_bytes, CLIB_CACHE_LINE_BYTES); return n_bytes; } always_inline u32 * vlib_frame_find_magic (vlib_frame_t * f, vlib_node_t * node) { void *p = f; p += vlib_frame_vector_byte_offset (node->scalar_size); p += (VLIB_FRAME_SIZE + VLIB_FRAME_SIZE_EXTRA) * node->vector_size; return p; } static vlib_frame_size_t * get_frame_size_info (vlib_node_main_t * nm, u32 n_scalar_bytes, u32 n_vector_bytes) { uword key = (n_scalar_bytes << 16) | n_vector_bytes; uword *p, i; p = hash_get (nm->frame_size_hash, key); if (p) i = p[0]; else { i = vec_len (nm->frame_sizes); vec_validate (nm->frame_sizes, i); hash_set (nm->frame_size_hash, key, i); } return vec_elt_at_index (nm->frame_sizes, i); } static u32 vlib_frame_alloc_to_node (vlib_main_t * vm, u32 to_node_index, u32 frame_flags) { vlib_node_main_t *nm = &vm->node_main; vlib_frame_size_t *fs; vlib_node_t *to_node; vlib_frame_t *f; u32 fi, l, n, scalar_size, vector_size; to_node = vlib_get_node (vm, to_node_index); scalar_size = to_node->scalar_size; vector_size = to_node->vector_size; fs = get_frame_size_info (nm, scalar_size, vector_size); n = vlib_frame_bytes (scalar_size, vector_size); if ((l = vec_len (fs->free_frame_indices)) > 0) { /* Allocate from end of free list. */ fi = fs->free_frame_indices[l - 1]; f = vlib_get_frame_no_check (vm, fi); _vec_len (fs->free_frame_indices) = l - 1; } else { f = clib_mem_alloc_aligned_no_fail (n, VLIB_FRAME_ALIGN); fi = vlib_frame_index_no_check (vm, f); } /* Poison frame when debugging. */ if (CLIB_DEBUG > 0) memset (f, 0xfe, n); /* Insert magic number. */ { u32 *magic; magic = vlib_frame_find_magic (f, to_node); *magic = VLIB_FRAME_MAGIC; } f->flags = VLIB_FRAME_IS_ALLOCATED | frame_flags; f->n_vectors = 0; f->scalar_size = scalar_size; f->vector_size = vector_size; fs->n_alloc_frames += 1; return fi; } /* Allocate a frame for from FROM_NODE to TO_NODE via TO_NEXT_INDEX. Returns frame index. */ static u32 vlib_frame_alloc (vlib_main_t * vm, vlib_node_runtime_t * from_node_runtime, u32 to_next_index) { vlib_node_t *from_node; from_node = vlib_get_node (vm, from_node_runtime->node_index); ASSERT (to_next_index < vec_len (from_node->next_nodes)); return vlib_frame_alloc_to_node (vm, from_node->next_nodes[to_next_index], /* frame_flags */ 0); } vlib_frame_t * vlib_get_frame_to_node (vlib_main_t * vm, u32 to_node_index) { u32 fi = vlib_frame_alloc_to_node (vm, to_node_index, /* frame_flags */ VLIB_FRAME_FREE_AFTER_DISPATCH); return vlib_get_frame (vm, fi); } void vlib_put_frame_to_node (vlib_main_t * vm, u32 to_node_index, vlib_frame_t * f) { vlib_pending_frame_t *p; vlib_node_t *to_node; if (f->n_vectors == 0) return; to_node = vlib_get_node (vm, to_node_index); vec_add2 (vm->node_main.pending_frames, p, 1); f->flags |= VLIB_FRAME_PENDING; p->frame_index = vlib_frame_index (vm, f); p->node_runtime_index = to_node->runtime_index; p->next_frame_index = VLIB_PENDING_FRAME_NO_NEXT_FRAME; } /* Free given frame. */ void