X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lnet%2Fklnds%2Fo2iblnd%2Fo2iblnd.c;h=82b8d10a017957935cd659fbcbc23136c9d1788c;hp=603cca268b18d88b61dba900800d7862ccfdca9f;hb=0024460d797490ae90a2221cb5d4648c9d4fac82;hpb=9a379e3226f1ec3be1a628f11a87e34f5d53abd4 diff --git a/lnet/klnds/o2iblnd/o2iblnd.c b/lnet/klnds/o2iblnd/o2iblnd.c index 603cca2..82b8d10 100644 --- a/lnet/klnds/o2iblnd/o2iblnd.c +++ b/lnet/klnds/o2iblnd/o2iblnd.c @@ -23,11 +23,10 @@ * 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.c * @@ -35,11 +34,13 @@ */ #include +#include + #include "o2iblnd.h" -static struct lnet_lnd the_o2iblnd; +static const struct lnet_lnd the_o2iblnd; -kib_data_t kiblnd_data; +struct kib_data kiblnd_data; static __u32 kiblnd_cksum (void *ptr, int nob) @@ -96,90 +97,88 @@ kiblnd_msgtype2str(int type) static int kiblnd_msgtype2size(int type) { - const int hdr_size = offsetof(kib_msg_t, ibm_u); + const int hdr_size = offsetof(struct kib_msg, ibm_u); switch (type) { case IBLND_MSG_CONNREQ: case IBLND_MSG_CONNACK: - return hdr_size + sizeof(kib_connparams_t); + return hdr_size + sizeof(struct kib_connparams); case IBLND_MSG_NOOP: return hdr_size; case IBLND_MSG_IMMEDIATE: - return offsetof(kib_msg_t, ibm_u.immediate.ibim_payload[0]); + return offsetof(struct kib_msg, ibm_u.immediate.ibim_payload[0]); case IBLND_MSG_PUT_REQ: - return hdr_size + sizeof(kib_putreq_msg_t); + return hdr_size + sizeof(struct kib_putreq_msg); case IBLND_MSG_PUT_ACK: - return hdr_size + sizeof(kib_putack_msg_t); + return hdr_size + sizeof(struct kib_putack_msg); case IBLND_MSG_GET_REQ: - return hdr_size + sizeof(kib_get_msg_t); + return hdr_size + sizeof(struct kib_get_msg); case IBLND_MSG_PUT_NAK: case IBLND_MSG_PUT_DONE: case IBLND_MSG_GET_DONE: - return hdr_size + sizeof(kib_completion_msg_t); + return hdr_size + sizeof(struct kib_completion_msg); default: return -1; } } -static int -kiblnd_unpack_rd(kib_msg_t *msg, int flip) +static int kiblnd_unpack_rd(struct kib_msg *msg, bool flip) { - kib_rdma_desc_t *rd; - int nob; - int n; - int i; + struct kib_rdma_desc *rd; + int nob; + int n; + int i; - LASSERT (msg->ibm_type == IBLND_MSG_GET_REQ || - msg->ibm_type == IBLND_MSG_PUT_ACK); + LASSERT(msg->ibm_type == IBLND_MSG_GET_REQ || + msg->ibm_type == IBLND_MSG_PUT_ACK); - rd = msg->ibm_type == IBLND_MSG_GET_REQ ? - &msg->ibm_u.get.ibgm_rd : - &msg->ibm_u.putack.ibpam_rd; + rd = msg->ibm_type == IBLND_MSG_GET_REQ ? + &msg->ibm_u.get.ibgm_rd : + &msg->ibm_u.putack.ibpam_rd; - if (flip) { - __swab32s(&rd->rd_key); - __swab32s(&rd->rd_nfrags); - } + if (flip) { + __swab32s(&rd->rd_key); + __swab32s(&rd->rd_nfrags); + } - n = rd->rd_nfrags; + n = rd->rd_nfrags; - if (n <= 0 || n > IBLND_MAX_RDMA_FRAGS) { - CERROR("Bad