1fi_msg(3) Libfabric v1.7.0 fi_msg(3)
2
3
4
6 fi_msg - Message data transfer operations
7
8 fi_recv / fi_recvv / fi_recvmsg
9 Post a buffer to receive an incoming message
10
11 fi_send / fi_sendv / fi_sendmsg
12 fi_inject / fi_senddata : Initiate an operation to send a message
13
15 #include <rdma/fi_endpoint.h>
16
17 ssize_t fi_recv(struct fid_ep *ep, void * buf, size_t len,
18 void *desc, fi_addr_t src_addr, void *context);
19
20 ssize_t fi_recvv(struct fid_ep *ep, const struct iovec *iov, void **desc,
21 size_t count, fi_addr_t src_addr, void *context);
22
23 ssize_t fi_recvmsg(struct fid_ep *ep, const struct fi_msg *msg,
24 uint64_t flags);
25
26 ssize_t fi_send(struct fid_ep *ep, const void *buf, size_t len,
27 void *desc, fi_addr_t dest_addr, void *context);
28
29 ssize_t fi_sendv(struct fid_ep *ep, const struct iovec *iov,
30 void **desc, size_t count, fi_addr_t dest_addr, void *context);
31
32 ssize_t fi_sendmsg(struct fid_ep *ep, const struct fi_msg *msg,
33 uint64_t flags);
34
35 ssize_t fi_inject(struct fid_ep *ep, const void *buf, size_t len,
36 fi_addr_t dest_addr);
37
38 ssize_t fi_senddata(struct fid_ep *ep, const void *buf, size_t len,
39 void *desc, uint64_t data, fi_addr_t dest_addr, void *context);
40
41 ssize_t fi_injectdata(struct fid_ep *ep, const void *buf, size_t len,
42 uint64_t data, fi_addr_t dest_addr);
43
45 ep Fabric endpoint on which to initiate send or post receive buf‐
46 fer.
47
48 buf Data buffer to send or receive.
49
50 len Length of data buffer to send or receive, specified in bytes.
51 Valid transfers are from 0 bytes up to the endpoint's
52 max_msg_size.
53
54 iov Vectored data buffer.
55
56 count Count of vectored data entries.
57
58 desc Descriptor associated with the data buffer
59
60 data Remote CQ data to transfer with the sent message.
61
62 dest_addr
63 Destination address for connectionless transfers. Ignored for
64 connected endpoints.
65
66 src_addr
67 Source address to receive from for connectionless transfers.
68 Applies only to connectionless endpoints with the FI_DIRECT‐
69 ED_RECV capability enabled, otherwise this field is ignored. If
70 set to FI_ADDR_UNSPEC, any source address may match.
71
72 msg Message descriptor for send and receive operations.
73
74 flags Additional flags to apply for the send or receive operation.
75
76 context
77 User specified pointer to associate with the operation. This
78 parameter is ignored if the operation will not generate a suc‐
79 cessful completion, unless an op flag specifies the context pa‐
80 rameter be used for required input.
81
83 The send functions -- fi_send, fi_sendv, fi_sendmsg, fi_inject, and
84 fi_senddata -- are used to transmit a message from one endpoint to an‐
85 other endpoint. The main difference between send functions are the
86 number and type of parameters that they accept as input. Otherwise,
87 they perform the same general function. Messages sent using fi_msg op‐
88 erations are received by a remote endpoint into a buffer posted to re‐
89 ceive such messages.
90
91 The receive functions -- fi_recv, fi_recvv, fi_recvmsg -- post a data
92 buffer to an endpoint to receive inbound messages. Similar to the send
93 operations, receive operations operate asynchronously. Users should
94 not touch the posted data buffer(s) until the receive operation has
95 completed.
96
97 An endpoint must be enabled before an application can post send or re‐
98 ceive operations to it. For connected endpoints, receive buffers may
99 be posted prior to connect or accept being called on the endpoint.
100 This ensures that buffers are available to receive incoming data imme‐
101 diately after the connection has been established.
102
103 Completed message operations are reported to the user through one or
104 more event collectors associated with the endpoint. Users provide con‐
105 text which are associated with each operation, and is returned to the
106 user as part of the event completion. See fi_cq for completion event
107 details.
108
109 fi_send
110 The call fi_send transfers the data contained in the user-specified da‐
111 ta buffer to a remote endpoint, with message boundaries being main‐
112 tained. For connection based endpoints (FI_EP_MSG) the local endpoint
113 must be connected to a remote endpoint or destination before fi_send is
114 called. Unless the endpoint has been configured differently, the data
115 buffer passed into fi_send must not be touched by the application until
116 the fi_send call completes asynchronously.
