X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lnet%2Fklnds%2Fo2iblnd%2Fo2iblnd.h;h=26956a37677c36912a558cc79152aecb709156c3;hp=25382c2d51fe53e256894569406e4ce31869696f;hb=HEAD;hpb=d8792a7dab933def57f6069296234ad48ea0da09 diff --git a/lnet/klnds/o2iblnd/o2iblnd.h b/lnet/klnds/o2iblnd/o2iblnd.h index 25382c2..4d01da9 100644 --- a/lnet/klnds/o2iblnd/o2iblnd.h +++ b/lnet/klnds/o2iblnd/o2iblnd.h @@ -23,24 +23,47 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2016, Intel Corporation. + * Copyright (c) 2011, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ - * Lustre is a trademark of Sun Microsystems, Inc. * * lnet/klnds/o2iblnd/o2iblnd.h * * Author: Eric Barton */ -#ifdef HAVE_COMPAT_RDMA +#include +#include + +#if defined(EXTERNAL_OFED_BUILD) && !defined(HAVE_OFED_IB_DMA_MAP_SG_SANE) +#undef CONFIG_INFINIBAND_VIRT_DMA +#endif + +#if defined(NEED_LOCKDEP_IS_HELD_DISCARD_CONST) \ + && defined(CONFIG_LOCKDEP) \ + && defined(lockdep_is_held) +#undef lockdep_is_held + #define lockdep_is_held(lock) \ + lock_is_held((struct lockdep_map *)&(lock)->dep_map) +#endif + +#ifdef HAVE_OFED_COMPAT_RDMA #include + +#ifdef LINUX_3_17_COMPAT_H +#undef NEED_KTIME_GET_REAL_NS +#endif + +#define HAVE_NLA_PUT_U64_64BIT 1 +#define HAVE_NLA_PARSE_6_PARAMS 1 +#define HAVE_NETLINK_EXTACK 1 + +/* MOFED has its own bitmap_alloc backport */ +#define HAVE_BITMAP_ALLOC 1 + #endif -#include -#include -#include #include #include #include @@ -60,9 +83,6 @@ #include #include #include -#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,32) -#include -#endif #include #include @@ -70,31 +90,44 @@ #include #include #include +#ifdef HAVE_OFED_FMR_POOL_API #include +#endif #define DEBUG_SUBSYSTEM S_LND -#include -#include #include +#include +#include "o2iblnd-idl.h" + +enum kiblnd_ni_lnd_tunables_attr { + LNET_NET_O2IBLND_TUNABLES_ATTR_UNSPEC = 0, + + LNET_NET_O2IBLND_TUNABLES_ATTR_HIW_PEER_CREDITS, + LNET_NET_O2IBLND_TUNABLES_ATTR_CONCURRENT_SENDS, + LNET_NET_O2IBLND_TUNABLES_ATTR_MAP_ON_DEMAND, + LNET_NET_O2IBLND_TUNABLES_ATTR_FMR_POOL_SIZE, + LNET_NET_O2IBLND_TUNABLES_ATTR_FMR_FLUSH_TRIGGER, + LNET_NET_O2IBLND_TUNABLES_ATTR_FMR_CACHE, + LNET_NET_O2IBLND_TUNABLES_ATTR_NTX, + LNET_NET_O2IBLND_TUNABLES_ATTR_CONNS_PER_PEER, + LNET_NET_O2IBLND_TUNABLES_ATTR_LND_TIMEOUT, + LNET_NET_O2IBLND_TUNABLES_ATTR_LND_TOS, + __LNET_NET_O2IBLND_TUNABLES_ATTR_MAX_PLUS_ONE, +}; -#define IBLND_PEER_HASH_SIZE 101 /* # peer_ni lists */ -/* # scheduler loops before reschedule */ -#define IBLND_RESCHED 100 +#define LNET_NET_O2IBLND_TUNABLES_ATTR_MAX (__LNET_NET_O2IBLND_TUNABLES_ATTR_MAX_PLUS_ONE - 1) +#define IBLND_PEER_HASH_BITS 7 /* log2 of # peer_ni lists */ #define IBLND_N_SCHED 2 #define IBLND_N_SCHED_HIGH 4 -typedef struct -{ +struct kib_tunables { int *kib_dev_failover; /* HCA failover */ unsigned int *kib_service; /* IB service number */ - int *kib_min_reconnect_interval; /* first failed connection retry... */ - int *kib_max_reconnect_interval; /* ...exponentially increasing to this */ - int *kib_cksum; /* checksum kib_msg_t? */ + int *kib_cksum; /* checksum struct kib_msg? */ int *kib_timeout; /* comms timeout (seconds) */ int *kib_keepalive; /* keepalive timeout (seconds) */ - int *kib_ntx; /* # tx descs */ char **kib_default_ipif; /* default IPoIB interface */ int *kib_retry_count; int *kib_rnr_retry_count; @@ -103,31 +136,29 @@ typedef struct int *kib_use_priv_port; /* use privileged port for active connect */ /* # threads on each CPT */ int *kib_nscheds; -} kib_tunables_t; + int *kib_wrq_sge; /* # sg elements per wrq */ + int *kib_use_fastreg_gaps; /* enable discontiguous fastreg fragment support */ +}; -extern kib_tunables_t kiblnd_tunables; +extern struct kib_tunables kiblnd_tunables; +extern struct lnet_ioctl_config_o2iblnd_tunables kib_default_tunables; #define IBLND_MSG_QUEUE_SIZE_V1 8 /* V1 only : # messages/RDMAs in-flight */ #define IBLND_CREDIT_HIGHWATER_V1 7 /* V1 only : when eagerly to return credits */ #define IBLND_CREDITS_DEFAULT 8 /* default # of peer_ni credits */ -#define IBLND_CREDITS_MAX ((typeof(((kib_msg_t*) 0)->ibm_credits)) - 1) /* Max # of peer_ni credits */ +#define IBLND_CREDITS_MAX ((typeof(((struct kib_msg *) 0)->ibm_credits)) - 1) /* Max # of peer_ni credits */ -/* when eagerly to return credits */ -#define IBLND_CREDITS_HIGHWATER(t, v) ((v) == IBLND_MSG_VERSION_1 ? \ - IBLND_CREDIT_HIGHWATER_V1 : \ - t->lnd_peercredits_hiw) - -#ifdef HAVE_RDMA_CREATE_ID_5ARG -# define kiblnd_rdma_create_id(cb, dev, ps, qpt) rdma_create_id(current->nsproxy->net_ns, \ - cb, dev, \ - ps, qpt) +#ifdef HAVE_OFED_RDMA_CREATE_ID_5ARG +# define kiblnd_rdma_create_id(ns, cb, dev, ps, qpt) \ + rdma_create_id((ns) ? (ns) : &init_net, cb, dev, ps, qpt) #else -# ifdef HAVE_RDMA_CREATE_ID_4ARG -# define kiblnd_rdma_create_id(cb, dev, ps, qpt) rdma_create_id(cb, dev, \ - ps, qpt) +# ifdef HAVE_OFED_RDMA_CREATE_ID_4ARG +# define kiblnd_rdma_create_id(ns, cb, dev, ps, qpt) \ + rdma_create_id(cb, dev, ps, qpt) # else -# define kiblnd_rdma_create_id(cb, dev, ps, qpt) rdma_create_id(cb, dev, ps) +# define kiblnd_rdma_create_id(ns, cb, dev, ps, qpt) \ + rdma_create_id(cb, dev, ps) # endif #endif @@ -135,8 +166,10 @@ extern kib_tunables_t kiblnd_tunables; #define IBLND_OOB_CAPABLE(v) ((v) != IBLND_MSG_VERSION_1) #define IBLND_OOB_MSGS(v) (IBLND_OOB_CAPABLE(v) ? 