| fi_eq(3) | @VERSION@ | fi_eq(3) |
fi_eq - Event queue operations
#include <rdma/fi_domain.h>
int fi_eq_open(struct fid_fabric *fabric, struct fi_eq_attr *attr,
struct fid_eq **eq, void *context);
int fi_close(struct fid *eq);
int fi_control(struct fid *eq, int command, void *arg);
ssize_t fi_eq_read(struct fid_eq *eq, uint32_t *event,
void *buf, size_t len, uint64_t flags);
ssize_t fi_eq_readerr(struct fid_eq *eq, struct fi_eq_err_entry *buf,
uint64_t flags);
ssize_t fi_eq_write(struct fid_eq *eq, uint32_t event,
const void *buf, size_t len, uint64_t flags);
ssize_t fi_eq_sread(struct fid_eq *eq, uint32_t *event,
void *buf, size_t len, int timeout, uint64_t flags);
const char * fi_eq_strerror(struct fid_eq *eq, int prov_errno,
const void *err_data, char *buf, size_t len);
Event queues are used to report events associated with control operations. They are associated with memory registration, address vectors, connection management, and fabric and domain level events. Reported events are either associated with a requested operation or affiliated with a call that registers for specific types of events, such as listening for connection requests.
fi_eq_open allocates a new event queue.
The properties and behavior of an event queue are defined by struct fi_eq_attr.
struct fi_eq_attr {
size_t size; /* # entries for EQ */
uint64_t flags; /* operation flags */
enum fi_wait_obj wait_obj; /* requested wait object */
int signaling_vector; /* interrupt affinity */
struct fid_wait *wait_set; /* optional wait set */
};
The fi_close call releases all resources associated with an event queue. Any events which remain on the EQ when it is closed are lost.
The EQ must not be bound to any other objects prior to being closed, otherwise the call will return -FI_EBUSY.
The fi_control call is used to access provider or implementation specific details of the event queue. Access to the EQ should be serialized across all calls when fi_control is invoked, as it may redirect the implementation of EQ operations. The following control commands are usable with an EQ.
struct fi_mutex_cond {
pthread_mutex_t *mutex;
pthread_cond_t *cond;
};
The fi_eq_read operations performs a non-blocking read of event data from the EQ. The format of the event data is based on the type of event retrieved from the EQ, with all events starting with a struct fi_eq_entry header. At most one event will be returned per EQ read operation. The number of bytes successfully read from the EQ is returned from the read. The FI_PEEK flag may be used to indicate that event data should be read from the EQ without being consumed. A subsequent read without the FI_PEEK flag would then remove the event from the EQ.
The following types of events may be reported to an EQ, along with information regarding the format associated with each event.
Control requests report their completion by inserting a struct fi_eq_entry into the EQ. The format of this structure is:
struct fi_eq_entry {
fid_t fid; /* fid associated with request */
void *context; /* operation context */
uint64_t data; /* completion-specific data */
};
For the completion of basic asynchronous control operations, the returned event will indicate the operation that has completed, and the fid will reference the fabric descriptor associated with the event. For memory registration, this will be an FI_MR_COMPLETE event and the fid_mr. Address resolution will reference an FI_AV_COMPLETE event and fid_av. Multicast joins will report an FI_JOIN_COMPLETE and fid_mc. The context field will be set to the context specified as part of the operation, if available, otherwise the context will be associated with the fabric descriptor. The data field will be set as described in the man page for the corresponding object type (e.g., see fi_av(3) for a description of how asynchronous address vector insertions are completed).
struct fi_eq_cm_entry {
fid_t fid; /* fid associated with request */
struct fi_info *info; /* endpoint information */
uint8_t data[]; /* app connection data */
};
A connection request (FI_CONNREQ) event indicates that a remote endpoint wishes to establish a new connection to a listening, or passive, endpoint. The fid is the passive endpoint. Information regarding the requested, active endpoint's capabilities and attributes are available from the info field. The application is responsible for freeing this structure by calling fi_freeinfo when it is no longer needed. The fi_info connreq field will reference the connection request associated with this event. To accept a connection, an endpoint must first be created by passing an fi_info structure referencing this connreq field to fi_endpoint(). This endpoint is then passed to fi_accept() to complete the acceptance of the connection attempt. Creating the endpoint is most easily accomplished by passing the fi_info returned as part of the CM event into fi_endpoint(). If the connection is to be rejected, the connreq is passed to fi_reject().
Any application data exchanged as part of the connection request is placed beyond the fi_eq_cm_entry structure. The amount of data available is application dependent and limited to the buffer space provided by the application when fi_eq_read is called. The amount of returned data may be calculated using the return value to fi_eq_read. Note that the amount of returned data is limited by the underlying connection protocol, and the length of any data returned may include protocol padding. As a result, the returned length may be larger than that specified by the connecting peer.