vlib_frame_free (vlib_main_t * vm, vlib_node_runtime_t * r, vlib_frame_t * f) { vlib_node_main_t *nm = &vm->node_main; vlib_node_t *node; vlib_frame_size_t *fs; u32 frame_index; ASSERT (f->flags & VLIB_FRAME_IS_ALLOCATED); node = vlib_get_node (vm, r->node_index); fs = get_frame_size_info (nm, node->scalar_size, node->vector_size); frame_index = vlib_frame_index (vm, f); ASSERT (f->flags & VLIB_FRAME_IS_ALLOCATED); /* No next frames may point to freed frame. */ if (CLIB_DEBUG > 0) { vlib_next_frame_t *nf; vec_foreach (nf, vm->node_main.next_frames) ASSERT (nf->frame_index != frame_index); } f->flags &= ~VLIB_FRAME_IS_ALLOCATED; vec_add1 (fs->free_frame_indices, frame_index); ASSERT (fs->n_alloc_frames > 0); fs->n_alloc_frames -= 1; } static clib_error_t * show_frame_stats (vlib_main_t * vm, unformat_input_t * input, vlib_cli_command_t * cmd) { vlib_node_main_t *nm = &vm->node_main; vlib_frame_size_t *fs; vlib_cli_output (vm, "%=6s%=12s%=12s", "Size", "# Alloc", "# Free"); vec_foreach (fs, nm->frame_sizes) { u32 n_alloc = fs->n_alloc_frames; u32 n_free = vec_len (fs->free_frame_indices); if (n_alloc + n_free > 0) vlib_cli_output (vm, "%=6d%=12d%=12d", fs - nm->frame_sizes, n_alloc, n_free); } return 0; } /* *INDENT-OFF* */ VLIB_CLI_COMMAND (show_frame_stats_cli, static) = { .path = "show vlib frame-allocation", .short_help = "Show node dispatch frame statistics", .function = show_frame_stats, }; /* *INDENT-ON* */ /* Change ownership of enqueue rights to given next node. */ static void vlib_next_frame_change_ownership (vlib_main_t * vm, vlib_node_runtime_t * node_runtime, u32 next_index) { vlib_node_main_t *nm = &vm->node_main; vlib_next_frame_t *next_frame; vlib_node_t *node, *next_node; node = vec_elt (nm->nodes, node_runtime->node_index); /* Only internal & input nodes are allowed to call other nodes. */ ASSERT (node->type == VLIB_NODE_TYPE_INTERNAL || node->type == VLIB_NODE_TYPE_INPUT || node->type == VLIB_NODE_TYPE_PROCESS); ASSERT (vec_len (node->next_nodes) == node_runtime->n_next_nodes); next_frame = vlib_node_runtime_get_next_frame (vm, node_runtime, next_index); next_node = vec_elt (nm->nodes, node->next_nodes[next_index]); if (next_node->owner_node_index != VLIB_INVALID_NODE_INDEX) { /* Get frame from previous owner. */ vlib_next_frame_t *owner_next_frame; vlib_next_frame_t tmp; owner_next_frame = vlib_node_get_next_frame (vm, next_node->owner_node_index, next_node->owner_next_index); /* Swap target next frame with owner's. */ tmp = owner_next_frame[0]; owner_next_frame[0] = next_frame[0]; next_frame[0] = tmp; /* * If next_frame is already pending, we have to track down * all pending frames and fix their next_frame_index fields. */ if (next_frame->flags & VLIB_FRAME_PENDING) { vlib_pending_frame_t *p; if (next_frame->frame_index != ~0) { vec_foreach (p, nm-
/*
 * Copyright (c) 2017 Cisco and/or its affiliates.
 * 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 <vppinfra/maplog.h>

