* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2015, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*/
#ifndef __UAPI_LNET_TYPES_H__
/** \addtogroup lnet
* @{ */
-#include <linux/types.h>
+#include <linux/lnet/lnet-idl.h>
+
/** \addtogroup lnet_addr
* @{ */
*/
#define LNET_RESERVED_PORTAL 0
-/**
- * Address of an end-point in an LNet network.
- *
- * A node can have multiple end-points and hence multiple addresses.
- * An LNet network can be a simple network (e.g. tcp0) or a network of
- * LNet networks connected by LNet routers. Therefore an end-point address
- * has two parts: network ID, and address within a network.
- *
- * \see LNET_NIDNET, LNET_NIDADDR, and LNET_MKNID.
- */
-typedef __u64 lnet_nid_t;
-
-/**
- * ID of a process in a node. Shortened as PID to distinguish from
- * lnet_process_id, the global process ID.
- */
-typedef __u32 lnet_pid_t;
-
/** wildcard NID that matches any end-point address */
#define LNET_NID_ANY ((lnet_nid_t) -1)
/** wildcard PID that matches any lnet_pid_t */
return (type << 16) | num;
}
-#define WIRE_ATTR __attribute__((packed))
-
-/* Packed version of struct lnet_process_id to transfer via network */
-typedef struct lnet_process_id_packed {
- lnet_nid_t nid;
- lnet_pid_t pid; /* node id / process id */
-} WIRE_ATTR lnet_process_id_packed;
-
-/* The wire handle's interface cookie only matches one network interface in
- * one epoch (i.e. new cookie when the interface restarts or the node
- * reboots). The object cookie only matches one object on that interface
- * during that object's lifetime (i.e. no cookie re-use). */
-typedef struct lnet_handle_wire {
- __u64 wh_interface_cookie;
- __u64 wh_object_cookie;
-} WIRE_ATTR lnet_handle_wire_t;
-
-typedef enum lnet_msg_type {
- LNET_MSG_ACK = 0,
- LNET_MSG_PUT,
- LNET_MSG_GET,
- LNET_MSG_REPLY,
- LNET_MSG_HELLO,
-} lnet_msg_type_t;
-
-/* The variant fields of the portals message header are aligned on an 8
- * byte boundary in the message header. Note that all types used in these
- * wire structs MUST be fixed size and the smaller types are placed at the
- * end. */
-struct lnet_ack {
- struct lnet_handle_wire dst_wmd;
- __u64 match_bits;
- __u32 mlength;
-} WIRE_ATTR;
-
-struct lnet_put {
- struct lnet_handle_wire ack_wmd;
- __u64 match_bits;
- __u64 hdr_data;
- __u32 ptl_index;
- __u32 offset;
-} WIRE_ATTR;
-
-struct lnet_get {
- struct lnet_handle_wire return_wmd;
- __u64 match_bits;
- __u32 ptl_index;
- __u32 src_offset;
- __u32 sink_length;
-} WIRE_ATTR;
-
-struct lnet_reply {
- struct lnet_handle_wire dst_wmd;
-} WIRE_ATTR;
-
-struct lnet_hello {
- __u64 incarnation;
- __u32 type;
-} WIRE_ATTR;
-
-typedef struct lnet_hdr {
- lnet_nid_t dest_nid;
- lnet_nid_t src_nid;
- lnet_pid_t dest_pid;
- lnet_pid_t src_pid;
- __u32 type; /* enum lnet_msg_type */
- __u32 payload_length; /* payload data to follow */
- /*<------__u64 aligned------->*/
- union {
- struct lnet_ack ack;
- struct lnet_put put;
- struct lnet_get get;
- struct lnet_reply reply;
- struct lnet_hello hello;
- } msg;
-} WIRE_ATTR lnet_hdr_t;
-
-/* A HELLO message contains a magic number and protocol version
- * code in the header's dest_nid, the peer's NID in the src_nid, and
- * LNET_MSG_HELLO in the type field. All other common fields are zero
- * (including payload_size; i.e. no payload).
- * This is for use by byte-stream LNDs (e.g. TCP/IP) to check the peer is
- * running the same protocol and to find out its NID. These LNDs should
- * exchange HELLO messages when a connection is first established. Individual
- * LNDs can put whatever else they fancy in lnet_hdr::msg.
