1fi_cntr(3) Libfabric v1.7.0 fi_cntr(3)
2
3
4
6 fi_cntr - Completion and event counter operations
7
8 fi_cntr_open / fi_close
9 Allocate/free a counter
10
11 fi_cntr_read
12 Read the current value of a counter
13
14 fi_cntr_readerr
15 Reads the number of operations which have completed in error.
16
17 fi_cntr_add
18 Increment a counter by a specified value
19
20 fi_cntr_set
21 Set a counter to a specified value
22
23 fi_cntr_wait
24 Wait for a counter to be greater or equal to a threshold value
25
27 #include <rdma/fi_domain.h>
28
29 int fi_cntr_open(struct fid_domain *domain, struct fi_cntr_attr *attr,
30 struct fid_cntr **cntr, void *context);
31
32 int fi_close(struct fid *cntr);
33
34 uint64_t fi_cntr_read(struct fid_cntr *cntr);
35
36 uint64_t fi_cntr_readerr(struct fid_cntr *cntr);
37
38 int fi_cntr_add(struct fid_cntr *cntr, uint64_t value);
39
40 int fi_cntr_adderr(struct fid_cntr *cntr, uint64_t value);
41
42 int fi_cntr_set(struct fid_cntr *cntr, uint64_t value);
43
44 int fi_cntr_seterr(struct fid_cntr *cntr, uint64_t value);
45
46 int fi_cntr_wait(struct fid_cntr *cntr, uint64_t threshold,
47 int timeout);
48
50 domain Fabric domain
51
52 cntr Fabric counter
53
54 attr Counter attributes
55
56 context
57 User specified context associated with the counter
58
59 value Value to increment or set counter
60
61 threshold
62 Value to compare counter against
63
64 timeout
65 Time in milliseconds to wait. A negative value indicates infi‐
66 nite timeout.
67
69 Counters record the number of requested operations that have completed.
70 Counters can provide a light-weight completion mechanism by suppressing
71 the generation of a full completion event. They are useful for appli‐
72 cations that only need to know the number of requests that have com‐
73 pleted, and not details about each request. For example, counters may
74 be useful for implementing credit based flow control or tracking the
75 number of remote processes which have responded to a request.
76
77 Counters typically only count successful completions. However, if an
78 operation completes in error, it may increment an associated error val‐
79 ue. That is, a counter actually stores two distinct values, with error
80 completions updating an error specific value.
81
82 fi_cntr_open
83 fi_cntr_open allocates a new fabric counter. The properties and behav‐
84 ior of the counter are defined by struct fi_cntr_attr.
85
86 struct fi_cntr_attr {
87 enum fi_cntr_events events; /* type of events to count */
88 enum fi_wait_obj wait_obj; /* requested wait object */
89 struct fid_wait *wait_set; /* optional wait set */
90 uint64_t flags; /* operation flags */
91 };
92
93 events A counter captures different types of events. The specific type
94 which is to counted are one of the following:
95
96 - FI_CNTR_EVENTS_COMP
97 The counter increments for every successful completion that oc‐
98 curs on an associated bound endpoint. The type of completions
99 -- sends and/or receives -- which are counted may be restricted
100 using control flags when binding the counter and the endpoint.
101 Counters increment on all successful completions, separately
102 from whether the operation generates an entry in an event queue.
103
104 wait_obj
105 Counters may be associated with a specific wait object. Wait
106 objects allow applications to block until the wait object is
107 signaled, indicating that a counter has reached a specific
108 threshold. Users may use fi_control to retrieve the underlying
109 wait object associated with a counter, in order to use it in
110 other system calls. The following values may be used to specify
111 the type of wait object associated with a counter: FI_WAIT_NONE,
112 FI_WAIT_UNSPEC, FI_WAIT_SET, FI_WAIT_FD, and FI_WAIT_MUTEX_COND.
113 The default is FI_WAIT_NONE.
114
115 - FI_WAIT_NONE
116 Used to indicate that the user will not block (wait) for events
117 on the counter.
118
119 - FI_WAIT_UNSPEC
120 Specifies that the user will only wait on the counter using fab‐
121 ric interface calls, such as fi_cntr_wait. In this case, the
122 underlying provider may select the most appropriate or highest
123 performing wait object available, including custom wait mecha‐
124 nisms. Applications that select FI_WAIT_UNSPEC are not guaran‐
125 teed to retrieve the underlying wait object.
126
127 - FI_WAIT_SET
128 Indicates that the event counter should use a wait set object to
129 wait for events. If specified, the wait_set field must refer‐
130 ence an existing wait set object.
