1fi_av(3) Libfabric v1.15.1 fi_av(3)
2
6 fi_av - Address vector operations
7 fi_av_open / fi_close
9 Open or close an address vector
10 fi_av_bind
12 Associate an address vector with an event queue.
13 fi_av_insert / fi_av_insertsvc / fi_av_remove
15 Insert/remove an address into/from the address vector.
16 fi_av_lookup
18 Retrieve an address stored in the address vector.
19 fi_av_straddr
21 Convert an address into a printable string.
22
24 #include <rdma/fi_domain.h>
25 int fi_av_open(struct fid_domain *domain, struct fi_av_attr *attr,
27 struct fid_av **av, void *context);
28 int fi_close(struct fid *av);
30 int fi_av_bind(struct fid_av *av, struct fid *eq, uint64_t flags);
32 int fi_av_insert(struct fid_av *av, void *addr, size_t count,
34 fi_addr_t *fi_addr, uint64_t flags, void *context);
35 int fi_av_insertsvc(struct fid_av *av, const char *node,
37 const char *service, fi_addr_t *fi_addr, uint64_t flags,
38 void *context);
39 int fi_av_insertsym(struct fid_av *av, const char *node,
41 size_t nodecnt, const char *service, size_t svccnt,
42 fi_addr_t *fi_addr, uint64_t flags, void *context);
43 int fi_av_remove(struct fid_av *av, fi_addr_t *fi_addr, size_t count,
45 uint64_t flags);
46 int fi_av_lookup(struct fid_av *av, fi_addr_t fi_addr,
48 void *addr, size_t *addrlen);
49 fi_addr_t fi_rx_addr(fi_addr_t fi_addr, int rx_index,
51 int rx_ctx_bits);
52 const char * fi_av_straddr(struct fid_av *av, const void *addr,
54 char *buf, size_t *len);
55
57 domain Resource domain
58 av Address vector
60 eq Event queue
62 attr Address vector attributes
64 context
66 User specified context associated with the address vector or in‐
67 sert operation.
68 addr Buffer containing one or more addresses to insert into address
70 vector.
71 addrlen
73 On input, specifies size of addr buffer. On output, stores num‐
74 ber of bytes written to addr buffer.
75 fi_addr
77 For insert, a reference to an array where returned fabric ad‐
78 dresses will be written. For remove, one or more fabric ad‐
79 dresses to remove.
80 count Number of addresses to insert/remove from an AV.
82 flags Additional flags to apply to the operation.
84
86 Address vectors are used to map higher level addresses, which may be
87 more natural for an application to use, into fabric specific addresses.
88 The mapping of addresses is fabric and provider specific, but may in‐
89 volve lengthy address resolution and fabric management protocols. AV
90 operations are synchronous by default, but may be set to operate asyn‐
91 chronously by specifying the FI_EVENT flag to fi_av_open. When re‐
92 questing asynchronous operation, the application must first bind an
93 event queue to the AV before inserting addresses. See the NOTES sec‐
94 tion for AV restrictions on duplicate addresses.
95 fi_av_open
97 fi_av_open allocates or opens an address vector. The properties and
98 behavior of the address vector are defined by struct fi_av_attr.
99 struct fi_av_attr {
101 enum fi_av_type type; /* type of AV */
102 int rx_ctx_bits; /* address bits to identify rx ctx */
103 size_t count; /* # entries for AV */
104 size_t ep_per_node; /* # endpoints per fabric address */
105 const char *name; /* system name of AV */
106 void *map_addr; /* base mmap address */
107 uint64_t flags; /* operation flags */
108 };
109 type An AV type corresponds to a conceptual implementation of an ad‐
111 dress vector. The type specifies how an application views data
112 stored in the AV, including how it may be accessed. Valid val‐
113 ues are:
114 - FI_AV_MAP
116 Addresses which are inserted into an AV are mapped to a native
117 fabric address for use by the application. The use of FI_AV_MAP
118 requires that an application store the returned fi_addr_t value
119 that is associated with each inserted address. The advantage of
120 using FI_AV_MAP is that the returned fi_addr_t value may contain
121 encoded address data, which is immediately available when pro‐
122 cessing data transfer requests. This can eliminate or reduce
123 the number of memory lookups needed when initiating a transfer.
