FD.io VPP  v20.09-64-g4f7b92f0a
Vector Packet Processing
vlib_process_doc.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2016 Cisco and/or its affiliates.
3  * Licensed under the Apache License, Version 2.0 (the "License");
4  * you may not use this file except in compliance with the License.
5  * You may obtain a copy of the License at:
6  *
7  * http://www.apache.org/licenses/LICENSE-2.0
8  *
9  * Unless required by applicable law or agreed to in writing, software
10  * distributed under the License is distributed on an "AS IS" BASIS,
11  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12  * See the License for the specific language governing permissions and
13  * limitations under the License.
14 */
15 
16 #error do not #include this file!
17 
18 /** \file
19 
20  Cooperative multi-tasking thread support.
21 
22  Vlib provides a lightweight cooperative multi-tasking thread
23  model. Context switching costs a setjmp/longjump pair. It's not
24  unreasonable to put vlib threads to sleep for 10us.
25 
26  The graph node scheduler invokes these processes in much the same
27  way as traditional vector-processing run-to-completion graph
28  nodes; plus-or-minus a setjmp/longjmp pair required to switch
29  stacks. Simply set the vlib_node_registration_t type field to
30  VLIB_NODE_TYPE_PROCESS. Process is a misnomer; these are threads.
31 
32  As of this writing, the default stack size is 2<<15;
33  32kb. Initialize the node registration's
34  process_log2_n_stack_bytes member as needed. The graph node
35  dispatcher makes some effort to detect stack overrun. We map a
36  no-access page below each thread stack.
37 
38  Process node dispatch functions are expected to be while(1) { }
39  loops which suspend when not otherwise occupied, and which must
40  not run for unreasonably long periods of time. Unreasonably long
41  is an application-dependent concept. Over the years, we have
42  constructed frame-size sensitive control-plane nodes which will
43  use a much higher fraction of the available CPU bandwidth when the
44  frame size is low. Classic example: modifying forwarding
45  tables. So long as the table-builder leaves the forwarding tables
46  in a valid state, one can suspend the table builder to avoid
47  dropping packets as a result of control-plane activity.
48 
49  Process nodes can suspend for fixed amounts of time, or until another
50  entity signals an event, or both. See the example below.
51 
52  When running in VLIB process context, one must pay strict attention to
53  loop invariant issues. If one walks a data structure and calls a
54  function which may suspend, one had best know by construction that it
55  cannot change. Often, it s best to simply make a snapshot copy of a
56  data structure, walk the copy at leisure, then free the copy.
57 
58  Here's an example:
59 
60  <code><pre>
61  \#define EXAMPLE_POLL_PERIOD 10.0
62 
63  static uword
64  example_process (vlib_main_t * vm, vlib_node_runtime_t * rt,
65  vlib_frame_t * f)
66  {
67  f64 poll_time_remaining;
68  uword event_type, *event_data = 0;
69 
70  poll_time_remaining = EXAMPLE_POLL_PERIOD;
71  while (1)
72  {
73  int i;
74 
75  // Sleep until next periodic call due,
76  // or until we receive event(s)
77  //
78  poll_time_remaining =
79  vlib_process_wait_for_event_or_clock (vm, poll_time_remaining);
80 
81  event_type = vlib_process_get_events (vm, &event_data);
82  switch (event_type)
83  {
84  case ~0: // no events => timeout
85  break;
86 
87  case EVENT1:
88  for (i = 0; i < vec_len (event_data); i++)
89  handle_event1 (mm, event_data[i]);
90  break;
91 
92  case EVENT2:
93  for (i = 0; i < vec_len (event_data); i++)
94  handle_event2 (vm, event_data[i]);
95  break;
96 
97  // ... and so forth for each event type
98 
99  default:
100  // This should never happen...
101  clib_warning ("BUG: unhandled event type %d",
102  event_type);
103  break;
104  }
105  vec_reset_length (event_data);
106 
107  // Timer expired, call periodic function
108  if (vlib_process_suspend_time_is_zero (poll_time_remaining))
109  {
110  example_periodic (vm);
111  poll_time_remaining = EXAMPLE_POLL_PERIOD;
112  }
113  }
114  // NOTREACHED
115  return 0;
116  }
117 
118  static VLIB_REGISTER_NODE (example_node) = {
119  .function = example_process,
120  .type = VLIB_NODE_TYPE_PROCESS,
121  .name = "example-process",
122  };
123  </pre></code>
124 
125  In this example, the VLIB process node waits for an event to
126  occur, or for 10 seconds to elapse. The code demuxes on the event
127  type, calling the appropriate handler function.
128 
129  Each call to vlib_process_get_events returns a vector of
130  per-event-type data passed to successive vlib_process_signal_event
131  calls; vec_len (event_data) >= 1. It is an error to process only
132  event_data[0].
133 
134  Resetting the event_data vector-length to 0 by calling
135  vec_reset_length (event_data) - instead of calling vec_free (...)
136  - means that the event scheme doesn t burn cycles continuously
137  allocating and freeing the event data vector. This is a common
138  coding pattern, well worth using when appropriate.
139 */
140 
141 /*
142  * fd.io coding-style-patch-verification: ON
143  *
144  * Local Variables:
145  * eval: (c-set-style "gnu")
146  * End:
147  */