131
132 - FI_WAIT_FD
133 Indicates that the counter should use a file descriptor as its
134 wait mechanism. A file descriptor wait object must be usable in
135 select, poll, and epoll routines. However, a provider may sig‐
136 nal an FD wait object by marking it as readable, writable, or
137 with an error.
138
139 - FI_WAIT_MUTEX_COND
140 Specifies that the counter should use a pthread mutex and cond
141 variable as a wait object.
142
143 wait_set
144 If wait_obj is FI_WAIT_SET, this field references a wait object
145 to which the event counter should attach. When an event is
146 added to the event counter, the corresponding wait set will be
147 signaled if all necessary conditions are met. The use of a
148 wait_set enables an optimized method of waiting for events
149 across multiple event counters. This field is ignored if
150 wait_obj is not FI_WAIT_SET.
151
152 flags Flags are reserved for future use, and must be set to 0.
153
154 fi_close
155 The fi_close call releases all resources associated with a counter.
156 When closing the counter, there must be no opened endpoints, transmit
157 contexts, receive contexts or memory regions associated with the
158 counter. If resources are still associated with the counter when at‐
159 tempting to close, the call will return -FI_EBUSY.
160
161 fi_cntr_control
162 The fi_cntr_control call is used to access provider or implementation
163 specific details of the counter. Access to the counter should be seri‐
164 alized across all calls when fi_cntr_control is invoked, as it may re‐
165 direct the implementation of counter operations. The following control
166 commands are usable with a counter:
167
168 FI_GETOPSFLAG (uint64_t *)
169 Returns the current default operational flags associated with
170 the counter.
171
172 FI_SETOPSFLAG (uint64_t *)
173 Modifies the current default operational flags associated with
174 the counter.
175
176 FI_GETWAIT (void **)
177 This command allows the user to retrieve the low-level wait ob‐
178 ject associated with the counter. The format of the wait-object
179 is specified during counter creation, through the counter at‐
180 tributes. See fi_eq.3 for addition details using control with
181 FI_GETWAIT.
182
183 fi_cntr_read
184 The fi_cntr_read call returns the current value of the counter.
185
186 fi_cntr_readerr
187 The read error call returns the number of operations that completed in
188 error and were unable to update the counter.
189
190 fi_cntr_add
191 This adds the user-specified value to the counter.
192
193 fi_cntr_adderr
194 This adds the user-specified value to the error value of the counter.
195
196 fi_cntr_set
197 This sets the counter to the specified value.
198
199 fi_cntr_seterr
200 This sets the error value of the counter to the specified value.
201
202 fi_cntr_wait
203 This call may be used to wait until the counter reaches the specified
204 threshold, or until an error or timeout occurs. Upon successful return
205 from this call, the counter will be greater than or equal to the input
206 threshold value.
207
208 If an operation associated with the counter encounters an error, it
209 will increment the error value associated with the counter. Any change
210 in a counter's error value will unblock any thread inside fi_cntr_wait.
211
212 If the call returns due to timeout, -FI_ETIMEDOUT will be returned.
213 The error value associated with the counter remains unchanged.
214
215 It is invalid for applications to call this function if the counter has
216 been configured with a wait object of FI_WAIT_NONE or FI_WAIT_SET.
217
219 Returns 0 on success. On error, a negative value corresponding to fab‐
220 ric errno is returned.
221
222 fi_cntr_read / fi_cntr_readerr
223 Returns the current value of the counter.
224
225 Fabric errno values are defined in rdma/fi_errno.h.
226
228 In order to support a variety of counter implementations, updates made
229 to counter values (e.g. fi_cntr_set or fi_cntr_add) may not be immedi‐
230 ately visible to counter read operations (i.e. fi_cntr_read or fi_cn‐
231 tr_readerr). A small, but undefined, delay may occur between the
232 counter changing and the reported value being updated. However, a fi‐
233 nal updated value will eventually be reflected in the read counter val‐
234 ue.
235
236 Additionally, applications should ensure that the value of a counter is
237 stable and not subject to change prior to calling fi_cntr_set or fi_cn‐
238 tr_seterr. Otherwise, the resulting value of the counter after fi_cn‐
239 tr_set / fi_cntr_seterr is undefined, as updates to the counter may be
240 lost. A counter value is considered stable if all previous updates us‐
241 ing fi_cntr_set / fi_cntr_seterr and results of related operations are
242 reflected in the observed value of the counter.
243
245 fi_getinfo(3), fi_endpoint(3), fi_domain(3), fi_eq(3), fi_poll(3)
246
248 OpenFabrics.
249
250
251
252Libfabric Programmer's Manual 2018-10-05 fi_cntr(3)