124 The disadvantage of FI_AV_MAP is the increase in memory usage
125 needed to store the returned addresses. Addresses are stored in
126 the AV using a provider specific mechanism, including, but not
127 limited to a tree, hash table, or maintained on the heap.
128 - FI_AV_TABLE
130 Addresses which are inserted into an AV of type FI_AV_TABLE are
131 accessible using a simple index. Conceptually, the AV may be
132 treated as an array of addresses, though the provider may imple‐
133 ment the AV using a variety of mechanisms. When FI_AV_TABLE is
134 used, the returned fi_addr_t is an index, with the index for an
135 inserted address the same as its insertion order into the table.
136 The index of the first address inserted into an FI_AV_TABLE will
137 be 0, and successive insertions will be given sequential in‐
138 dices. Sequential indices will be assigned across insertion
139 calls on the same AV.
140 - FI_AV_UNSPEC
142 Provider will choose its preferred AV type. The AV type used
143 will be returned through the type field in fi_av_attr.
144 Receive Context Bits (rx_ctx_bits)
146 The receive context bits field is only for use with scalable
147 endpoints. It indicates the number of bits reserved in a re‐
148 turned fi_addr_t, which will be used to identify a specific tar‐
149 get receive context. See fi_rx_addr() and fi_endpoint(3) for
150 additional details on receive contexts. The requested number of
151 bits should be selected such that 2 ^ rx_ctx_bits >= rx_ctx_cnt
152 for the endpoint.
153 count Indicates the expected number of addresses that will be inserted
155 into the AV. The provider uses this to optimize resource allo‐
156 cations.
157 ep_per_node
159 This field indicates the number of endpoints that will be asso‐
160 ciated with a specific fabric, or network, address. If the num‐
161 ber of endpoints per node is unknown, this value should be set
162 to 0. The provider uses this value to optimize resource alloca‐
163 tions. For example, distributed, parallel applications may set
164 this to the number of processes allocated per node, times the
165 number of endpoints each process will open.
166 name An optional system name associated with the address vector to
168 create or open. Address vectors may be shared across multiple
169 processes which access the same named domain on the same node.
170 The name field allows the underlying provider to identify a
171 shared AV.
172 If the name field is non-NULL and the AV is not opened for read-only
174 access, a named AV will be created, if it does not already exist.
175 map_addr
177 The map_addr determines the base fi_addr_t address that a
178 provider should use when sharing an AV of type FI_AV_MAP between
179 processes. Processes that provide the same value for map_addr
180 to a shared AV may use the same fi_addr_t values returned from
181 an fi_av_insert call.
182 The map_addr may be used by the provider to mmap memory allocated for a
184 shared AV between processes; however, the provider is not required to
185 use the map_addr in this fashion. The only requirement is that an
186 fi_addr_t returned as part of an fi_av_insert call on one process is
187 usable on another process which opens an AV of the same name at the
188 same map_addr value. The relationship between the map_addr and any re‐
189 turned fi_addr_t is not defined.
190 If name is non-NULL and map_addr is 0, then the map_addr used by the
192 provider will be returned through the attribute structure. The map_ad‐
193 dr field is ignored if name is NULL.
194 flags The following flags may be used when opening an AV.
196 - FI_EVENT
198 When the flag FI_EVENT is specified, all insert operations on
199 this AV will occur asynchronously. There will be one EQ error
200 entry generated for each failed address insertion, followed by
201 one non-error event indicating that the insertion operation has
202 completed. There will always be one non-error completion event
203 for each insert operation, even if all addresses fail. The con‐
204 text field in all completions will be the context specified to
205 the insert call, and the data field in the final completion en‐
206 try will report the number of addresses successfully inserted.
