X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lnet%2Fklnds%2Fo2iblnd%2Fo2iblnd.h;h=3e4013f7939fc8ee14f8781ba99c07dede97c13a;hp=9bb0b60b4f175760bf92ba756f7abfea85be167f;hb=HEAD;hpb=7308662efc02fde077216f54728ecf278f31311b;ds=sidebyside diff --git a/lnet/klnds/o2iblnd/o2iblnd.h b/lnet/klnds/o2iblnd/o2iblnd.h index 9bb0b60..4d01da9 100644 --- a/lnet/klnds/o2iblnd/o2iblnd.h +++ b/lnet/klnds/o2iblnd/o2iblnd.h @@ -27,24 +27,43 @@ */ /* * 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 @@ -71,17 +90,35 @@ #include #include #include +#ifdef HAVE_OFED_FMR_POOL_API #include +#endif #define DEBUG_SUBSYSTEM S_LND -#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 @@ -104,6 +141,7 @@ struct kib_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 */ @@ -111,16 +149,11 @@ extern struct kib_tunables kiblnd_tunables; #define IBLND_CREDITS_DEFAULT 8 /* default # 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, conn) ((conn->ibc_version) == IBLND_MSG_VERSION_1 ? \ - IBLND_CREDIT_HIGHWATER_V1 : \ - min(t->lnd_peercredits_hiw, (__u32)conn->ibc_queue_depth - 1)) - -#ifdef HAVE_RDMA_CREATE_ID_5ARG +#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 +# 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 @@ -133,8 +166,10 @@ extern struct kib_tunables 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... */ @@ -169,9 +204,15 @@ struct kib_hca_dev; 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 */ @@ -200,7 +241,7 @@ struct kib_hca_dev { __u64 ibh_page_mask; /* page mask of current HCA */ __u64 ibh_mr_size; /* size of MR */ int ibh_max_qp_wr; /* maximum work requests size */ -#ifdef HAVE_IB_GET_DMA_MR +#ifdef HAVE_OFED_IB_GET_DMA_MR struct ib_mr *ibh_mrs; /* global MR */ #endif struct ib_pd *ibh_pd; /* PD */ @@ -313,7 +354,7 @@ struct kib_fmr_poolset { 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; }; @@ -322,7 +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; @@ -330,34 +371,43 @@ struct kib_fast_reg_descriptor { /* For fast registration */ #endif struct ib_mr *frd_mr; bool frd_valid; + bool frd_posted; }; struct kib_fmr_pool { struct list_head fpo_list; /* chain on pool list */ struct kib_hca_dev *fpo_hdev; /* device for 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 }; + 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 */ - bool fpo_is_fmr; /* True if FMR pools allocated */ }; 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; }; +#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) @@ -366,6 +416,8 @@ struct kib_fmr { 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; @@ -414,9 +466,7 @@ struct kib_data { /* 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 */ @@ -427,6 +477,8 @@ struct kib_data { 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. @@ -444,121 +496,6 @@ struct kib_data { #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). - */ - -struct kib_connparams { - __u16 ibcp_queue_depth; - __u16 ibcp_max_frags; - __u32 ibcp_max_msg_size; -} WIRE_ATTR; - -struct kib_immediate_msg { - struct lnet_hdr ibim_hdr; /* portals header */ - char ibim_payload[0];/* piggy-backed payload */ -} WIRE_ATTR; - -struct kib_rdma_frag { - __u32 rf_nob; /* # bytes this frag */ - __u64 rf_addr; /* CAVEAT EMPTOR: misaligned!! */ -} WIRE_ATTR; - -struct kib_rdma_desc { - __u32 rd_key; /* local/remote key */ - __u32 rd_nfrags; /* # fragments */ - struct kib_rdma_frag rd_frags[0]; /* buffer frags */ -} WIRE_ATTR; - -struct kib_putreq_msg { - struct lnet_hdr ibprm_hdr; /* portals header */ - __u64 ibprm_cookie; /* opaque completion cookie */ -} WIRE_ATTR; - -struct kib_putack_msg { - __u64 ibpam_src_cookie; /* reflected completion cookie */ - __u64 ibpam_dst_cookie; /* opaque completion cookie */ - struct kib_rdma_desc ibpam_rd; /* sender's sink buffer */ -} WIRE_ATTR; - -struct kib_get_msg { - struct lnet_hdr