117
118 fi_sendv
119 The fi_sendv call adds support for a scatter-gather list to fi_send.
120 The fi_sendv transfers the set of data buffers referenced by the iov
121 parameter to a remote endpoint as a single message.
122
123 fi_sendmsg
124 The fi_sendmsg call supports data transfers over both connected and un‐
125 connected endpoints, with the ability to control the send operation per
126 call through the use of flags. The fi_sendmsg function takes a
127 struct fi_msg as input.
128
129 struct fi_msg {
130 const struct iovec *msg_iov; /* scatter-gather array */
131 void **desc; /* local request descriptors */
132 size_t iov_count;/* # elements in iov */
133 fi_addr_t addr; /* optional endpoint address */
134 void *context; /* user-defined context */
135 uint64_t data; /* optional message data */
136 };
137
138 fi_inject
139 The send inject call is an optimized version of fi_send. The fi_inject
140 function behaves as if the FI_INJECT transfer flag were set, and
141 FI_COMPLETION were not. That is, the data buffer is available for re‐
142 use immediately on returning from fi_inject, and no completion event
143 will be generated for this send. The completion event will be sup‐
144 pressed even if the CQ was bound without FI_SELECTIVE_COMPLETION or the
145 endpoint's op_flags contain FI_COMPLETION. See the flags discussion
146 below for more details. The requested message size that can be used
147 with fi_inject is limited by inject_size.
148
149 fi_senddata
150 The send data call is similar to fi_send, but allows for the sending of
151 remote CQ data (see FI_REMOTE_CQ_DATA flag) as part of the transfer.
152
153 fi_injectdata
154 The inject data call is similar to fi_inject, but allows for the send‐
155 ing of remote CQ data (see FI_REMOTE_CQ_DATA flag) as part of the
156 transfer.
157
158 fi_recv
159 The fi_recv call posts a data buffer to the receive queue of the corre‐
160 sponding endpoint. Posted receives are searched in the order in which
161 they were posted in order to match sends. Message boundaries are main‐
162 tained. The order in which the receives complete is dependent on the
163 endpoint type and protocol. For unconnected endpoints, the src_addr
164 parameter can be used to indicate that a buffer should be posted to re‐
165 ceive incoming data from a specific remote endpoint.
166
167 fi_recvv
168 The fi_recvv call adds support for a scatter-gather list to fi_recv.
169 The fi_recvv posts the set of data buffers referenced by the iov param‐
170 eter to a receive incoming data.
171
172 fi_recvmsg
173 The fi_recvmsg call supports posting buffers over both connected and
174 unconnected endpoints, with the ability to control the receive opera‐
175 tion per call through the use of flags. The fi_recvmsg function takes
176 a struct fi_msg as input.
177
179 The fi_recvmsg and fi_sendmsg calls allow the user to specify flags
180 which can change the default message handling of the endpoint. Flags
181 specified with fi_recvmsg / fi_sendmsg override most flags previously
182 configured with the endpoint, except where noted (see fi_endpoint.3).
183 The following list of flags are usable with fi_recvmsg and/or
184 fi_sendmsg.
185
186 FI_REMOTE_CQ_DATA
187 Applies to fi_sendmsg and fi_senddata. Indicates that remote CQ
188 data is available and should be sent as part of the request.
189 See fi_getinfo for additional details on FI_REMOTE_CQ_DATA.
190
191 FI_CLAIM
192 Applies to posted receive operations for endpoints configured
193 for FI_BUFFERED_RECV or FI_VARIABLE_MSG. This flag is used to
194 retrieve a message that was buffered by the provider. See the
195 Buffered Receives section for details.
196
197 FI_COMPLETION
198 Indicates that a completion entry should be generated for the
199 specified operation. The endpoint must be bound to a completion
200 queue with FI_SELECTIVE_COMPLETION that corresponds to the spec‐
201 ified operation, or this flag is ignored.
202
203 FI_DISCARD
204 Applies to posted receive operations for endpoints configured
205 for FI_BUFFERED_RECV or FI_VARIABLE_MSG. This flag is used to
206 free a message that was buffered by the provider. See the
207 Buffered Receives section for details.
208
209 FI_MORE
210 Indicates that the user has additional requests that will imme‐
211 diately be posted after the current call returns. Use of this
212 flag may improve performance by enabling the provider to opti‐
213 mize its access to the fabric hardware.
214
215 FI_INJECT
216 Applies to fi_sendmsg. Indicates that the outbound data buffer
217 should be returned to user immediately after the send call re‐
218 turns, even if the operation is handled asynchronously. This
219 may require that the underlying provider implementation copy the
220 data into a local buffer and transfer out of that buffer. This
221 flag can only be used with messages smaller than inject_size.