clib_maplog_main_t maplog_main;

typedef struct
{
  u64 serial_number;
  u64 junk[7];
} test_entry_t;

typedef enum
{
  TEST_NORMAL,
  TEST_CIRCULAR,
} test_type_t;

static void
process_maplog_records (clib_maplog_header_t * h,
			test_entry_t * e, u64 records_this_file)
{
  static int print_header;
  int i = 0;

  if (print_header == 0)
    {
      print_header = 1;
      fformat (stdout, "%U", format_maplog_header, h, 1 /* verbose */ );
    }

  while (records_this_file--)
    {
      /* Padding at the end of a damaged log? */
      if (e->serial_number == 0ULL)
	break;
      fformat (stdout, "%4lld ", e->serial_number);
      if (++i == 8)
	{
	  fformat (stdout, "\n");
	  i = 0;
	}
      e++;
    }
  fformat (stdout, "\n--------------\n");
}

int
test_maplog_main (unformat_input_t * input)
{
  clib_maplog_main_t *mm = &maplog_main;
  clib_maplog_init_args_t _a, *a = &_a;
  int rv;
  int i, limit;
  test_entry_t *t;
  int noclose = 0;
  test_type_t which = TEST_NORMAL;

  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
    {
      if (unformat (input, "noclose"))
	noclose = 1;
      else if (unformat (input, "circular"))
	which = TEST_CIRCULAR;
      else
	clib_warning ("unknown input '%U'", format_unformat_error, input);
    }

  clib_memset (a, 0, sizeof (*a));
  a->mm = mm;
  a->file_basename = "/tmp/maplog_test";
  a->file_size_in_bytes = 4096;
  a->record_size_in_bytes = sizeof (test_entry_t);
  a->application_id = 1;
  a->application_major_version = 1;
  a->application_minor_version = 0;
  a->application_patch_version = 0;
  a->maplog_is_circular = (which == TEST_CIRCULAR) ? 1 : 0;

  rv = clib_maplog_init (a);

  if (rv)
    {
      clib_warning ("clib_maplog_init returned %d", rv);
      exit (1);
    }

  limit = (which == TEST_CIRCULAR) ? (64 + 2) : 64 * 5;

  for (i = 0; i < limit; i++)
    {
      t = clib_maplog_get_entry (mm);
      t->serial_number = i + 1;
    }

  if (noclose)
    clib_memset (mm, 0, sizeof (*mm));
  else
    clib_maplog_close (mm);

  clib_maplog_process ("/tmp/maplog_test", process_maplog_records);

  return 0;
}

#ifdef CLIB_UNIX
int
main (int argc, char *argv[])
{
  unformat_input_t i;
  int ret;

  clib_mem_init (0, 64ULL << 20);

  unformat_init_command_line (&i, argv);
  ret = test_maplog_main (&i);
  unformat_free (&i);