2 : 0) -#define IBLND_MSG_SIZE (4<<10) /* max size of queued messages (inc hdr) */ -#define IBLND_MAX_RDMA_FRAGS LNET_MAX_IOV /* max # of fragments supported */ +/* max size of queued messages (inc hdr) */ +#define IBLND_MSG_SIZE (4<<10) +/* max # of fragments supported. + 1 for unaligned case */ +#define IBLND_MAX_RDMA_FRAGS (LNET_MAX_IOV + 1) /************************/ /* derived constants... */ @@ -155,10 +188,9 @@ extern kib_tunables_t kiblnd_tunables; /* WRs and CQEs (per connection) */ #define IBLND_RECV_WRS(c) IBLND_RX_MSGS(c) -#define IBLND_SEND_WRS(c) \ - ((c->ibc_max_frags + 1) * kiblnd_concurrent_sends(c->ibc_version, \ - c->ibc_peer->ibp_ni)) -#define IBLND_CQ_ENTRIES(c) (IBLND_RECV_WRS(c) + IBLND_SEND_WRS(c)) + +/* 2 = LNet msg + Transfer chain */ +#define IBLND_CQ_ENTRIES(c) (IBLND_RECV_WRS(c) + kiblnd_send_wrs(c)) struct kib_hca_dev; @@ -169,8 +201,19 @@ struct kib_hca_dev; #define KIB_IFNAME_SIZE 256 #endif -typedef struct -{ +enum kib_dev_caps { + IBLND_DEV_CAPS_FASTREG_ENABLED = BIT(0), + IBLND_DEV_CAPS_FASTREG_GAPS_SUPPORT = BIT(1), +#ifdef HAVE_OFED_FMR_POOL_API + IBLND_DEV_CAPS_FMR_ENABLED = BIT(2), +#endif +}; + +#define IS_FAST_REG_DEV(dev) \ + ((dev)->ibd_dev_caps & IBLND_DEV_CAPS_FASTREG_ENABLED) + + +struct kib_dev { struct list_head ibd_list; /* chain on kib_devs */ struct list_head ibd_fail_list; /* chain on kib_failed_devs */ __u32 ibd_ifip; /* IPoIB interface IP */ @@ -178,7 +221,7 @@ typedef struct char ibd_ifname[KIB_IFNAME_SIZE]; int ibd_nnets; /* # nets extant */ - cfs_time_t ibd_next_failover; + time64_t ibd_next_failover; /* # failover failures */ int ibd_failed_failover; /* failover in progress */ @@ -187,33 +230,41 @@ typedef struct unsigned int ibd_can_failover; struct list_head ibd_nets; struct kib_hca_dev *ibd_hdev; -} kib_dev_t; + enum kib_dev_caps ibd_dev_caps; +}; -typedef struct kib_hca_dev -{ +struct kib_hca_dev { struct rdma_cm_id *ibh_cmid; /* listener cmid */ struct ib_device *ibh_ibdev; /* IB device */ int ibh_page_shift; /* page shift of current HCA */ int ibh_page_size; /* page size of current HCA */ __u64 ibh_page_mask; /* page mask of current HCA */ - int ibh_mr_shift; /* bits shift of max MR size */ __u64 ibh_mr_size; /* size of MR */ + int ibh_max_qp_wr; /* maximum work requests size */ +#ifdef HAVE_OFED_IB_GET_DMA_MR struct ib_mr *ibh_mrs; /* global MR */ +#endif struct ib_pd *ibh_pd; /* PD */ - kib_dev_t *ibh_dev; /* owner */ + u8 ibh_port; /* port number */ + struct ib_event_handler + ibh_event_handler; /* IB event handler */ + int ibh_state; /* device status */ +#define IBLND_DEV_PORT_DOWN 0 +#define IBLND_DEV_PORT_ACTIVE 1 +#define IBLND_DEV_FATAL 2 + struct kib_dev *ibh_dev; /* owner */ atomic_t ibh_ref; /* refcount */ -} kib_hca_dev_t; +}; /** # of seconds to keep pool alive */ #define IBLND_POOL_DEADLINE 300 /** # of seconds to retry if allocation failed */ #define IBLND_POOL_RETRY 1 -typedef struct -{ +struct kib_pages { int ibp_npages; /* # pages */ struct page *ibp_pages[0]; /* page array */ -} kib_pages_t; +}; struct kib_pool; struct kib_poolset; @@ -228,8 +279,7 @@ struct kib_net; #define IBLND_POOL_NAME_LEN 32 -typedef struct kib_poolset -{ +struct kib_poolset { /* serialize */ spinlock_t ps_lock; /* network it belongs to */ @@ -241,7 +291,7 @@ typedef struct kib_poolset /* failed pool list */ struct list_head ps_failed_pool_list; /* time stamp for retry if failed to allocate */ - cfs_time_t ps_next_retry; + time64_t ps_next_retry; /* is allocating new pool */ int ps_increasing; /* new pool size */ @@ -257,40 +307,38 @@ typedef struct kib_poolset kib_ps_node_init_t ps_node_init; /* finalize node */ kib_ps_node_fini_t ps_node_fini; -} kib_poolset_t; +}; -typedef struct kib_pool -{ +struct kib_pool { /* chain on pool list */ struct list_head po_list; /* pre-allocated node */ struct list_head po_free_list; /* pool_set of this pool */ - kib_poolset_t *po_owner; + struct kib_poolset *po_owner; /* deadline of this pool */ - cfs_time_t po_deadline; + time64_t po_deadline; /* # of elements in use */ int po_allocated; /* pool is created on failed HCA */ int po_failed; /* # of pre-allocated elements */ int po_size; -} kib_pool_t; +}; -typedef struct { - kib_poolset_t tps_poolset; /* pool-set */ +struct kib_tx_poolset { + struct kib_poolset tps_poolset; /* pool-set */ __u64 tps_next_tx_cookie; /* cookie of TX */ -} kib_tx_poolset_t; +}; -typedef struct { - kib_pool_t tpo_pool; /* pool */ +struct kib_tx_pool { + struct kib_pool tpo_pool; /* pool */ struct kib_hca_dev *tpo_hdev; /* device for this pool */ struct kib_tx *tpo_tx_descs; /* all the tx descriptors */ - kib_pages_t *tpo_tx_pages; /* premapped tx msg pages */ -} kib_tx_pool_t; + struct kib_pages *tpo_tx_pages; /* premapped tx msg pages */ +}; -typedef struct -{ +struct kib_fmr_poolset { spinlock_t fps_lock; /* serialize */ struct kib_net *fps_net; /* IB network */ struct list_head fps_pool_list; /* FMR pool list */ @@ -303,10 +351,10 @@ typedef struct /* is allocating new pool */ int fps_increasing; /* time stamp for retry if failed to allocate */ - cfs_time_t fps_next_retry; -} kib_fmr_poolset_t; + time64_t fps_next_retry; +}; -#ifndef HAVE_IB_RDMA_WR +#ifndef HAVE_OFED_IB_RDMA_WR struct ib_rdma_wr { struct ib_send_wr wr; }; @@ -315,7 +363,7 @@ struct ib_rdma_wr { struct kib_fast_reg_descriptor { /* For fast registration */ struct list_head frd_list; struct ib_rdma_wr frd_inv_wr; -#ifdef HAVE_IB_MAP_MR_SG +#ifdef HAVE_OFED_IB_MAP_MR_SG