207 If an error occurs during the asynchronous insertion, an error
208 completion entry is returned (see fi_eq(3) for a discussion of
209 the fi_eq_err_entry error completion struct). The context field
210 of the error completion will be the context that was specified
211 in the insert call; the data field will contain the index of the
212 failed address. There will be one error completion returned for
213 each address that fails to insert into the AV.
214 If an AV is opened with FI_EVENT, any insertions attempted before an EQ
216 is bound to the AV will fail with -FI_ENOEQ.
217 Error completions for failed insertions will contain the index of the
219 failed address in the index field of the error completion entry.
220 Note that the order of delivery of insert completions may not match the
222 order in which the calls to fi_av_insert were made. The only guarantee
223 is that all error completions for a given call to fi_av_insert will
224 precede the single associated non-error completion. • .RS 2
225 FI_READ
227 Opens an AV for read-only access. An AV opened for read-only
228 access must be named (name attribute specified), and the AV must
229 exist.
230 • .RS 2
231 FI_SYMMETRIC
233 Indicates that each node will be associated with the same number
234 of endpoints, the same transport addresses will be allocated on
235 each node, and the transport addresses will be sequential. This
236 feature targets distributed applications on large fabrics and
237 allows for highly-optimized storage of remote endpoint address‐
238 ing.
239 fi_close
241 The fi_close call is used to release all resources associated with an
242 address vector. Note that any events queued on an event queue refer‐
243 encing the AV are left untouched. It is recommended that callers re‐
244 trieve all events associated with the AV before closing it.
245 When closing the address vector, there must be no opened endpoints as‐
247 sociated with the AV. If resources are still associated with the AV
248 when attempting to close, the call will return -FI_EBUSY.
249 fi_av_bind
251 Associates an event queue with the AV. If an AV has been opened with
252 FI_EVENT, then an event queue must be bound to the AV before any inser‐
253 tion calls are attempted. Any calls to insert addresses before an
254 event queue has been bound will fail with -FI_ENOEQ. Flags are re‐
255 served for future use and must be 0.
256 fi_av_insert
258 The fi_av_insert call inserts zero or more addresses into an AV. The
259 number of addresses is specified through the count parameter. The addr
260 parameter references an array of addresses to insert into the AV. Ad‐
261 dresses inserted into an address vector must be in the same format as
262 specified in the addr_format field of the fi_info struct provided when
263 opening the corresponding domain. When using the FI_ADDR_STR format,
264 the addr parameter should reference an array of strings (char **).
265 For AV’s of type FI_AV_MAP, once inserted addresses have been mapped,
267 the mapped values are written into the buffer referenced by fi_addr.
268 The fi_addr buffer must remain valid until the AV insertion has com‐
269 pleted and an event has been generated to an associated event queue.
270 The value of the returned fi_addr should be considered opaque by the
271 application for AVs of type FI_AV_MAP. The returned value may point to
272 an internal structure or a provider specific encoding of low-level ad‐
273 dressing data, for example. In the latter case, use of FI_AV_MAP may
274 be able to avoid memory references during data transfer operations.
275 For AV’s of type FI_AV_TABLE, addresses are placed into the table in
277 order. An address is inserted at the lowest index that corresponds to
278 an unused table location, with indices starting at 0. That is, the
279 first address inserted may be referenced at index 0, the second at in‐
280 dex 1, and so forth. When addresses are inserted into an AV table, the
281 assigned fi_addr values will be simple indices corresponding to the en‐
282 try into the table where the address was inserted. Index values accu‐
283 mulate across successive insert calls in the order the calls are made,
284 not necessarily in the order the insertions complete.