- */
-typedef struct lnet_magicversion {
- __u32 magic; /* LNET_PROTO_TCP_MAGIC */
- __u16 version_major; /* increment on incompatible change */
- __u16 version_minor; /* increment on compatible change */
-} WIRE_ATTR lnet_magic_version_t;
-
-/* PROTO MAGIC for LNDs */
-#define LNET_PROTO_IB_MAGIC 0x0be91b91
-#define LNET_PROTO_GNI_MAGIC 0xb00fbabe /* ask Kim */
-#define LNET_PROTO_TCP_MAGIC 0xeebc0ded
-#define LNET_PROTO_ACCEPTOR_MAGIC 0xacce7100
-#define LNET_PROTO_PING_MAGIC 0x70696E67 /* 'ping' */
-
-/* Placeholder for a future "unified" protocol across all LNDs */
-/* Current LNDs that receive a request with this magic will respond
- * with a "stub" reply using their current protocol */
-#define LNET_PROTO_MAGIC 0x45726963 /* ! */
-
-#define LNET_PROTO_TCP_VERSION_MAJOR 1
-#define LNET_PROTO_TCP_VERSION_MINOR 0
-
-/* Acceptor connection request */
-typedef struct lnet_acceptor_connreq {
- __u32 acr_magic; /* PTL_ACCEPTOR_PROTO_MAGIC */
- __u32 acr_version; /* protocol version */
- __u64 acr_nid; /* target NID */
-} WIRE_ATTR lnet_acceptor_connreq_t;
-
-#define LNET_PROTO_ACCEPTOR_VERSION 1
-
-typedef struct lnet_counters {
- __u32 msgs_alloc;
- __u32 msgs_max;
- __u32 errors;
- __u32 send_count;
- __u32 recv_count;
- __u32 route_count;
- __u32 drop_count;
- __u64 send_length;
- __u64 recv_length;
- __u64 route_length;
- __u64 drop_length;
-} WIRE_ATTR lnet_counters_t;
-
-#define LNET_NI_STATUS_UP 0x15aac0de
-#define LNET_NI_STATUS_DOWN 0xdeadface
-#define LNET_NI_STATUS_INVALID 0x00000000
-
-struct lnet_ni_status {
- lnet_nid_t ns_nid;
- __u32 ns_status;
- __u32 ns_unused;
-} WIRE_ATTR;
-
-/*
- * NB: value of these features equal to LNET_PROTO_PING_VERSION_x
- * of old LNet, so there shouldn't be any compatibility issue
- */
-#define LNET_PING_FEAT_INVAL (0) /* no feature */
-#define LNET_PING_FEAT_BASE (1 << 0) /* just a ping */
-#define LNET_PING_FEAT_NI_STATUS (1 << 1) /* return NI status */
-#define LNET_PING_FEAT_RTE_DISABLED (1 << 2) /* Routing enabled */
-#define LNET_PING_FEAT_MULTI_RAIL (1 << 3) /* Multi-Rail aware */
-#define LNET_PING_FEAT_DISCOVERY (1 << 4) /* Supports Discovery */
-
-/*
- * All ping feature bits fit to hit the wire.
- * In lnet_assert_wire_constants() this is compared against its open-coded
- * value, and in lnet_ping_target_update() it is used to verify that no
- * unknown bits have been set.
- * New feature bits can be added, just be aware that this does change the
- * over-the-wire protocol.