  return ret;
}
#endif /* CLIB_UNIX */


/*
 * fd.io coding-style-patch-verification: ON
 *
 * Local Variables:
 * eval: (c-set-style "gnu")
 * End:
 */
ing = 1; #endif } else if (dispatch_state == VLIB_NODE_STATE_POLLING && v <= nm->interrupt_threshold_vector_length) { vlib_node_t *n = vlib_get_node (vm, node->node_index); if (node->flags & VLIB_NODE_FLAG_SWITCH_FROM_POLLING_TO_INTERRUPT_MODE) { /* Switch to interrupt mode after dispatch in polling one more time. This allows driver to re-enable interrupts. */ n->state = VLIB_NODE_STATE_INTERRUPT; node->state = VLIB_NODE_STATE_INTERRUPT; node->flags &= ~VLIB_NODE_FLAG_SWITCH_FROM_INTERRUPT_TO_POLLING_MODE; nm->input_node_counts_by_state[VLIB_NODE_STATE_POLLING] -= 1; nm->input_node_counts_by_state[VLIB_NODE_STATE_INTERRUPT] += 1; } else { node->flags |= VLIB_NODE_FLAG_SWITCH_FROM_POLLING_TO_INTERRUPT_MODE; #ifdef DISPATCH_NODE_ELOG_REQUIRED ed = ELOG_TRACK_DATA (&vlib_global_main.elog_main, e, w->elog_track); ed->node_name = n->name_elog_string; ed->vector_length = v; ed->is_polling = 0; #endif } } } } return t; } static u64 dispatch_pending_node (vlib_main_t * vm, uword pending_frame_index, u64 last_time_stamp) { vlib_node_main_t *nm = &vm->node_main; vlib_frame_t *f; vlib_next_frame_t *nf, nf_dummy; vlib_node_runtime_t *n; u32 restore_frame_index; vlib_pending_frame_t *p; /* See comment below about dangling references to nm->pending_frames */ p = nm->pending_frames + pending_frame_index; n = vec_elt_at_index (nm->nodes_by_type[VLIB_NODE_TYPE_INTERNAL], p->node_runtime_index); f = vlib_get_frame (vm, p->frame_index); if (p->next_frame_index == VLIB_PENDING_FRAME_NO_NEXT_FRAME) { /* No next frame: so use dummy on stack. */ nf = &nf_dummy; nf->flags = f->flags & VLIB_NODE_FLAG_TRACE; nf->frame_index = ~p->frame_index; } else nf = vec_elt_at_index (nm->next_frames, p->next_frame_index); ASSERT (f->flags & VLIB_FRAME_IS_ALLOCATED); /* Force allocation of new frame while current frame is being dispatched. */ restore_frame_index = ~0; if (nf->frame_index == p->frame_index) { nf->frame_index = ~0; nf->flags &= ~VLIB_FRAME_IS_ALLOCATED; if (!(n->flags & VLIB_NODE_FLAG_FRAME_NO_FREE_AFTER_DISPATCH)) restore_frame_index = p->frame_index; } /* Frame must be pending. */ ASSERT (f->flags & VLIB_FRAME_PENDING); ASSERT (f->n_vectors > 0); /* Copy trace flag from next frame to node. Trace flag indicates that at least one vector in the dispatched frame is traced. */ n->flags &= ~VLIB_NODE_FLAG_TRACE; n->flags |= (nf->flags & VLIB_FRAME_TRACE) ? VLIB_NODE_FLAG_TRACE : 0; nf->flags &= ~VLIB_FRAME_TRACE; last_time_stamp = dispatch_node (vm, n, VLIB_NODE_TYPE_INTERNAL, VLIB_NODE_STATE_POLLING, f, last_time_stamp); f->flags &= ~VLIB_FRAME_PENDING; /* Frame is ready to be used again, so restore it. */ if (restore_frame_index != ~0) { /* * We musn't restore a frame that is flagged to be freed. This * shouldn't happen since frames to be freed post dispatch are * those used when the to-node frame becomes full i.e. they form a * sort of queue of frames to a single node. If we get here then * the to-node frame and the pending frame *were* the same, and so * we removed the to-node frame. Therefore this frame is no * longer part of the queue for that node and hence it cannot be * it's overspill. */ ASSERT (!