If a connection request has been accepted, an FI_CONNECTED event will be generated on both sides of the connection. The active side -- one that called fi_connect() -- may receive user data as part of the FI_CONNECTED event. The user data is passed to the connection manager on the passive side through the fi_accept call. User data is not provided with an FI_CONNECTED event on the listening side of the connection.
Notification that a remote peer has disconnected from an active endpoint is done through the FI_SHUTDOWN event. Shutdown notification uses struct fi_eq_cm_entry as declared above. The fid field for a shutdown notification refers to the active endpoint's fid_ep.
Asynchronous errors may be reported for completion queues and endpoints of all types. CQ errors can result when resource management has been disabled, and the provider has detected a queue overrun. Endpoint errors may be result of numerous actions, but are often associated with a failed operation. Operations may fail because of buffer overruns, invalid permissions, incorrect memory access keys, network routing failures, network reach-ability issues, etc.
Asynchronous errors are reported using struct fi_eq_err_entry, as defined below. The fabric descriptor (fid) associated with the error is provided as part of the error data. An error code is also available to determine the cause of the error.
The fi_eq_sread call is the blocking (or synchronous) equivalent to fi_eq_read. It behaves is similar to the non-blocking call, with the exception that the calls will not return until either an event has been read from the EQ or an error or timeout occurs. Specifying a negative timeout means an infinite timeout.
Threads blocking in this function will return to the caller if they are signaled by some external source. This is true even if the timeout has not occurred or was specified as infinite.
It is invalid for applications to call this function if the EQ has been configured with a wait object of FI_WAIT_NONE or FI_WAIT_SET.
The read error function, fi_eq_readerr, retrieves information regarding any asynchronous operation which has completed with an unexpected error. fi_eq_readerr is a non-blocking call, returning immediately whether an error completion was found or not.
EQs are optimized to report operations which have completed successfully. Operations which fail are reported 'out of band'. Such operations are retrieved using the fi_eq_readerr function. When an operation that completes with an unexpected error is inserted into an EQ, it is placed into a temporary error queue. Attempting to read from an EQ while an item is in the error queue results in an FI_EAVAIL failure. Applications may use this return code to determine when to call fi_eq_readerr.
Error information is reported to the user through struct fi_eq_err_entry. The format of this structure is defined below.
struct fi_eq_err_entry {
fid_t fid; /* fid associated with error */
void *context; /* operation context */
uint64_t data; /* completion-specific data */
int err; /* positive error code */
int prov_errno; /* provider error code */
void *err_data; /* additional error data */
size_t err_data_size; /* size of err_data */
};
The fid will reference the fabric descriptor associated with the event. For memory registration, this will be the fid_mr, address resolution will reference a fid_av, and CM events will refer to a fid_ep. The context field will be set to the context specified as part of the operation.
The data field will be set as described in the man page for the corresponding object type (e.g., see fi_av(3) for a description of how asynchronous address vector insertions are completed).
The general reason for the error is provided through the err field. Provider or operational specific error information may also be available through the prov_errno and err_data fields. Users may call fi_eq_strerror to convert provider specific error information into a printable string for debugging purposes.
On input, err_data_size indicates the size of the err_data buffer in bytes. On output, err_data_size will be set to the number of bytes copied to the err_data buffer. The err_data information is typically used with fi_eq_strerror to provide details about the type of error that occurred.
For compatibility purposes, if err_data_size is 0 on input, or the fabric was opened with release < 1.5, err_data will be set to a data buffer owned by the provider. The contents of the buffer will remain valid until a subsequent read call against the EQ. Applications must serialize access to the EQ when processing errors to ensure that the buffer referenced by err_data does not change.
The EQ entry data structures share many of the same fields. The meanings are the same or similar for all EQ structure formats, with specific details described below.
For compatibility purposes, if err_data_size is 0 on input, or the fabric was opened with release < 1.5, err_data will be set to a data buffer owned by the provider. The contents of the buffer will remain valid until a subsequent read call against the EQ. Applications must serialize access to the EQ when processing errors to ensure that the buffer referenced by err_data does no change.
If an event queue has been overrun, it will be placed into an 'overrun' state. Write operations against an overrun EQ will fail with -FI_EOVERRUN. Read operations will continue to return any valid, non-corrupted events, if available. After all valid events have been retrieved, any attempt to read the EQ will result in it returning an FI_EOVERRUN error event. Overrun event queues are considered fatal and may not be used to report additional events once the overrun occurs.
Fabric errno values are defined in rdma/fi_errno.h.
fi_getinfo(3), fi_endpoint(3), fi_domain(3), fi_cntr(3), fi_poll(3)
OpenFabrics.
| 2019-12-13 | Libfabric Programmer's Manual |