285 Because insertions occur at a pre-determined index, the fi_addr parame‐
287 ter may be NULL. If fi_addr is non-NULL, it must reference an array of
288 fi_addr_t, and the buffer must remain valid until the insertion opera‐
289 tion completes. Note that if fi_addr is NULL and synchronous operation
290 is requested without using FI_SYNC_ERR flag, individual insertion fail‐
291 ures cannot be reported and the application must use other calls, such
292 as fi_av_lookup to learn which specific addresses failed to insert.
293 Since fi_av_remove is provider-specific, it is recommended that calls
294 to fi_av_insert following a call to fi_av_remove always reference a
295 valid buffer in the fi_addr parameter. Otherwise it may be difficult
296 to determine what the next assigned index will be.
297 flags The following flag may be passed to AV insertion calls:
299 fi_av_insert, fi_av_insertsvc, or fi_av_insertsym.
300 - FI_MORE
302 In order to allow optimized address insertion, the application
303 may specify the FI_MORE flag to the insert call to give a hint
304 to the provider that more insertion requests will follow, allow‐
305 ing the provider to aggregate insertion requests if desired. An
306 application may make any number of insertion calls with FI_MORE
307 set, provided that they are followed by an insertion call with‐
308 out FI_MORE. This signifies to the provider that the insertion
309 list is complete. Providers are free to ignore FI_MORE.
310 - FI_SYNC_ERR
312 This flag applies to synchronous insertions only, and is used to
313 retrieve error details of failed insertions. If set, the con‐
314 text parameter of insertion calls references an array of inte‐
315 gers, with context set to address of the first element of the
316 array. The resulting status of attempting to insert each ad‐
317 dress will be written to the corresponding array location. Suc‐
318 cessful insertions will be updated to 0. Failures will contain
319 a fabric errno code.
320 fi_av_insertsvc
322 The fi_av_insertsvc call behaves similar to fi_av_insert, but allows
323 the application to specify the node and service names, similar to the
324 fi_getinfo inputs, rather than an encoded address. The node and ser‐
325 vice parameters are defined the same as fi_getinfo(3). Node should be
326 a string that corresponds to a hostname or network address. The ser‐
327 vice string corresponds to a textual representation of a transport ad‐
328 dress. Applications may also pass in an FI_ADDR_STR formatted address
329 as the node parameter. In such cases, the service parameter must be
330 NULL. See fi_getinfo.3 for details on using FI_ADDR_STR. Supported
331 flags are the same as for fi_av_insert.
332 fi_av_insertsym
334 fi_av_insertsym performs a symmetric insert that inserts a sequential
335 range of nodes and/or service addresses into an AV. The svccnt parame‐
336 ter indicates the number of transport (endpoint) addresses to insert
337 into the AV for each node address, with the service parameter specify‐
338 ing the starting transport address. Inserted transport addresses will
339 be of the range {service, service + svccnt - 1}, inclusive. All ser‐
340 vice addresses for a node will be inserted before the next node is in‐
341 serted.
342 The nodecnt parameter indicates the number of node (network) addresses
344 to insert into the AV, with the node parameter specifying the starting
345 node address. Inserted node addresses will be of the range {node, node
346 + nodecnt - 1}, inclusive. If node is a non-numeric string, such as a
347 hostname, it must contain a numeric suffix if nodecnt > 1.
348 As an example, if node = “10.1.1.1”, nodecnt = 2, service = “5000”, and
350 svccnt = 2, the following addresses will be inserted into the AV in the
351 order shown: 10.1.1.1:5000, 10.1.1.1:5001, 10.1.1.2:5000,
352 10.1.1.2:5001. If node were replaced by the hostname “host10”, the ad‐
353 dresses would be: host10:5000, host10:5001, host11:5000, host11:5001.
354 The total number of inserted addresses will be nodecnt x svccnt.
356 Supported flags are the same as for fi_av_insert.
358 fi_av_remove
360 fi_av_remove removes a set of addresses from an address vector. All
361 resources associated with the indicated addresses are released. The
362 removed address - either the mapped address (in the case of FI_AV_MAP)
363 or index (FI_AV_TABLE) - is invalid until it is returned again by a new
364 fi_av_insert.