struct ib_reg_wr frd_fastreg_wr; #else struct ib_rdma_wr frd_fastreg_wr; @@ -323,38 +371,55 @@ struct kib_fast_reg_descriptor { /* For fast registration */ #endif struct ib_mr *frd_mr; bool frd_valid; + bool frd_posted; }; -typedef struct -{ +struct kib_fmr_pool { struct list_head fpo_list; /* chain on pool list */ struct kib_hca_dev *fpo_hdev; /* device for this pool */ - kib_fmr_poolset_t *fpo_owner; /* owner of this pool */ + struct kib_fmr_poolset *fpo_owner; /* owner of this pool */ +#ifdef HAVE_OFED_FMR_POOL_API union { struct { struct ib_fmr_pool *fpo_fmr_pool; /* IB FMR pool */ } fmr; +#endif struct { /* For fast registration */ struct list_head fpo_pool_list; int fpo_pool_size; } fast_reg; +#ifdef HAVE_OFED_FMR_POOL_API }; - cfs_time_t fpo_deadline; /* deadline of this pool */ + bool fpo_is_fmr; /* True if FMR pools allocated */ +#endif + time64_t fpo_deadline; /* deadline of this pool */ int fpo_failed; /* fmr pool is failed */ int fpo_map_count; /* # of mapped FMR */ - int fpo_is_fmr; -} kib_fmr_pool_t; +}; -typedef struct { - kib_fmr_pool_t *fmr_pool; /* pool of FMR */ +struct kib_fmr { + struct kib_fmr_pool *fmr_pool; /* pool of FMR */ +#ifdef HAVE_OFED_FMR_POOL_API struct ib_pool_fmr *fmr_pfmr; /* IB pool fmr */ +#endif /* HAVE_OFED_FMR_POOL_API */ struct kib_fast_reg_descriptor *fmr_frd; u32 fmr_key; -} kib_fmr_t; +}; -typedef struct kib_net -{ - /* chain on kib_dev_t::ibd_nets */ +#ifdef HAVE_OFED_FMR_POOL_API + +#ifdef HAVE_ORACLE_OFED_EXTENSIONS +#define kib_fmr_pool_map(pool, pgs, n, iov) \ + ib_fmr_pool_map_phys((pool), (pgs), (n), (iov), NULL) +#else +#define kib_fmr_pool_map(pool, pgs, n, iov) \ + ib_fmr_pool_map_phys((pool), (pgs), (n), (iov)) +#endif + +#endif /* HAVE_OFED_FMR_POOL_API */ + +struct kib_net { + /* chain on struct kib_dev::ibd_nets */ struct list_head ibn_list; __u64 ibn_incarnation;/* my epoch */ int ibn_init; /* initialisation state */ @@ -363,11 +428,12 @@ typedef struct kib_net atomic_t ibn_npeers; /* # peers extant */ atomic_t ibn_nconns; /* # connections extant */ - kib_tx_poolset_t **ibn_tx_ps; /* tx pool-set */ - kib_fmr_poolset_t **ibn_fmr_ps; /* fmr pool-set */ + struct kib_tx_poolset **ibn_tx_ps; /* tx pool-set */ + struct kib_fmr_poolset **ibn_fmr_ps; /* fmr pool-set */ - kib_dev_t *ibn_dev; /* underlying IB device */ -} kib_net_t; + struct kib_dev *ibn_dev; /* underlying IB device */ + struct lnet_ni *ibn_ni; /* LNet interface */ +}; #define KIB_THREAD_SHIFT 16 #define KIB_THREAD_ID(cpt, tid) ((cpt) << KIB_THREAD_SHIFT | (tid)) @@ -388,8 +454,7 @@ struct kib_sched_info { int ibs_cpt; /* CPT id */ }; -typedef struct -{ +struct kib_data { int kib_init; /* initialisation state */ int kib_shutdown; /* shut down? */ struct list_head kib_devs; /* IB devices extant */ @@ -401,9 +466,7 @@ typedef struct /* stabilize net/dev/peer_ni/conn ops */ rwlock_t kib_global_lock; /* hash table of all my known peers */ - struct list_head *kib_peers; - /* size of kib_peers */ - int kib_peer_hash_size; + DECLARE_HASHTABLE(kib_peers, IBLND_PEER_HASH_BITS); /* the connd task (serialisation assertions) */ void *kib_connd; /* connections to setup/teardown */ @@ -414,6 +477,8 @@ typedef struct struct list_head kib_reconn_list; /* peers wait for reconnection */ struct list_head kib_reconn_wait; + /* connections wait for completion */ + struct list_head kib_connd_waits; /* * The second that peers are pulled out from \a kib_reconn_wait * for reconnection. @@ -425,224 +490,103 @@ typedef struct struct ib_qp_attr kib_error_qpa; /* QP->ERROR */ /* percpt data for schedulers */ struct kib_sched_info **kib_scheds; -} kib_data_t; +}; #define IBLND_INIT_NOTHING 0 #define IBLND_INIT_DATA 1 #define IBLND_INIT_ALL 2 -/************************************************************************ - * IB Wire message format. - * These are sent in sender's byte order (i.e. receiver flips). - */ - -typedef struct kib_connparams -{ - __u16 ibcp_queue_depth; - __u16 ibcp_max_frags; - __u32 ibcp_max_msg_size; -} WIRE_ATTR kib_connparams_t; - -typedef struct -{ - lnet_hdr_t ibim_hdr; /* portals header */ - char ibim_payload[0]; /* piggy-backed payload */ -} WIRE_ATTR kib_immediate_msg_t; - -typedef struct -{ - __u32 rf_nob; /* # bytes this frag */ - __u64 rf_addr; /* CAVEAT EMPTOR: misaligned!! */ -} WIRE_ATTR kib_rdma_frag_t; - -typedef struct -{ - __u32 rd_key; /* local/remote key */ - __u32 rd_nfrags; /* # fragments */ - kib_rdma_frag_t rd_frags[0]; /* buffer frags */ -} WIRE_ATTR kib_rdma_desc_t; - -typedef struct -{ - lnet_hdr_t ibprm_hdr; /* portals header */ - __u64 ibprm_cookie; /* opaque completion cookie */ -} WIRE_ATTR kib_putreq_msg_t; - -typedef struct -{ - __u64 ibpam_src_cookie; /* reflected completion cookie */ - __u64 ibpam_dst_cookie; /* opaque completion cookie */ - kib_rdma_desc_t ibpam_rd; /* sender's sink buffer */ -} WIRE_ATTR kib_putack_msg_t; - -typedef struct -{ - lnet_hdr_t ibgm_hdr; /* portals header */ - __u64 ibgm_cookie; /* opaque completion cookie */ - kib_rdma_desc_t ibgm_rd; /* rdma descriptor */ -} WIRE_ATTR kib_get_msg_t; - -typedef struct -{ - __u64 ibcm_cookie; /* opaque completion cookie */ - __s32 ibcm_status; /* < 0 failure: >= 0 length */ -} WIRE_ATTR kib_completion_msg_t; - -typedef struct -{ - /* First 2 fields fixed FOR ALL TIME */ - __u32 ibm_magic; /* I'm an ibnal message */ - __u16 ibm_version; /* this is my version number */ - - __u8 ibm_type; /* msg type */ - __u8 ibm_credits; /* returned credits */ - __u32 ibm_nob; /* # bytes in whole message */ - __u32 ibm_cksum; /* checksum (0 == no checksum) */ - __u64 ibm_srcnid; /* sender's NID */ - __u64 ibm_srcstamp; /* sender's incarnation */ - __u64 ibm_dstnid; /* destination's NID */ - __u64 ibm_dststamp; /* destination's incarnation */ - - union { - kib_connparams_t connparams; - kib_immediate_msg_t immediate; - kib_putreq_msg_t putreq; - kib_putack_msg_t putack; - kib_get_msg_t get; - kib_completion_msg_t completion; - } WIRE_ATTR ibm_u; -} WIRE_ATTR kib_msg_t; - -#define IBLND_MSG_MAGIC LNET_PROTO_IB_MAGIC /* unique magic */ - -#define IBLND_MSG_VERSION_1 0x11 -#define IBLND_MSG_VERSION_2 0x12 -#define IBLND_MSG_VERSION IBLND_MSG_VERSION_2 - -#define IBLND_MSG_CONNREQ 0xc0 /* connection request */ -#define IBLND_MSG_CONNACK 0xc1 /* connection acknowledge */ -#define IBLND_MSG_NOOP 0xd0 /* nothing (just credits) */ -#define IBLND_MSG_IMMEDIATE 0xd1 /* immediate */ -#define IBLND_MSG_PUT_REQ 0xd2 /* putreq (src->sink) */ -#define IBLND_MSG_PUT_NAK 0xd3 /* completion (sink->src) */ -#define IBLND_MSG_PUT_ACK 0xd4 /* putack (sink->src) */ -#define IBLND_MSG_PUT_DONE 0xd5 /* completion (src->sink) */ -#define IBLND_MSG_GET_REQ 0xd6 /* getreq (sink->src) */ -#define IBLND_MSG_GET_DONE 0xd7 /* completion (src->sink: all OK) */ - -typedef struct { - __u32 ibr_magic; /* sender's magic */ - __u16 ibr_version; /* sender's version */ - __u8 ibr_why; /* reject reason */ - __u8 ibr_padding; /* padding */ - __u64 ibr_incarnation; /* incarnation of peer_ni */ - kib_connparams_t ibr_cp; /* connection parameters */ -} WIRE_ATTR kib_rej_t; - -/* connection rejection reasons */ -#define IBLND_REJECT_CONN_RACE 1 /* You lost connection race */ -#define IBLND_REJECT_NO_RESOURCES 2 /* Out of memory/conns etc */ -#define IBLND_REJECT_FATAL 3 /* Anything else */ - -#define IBLND_REJECT_CONN_UNCOMPAT 4 /* incompatible version peer_ni */ -#define IBLND_REJECT_CONN_STALE 5 /* stale peer_ni */ - -/* peer_ni's rdma frags doesn't match mine */ -#define IBLND_REJECT_RDMA_FRAGS 6 -/* peer_ni's msg queue size doesn't match mine */ -#define IBLND_REJECT_MSG_QUEUE_SIZE 7 - -/***********************************************************************/ - -typedef struct kib_rx /* receive message */ -{ +struct kib_rx { /* receive message */ /* queue for attention */ struct list_head rx_list; /* owning conn */ struct kib_conn *rx_conn; /* # bytes received (-1 while posted) */ int rx_nob; - /* completion status */ - enum ib_wc_status rx_status; /* message buffer (host vaddr) */ - kib_msg_t *rx_msg; + struct kib_msg *rx_msg; /* message buffer (I/O addr) */ __u64 rx_msgaddr; /* for dma_unmap_single() */ - DECLARE_PCI_UNMAP_ADDR(rx_msgunmap); + DEFINE_DMA_UNMAP_ADDR(rx_msgunmap); /* receive work item... */ struct ib_recv_wr rx_wrq; /* ...and its memory */ struct ib_sge rx_sge; -} kib_rx_t; +}; #define IBLND_POSTRX_DONT_POST 0 /* don't post */ #define IBLND_POSTRX_NO_CREDIT 1 /* post: no credits */ #define IBLND_POSTRX_PEER_CREDIT 2 /* post: give peer_ni back 1 credit */ #define IBLND_POSTRX_RSRVD_CREDIT 3 /* post: give myself back 1 reserved credit */ -typedef struct kib_tx /* transmit message */ -{ +struct kib_tx { /* transmit message */ /* queue on idle_txs ibc_tx_queue etc. */ struct list_head tx_list; /* pool I'm from */ - kib_tx_pool_t *tx_pool; + struct kib_tx_pool *tx_pool; /* owning conn */ struct kib_conn *tx_conn; /* # tx callbacks outstanding */ short tx_sending; /* queued for sending */ - short tx_queued; + unsigned long tx_queued:1, /* waiting for peer_ni */ - short tx_waiting; + tx_waiting:1, + /* force RDMA */ + tx_gpu:1; /* LNET completion status */ int tx_status; + /* health status of the transmit */ + enum lnet_msg_hstatus tx_hstatus; /* completion deadline */ - unsigned long tx_deadline; + ktime_t tx_deadline; /* completion cookie */ __u64 tx_cookie; /* lnet msgs to finalize on completion */ - lnet_msg_t *tx_lntmsg[2]; + struct lnet_msg *tx_lntmsg[2]; /* message buffer (host vaddr) */ - kib_msg_t *tx_msg; + struct kib_msg *tx_msg; /* message buffer (I/O addr) */ __u64 tx_msgaddr; /* for dma_unmap_single() */ - DECLARE_PCI_UNMAP_ADDR(tx_msgunmap); + DEFINE_DMA_UNMAP_ADDR(tx_msgunmap); /* # send work items */ int tx_nwrq; + /* # used scatter/gather elements */ + int tx_nsge; /* send work items... */ struct ib_rdma_wr *tx_wrq; /* ...and their memory */ struct ib_sge *tx_sge; /* rdma descriptor */ - kib_rdma_desc_t *tx_rd; + struct kib_rdma_desc *tx_rd; /* # entries in... */ int tx_nfrags; /* dma_map_sg descriptor */ struct scatterlist *tx_frags; /* rdma phys page addrs */ __u64 *tx_pages; + /* gaps in fragments */ + bool tx_gaps; /* FMR */ - kib_fmr_t fmr; + struct kib_fmr tx_fmr; /* dma direction */ int tx_dmadir; -} kib_tx_t; +}; -typedef struct kib_connvars -{ +struct kib_connvars { /* connection-in-progress variables */ - kib_msg_t cv_msg; -} kib_connvars_t; + struct kib_msg cv_msg; +}; -typedef struct kib_conn -{ +struct kib_conn { /* scheduler information */ struct kib_sched_info *ibc_sched; /* owning peer_ni */ - struct kib_peer *ibc_peer; + struct kib_peer_ni *ibc_peer; /* HCA bound on */ - kib_hca_dev_t *ibc_hdev; + struct kib_hca_dev *ibc_hdev; /* stash on peer_ni's conn list */ struct list_head ibc_list; /* schedule for attention */ @@ -673,6 +617,8 @@ typedef struct kib_conn __u16 ibc_queue_depth; /* connections max frags */ __u16 ibc_max_frags; + /* count of timeout txs waiting on cq */ + __u16 ibc_waits; /* receive buffers owned */ unsigned int ibc_nrx:16; /* scheduled for attention */ @@ -680,7 +626,7 @@ typedef struct kib_conn /* CQ callback fired */ unsigned int ibc_ready:1; /* time of last send */ - unsigned long ibc_last_send; + ktime_t ibc_last_send; /** link chain for kiblnd_check_conns only */ struct list_head ibc_connd_list; /** rxs completed before ESTABLISHED */ @@ -695,12 +641,14 @@ typedef struct kib_conn struct list_head ibc_tx_queue_rsrvd; /* active tx awaiting completion */ struct list_head ibc_active_txs; + /* zombie tx awaiting done */ + struct list_head ibc_zombie_txs; /* serialise */ spinlock_t ibc_lock; /* the rx descs */ - kib_rx_t *ibc_rxs; + struct kib_rx *ibc_rxs; /* premapped rx msg pages */ - kib_pages_t *ibc_rx_pages; + struct kib_pages *ibc_rx_pages; /* CM id */ struct rdma_cm_id *ibc_cmid; @@ -708,8 +656,8 @@ typedef struct kib_conn struct ib_cq *ibc_cq; /* in-progress connection state */ - kib_connvars_t *ibc_connvars; -} kib_conn_t; + struct kib_connvars *ibc_connvars; +}; #define IBLND_CONN_INIT 0 /* being initialised */ #define IBLND_CONN_ACTIVE_CONNECT 1 /* active sending req */ @@ -718,24 +666,25 @@ typedef struct kib_conn #define IBLND_CONN_CLOSING 4 /* being closed */ #define IBLND_CONN_DISCONNECTED 5 /* disconnected */ -typedef struct kib_peer -{ - /* stash on global peer_ni list */ - struct list_head ibp_list; +struct kib_peer_ni { + /* on peer_ni hash chain */ + struct hlist_node ibp_list; /* who's on the other end(s) */ lnet_nid_t ibp_nid; /* LNet interface */ - lnet_ni_t *ibp_ni; + struct lnet_ni *ibp_ni; /* all active connections */ struct list_head ibp_conns; + /* next connection to send on for round robin */ + struct kib_conn *ibp_next_conn; /* msgs waiting for a conn */ struct list_head ibp_tx_queue; /* incarnation of peer_ni */ __u64 ibp_incarnation; - /* when (in jiffies) I was last alive */ - cfs_time_t ibp_last_alive; + /* when (in seconds) I was last alive */ + time64_t ibp_last_alive; /* # users */ - atomic_t ibp_refcount; + struct kref ibp_kref; /* version of peer_ni */ __u16 ibp_version; /* current passive connection attempts */ @@ -743,7 +692,7 @@ typedef struct kib_peer /* current active connection attempts */ unsigned short ibp_connecting; /* reconnect this peer_ni later */ - unsigned short ibp_reconnecting:1; + unsigned char ibp_reconnecting; /* counter of how many times we triggered a conn race */ unsigned char ibp_races; /* # consecutive reconnection attempts to this peer */ @@ -754,9 +703,13 @@ typedef struct kib_peer __u16 ibp_max_frags; /* max_peer_credits */ __u16 ibp_queue_depth; -} kib_peer_ni_t; + /* reduced value which allows conn to be created if max fails */ + __u16 ibp_queue_depth_mod; + /* Number of connections allocated. */ + atomic_t ibp_nconns; +}; -#ifndef HAVE_IB_INC_RKEY +#ifndef HAVE_OFED_IB_INC_RKEY /** * ib_inc_rkey - increments the key portion of the given rkey. Can be used * for calculating a new rkey for type 2 memory windows. @@ -769,30 +722,16 @@ static inline u32 ib_inc_rkey(u32 rkey) } #endif -extern kib_data_t kiblnd_data; +extern struct kib_data kiblnd_data; -extern void kiblnd_hdev_destroy(kib_hca_dev_t *hdev); +extern void kiblnd_hdev_destroy(struct kib_hca_dev *hdev); int kiblnd_msg_queue_size(int version, struct lnet_ni *ni); -/* max # of fragments configured by user */ -static inline int -kiblnd_cfg_rdma_frags(struct lnet_ni *ni) -{ - struct lnet_ioctl_config_o2iblnd_tunables *tunables; - int mod; - - tunables = &ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib; - mod = tunables->lnd_map_on_demand; - return mod != 0 ? mod : IBLND_MAX_RDMA_FRAGS; -} - -static inline int -kiblnd_rdma_frags(int version, struct lnet_ni *ni) +static inline int kiblnd_timeout(void) { - return version == IBLND_MSG_VERSION_1 ? - IBLND_MAX_RDMA_FRAGS : - kiblnd_cfg_rdma_frags(ni); + return *kiblnd_tunables.kib_timeout ? *kiblnd_tunables.kib_timeout : + lnet_get_lnd_timeout(); } static inline int @@ -816,14 +755,14 @@ kiblnd_concurrent_sends(int version, struct lnet_ni *ni) } static inline void -kiblnd_hdev_addref_locked(kib_hca_dev_t *hdev) +kiblnd_hdev_addref_locked(struct kib_hca_dev *hdev) { LASSERT(atomic_read(&hdev->ibh_ref) > 0); atomic_inc(&hdev->ibh_ref); } static inline void -kiblnd_hdev_decref(kib_hca_dev_t *hdev) +kiblnd_hdev_decref(struct kib_hca_dev *hdev) { LASSERT(atomic_read(&hdev->ibh_ref) > 0); if (atomic_dec_and_test(&hdev->ibh_ref)) @@ -831,7 +770,7 @@ kiblnd_hdev_decref(kib_hca_dev_t *hdev) } static inline int -kiblnd_dev_can_failover(kib_dev_t *dev) +kiblnd_dev_can_failover(struct kib_dev *dev) { if (!list_empty(&dev->ibd_fail_list)) /* already scheduled */ return 0; @@ -845,49 +784,52 @@ kiblnd_dev_can_failover(kib_dev_t *dev) return dev->ibd_can_failover; } -#define kiblnd_conn_addref(conn) \ -do { \ - CDEBUG(D_NET, "conn[%p] (%d)++\n", \ - (conn), atomic_read(&(conn)->ibc_refcount)); \ - atomic_inc(&(conn)->ibc_refcount); \ -} while (0) - -#define kiblnd_conn_decref(conn) \ -do { \ - unsigned long flags; \ - \ - CDEBUG(D_NET, "conn[%p] (%d)--\n", \ - (conn), atomic_read(&(conn)->ibc_refcount)); \ - LASSERT_ATOMIC_POS(&(conn)->ibc_refcount); \ - if (atomic_dec_and_test(&(conn)->ibc_refcount)) { \ - spin_lock_irqsave(&kiblnd_data.kib_connd_lock, flags); \ - list_add_tail(&(conn)->ibc_list, \ - &kiblnd_data.kib_connd_zombies); \ - wake_up(&kiblnd_data.kib_connd_waitq); \ - spin_unlock_irqrestore(&kiblnd_data.kib_connd_lock, flags);\ - } \ -} while (0) - -#define kiblnd_peer_addref(peer_ni) \ -do { \ - CDEBUG(D_NET, "peer_ni[%p] -> %s (%d)++\n", \ - (peer_ni), libcfs_nid2str((peer_ni)->ibp_nid), \ - atomic_read (&(peer_ni)->ibp_refcount)); \ - atomic_inc(&(peer_ni)->ibp_refcount); \ -} while (0) - -#define kiblnd_peer_decref(peer_ni) \ -do { \ - CDEBUG(D_NET, "peer_ni[%p] -> %s (%d)--\n", \ - (peer_ni), libcfs_nid2str((peer_ni)->ibp_nid), \ - atomic_read (&(peer_ni)->ibp_refcount)); \ - LASSERT_ATOMIC_POS(&(peer_ni)->ibp_refcount); \ - if (atomic_dec_and_test(&(peer_ni)->ibp_refcount)) \ - kiblnd_destroy_peer(peer_ni); \ -} while (0) +static inline void kiblnd_conn_addref(struct kib_conn *conn) +{ +#ifdef O2IBLND_CONN_REFCOUNT_DEBUG + CDEBUG(D_NET, "conn[%p] (%d)++\n", + (conn), atomic_read(&(conn)->ibc_refcount)); +#endif + atomic_inc(&(conn)->ibc_refcount); +} + +static inline void kiblnd_conn_decref(struct kib_conn *conn) +{ + unsigned long flags; +#ifdef O2IBLND_CONN_REFCOUNT_DEBUG + CDEBUG(D_NET, "conn[%p] (%d)--\n", + (conn), atomic_read(&(conn)->ibc_refcount)); +#endif + LASSERT(atomic_read(&(conn)->ibc_refcount) > 0); + if (atomic_dec_and_test(&(conn)->ibc_refcount)) { + spin_lock_irqsave(&kiblnd_data.kib_connd_lock, flags); + list_add_tail(&(conn)->ibc_list, + &kiblnd_data.kib_connd_zombies); + wake_up(&kiblnd_data.kib_connd_waitq); + spin_unlock_irqrestore(&kiblnd_data.kib_connd_lock, flags); + } +} + +void kiblnd_destroy_peer(struct kref *kref); + +static inline void kiblnd_peer_addref(struct kib_peer_ni *peer_ni) +{ + CDEBUG(D_NET, "peer_ni[%p] -> %s (%d)++\n", + peer_ni, libcfs_nid2str(peer_ni->ibp_nid), + kref_read(&peer_ni->ibp_kref)); + kref_get(&(peer_ni)->ibp_kref); +} + +static inline void kiblnd_peer_decref(struct kib_peer_ni *peer_ni) +{ + CDEBUG(D_NET, "peer_ni[%p] -> %s (%d)--\n", + peer_ni, libcfs_nid2str(peer_ni->ibp_nid), + kref_read(&peer_ni->ibp_kref)); + kref_put(&peer_ni->ibp_kref, kiblnd_destroy_peer); +} static inline bool -kiblnd_peer_connecting(kib_peer_ni_t *peer_ni) +kiblnd_peer_connecting(struct kib_peer_ni *peer_ni) { return peer_ni->ibp_connecting != 0 || peer_ni->ibp_reconnecting != 0 || @@ -895,56 +837,77 @@ kiblnd_peer_connecting(kib_peer_ni_t *peer_ni) } static inline bool -kiblnd_peer_idle(kib_peer_ni_t *peer_ni) +kiblnd_peer_idle(struct kib_peer_ni *peer_ni) { return !kiblnd_peer_connecting(peer_ni) && list_empty(&peer_ni->ibp_conns); } -static inline struct list_head * -kiblnd_nid2peerlist (lnet_nid_t nid) -{ - unsigned int hash = - ((unsigned int)nid) % kiblnd_data.kib_peer_hash_size; - - return &kiblnd_data.kib_peers[hash]; -} - static inline int -kiblnd_peer_active (kib_peer_ni_t *peer_ni) +kiblnd_peer_active(struct kib_peer_ni *peer_ni) { /* Am I in the peer_ni hash table? */ - return !list_empty(&peer_ni->ibp_list); + return !hlist_unhashed(&peer_ni->ibp_list); } -static inline kib_conn_t * -kiblnd_get_conn_locked (kib_peer_ni_t *peer_ni) +static inline struct kib_conn * +kiblnd_get_conn_locked(struct kib_peer_ni *peer_ni) { + struct list_head *next; + LASSERT(!list_empty(&peer_ni->ibp_conns)); - /* just return the first connection */ - return list_entry(peer_ni->ibp_conns.next, kib_conn_t, ibc_list); + /* Advance to next connection, be sure to skip the head node */ + if (!peer_ni->ibp_next_conn || + peer_ni->ibp_next_conn->ibc_list.next == &peer_ni->ibp_conns) + next = peer_ni->ibp_conns.next; + else + next = peer_ni->ibp_next_conn->ibc_list.next; + peer_ni->ibp_next_conn = list_entry(next, struct kib_conn, ibc_list); + + return peer_ni->ibp_next_conn; } static inline int -kiblnd_send_keepalive(kib_conn_t *conn) +kiblnd_send_keepalive(struct kib_conn *conn) { + s64 keepalive_ns = *kiblnd_tunables.kib_keepalive * NSEC_PER_SEC; + return (*kiblnd_tunables.kib_keepalive > 0) && - cfs_time_after(jiffies, conn->ibc_last_send + - msecs_to_jiffies(*kiblnd_tunables.kib_keepalive * - MSEC_PER_SEC)); + ktime_after(ktime_get(), + ktime_add_ns(conn->ibc_last_send, keepalive_ns)); } +/* when to return credits eagerly */ static inline int -kiblnd_need_noop(kib_conn_t *conn) +kiblnd_credits_highwater(struct lnet_ioctl_config_o2iblnd_tunables *t, + struct lnet_ioctl_config_lnd_cmn_tunables *nt, + struct kib_conn *conn) { - lnet_ni_t *ni = conn->ibc_peer->ibp_ni; + int credits_hiw = IBLND_CREDIT_HIGHWATER_V1; + + if ((conn->ibc_version) == IBLND_MSG_VERSION_1) + return credits_hiw; + + /* if queue depth is negotiated down, calculate hiw proportionally */ + credits_hiw = (conn->ibc_queue_depth * t->lnd_peercredits_hiw) / + nt->lct_peer_tx_credits; + + return credits_hiw; +} + +static inline int +kiblnd_need_noop(struct kib_conn *conn) +{ + struct lnet_ni *ni = conn->ibc_peer->ibp_ni; struct lnet_ioctl_config_o2iblnd_tunables *tunables; + struct lnet_ioctl_config_lnd_cmn_tunables *net_tunables; LASSERT(conn->ibc_state >= IBLND_CONN_ESTABLISHED); tunables = &ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib; + net_tunables = &ni->ni_net->net_tunables; if (conn->ibc_outstanding_credits < - IBLND_CREDITS_HIGHWATER(tunables, conn->ibc_version) && + kiblnd_credits_highwater(tunables, net_tunables, conn) && !kiblnd_send_keepalive(conn)) return 0; /* No need to send NOOP */ @@ -971,14 +934,14 @@ kiblnd_need_noop(kib_conn_t *conn) } static inline void -kiblnd_abort_receives(kib_conn_t *conn) +kiblnd_abort_receives(struct kib_conn *conn) { ib_modify_qp(conn->ibc_cmid->qp, &kiblnd_data.kib_error_qpa, IB_QP_STATE); } static inline const char * -kiblnd_queue2str(kib_conn_t *conn, struct list_head *q) +kiblnd_queue2str(struct kib_conn *conn, struct list_head *q) { if (q == &conn->ibc_tx_queue) return "tx_queue"; @@ -1029,21 +992,21 @@ kiblnd_wreqid2type (__u64 wreqid) } static inline void -kiblnd_set_conn_state (kib_conn_t *conn, int state) +kiblnd_set_conn_state(struct kib_conn *conn, int state) { conn->ibc_state = state; smp_mb(); } static inline void -kiblnd_init_msg (kib_msg_t *msg, int type, int body_nob) +kiblnd_init_msg(struct kib_msg *msg, int type, int body_nob) { msg->ibm_type = type; - msg->ibm_nob = offsetof(kib_msg_t, ibm_u) + body_nob; + msg->ibm_nob = offsetof(struct kib_msg, ibm_u) + body_nob; } static inline int -kiblnd_rd_size (kib_rdma_desc_t *rd) +kiblnd_rd_size(struct kib_rdma_desc *rd) { int i; int size; @@ -1055,25 +1018,25 @@ kiblnd_rd_size (kib_rdma_desc_t *rd) } static inline __u64 -kiblnd_rd_frag_addr(kib_rdma_desc_t *rd, int index) +kiblnd_rd_frag_addr(struct kib_rdma_desc *rd, int index) { return rd->rd_frags[index].rf_addr; } -static inline __u32 -kiblnd_rd_frag_size(kib_rdma_desc_t *rd, int index) +static inline int +kiblnd_rd_frag_size(struct kib_rdma_desc *rd, int index) { return rd->rd_frags[index].rf_nob; } static inline __u32 -kiblnd_rd_frag_key(kib_rdma_desc_t *rd, int index) +kiblnd_rd_frag_key(struct kib_rdma_desc *rd, int index) { return rd->rd_key; } static inline int -kiblnd_rd_consume_frag(kib_rdma_desc_t *rd, int index, __u32 nob) +kiblnd_rd_consume_frag(struct kib_rdma_desc *rd, int index, __u32 nob) { if (nob < rd->rd_frags[index].rf_nob) { rd->rd_frags[index].rf_addr += nob; @@ -1086,14 +1049,14 @@ kiblnd_rd_consume_frag(kib_rdma_desc_t *rd, int index, __u32 nob) } static inline int -kiblnd_rd_msg_size(kib_rdma_desc_t *rd, int msgtype, int n) +kiblnd_rd_msg_size(struct kib_rdma_desc *rd, int msgtype, int n) { LASSERT (msgtype == IBLND_MSG_GET_REQ || msgtype == IBLND_MSG_PUT_ACK); return msgtype == IBLND_MSG_GET_REQ ? - offsetof(kib_get_msg_t, ibgm_rd.rd_frags[n]) : - offsetof(kib_putack_msg_t, ibpam_rd.rd_frags[n]); + offsetof(struct kib_get_msg, ibgm_rd.rd_frags[n]) : + offsetof(struct kib_putack_msg, ibpam_rd.rd_frags[n]); } static inline __u64 @@ -1119,20 +1082,40 @@ static inline void kiblnd_dma_unmap_single(struct ib_device *dev, #define KIBLND_UNMAP_ADDR_SET(p, m, a) do {} while (0) #define KIBLND_UNMAP_ADDR(p, m, a) (a) -static inline int kiblnd_dma_map_sg(struct ib_device *dev, - struct scatterlist *sg, int nents, - enum dma_data_direction direction) +static inline +int kiblnd_dma_map_sg(struct kib_hca_dev *hdev, struct kib_tx *tx) { - return ib_dma_map_sg(dev, sg, nents, direction); + struct scatterlist *sg = tx->tx_frags; + int nents = tx->tx_nfrags; + enum dma_data_direction direction = tx->tx_dmadir; + + if (tx->tx_gpu) + return lnet_rdma_map_sg_attrs(hdev->ibh_ibdev->dma_device, + sg, nents, direction); + + return ib_dma_map_sg(hdev->ibh_ibdev, sg, nents, direction); } -static inline void kiblnd_dma_unmap_sg(struct ib_device *dev, - struct scatterlist *sg, int nents, - enum dma_data_direction direction) +static inline +void kiblnd_dma_unmap_sg(struct kib_hca_dev *hdev, struct kib_tx *tx) { - ib_dma_unmap_sg(dev, sg, nents, direction); + struct scatterlist *sg = tx->tx_frags; + int nents = tx->tx_nfrags; + enum dma_data_direction direction = tx->tx_dmadir; + + if (tx->tx_gpu) + lnet_rdma_unmap_sg(hdev->ibh_ibdev->dma_device, + sg, nents, direction); + else + ib_dma_unmap_sg(hdev->ibh_ibdev, sg, nents, direction); } +#ifndef HAVE_OFED_IB_SG_DMA_ADDRESS +#include +#define ib_sg_dma_address(dev, sg) sg_dma_address(sg) +#define ib_sg_dma_len(dev, sg) sg_dma_len(sg) +#endif + static inline __u64 kiblnd_sg_dma_address(struct ib_device *dev, struct scatterlist *sg) { @@ -1145,6 +1128,10 @@ static inline unsigned int kiblnd_sg_dma_len(struct ib_device *dev, return ib_sg_dma_len(dev, sg); } +#ifndef HAVE_OFED_RDMA_CONNECT_LOCKED +#define rdma_connect_locked(cmid, cpp) rdma_connect(cmid, cpp) +#endif + /* XXX We use KIBLND_CONN_PARAM(e) as writable buffer, it's not strictly * right because OFED1.2 defines it as const, to use it we have to add * (void *) cast to overcome "const" */ @@ -1152,63 +1139,106 @@ static inline unsigned int kiblnd_sg_dma_len(struct ib_device *dev, #define KIBLND_CONN_PARAM(e) ((e)->param.conn.private_data) #define KIBLND_CONN_PARAM_LEN(e) ((e)->param.conn.private_data_len) -struct ib_mr *kiblnd_find_rd_dma_mr(struct lnet_ni *ni, kib_rdma_desc_t *rd, - int negotiated_nfrags); -void kiblnd_map_rx_descs(kib_conn_t *conn); -void kiblnd_unmap_rx_descs(kib_conn_t *conn); -void kiblnd_pool_free_node(kib_pool_t *pool, struct list_head *node); -struct list_head *kiblnd_pool_alloc_node(kib_poolset_t *ps); +void kiblnd_abort_txs(struct kib_conn *conn, struct list_head *txs); +void kiblnd_map_rx_descs(struct kib_conn *conn); +void kiblnd_unmap_rx_descs(struct kib_conn *conn); +void kiblnd_pool_free_node(struct kib_pool *pool, struct list_head *node); +struct list_head *kiblnd_pool_alloc_node(struct kib_poolset *ps); -int kiblnd_fmr_pool_map(kib_fmr_poolset_t *fps, kib_tx_t *tx, - kib_rdma_desc_t *rd, __u32 nob, __u64 