ibgm_hdr; /* portals header */ - __u64 ibgm_cookie; /* opaque completion cookie */ - struct kib_rdma_desc ibgm_rd; /* rdma descriptor */ -} WIRE_ATTR; - -struct kib_completion_msg { - __u64 ibcm_cookie; /* opaque completion cookie */ - __s32 ibcm_status; /* < 0 failure: >= 0 length */ -} WIRE_ATTR; - -struct kib_msg { - /* 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 { - struct kib_connparams connparams; - struct kib_immediate_msg immediate; - struct kib_putreq_msg putreq; - struct kib_putack_msg putack; - struct kib_get_msg get; - struct kib_completion_msg completion; - } WIRE_ATTR ibm_u; -} WIRE_ATTR; - -#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) */ - -struct kib_rej { - __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 */ - struct kib_connparams ibr_cp; /* connection parameters */ -} WIRE_ATTR; - -/* 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 -#define IBLND_REJECT_INVALID_SRV_ID 8 - -/***********************************************************************/ - struct kib_rx { /* receive message */ /* queue for attention */ struct list_head rx_list; @@ -593,9 +530,11 @@ struct kib_tx { /* transmit message */ /* # 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 */ @@ -612,8 +551,6 @@ struct kib_tx { /* transmit message */ __u64 tx_msgaddr; /* for dma_unmap_single() */ DEFINE_DMA_UNMAP_ADDR(tx_msgunmap); - /** sge for tx_msgaddr */ - struct ib_sge tx_msgsge; /* # send work items */ int tx_nwrq; /* # used scatter/gather elements */ @@ -636,8 +573,6 @@ struct kib_tx { /* transmit message */ struct kib_fmr tx_fmr; /* dma direction */ int tx_dmadir; - /* time when tx added on ibc_active_txs */ - ktime_t tx_on_activeq; }; struct kib_connvars { @@ -682,6 +617,8 @@ 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 */ @@ -730,8 +667,8 @@ struct kib_conn { #define IBLND_CONN_DISCONNECTED 5 /* disconnected */ struct kib_peer_ni { - /* stash on global peer_ni list */ - struct list_head ibp_list; + /* on peer_ni hash chain */ + struct hlist_node ibp_list; /* who's on the other end(s) */ lnet_nid_t ibp_nid; /* LNet interface */ @@ -747,7 +684,7 @@ struct kib_peer_ni { /* 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 */ @@ -766,9 +703,13 @@ struct kib_peer_ni { __u16 ibp_max_frags; /* max_peer_credits */ __u16 ibp_queue_depth; + /* 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. @@ -843,46 +784,49 @@ kiblnd_dev_can_failover(struct kib_dev *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(struct kib_peer_ni *peer_ni) @@ -898,20 +842,11 @@ 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(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 struct kib_conn * @@ -942,17 +877,37 @@ kiblnd_send_keepalive(struct kib_conn *conn) ktime_add_ns(conn->ibc_last_send, keepalive_ns)); } +/* when to return credits eagerly */ +static inline int +kiblnd_credits_highwater(struct lnet_ioctl_config_o2iblnd_tunables *t, + struct lnet_ioctl_config_lnd_cmn_tunables *nt, + struct kib_conn *conn) +{ + 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) && + kiblnd_credits_highwater(tunables, net_tunables, conn) && !kiblnd_send_keepalive(conn)) return 0; /* No need to send NOOP */ @@ -1127,21 +1082,35 @@ 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_IB_SG_DMA_ADDRESS +#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) @@ -1159,6 +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" */ @@ -1182,7 +1155,15 @@ 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(struct kib_pages **pp, int cpt, int npages); @@ -1194,7 +1175,6 @@ int kiblnd_translate_mtu(int value); 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); -void kiblnd_destroy_peer(struct kib_peer_ni *peer); 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); @@ -1228,4 +1208,37 @@ 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 \ + ); \ +})