365 The behavior of operations in progress that reference the removed ad‐
367 dresses is undefined.
368 The use of fi_av_remove is an optimization that applications may use to
370 free memory allocated with addresses that will no longer be accessed.
371 Inserted addresses are not required to be removed. fi_av_close will
372 automatically cleanup any resources associated with addresses remaining
373 in the AV when it is invoked.
374 Flags are reserved for future use and must be 0.
376 fi_av_lookup
378 This call returns the address stored in the address vector that corre‐
379 sponds to the given fi_addr. The returned address is the same format
380 as those stored by the AV. On input, the addrlen parameter should in‐
381 dicate the size of the addr buffer. If the actual address is larger
382 than what can fit into the buffer, it will be truncated. On output,
383 addrlen is set to the size of the buffer needed to store the address,
384 which may be larger than the input value.
385 fi_rx_addr
387 This function is used to convert an endpoint address, returned by
388 fi_av_insert, into an address that specifies a target receive context.
389 The specified fi_addr parameter must either be a value returned from
390 fi_av_insert, in the case of FI_AV_MAP, or an index, in the case of
391 FI_AV_TABLE. The value for rx_ctx_bits must match that specified in
392 the AV attributes for the given address.
393 Connected endpoints that support multiple receive contexts, but are not
395 associated with address vectors should specify FI_ADDR_NOTAVAIL for the
396 fi_addr parameter.
397 fi_av_straddr
399 The fi_av_straddr function converts the provided address into a print‐
400 able string. The specified address must be of the same format as those
401 stored by the AV, though the address itself is not required to have
402 been inserted. On input, the len parameter should specify the size of
403 the buffer referenced by buf. On output, addrlen is set to the size of
404 the buffer needed to store the address. This size may be larger than
405 the input len. If the provided buffer is too small, the results will
406 be truncated. fi_av_straddr returns a pointer to buf.
407
409 An AV should only store a single instance of an address. Attempting to
410 insert a duplicate copy of the same address into an AV may result in
411 undefined behavior, depending on the provider implementation.
412 Providers are not required to check for duplicates, as doing so could
413 incur significant overhead to the insertion process. For portability,
414 applications may need to track which peer addresses have been inserted
415 into a given AV in order to avoid duplicate entries. However,
416 providers are required to support the removal, followed by the re-in‐
417 sertion of an address. Only duplicate insertions are restricted.
418 Providers may implement AV’s using a variety of mechanisms. Specifi‐
420 cally, a provider may begin resolving inserted addresses as soon as
421 they have been added to an AV, even if asynchronous operation has been
422 specified. Similarly, a provider may lazily release resources from re‐
423 moved entries.
424
426 Insertion calls for an AV opened for synchronous operation will return
427 the number of addresses that were successfully inserted. In the case
428 of failure, the return value will be less than the number of addresses
429 that was specified.
430 Insertion calls for an AV opened for asynchronous operation (with
432 FI_EVENT flag specified) will return 0 if the operation was successful‐
433 ly initiated. In the case of failure, a negative fabric errno will be
434 returned. Providers are allowed to abort insertion operations in the
435 case of an error. Addresses that are not inserted because they were
436 aborted will fail with an error code of FI_ECANCELED.
437 In both the synchronous and asynchronous modes of operation, the fi_ad‐
439 dr buffer associated with a failed or aborted insertion will be set to
440 FI_ADDR_NOTAVAIL.
441 All other calls return 0 on success, or a negative value corresponding
443 to fabric errno on error. Fabric errno values are defined in rd‐
444 ma/fi_errno.h.
445
447 fi_getinfo(3), fi_endpoint(3), fi_domain(3), fi_eq(3)
448
450 OpenFabrics.
451Libfabric Programmer’s Manual 2022-01-22 fi_av(3)