iov, - kib_fmr_t *fmr); -void kiblnd_fmr_pool_unmap(kib_fmr_t *fmr, int status); +int kiblnd_fmr_pool_map(struct kib_fmr_poolset *fps, struct kib_tx *tx, + struct kib_rdma_desc *rd, u32 nob, u64 iov, + struct kib_fmr *fmr); +void kiblnd_fmr_pool_unmap(struct kib_fmr *fmr, int status); int kiblnd_tunables_setup(struct lnet_ni *ni); int kiblnd_tunables_init(void); int kiblnd_connd (void *arg); int kiblnd_scheduler(void *arg); -int kiblnd_thread_start(int (*fn)(void *arg), void *arg, char *name); +#define kiblnd_thread_start(fn, data, namefmt, arg...) \ + ({ \ + struct task_struct *__task = kthread_run(fn, data, \ + namefmt, ##arg); \ + if (!IS_ERR(__task)) \ + atomic_inc(&kiblnd_data.kib_nthreads); \ + PTR_ERR_OR_ZERO(__task); \ + }) + int kiblnd_failover_thread (void *arg); -int kiblnd_alloc_pages(kib_pages_t **pp, int cpt, int npages); +int kiblnd_alloc_pages(struct kib_pages **pp, int cpt, int npages); int kiblnd_cm_callback(struct rdma_cm_id *cmid, struct rdma_cm_event *event); int kiblnd_translate_mtu(int value); -int kiblnd_dev_failover(kib_dev_t *dev); -int kiblnd_create_peer(lnet_ni_t *ni, kib_peer_ni_t **peerp, lnet_nid_t nid); -void kiblnd_destroy_peer (kib_peer_ni_t *peer); -bool kiblnd_reconnect_peer(kib_peer_ni_t *peer); -void kiblnd_destroy_dev (kib_dev_t *dev); -void kiblnd_unlink_peer_locked (kib_peer_ni_t *peer_ni); -kib_peer_ni_t *kiblnd_find_peer_locked(struct lnet_ni *ni, lnet_nid_t nid); -int kiblnd_close_stale_conns_locked (kib_peer_ni_t *peer_ni, - int version, __u64 incarnation); -int kiblnd_close_peer_conns_locked (kib_peer_ni_t *peer_ni, int why); - -kib_conn_t *kiblnd_create_conn(kib_peer_ni_t *peer_ni, struct rdma_cm_id *cmid, - int state, int version); -void kiblnd_destroy_conn(kib_conn_t *conn, bool free_conn); -void kiblnd_close_conn (kib_conn_t *conn, int error); -void kiblnd_close_conn_locked (kib_conn_t *conn, int error); - -void kiblnd_launch_tx (lnet_ni_t *ni, kib_tx_t *tx, lnet_nid_t nid); -void kiblnd_txlist_done(lnet_ni_t *ni, struct list_head *txlist, int status); +int kiblnd_dev_failover(struct kib_dev *dev, struct net *ns); +int kiblnd_create_peer(struct lnet_ni *ni, struct kib_peer_ni **peerp, + lnet_nid_t nid); +bool kiblnd_reconnect_peer(struct kib_peer_ni *peer); +void kiblnd_destroy_dev(struct kib_dev *dev); +void kiblnd_unlink_peer_locked(struct kib_peer_ni *peer_ni); +struct kib_peer_ni *kiblnd_find_peer_locked(struct lnet_ni *ni, lnet_nid_t nid); +int kiblnd_close_stale_conns_locked(struct kib_peer_ni *peer_ni, + int version, u64 incarnation); +int kiblnd_close_peer_conns_locked(struct kib_peer_ni *peer_ni, int why); + +struct kib_conn *kiblnd_create_conn(struct kib_peer_ni *peer_ni, + struct rdma_cm_id *cmid, + int state, int version); +void kiblnd_destroy_conn(struct kib_conn *conn); +void kiblnd_close_conn(struct kib_conn *conn, int error); +void kiblnd_close_conn_locked(struct kib_conn *conn, int error); + +void kiblnd_launch_tx(struct lnet_ni *ni, struct kib_tx *tx, lnet_nid_t nid); +void kiblnd_txlist_done(struct list_head *txlist, int status, + enum lnet_msg_hstatus hstatus); void kiblnd_qp_event(struct ib_event *event, void *arg); void kiblnd_cq_event(struct ib_event *event, void *arg); void kiblnd_cq_completion(struct ib_cq *cq, void *arg); -void kiblnd_pack_msg (lnet_ni_t *ni, kib_msg_t *msg, int version, - int credits, lnet_nid_t dstnid, __u64 dststamp); -int kiblnd_unpack_msg(kib_msg_t *msg, int nob); -int kiblnd_post_rx (kib_rx_t *rx, int credit); - -int kiblnd_send(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg); -int kiblnd_recv(lnet_ni_t *ni, void *private, lnet_msg_t *lntmsg, int delayed, - unsigned int niov, struct kvec *iov, lnet_kiov_t *kiov, - unsigned int offset, unsigned int mlen, unsigned int rlen); +void kiblnd_pack_msg(struct lnet_ni *ni, struct kib_msg *msg, int version, + int credits, lnet_nid_t dstnid, __u64 dststamp); +int kiblnd_unpack_msg(struct kib_msg *msg, int nob); +int kiblnd_post_rx(struct kib_rx *rx, int credit); + +int kiblnd_send(struct lnet_ni *ni, void *private, struct lnet_msg *lntmsg); +int kiblnd_recv(struct lnet_ni *ni, void *private, struct lnet_msg *lntmsg, + int delayed, unsigned int niov, + struct bio_vec *kiov, unsigned int offset, unsigned int mlen, + unsigned int rlen); +unsigned int kiblnd_get_dev_prio(struct lnet_ni *ni, unsigned int dev_idx); + +#if LINUX_VERSION_CODE < KERNEL_VERSION(3, 11, 0) +#undef netdev_notifier_info_to_dev +#define netdev_notifier_info_to_dev(ndev) ndev +#endif +#define kiblnd_dump_conn_dbg(conn) \ +({ \ + if (conn && conn->ibc_cmid) \ + CDEBUG(D_NET, "conn %p state %d nposted %d/%d c/o/r %d/%d/%d ce %d : cm_id %p qp_num 0x%x device_name %s\n", \ + conn, \ + conn->ibc_state, \ + conn->ibc_noops_posted, \ + conn->ibc_nsends_posted, \ + conn->ibc_credits, \ + conn->ibc_outstanding_credits, \ + conn->ibc_reserved_credits, \ + conn->ibc_comms_error, \ + conn->ibc_cmid, \ + conn->ibc_cmid->qp ? conn->ibc_cmid->qp->qp_num : 0, \ + conn->ibc_cmid->qp ? (conn->ibc_cmid->qp->device ? dev_name(&conn->ibc_cmid->qp->device->dev) : "NULL") : "NULL"); \ + else if (conn) \ + CDEBUG(D_NET, "conn %p state %d nposted %d/%d c/o/r %d/%d/%d ce %d : cm_id NULL\n", \ + conn, \ + conn->ibc_state, \ + conn->ibc_noops_posted, \ + conn->ibc_nsends_posted, \ + conn->ibc_credits, \ + conn->ibc_outstanding_credits, \ + conn->ibc_reserved_credits, \ + conn->ibc_comms_error \ + ); \ +})