nfrags: %d, should be 0 < n <= %d\n", - n, IBLND_MAX_RDMA_FRAGS); - return 1; - } + if (n <= 0 || n > IBLND_MAX_RDMA_FRAGS) { + CERROR("Bad nfrags: %d, should be 0 < n <= %d\n", + n, IBLND_MAX_RDMA_FRAGS); + return 1; + } - nob = offsetof (kib_msg_t, ibm_u) + - kiblnd_rd_msg_size(rd, msg->ibm_type, n); + nob = offsetof(struct kib_msg, ibm_u) + + kiblnd_rd_msg_size(rd, msg->ibm_type, n); - if (msg->ibm_nob < nob) { - CERROR("Short %s: %d(%d)\n", - kiblnd_msgtype2str(msg->ibm_type), msg->ibm_nob, nob); - return 1; - } + if (msg->ibm_nob < nob) { + CERROR("Short %s: %d(%d)\n", + kiblnd_msgtype2str(msg->ibm_type), msg->ibm_nob, nob); + return 1; + } - if (!flip) - return 0; + if (!flip) + return 0; - for (i = 0; i < n; i++) { - __swab32s(&rd->rd_frags[i].rf_nob); - __swab64s(&rd->rd_frags[i].rf_addr); - } + for (i = 0; i < n; i++) { + __swab32s(&rd->rd_frags[i].rf_nob); + __swab64s(&rd->rd_frags[i].rf_addr); + } - return 0; + return 0; } -void -kiblnd_pack_msg(struct lnet_ni *ni, kib_msg_t *msg, int version, - int credits, lnet_nid_t dstnid, __u64 dststamp) +void kiblnd_pack_msg(struct lnet_ni *ni, struct kib_msg *msg, int version, + int credits, lnet_nid_t dstnid, __u64 dststamp) { - kib_net_t *net = ni->ni_data; + struct kib_net *net = ni->ni_data; /* CAVEAT EMPTOR! all message fields not set here should have been * initialised previously. */ @@ -200,134 +199,135 @@ kiblnd_pack_msg(struct lnet_ni *ni, kib_msg_t *msg, int version, } } -int -kiblnd_unpack_msg(kib_msg_t *msg, int nob) -{ - const int hdr_size = offsetof(kib_msg_t, ibm_u); - __u32 msg_cksum; - __u16 version; - int msg_nob; - int flip; - - /* 6 bytes are enough to have received magic + version */ - if (nob < 6) { - CERROR("Short message: %d\n", nob); - return -EPROTO; - } +int kiblnd_unpack_msg(struct kib_msg *msg, int nob) +{ + const int hdr_size = offsetof(struct kib_msg, ibm_u); + __u32 msg_cksum; + __u16 version; + int msg_nob; + bool flip; - if (msg->ibm_magic == IBLND_MSG_MAGIC) { - flip = 0; - } else if (msg->ibm_magic == __swab32(IBLND_MSG_MAGIC)) { - flip = 1; - } else { - CERROR("Bad magic: %08x\n", msg->ibm_magic); - return -EPROTO; - } + /* 6 bytes are enough to have received magic + version */ + if (nob < 6) { + CERROR("Short message: %d\n", nob); + return -EPROTO; + } - version = flip ? __swab16(msg->ibm_version) : msg->ibm_version; - if (version != IBLND_MSG_VERSION && - version != IBLND_MSG_VERSION_1) { - CERROR("Bad version: %x\n", version); - return -EPROTO; - } + if (msg->ibm_magic == IBLND_MSG_MAGIC) { + flip = false; + } else if (msg->ibm_magic == __swab32(IBLND_MSG_MAGIC)) { + flip = true; + } else { + CERROR("Bad magic: %08x\n", msg->ibm_magic); + return -EPROTO; + } - if (nob < hdr_size) { - CERROR("Short