- */
-#define LNET_PING_FEAT_BITS (LNET_PING_FEAT_BASE | \
- LNET_PING_FEAT_NI_STATUS | \
- LNET_PING_FEAT_RTE_DISABLED | \
- LNET_PING_FEAT_MULTI_RAIL | \
- LNET_PING_FEAT_DISCOVERY)
-
-struct lnet_ping_info {
- __u32 pi_magic;
- __u32 pi_features;
- lnet_pid_t pi_pid;
- __u32 pi_nnis;
- struct lnet_ni_status pi_ni[0];
-} WIRE_ATTR;
-
-#define LNET_PING_INFO_SIZE(NNIDS) \
- offsetof(struct lnet_ping_info, pi_ni[NNIDS])
-#define LNET_PING_INFO_LONI(PINFO) ((PINFO)->pi_ni[0].ns_nid)
-#define LNET_PING_INFO_SEQNO(PINFO) ((PINFO)->pi_ni[0].ns_status)
+/** The lolnd NID (i.e. myself) */
+#define LNET_NID_LO_0 LNET_MKNID(LNET_MKNET(LOLND, 0), 0)
+
+#define LNET_NET_ANY LNET_NIDNET(LNET_NID_ANY)
+
+struct lnet_counters_health {
+ __u32 lch_rst_alloc;
+ __u32 lch_resend_count;
+ __u32 lch_response_timeout_count;
+ __u32 lch_local_interrupt_count;
+ __u32 lch_local_dropped_count;
+ __u32 lch_local_aborted_count;
+ __u32 lch_local_no_route_count;
+ __u32 lch_local_timeout_count;
+ __u32 lch_local_error_count;
+ __u32 lch_remote_dropped_count;
+ __u32 lch_remote_error_count;
+ __u32 lch_remote_timeout_count;
+ __u32 lch_network_timeout_count;
+};
+
+struct lnet_counters {
+ struct lnet_counters_common lct_common;
+ struct lnet_counters_health lct_health;
+};
/*
* This is a hard-coded limit on the number of interfaces supported by
/**
* Objects maintained by the LNet are accessed through handles. Handle types
* have names of the form lnet_handle_xx, where xx is one of the two letter
- * object type codes ('eq' for event queue, 'md' for memory descriptor, and
+ * object type codes ('md' for memory descriptor, and
* 'me' for match entry). Each type of object is given a unique handle type
* to enhance type checking.
*/
-#define LNET_WIRE_HANDLE_COOKIE_NONE (-1)
+#define LNET_WIRE_HANDLE_COOKIE_NONE (~0ULL)
-typedef struct lnet_handle_eq {
+struct lnet_handle_md {
__u64 cookie;
-} lnet_handle_eq_t;
-
-/**
- * Invalidate eq handle \a h.
- */
-static inline void LNetInvalidateEQHandle(struct lnet_handle_eq *h)
-{
- h->cookie = LNET_WIRE_HANDLE_COOKIE_NONE;
-}
-
-/**
- * Check whether eq handle \a h is invalid.
- *
- * \return 1 if handle is invalid, 0 if valid.
- */
-static inline int LNetEQHandleIsInvalid(struct lnet_handle_eq h)
-{
- return (LNET_WIRE_HANDLE_COOKIE_NONE == h.cookie);
-}
-
-typedef struct lnet_handle_md {
- __u64 cookie;
-} lnet_handle_md_t;
+};
/**
* Invalidate md handle \a h.
}
/**
- * Check whether eq handle \a h is invalid.
+ * Check whether handler \a h is invalid.
*
* \return 1 if handle is invalid, 0 if valid.
*/
return (LNET_WIRE_HANDLE_COOKIE_NONE == h.cookie);
}
-typedef struct lnet_handle_me {
- __u64 cookie;
-} lnet_handle_me_t;
-
/**
* Global process ID.
*/
-typedef struct lnet_process_id {
+struct lnet_process_id {
/** node id */
lnet_nid_t nid;
/** process id */
lnet_pid_t pid;
-} lnet_process_id_t;
+};
/** @} lnet_addr */
/** \addtogroup lnet_me
* Specifies whether the match entry or memory descriptor should be unlinked
* automatically (LNET_UNLINK) or not (LNET_RETAIN).
*/
-typedef enum lnet_unlink {
+enum lnet_unlink {
LNET_RETAIN = 0,
LNET_UNLINK
-} lnet_unlink_t;
+};
/**
* Values of the type enum lnet_ins_pos are used to control where a new match
* LNET_INS_AFTER is used to insert the new entry after the current entry
* or after the last item in the list.
*/
-typedef enum lnet_ins_pos {
+enum lnet_ins_pos {
/** insert ME before current position or head of the list */
LNET_INS_BEFORE,
/** insert ME after current position or tail of the list */
LNET_INS_AFTER,
/** attach ME at tail of local CPU partition ME list */
LNET_INS_LOCAL
-} lnet_ins_pos;
+};
/** @} lnet_me */
* @{ */
/**
+ * Event queue handler function type.