(f->flags & VLIB_FRAME_FREE_AFTER_DISPATCH)); /* * NB: dispatching node n can result in the creation and scheduling * of new frames, and hence in the reallocation of nm->pending_frames. * Recompute p, or no supper. This was broken for more than 10 years. */ p = nm->pending_frames + pending_frame_index; /* * p->next_frame_index can change during node dispatch if node * function decides to change graph hook up. */ nf = vec_elt_at_index (nm->next_frames, p->next_frame_index); nf->flags |= VLIB_FRAME_IS_ALLOCATED; if (~0 == nf->frame_index) { /* no new frame has been assigned to this node, use the saved one */ nf->frame_index = restore_frame_index; f->n_vectors = 0; } else { /* The node has gained a frame, implying packets from the current frame were re-queued to this same node. we don't need the saved one anymore */ vlib_frame_free (vm, n, f); } } else { if (f->flags & VLIB_FRAME_FREE_AFTER_DISPATCH) { ASSERT (!(n->flags & VLIB_NODE_FLAG_FRAME_NO_FREE_AFTER_DISPATCH)); vlib_frame_free (vm, n, f); } } return last_time_stamp; } always_inline uword vlib_process_stack_is_valid (vlib_process_t * p) { return p->stack[0] == VLIB_PROCESS_STACK_MAGIC; } typedef struct { vlib_main_t *vm; vlib_process_t *process; vlib_frame_t *frame; } vlib_process_bootstrap_args_t; /* Called in process stack. */ static uword vlib_process_bootstrap (uword _a) { vlib_process_bootstrap_args_t *a; vlib_main_t *vm; vlib_node_runtime_t *node; vlib_frame_t *f; vlib_process_t *p; uword n; a = uword_to_pointer (_a, vlib_process_bootstrap_args_t *); vm = a->vm; p = a->process; f = a->frame; node = &p->node_runtime; n = node->function (vm, node, f); ASSERT (vlib_process_stack_is_valid (p)); clib_longjmp (&p->return_longjmp, n); return n; } /* Called in main stack. */ static_always_inline uword vlib_process_startup (vlib_main_t * vm, vlib_process_t * p, vlib_frame_t * f) { vlib_process_bootstrap_args_t a; uword r; a.vm = vm; a.process = p; a.frame = f; r = clib_setjmp (&p->return_longjmp, VLIB_PROCESS_RETURN_LONGJMP_RETURN); if (r == VLIB_PROCESS_RETURN_LONGJMP_RETURN) r = clib_calljmp (vlib_process_bootstrap, pointer_to_uword (&a), (void *) p->stack + (1 << p->log2_n_stack_bytes)); return r; } static_always_inline uword vlib_process_resume (vlib_process_t * p) { uword r; p->flags &= ~(VLIB_PROCESS_IS_SUSPENDED_WAITING_FOR_CLOCK | VLIB_PROCESS_IS_SUSPENDED_WAITING_FOR_EVENT | VLIB_PROCESS_RESUME_PENDING); r = clib_setjmp (&p->return_longjmp, VLIB_PROCESS_RETURN_LONGJMP_RETURN); if (r == VLIB_PROCESS_RETURN_LONGJMP_RETURN) clib_longjmp (&p->resume_longjmp, VLIB_PROCESS_RESUME_LONGJMP_RESUME); return r; } static u64 dispatch_process (vlib_main_t * vm, vlib_process_t * p, vlib_frame_t * f, u64 last_time_stamp) { vlib_node_main_t *nm = &vm->node_main; vlib_node_runtime_t *node_runtime = &p->node_runtime; vlib_node_t *node = vlib_get_node (vm, node_runtime->node_index); u64 t; uword n_vectors, is_suspend; if (node->state != VLIB_NODE_STATE_POLLING || (p->flags & (VLIB_PROCESS_IS_SUSPENDED_WAITING_FOR_CLOCK | VLIB_PROCESS_IS_SUSPENDED_WAITING_FOR_EVENT))) return last_time_stamp; p->flags |= VLIB_PROCESS_IS_RUNNING; t = last_time_stamp; vlib_elog_main_loop_event (vm, node_runtime->node_index, t, f ? f->n_vectors : 0, /* is_after */ 0); /* Save away current process for suspend. */ nm->current_process_index = node->runtime_index; n_vectors = vlib_process_startup (vm, p, f); nm->current_process_index = ~0; ASSERT (n_vectors != VLIB_PROCESS_RETURN_LONGJMP_RETURN); is_suspend = n_vectors == VLIB_PROCESS_RETURN_LONGJMP_SUSPEND; if (is_suspend) { vlib_pending_frame_t *pf; n_vectors = 0; pool_get (nm->suspended_process_frames, pf); pf->node_runtime_index = node->runtime_index; pf->frame_index = f ? vlib_frame_index (vm, f) : ~0; pf->next_frame_index = ~0; p->n_suspends += 1; p->suspended_process_frame_index = pf - nm->suspended_process_frames; if (p->flags & VLIB_PROCESS_IS_SUSPENDED_WAITING_FOR_CLOCK) { TWT (tw_timer_wheel) * tw = (TWT (tw_timer_wheel) *) nm->timing_wheel; p->stop_timer_handle = TW (tw_timer_start) (tw, vlib_timing_wheel_data_set_suspended_process (node->runtime_index) /* [sic] pool