222
223 FI_MULTI_RECV
224 Applies to posted receive operations. This flag allows the user
225 to post a single buffer that will receive multiple incoming mes‐
226 sages. Received messages will be packed into the receive buffer
227 until the buffer has been consumed. Use of this flag may cause
228 a single posted receive operation to generate multiple events as
229 messages are placed into the buffer. The placement of received
230 data into the buffer may be subjected to provider specific
231 alignment restrictions.
232
233 The buffer will be released by the provider when the available buffer
234 space falls below the specified minimum (see FI_OPT_MIN_MULTI_RECV).
235 Note that an entry to the associated receive completion queue will al‐
236 ways be generated when the buffer has been consumed, even if other re‐
237 ceive completions have been suppressed (i.e. the Rx context has been
238 configured for FI_SELECTIVE_COMPLETION). See the FI_MULTI_RECV comple‐
239 tion flag fi_cq(3).
240
241 FI_INJECT_COMPLETE
242 Applies to fi_sendmsg. Indicates that a completion should be
243 generated when the source buffer(s) may be reused.
244
245 FI_TRANSMIT_COMPLETE
246 Applies to fi_sendmsg. Indicates that a completion should not
247 be generated until the operation has been successfully transmit‐
248 ted and is no longer being tracked by the provider.
249
250 FI_DELIVERY_COMPLETE
251 Applies to fi_sendmsg. Indicates that a completion should be
252 generated when the operation has been processed by the destina‐
253 tion.
254
255 FI_FENCE
256 Applies to transmits. Indicates that the requested operation,
257 also known as the fenced operation, and any operation posted af‐
258 ter the fenced operation will be deferred until all previous op‐
259 erations targeting the same peer endpoint have completed. Oper‐
260 ations posted after the fencing will see and/or replace the re‐
261 sults of any operations initiated prior to the fenced operation.
262
263 The ordering of operations starting at the posting of the fenced opera‐
264 tion (inclusive) to the posting of a subsequent fenced operation (ex‐
265 clusive) is controlled by the endpoint's ordering semantics.
266
267 FI_MULTICAST
268 Applies to transmits. This flag indicates that the address
269 specified as the data transfer destination is a multicast ad‐
270 dress. This flag must be used in all multicast transfers, in
271 conjunction with a multicast fi_addr_t.
272
274 Buffered receives indicate that the networking layer allocates and man‐
275 ages the data buffers used to receive network data transfers. As a re‐
276 sult, received messages must be copied from the network buffers into
277 application buffers for processing. However, applications can avoid
278 this copy if they are able to process the message in place (directly
279 from the networking buffers).
280
281 Handling buffered receives differs based on the size of the message be‐
282 ing sent. In general, smaller messages are passed directly to the ap‐
283 plication for processing. However, for large messages, an application
284 will only receive the start of the message and must claim the rest.
285 The details for how small messages are reported and large messages may
286 be claimed are described below.
287
288 When a provider receives a message, it will write an entry to the com‐
289 pletion queue associated with the receiving endpoint. For discussion
290 purposes, the completion queue is assumed to be configured for
291 FI_CQ_FORMAT_DATA. Since buffered receives are not associated with ap‐
292 plication posted buffers, the CQ entry op_context will point to a
293 struct fi_recv_context.
294
295 struct fi_recv_context {
296 struct fid_ep *ep;
297 void *context;
298 };
299
300 The 'ep' field will point to the receiving endpoint or Rx context, and
301 'context' will be NULL. The CQ entry's 'buf' will point to a provider
302 managed buffer where the start of the received message is located, and
303 'len' will be set to the total size of the message.
304
305 The maximum sized message that a provider can buffer is limited by an
306 FI_OPT_BUFFERED_LIMIT. This threshold can be obtained and may be ad‐
307 justed by the application using the fi_getopt and fi_setopt calls, re‐
308 spectively. Any adjustments must be made prior to enabling the end‐
309 point. The CQ entry 'buf' will point to a buffer of received data. If
310 the sent message is larger than the buffered amount, the CQ entry
311 'flags' will have the FI_MORE bit set. When the FI_MORE bit is set,
312 'buf' will reference at least FI_OPT_BUFFERED_MIN bytes of data (see
313 fi_endpoint.3 for more info).
314
315 After being notified that a buffered receive has arrived, applications
316 must either claim or discard the message. Typically, small messages
317 are processed and discarded, while large messages are claimed. Howev‐
318 er, an application is free to claim or discard any message regardless
319 of message size.
320
321 To claim a message, an application must post a receive operation with
322 the FI_CLAIM flag set. The struct fi_recv_context returned as part of
323 the notification must be provided as the receive operation's context.