+ *
+ * The EQ handler runs for each event that is deposited into the EQ. The
+ * handler is supplied with a pointer to the event that triggered the
+ * handler invocation.
+ *
+ * The handler must not block, must be reentrant, and must not call any LNet
+ * API functions. It should return as quickly as possible.
+ */
+struct lnet_event;
+typedef void (*lnet_handler_t)(struct lnet_event *event);
+
+/**
* Defines the visible parts of a memory descriptor. Values of this type
* are used to initialize memory descriptors.
*/
-typedef struct lnet_md {
+struct lnet_md {
/**
* Specify the memory region associated with the memory descriptor.
* If the options field has:
* - LNET_MD_KIOV bit set: The start field points to the starting
- * address of an array of lnet_kiov_t and the length field specifies
+ * address of an array of struct bio_vec and the length field specifies
* the number of entries in the array. The length can't be bigger
- * than LNET_MAX_IOV. The lnet_kiov_t is used to describe page-based
+ * than LNET_MAX_IOV. The struct bio_vec is used to describe page-based
* fragments that are not necessarily mapped in virtal memory.
- * - LNET_MD_IOVEC bit set: The start field points to the starting
- * address of an array of struct kvec and the length field specifies
- * the number of entries in the array. The length can't be bigger
- * than LNET_MAX_IOV. The struct kvec is used to describe fragments
- * that have virtual addresses.
* - Otherwise: The memory region is contiguous. The start field
* specifies the starting address for the memory region and the
* length field specifies its length.
* acknowledgment. Acknowledgments are never sent for GET operations.
* The data sent in the REPLY serves as an implicit acknowledgment.
* - LNET_MD_KIOV: The start and length fields specify an array of
- * lnet_kiov_t.
- * - LNET_MD_IOVEC: The start and length fields specify an array of
- * struct iovec.
+ * struct bio_vec.
* - LNET_MD_MAX_SIZE: The max_size field is valid.
* - LNET_MD_BULK_HANDLE: The bulk_handle field is valid.
+ * - LNET_MD_TRACK_RESPONSE: Enable response tracking on this MD
+ * regardless of the value of the lnet_response_tracking param.
+ * - LNET_MD_NO_TRACK_RESPONSE: Disable response tracking on this MD
+ * regardless of the value of the lnet_response_tracking param.
*
* Note:
- * - LNET_MD_KIOV or LNET_MD_IOVEC allows for a scatter/gather
- * capability for memory descriptors. They can't be both set.
+ * - LNET_MD_KIOV allows for a scatter/gather capability for memory
+ * descriptors.
* - When LNET_MD_MAX_SIZE is set, the total length of the memory
* region (i.e. sum of all fragment lengths) must not be less than
* \a max_size.
*/
void *user_ptr;
/**
- * A handle for the event queue used to log the operations performed on
- * the memory region. If this argument is a NULL handle (i.e. nullified
- * by LNetInvalidateHandle()), operations performed on this memory
- * descriptor are not logged.
+ * The event handler used to log the operations performed on
+ * the memory region. If this argument is NULL operations
+ * performed on this memory descriptor are not logged.
*/
- struct lnet_handle_eq eq_handle;
+ lnet_handler_t handler;
/**
* The bulk MD handle which was registered to describe the buffers
* either to be used to transfer data to the peer or receive data
* if the LNET_MD_BULK_HANDLE option is set.
*/
struct lnet_handle_md bulk_handle;
-} lnet_md_t;
+};
/* Max Transfer Unit (minimum supported everywhere).
* CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks)
#define LNET_MTU_BITS 20
#define LNET_MTU (1 << LNET_MTU_BITS)
-/** limit on the number of fragments in discontiguous MDs */
-#define LNET_MAX_IOV 256
-
/**
* Options for the MD structure. See struct lnet_md::options.