idex */ , 0 /* timer_id */ , p->resume_clock_interval); } } else p->flags &= ~VLIB_PROCESS_IS_RUNNING; t = clib_cpu_time_now (); vlib_elog_main_loop_event (vm, node_runtime->node_index, t, is_suspend, /* is_after */ 1); vlib_process_update_stats (vm, p, /* n_calls */ !is_suspend, /* n_vectors */ n_vectors, /* n_clocks */ t - last_time_stamp); return t; } void vlib_start_process (vlib_main_t * vm, uword process_index) { vlib_node_main_t *nm = &vm->node_main; vlib_process_t *p = vec_elt (nm->processes, process_index); dispatch_process (vm, p, /* frame */ 0, /* cpu_time_now */ 0); } static u64 dispatch_suspended_process (vlib_main_t * vm, uword process_index, u64 last_time_stamp) { vlib_node_main_t *nm = &vm->node_main; vlib_node_runtime_t *node_runtime; vlib_node_t *node; vlib_frame_t *f; vlib_process_t *p; vlib_pending_frame_t *pf; u64 t, n_vectors, is_suspend; t = last_time_stamp; p = vec_elt (nm->processes, process_index); if (PREDICT_FALSE (!(p->flags & VLIB_PROCESS_IS_RUNNING))) return last_time_stamp; ASSERT (p->flags & (VLIB_PROCESS_IS_SUSPENDED_WAITING_FOR_CLOCK | VLIB_PROCESS_IS_SUSPENDED_WAITING_FOR_EVENT)); pf = pool_elt_at_index (nm->suspended_process_frames, p->suspended_process_frame_index); node_runtime = &p->node_runtime; node = vlib_get_node (vm, node_runtime->node_index); f = pf->frame_index != ~0 ? vlib_get_frame (vm, pf->frame_index) : 0; vlib_elog_main_loop_event (vm, node_runtime->node_index, t, f ? f->n_vectors : 0, /* is_after */ 0); /* Save away current process for suspend. */ nm->current_process_index = node->runtime_index; n_vectors = vlib_process_resume (p); t = clib_cpu_time_now (); nm->current_process_index = ~0; is_suspend = n_vectors == VLIB_PROCESS_RETURN_LONGJMP_SUSPEND; if (is_suspend) { /* Suspend it again. */ n_vectors = 0; p->n_suspends += 1; if (p->flags & VLIB_PROCESS_IS_SUSPENDED_WAITING_FOR_CLOCK) { p->stop_timer_handle = TW (tw_timer_start) ((TWT (tw_timer_wheel) *) nm->timing_wheel, vlib_timing_wheel_data_set_suspended_process (node->runtime_index) /* [sic] pool idex */ , 0 /* timer_id */ , p->resume_clock_interval); } } else { p->flags &= ~VLIB_PROCESS_IS_RUNNING; p->suspended_process_frame_index = ~0; pool_put (nm->suspended_process_frames, pf); } t = clib_cpu_time_now (); vlib_elog_main_loop_event (vm, node_runtime->node_index, t, !is_suspend, /* is_after */ 1); vlib_process_update_stats (vm, p, /* n_calls */ !is_suspend, /* n_vectors */ n_vectors, /* n_clocks */ t - last_time_stamp); return t; } static_always_inline void vlib_main_or_worker_loop (vlib_main_t * vm, int is_main) { vlib_node_main_t *nm = &vm->node_main; vlib_thread_main_t *tm = vlib_get_thread_main (); uword i; u64 cpu_time_now; vlib_frame_queue_main_t *fqm; u32 *last_node_runtime_indices = 0; /* Initialize pending node vector. */ if (is_main) { vec_resize (nm->pending_frames, 32); _vec_len (nm->pending_frames) = 0; } /* Mark time of main loop start. */ if (is_main) { cpu_time_now = vm->clib_time.last_cpu_time; vm->cpu_time_main_loop_start = cpu_time_now; } else cpu_time_now = clib_cpu_time_now (); /* Pre-allocate interupt runtime indices and lock. */ vec_alloc (nm->pending_interrupt_node_runtime_indices, 32); vec_alloc (last_node_runtime_indices, 32); if (!is_main) clib_spinlock_init (&nm->pending_interrupt_lock); /* Pre-allocate expired nodes. */ if (!nm->polling_threshold_vector_length) nm->polling_threshold_vector_length = 10; if (!nm->interrupt_threshold_vector_length) nm->interrupt_threshold_vector_length = 5; /* Start all processes. */ if (is_main) { uword i; nm->current_process_index = ~0; for (i = 0; i < vec_len (nm->processes); i++) cpu_time_now = dispatch_process (vm, nm->processes[i], /* frame */ 