324 The struct fi_recv_context contains a 'context' field. Applications
325 may modify this field prior to claiming the message. When the claim
326 operation completes, a standard receive completion entry will be gener‐
327 ated on the completion queue. The 'context' of the associated CQ entry
328 will be set to the 'context' value passed in through the fi_recv_con‐
329 text structure, and the CQ entry flags will have the FI_CLAIM bit set.
330
331 Buffered receives that are not claimed must be discarded by the appli‐
332 cation when it is done processing the CQ entry data. To discard a mes‐
333 sage, an application must post a receive operation with the FI_DISCARD
334 flag set. The struct fi_recv_context returned as part of the notifica‐
335 tion must be provided as the receive operation's context. When the
336 FI_DISCARD flag is set for a receive operation, the receive input buf‐
337 fer(s) and length parameters are ignored.
338
339 IMPORTANT: Buffered receives must be claimed or discarded in a timely
340 manner. Failure to do so may result in increased memory usage for net‐
341 work buffering or communication stalls. Once a buffered receive has
342 been claimed or discarded, the original CQ entry 'buf' or struct
343 fi_recv_context data may no longer be accessed by the application.
344
345 The use of the FI_CLAIM and FI_DISCARD operation flags is also de‐
346 scribed with respect to tagged message transfers in fi_tagged.3.
347 Buffered receives of tagged messages will include the message tag as
348 part of the CQ entry, if available.
349
350 The handling of buffered receives follows all message ordering restric‐
351 tions assigned to an endpoint. For example, completions may indicate
352 the order in which received messages arrived at the receiver based on
353 the endpoint attributes.
354
356 Variable length messages, or simply variable messages, are transfers
357 where the size of the message is unknown to the receiver prior to the
358 message being sent. It indicates that the recipient of a message does
359 not know the amount of data to expect prior to the message arriving.
360 It is most commonly used when the size of message transfers varies
361 greatly, with very large messages interspersed with much smaller mes‐
362 sages, making receive side message buffering difficult to manage.
363 Variable messages are not subject to max message length restrictions
364 (i.e. struct fi_ep_attr::max_msg_size limits), and may be up to the
365 maximum value of size_t (e.g. SIZE_MAX) in length.
366
367 Variable length messages support requests that the provider allocate
368 and manage the network message buffers. As a result, the application
369 requirements and provider behavior is identical as those defined for
370 supporting the FI_BUFFERED_RECV mode bit. See the Buffered Receive
371 section above for details. The main difference is that buffered re‐
372 ceives are limited by the fi_ep_attr::max_msg_size threshold, whereas
373 variable length messages are not.
374
375 Support for variable messages is indicated through the FI_VARIABLE_MSG
376 capability bit.
377
379 If an endpoint has been configured with FI_MSG_PREFIX, the application
380 must include buffer space of size msg_prefix_size, as specified by the
381 endpoint attributes. The prefix buffer must occur at the start of the
382 data referenced by the buf parameter, or be referenced by the first IO
383 vector. Message prefix space cannot be split between multiple IO vec‐
384 tors. The size of the prefix buffer should be included as part of the
385 total buffer length.
386
388 Returns 0 on success. On error, a negative value corresponding to fab‐
389 ric errno is returned. Fabric errno values are defined in rdma/fi_er‐
390 rno.h.
391
392 See the discussion below for details handling FI_EAGAIN.
393
395 -FI_EAGAIN
396 Indicates that the underlying provider currently lacks the re‐
397 sources needed to initiate the requested operation. The reasons
398 for a provider returning FI_EAGAIN are varied. However, common
399 reasons include insufficient internal buffering or full process‐
400 ing queues.
401
402 Insufficient internal buffering is often associated with operations
403 that use FI_INJECT. In such cases, additional buffering may become
404 available as posted operations complete.
405
406 Full processing queues may be a temporary state related to local pro‐
407 cessing (for example, a large message is being transferred), or may be
408 the result of flow control. In the latter case, the queues may remain
409 blocked until additional resources are made available at the remote
410 side of the transfer.
411
412 In all cases, the operation may be retried after additional resources
413 become available. It is strongly recommended that applications check
414 for transmit and receive completions after receiving FI_EAGAIN as a re‐
415 turn value, independent of the operation which failed. This is partic‐
416 ularly important in cases where manual progress is employed, as ac‐
417 knowledgements or flow control messages may need to be processed in or‐
418 der to resume execution.
419
421 fi_getinfo(3), fi_endpoint(3), fi_domain(3), fi_cq(3)
422
424 OpenFabrics.
425
426
427
428Libfabric Programmer's Manual 2018-11-28 fi_msg(3)