*/
/** See struct lnet_md::options. */
#define LNET_MD_ACK_DISABLE (1 << 5)
/** See struct lnet_md::options. */
-#define LNET_MD_IOVEC (1 << 6)
+/* deprecated #define LNET_MD_IOVEC (1 << 6) */
/** See struct lnet_md::options. */
#define LNET_MD_MAX_SIZE (1 << 7)
/** See struct lnet_md::options. */
#define LNET_MD_KIOV (1 << 8)
/** See struct lnet_md::options. */
#define LNET_MD_BULK_HANDLE (1 << 9)
-
-/* For compatibility with Cray Portals */
-#define LNET_MD_PHYS 0
+/** See struct lnet_md::options. */
+#define LNET_MD_TRACK_RESPONSE (1 << 10)
+/** See struct lnet_md::options. */
+#define LNET_MD_NO_TRACK_RESPONSE (1 << 11)
/** Infinite threshold on MD operations. See struct lnet_md::threshold */
#define LNET_MD_THRESH_INF (-1)
-/**
- * A page-based fragment of a MD.
- */
-typedef struct {
- /** Pointer to the page where the fragment resides */
- struct page *kiov_page;
- /** Length in bytes of the fragment */
- unsigned int kiov_len;
- /**
- * Starting offset of the fragment within the page. Note that the
- * end of the fragment must not pass the end of the page; i.e.,
- * kiov_len + kiov_offset <= PAGE_SIZE.
- */
- unsigned int kiov_offset;
-} lnet_kiov_t;
/** @} lnet_md */
/** \addtogroup lnet_eq
/**
* Six types of events can be logged in an event queue.
*/
-typedef enum lnet_event_kind {
+enum lnet_event_kind {
/** An incoming GET operation has completed on the MD. */
LNET_EVENT_GET = 1,
/**
* \see LNetMDUnlink
*/
LNET_EVENT_UNLINK,
-} lnet_event_kind_t;
+};
#define LNET_SEQ_GT(a, b) (((signed long)((a) - (b))) > 0)
/**
* Information about an event on a MD.
*/
-typedef struct lnet_event {
+struct lnet_event {
/** The identifier (nid, pid) of the target. */
struct lnet_process_id target;
/** The identifier (nid, pid) of the initiator. */
*/
struct lnet_handle_md md_handle;
/**
- * A snapshot of the state of the MD immediately after the event has
- * been processed. In particular, the threshold field in md will
- * reflect the value of the threshold after the operation occurred.
+ * A snapshot of relevant state of the MD immediately after the event
+ * has been processed.
*/
- struct lnet_md md;
+ void *md_start;
+ void *md_user_ptr;
+ unsigned int md_options;
/**
* 64 bits of out-of-band user data. Only valid for LNET_EVENT_PUT.
* \see LNetPut
* to each event.
*/
volatile unsigned long sequence;
-} lnet_event_t;
-
-/**
- * Event queue handler function type.
- *
- * The EQ handler runs for each event that is deposited into the EQ. The
- * handler is supplied with a pointer to the event that triggered the
- * handler invocation.
- *
- * The handler must not block, must be reentrant, and must not call any LNet
- * API functions. It should return as quickly as possible.
- */
-typedef void (*lnet_eq_handler_t)(struct lnet_event *event);
-#define LNET_EQ_HANDLER_NONE NULL
-/** @} lnet_eq */
+};
/** \addtogroup lnet_data
* @{ */
* \see struct lnet_md::options for the discussion on LNET_MD_ACK_DISABLE
* by which acknowledgments can be disabled for a MD.
*/
-typedef enum lnet_ack_req {
+enum lnet_ack_req {
/** Request an acknowledgment */
LNET_ACK_REQ,
/** Request that no acknowledgment should be generated. */
LNET_NOACK_REQ
-} lnet_ack_req_t;
+};
+
+/**
+ * UDSP action types. There are two available actions:
+ * 1. PRIORITY - set priority of matching LNet constructs
+ * 2. PREFERRED LIST - set preferred list of matching LNet constructs
+ */
+enum lnet_udsp_action_type {
+ EN_LNET_UDSP_ACTION_NONE = 0,
+ /** assign a priority to matching constructs */
+ EN_LNET_UDSP_ACTION_PRIORITY = 1,
+ /** assign a preferred list of NIDs to matching constructs */
+ EN_LNET_UDSP_ACTION_PREFERRED_LIST = 2,
+};
+
/** @} lnet_data */
/** @} lnet */