0, cpu_time_now); } while (1) { vlib_node_runtime_t *n; if (!is_main) { vlib_worker_thread_barrier_check (); vec_foreach (fqm, tm->frame_queue_mains) vlib_frame_queue_dequeue (vm, fqm); } /* Process pre-input nodes. */ if (is_main) vec_foreach (n, nm->nodes_by_type[VLIB_NODE_TYPE_PRE_INPUT]) cpu_time_now = dispatch_node (vm, n, VLIB_NODE_TYPE_PRE_INPUT, VLIB_NODE_STATE_POLLING, /* frame */ 0, cpu_time_now); /* Next process input nodes. */ vec_foreach (n, nm->nodes_by_type[VLIB_NODE_TYPE_INPUT]) cpu_time_now = dispatch_node (vm, n, VLIB_NODE_TYPE_INPUT, VLIB_NODE_STATE_POLLING, /* frame */ 0, cpu_time_now); if (PREDICT_TRUE (is_main && vm->queue_signal_pending == 0)) vm->queue_signal_callback (vm); /* Next handle interrupts. */ { uword l = _vec_len (nm->pending_interrupt_node_runtime_indices); uword i; if (l > 0) { u32 *tmp; if (!is_main) clib_spinlock_lock (&nm->pending_interrupt_lock); tmp = nm->pending_interrupt_node_runtime_indices; nm->pending_interrupt_node_runtime_indices = last_node_runtime_indices; last_node_runtime_indices = tmp; _vec_len (last_node_runtime_indices) = 0; if (!is_main) clib_spinlock_unlock (&nm->pending_interrupt_lock); for (i = 0; i < l; i++) { n = vec_elt_at_index (nm->nodes_by_type[VLIB_NODE_TYPE_INPUT], last_node_runtime_indices[i]); cpu_time_now = dispatch_node (vm, n, VLIB_NODE_TYPE_INPUT, VLIB_NODE_STATE_INTERRUPT, /* frame */ 0, cpu_time_now); } } } if (is_main) { /* Check if process nodes have expired from timing wheel. */ ASSERT (nm->data_from_advancing_timing_wheel != 0); nm->data_from_advancing_timing_wheel = TW (tw_timer_expire_timers_vec) ((TWT (tw_timer_wheel) *) nm->timing_wheel, vlib_time_now (vm), nm->data_from_advancing_timing_wheel); ASSERT (nm->data_from_advancing_timing_wheel != 0); if (PREDICT_FALSE (_vec_len (nm->data_from_advancing_timing_wheel) > 0)) { uword i; processes_timing_wheel_data: for (i = 0; i < _vec_len (nm->data_from_advancing_timing_wheel); i++) { u32 d = nm->data_from_advancing_timing_wheel[i]; u32 di = vlib_timing_wheel_data_get_index (d); if (vlib_timing_wheel_data_is_timed_event (d)) { vlib_signal_timed_event_data_t *te = pool_elt_at_index (nm->signal_timed_event_data_pool, di); vlib_node_t *n = vlib_get_node (vm, te->process_node_index); vlib_process_t *p = vec_elt (nm->processes, n->runtime_index); void *data; data = vlib_process_signal_event_helper (nm, n, p, te->event_type_index, te->n_data_elts, te->n_data_elt_bytes); if (te->n_data_bytes < sizeof (te->inline_event_data)) clib_memcpy (data, te->inline_event_data, te->n_data_bytes); else { clib_memcpy (data, te->event_data_as_vector, te->n_data_bytes); vec_free (te->event_data_as_vector); } pool_put (nm->signal_timed_event_data_pool, te); } else { cpu_time_now = clib_cpu_time_now (); cpu_time_now = dispatch_suspended_process (vm, di, cpu_time_now); } } _vec_len (nm->data_from_advancing_timing_wheel) = 0; } } /* Input nodes may have added work to the pending vector. Process pending vector until there is nothing left. All pending vectors will be processed from input -> output. */ for (i = 0; i < _vec_len (nm->pending_frames); i++) cpu_time_now = dispatch_pending_node (vm, i, cpu_time_now); /* Reset pending vector for next iteration. */ _vec_len (nm->pending_frames) = 0; /* Pending internal nodes may resume processes. */ if (is_main && _vec_len (nm->data_from_advancing_timing_wheel) > 0) goto processes_timing_wheel_data; vlib_increment_main_loop_counter (vm); /* Record time stamp in case there are no enabled nodes and above calls do not update time stamp. */ cpu_time_now = clib_cpu_time_now (); } } static void vlib_main_loop (vlib_main_t * vm) { vlib_main_or_worker_loop (vm, /* is_main */ 1); } void vlib_worker_loop (vlib_main_t * vm) { vlib_main_or_worker_loop (vm, /* is_main */ 0); } vlib_main_t vlib_global_main; static clib_error_t * vlib_main_configure (vlib_main_t * vm, unformat_input_t * input) { int turn_on_mem_trace = 0; while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) { if (unformat (input, "memory-trace")) turn_on_mem_trace = 1; else if (unformat (input, "elog-events %d", &vm->elog_main.event_ring_size)) ; else if (unformat (input, "elog-post-mortem-dump")) vm->elog_post_mortem_dump = 1; else return unformat_parse_error (input); } unformat_free (input); /* Enable memory trace as early as possible. */ if (turn_on_mem_trace) clib_mem_trace (1); return 0; } VLIB_EARLY_CONFIG_FUNCTION (vlib_main_configure, "vlib"); static void dummy_queue_signal_callback (vlib_main_t * vm) { } /* Main function. */ int vlib_main (vlib_main_t * volatile vm, unformat_input_t * input) { clib_error_t *volatile error; vlib_node_main_t *nm = &vm->node_main; vm->queue_signal_callback = dummy_queue_signal_callback; clib_time_init (&vm->clib_time); /* Turn on event log. */ if (!vm->elog_main.event_ring_size) vm->elog_main.event_ring_size = 128 << 10; elog_init (&vm->elog_main, vm->elog_main.event_ring_size); elog_enable_disable (&vm->elog_main, 1); /* Default name. */ if (!vm->name) vm->name = "VLIB"; if ((error = unix_physmem_init (vm))) { clib_error_report (error); goto done; } if ((error = vlib_buffer_main_init (vm))) { clib_error_report (error); goto done; } if ((error = vlib_thread_init (vm))) { clib_error_report (error); goto done; } /* Register static nodes so that init functions may use them. */ vlib_register_all_static_nodes (vm); /* Set seed for random number generator. Allow user to specify seed to make random sequence deterministic. */ if (!unformat (input, "seed %wd", &vm->random_seed)) vm->random_seed = clib_cpu_time_now (); clib_random_buffer_init (&vm->random_buffer, vm->random_seed); /* Initialize node graph. */ if ((error = vlib_node_main_init (vm))) { /* Arrange for graph hook up error to not be fatal when debugging. */ if (CLIB_DEBUG > 0) clib_error_report (error); else goto done; } /* See unix/main.c; most likely already set up */ if (vm->init_functions_called == 0) vm->init_functions_called = hash_create (0, /* value bytes */ 0); if ((error = vlib_call_all_init_functions (vm))) goto done; /* Create default buffer free list. */ vlib_buffer_get_or_create_free_list (vm, VLIB_BUFFER_DEFAULT_FREE_LIST_BYTES, "default"); nm->timing_wheel = clib_mem_alloc_aligned (sizeof (TWT (tw_timer_wheel)), CLIB_CACHE_LINE_BYTES); vec_validate (nm->data_from_advancing_timing_wheel, 10); _vec_len (nm->data_from_advancing_timing_wheel) = 0; /* Create the process timing wheel */ TW (tw_timer_wheel_init) ((TWT (tw_timer_wheel) *) nm->timing_wheel, 0 /* no callback */ , 10e-6 /* timer period 10us */ , ~0 /* max expirations per call */ ); switch (clib_setjmp (&vm->main_loop_exit, VLIB_MAIN_LOOP_EXIT_NONE)) { case VLIB_MAIN_LOOP_EXIT_NONE: vm->main_loop_exit_set = 1; break; case VLIB_MAIN_LOOP_EXIT_CLI: goto done; default: error = vm->main_loop_error; goto done; } if ((error = vlib_call_all_config_functions (vm, input, 0 /* is_early */ ))) goto done; /* Call all main loop enter functions. */ { clib_error_t *sub_error; sub_error = vlib_call_all_main_loop_enter_functions (vm); if (sub_error) clib_error_report (sub_error); } vlib_main_loop (vm); done: /* Call all exit functions. */ { clib_error_t *sub_error; sub_error = vlib_call_all_main_loop_exit_functions (vm); if (sub_error) clib_error_report (sub_error); } if (error) clib_error_report (error); return 0; } /* * fd.io coding-style-patch-verification: ON * * Local Variables: * eval: (c-set-style "gnu") * End: */