message: %d\n", nob); - return -EPROTO; - } + version = flip ? __swab16(msg->ibm_version) : msg->ibm_version; + if (version != IBLND_MSG_VERSION && + version != IBLND_MSG_VERSION_1) { + CERROR("Bad version: %x\n", version); + return -EPROTO; + } - msg_nob = flip ? __swab32(msg->ibm_nob) : msg->ibm_nob; - if (msg_nob > nob) { - CERROR("Short message: got %d, wanted %d\n", nob, msg_nob); - return -EPROTO; - } + if (nob < hdr_size) { + CERROR("Short message: %d\n", nob); + return -EPROTO; + } - /* checksum must be computed with ibm_cksum zero and BEFORE anything - * gets flipped */ - msg_cksum = flip ? __swab32(msg->ibm_cksum) : msg->ibm_cksum; - msg->ibm_cksum = 0; - if (msg_cksum != 0 && - msg_cksum != kiblnd_cksum(msg, msg_nob)) { - CERROR("Bad checksum\n"); - return -EPROTO; - } + msg_nob = flip ? __swab32(msg->ibm_nob) : msg->ibm_nob; + if (msg_nob > nob) { + CERROR("Short message: got %d, wanted %d\n", nob, msg_nob); + return -EPROTO; + } - msg->ibm_cksum = msg_cksum; - - if (flip) { - /* leave magic unflipped as a clue to peer_ni endianness */ - msg->ibm_version = version; - CLASSERT (sizeof(msg->ibm_type) == 1); - CLASSERT (sizeof(msg->ibm_credits) == 1); - msg->ibm_nob = msg_nob; - __swab64s(&msg->ibm_srcnid); - __swab64s(&msg->ibm_srcstamp); - __swab64s(&msg->ibm_dstnid); - __swab64s(&msg->ibm_dststamp); - } + /* checksum must be computed with ibm_cksum zero and BEFORE anything + * gets flipped + */ + msg_cksum = flip ? __swab32(msg->ibm_cksum) : msg->ibm_cksum; + msg->ibm_cksum = 0; + if (msg_cksum != 0 && + msg_cksum != kiblnd_cksum(msg, msg_nob)) { + CERROR("Bad checksum\n"); + return -EPROTO; + } - if (msg->ibm_srcnid == LNET_NID_ANY) { - CERROR("Bad src nid: %s\n", libcfs_nid2str(msg->ibm_srcnid)); - return -EPROTO; - } + msg->ibm_cksum = msg_cksum; - if (msg_nob < kiblnd_msgtype2size(msg->ibm_type)) { - CERROR("Short %s: %d(%d)\n", kiblnd_msgtype2str(msg->ibm_type), - msg_nob, kiblnd_msgtype2size(msg->ibm_type)); - return -EPROTO; - } + if (flip) { + /* leave magic unflipped as a clue to peer_ni endianness */ + msg->ibm_version = version; + BUILD_BUG_ON(sizeof(msg->ibm_type) != 1); + BUILD_BUG_ON(sizeof(msg->ibm_credits) != 1); + msg->ibm_nob = msg_nob; + __swab64s(&msg->ibm_srcnid); + __swab64s(&msg->ibm_srcstamp); + __swab64s(&msg->ibm_dstnid); + __swab64s(&msg->ibm_dststamp); + } - switch (msg->ibm_type) { - default: - CERROR("Unknown message type %x\n", msg->ibm_type); - return -EPROTO; + if (msg->ibm_srcnid == LNET_NID_ANY) { + CERROR("Bad src nid: %s\n", libcfs_nid2str(msg->ibm_srcnid)); + return -EPROTO; + } - case IBLND_MSG_NOOP: - case IBLND_MSG_IMMEDIATE: - case IBLND_MSG_PUT_REQ: - break; + if (msg_nob < kiblnd_msgtype2size(msg->ibm_type)) { + CERROR("Short %s: %d(%d)\n", kiblnd_msgtype2str(msg->ibm_type), + msg_nob, kiblnd_msgtype2size(msg->ibm_type)); + return -EPROTO; + } - case IBLND_MSG_PUT_ACK: - case IBLND_MSG_GET_REQ: - if (kiblnd_unpack_rd(msg, flip)) - return -EPROTO; - break; + switch (msg->ibm_type) { + default: + CERROR("Unknown message type %x\n", msg->ibm_type); + return -EPROTO; - case IBLND_MSG_PUT_NAK: - case IBLND_MSG_PUT_DONE: - case IBLND_MSG_GET_DONE: - if (flip) - __swab32s(&msg->ibm_u.completion.ibcm_status); - break; + case IBLND_MSG_NOOP: + case IBLND_MSG_IMMEDIATE: + case IBLND_MSG_PUT_REQ: + break; - case IBLND_MSG_CONNREQ: - case IBLND_MSG_CONNACK: - if (flip) { - __swab16s(&msg->ibm_u.connparams.ibcp_queue_depth); - __swab16s(&msg->ibm_u.connparams.ibcp_max_frags); - __swab32s(&msg->ibm_u.connparams.ibcp_max_msg_size); - } - break; - } - return 0; + case IBLND_MSG_PUT_ACK: + case IBLND_MSG_GET_REQ: + if (kiblnd_unpack_rd(msg, flip)) + return -EPROTO; + break; + + case IBLND_MSG_PUT_NAK: + case IBLND_MSG_PUT_DONE: + case IBLND_MSG_GET_DONE: + if (flip) + __swab32s(&msg->ibm_u.completion.ibcm_status); + break; + + case IBLND_MSG_CONNREQ: + case IBLND_MSG_CONNACK: + if (flip) { + __swab16s(&msg->ibm_u.connparams.ibcp_queue_depth); + __swab16s(&msg->ibm_u.connparams.ibcp_max_frags); + __swab32s(&msg->ibm_u.connparams.ibcp_max_msg_size); + } + break; + } + return 0; } int -kiblnd_create_peer(struct lnet_ni *ni, kib_peer_ni_t **peerp, lnet_nid_t nid) +kiblnd_create_peer(struct lnet_ni *ni, struct kib_peer_ni **peerp, + lnet_nid_t nid) { - kib_peer_ni_t *peer_ni; - kib_net_t *net = ni->ni_data; - int cpt = lnet_cpt_of_nid(nid, ni); - unsigned long flags; + struct kib_peer_ni *peer_ni; + struct kib_net *net = ni->ni_data; + int cpt = lnet_cpt_of_nid(nid, ni); + unsigned long flags; LASSERT(net != NULL); LASSERT(nid != LNET_NID_ANY); LIBCFS_CPT_ALLOC(peer_ni, lnet_cpt_table(), cpt, sizeof(*peer_ni)); - if (peer_ni == NULL) { - CERROR("Cannot allocate peer_ni\n"); - return -ENOMEM; - } + if (!peer_ni) { + CERROR("Cannot allocate peer_ni\n"); + return -ENOMEM; + } peer_ni->ibp_ni = ni; peer_ni->ibp_nid = nid; @@ -335,9 +335,10 @@ kiblnd_create_peer(struct lnet_ni *ni, kib_peer_ni_t **peerp, lnet_nid_t nid) peer_ni->ibp_last_alive = 0; peer_ni->ibp_max_frags = IBLND_MAX_RDMA_FRAGS; peer_ni->ibp_queue_depth = ni->ni_net->net_tunables.lct_peer_tx_credits; + peer_ni->ibp_queue_depth_mod = 0; /* try to use the default */ atomic_set(&peer_ni->ibp_refcount, 1); /* 1 ref for caller */ - INIT_LIST_HEAD(&peer_ni->ibp_list); /* not in the peer_ni table yet */ + INIT_HLIST_NODE(&peer_ni->ibp_list); INIT_LIST_HEAD(&peer_ni->ibp_conns); INIT_LIST_HEAD(&peer_ni->ibp_tx_queue); @@ -356,9 +357,9 @@ kiblnd_create_peer(struct lnet_ni *ni, kib_peer_ni_t **peerp, lnet_nid_t nid) } void -kiblnd_destroy_peer (kib_peer_ni_t *peer_ni) +kiblnd_destroy_peer(struct kib_peer_ni *peer_ni) { - kib_net_t *net = peer_ni->ibp_ni->ni_data; + struct kib_net *net = peer_ni->ibp_ni->ni_data; LASSERT(net != NULL); LASSERT (atomic_read(&peer_ni->ibp_refcount) == 0); @@ -371,22 +372,22 @@ kiblnd_destroy_peer (kib_peer_ni_t *peer_ni) /* NB a peer_ni's connections keep a reference on their peer_ni until * they are destroyed, so we can be assured that _all_ state to do * with this peer_ni has been cleaned up when its refcount drops to - * zero. */ - atomic_dec(&net->ibn_npeers); + * zero. + */ + if (atomic_dec_and_test(&net->ibn_npeers)) + wake_up_var(&net->ibn_npeers); } -kib_peer_ni_t * +struct kib_peer_ni * kiblnd_find_peer_locked(struct lnet_ni *ni, lnet_nid_t nid) { /* the caller is responsible for accounting the additional reference - * that this creates */ - struct list_head *peer_list = kiblnd_nid2peerlist(nid); - struct list_head *tmp; - kib_peer_ni_t *peer_ni; - - list_for_each(tmp, peer_list) { + * that this creates + */ + struct kib_peer_ni *peer_ni; - peer_ni = list_entry(tmp, kib_peer_ni_t, ibp_list); + hash_for_each_possible(kiblnd_data.kib_peers, peer_ni, + ibp_list, nid) { LASSERT(!kiblnd_peer_idle(peer_ni)); /* @@ -409,47 +410,40 @@ kiblnd_find_peer_locked(struct lnet_ni *ni, lnet_nid_t nid) } void -kiblnd_unlink_peer_locked (kib_peer_ni_t *peer_ni) +kiblnd_unlink_peer_locked(struct kib_peer_ni *peer_ni) { LASSERT(list_empty(&peer_ni->ibp_conns)); - LASSERT (kiblnd_peer_active(peer_ni)); - list_del_init(&peer_ni->ibp_list); - /* lose peerlist's ref */ - kiblnd_peer_decref(peer_ni); + LASSERT(kiblnd_peer_active(peer_ni)); + hlist_del_init(&peer_ni->ibp_list); + /* lose peerlist's ref */ + kiblnd_peer_decref(peer_ni); } static int kiblnd_get_peer_info(struct lnet_ni *ni, int index, lnet_nid_t *nidp, int *count) { - kib_peer_ni_t *peer_ni; - struct list_head *ptmp; + struct kib_peer_ni *peer_ni; int i; unsigned long flags; read_lock_irqsave(&kiblnd_data.kib_global_lock, flags); - for (i = 0; i < kiblnd_data.kib_peer_hash_size; i++) { - - list_for_each(ptmp, &kiblnd_data.kib_peers[i]) { - - peer_ni = list_entry(ptmp, kib_peer_ni_t, ibp_list); - LASSERT(!kiblnd_peer_idle(peer_ni)); + hash_for_each(kiblnd_data.kib_peers, i, peer_ni, ibp_list) { + LASSERT(!kiblnd_peer_idle(peer_ni)); - if (peer_ni->ibp_ni != ni) - continue; + if (peer_ni->ibp_ni != ni) + continue; - if (index-- > 0) - continue; + if (index-- > 0) + continue; - *nidp = peer_ni->ibp_nid; - *count = atomic_read(&peer_ni->ibp_refcount); + *nidp = peer_ni->ibp_nid; + *count = atomic_read(&peer_ni->ibp_refcount); - read_unlock_irqrestore(&kiblnd_data.kib_global_lock, - flags); - return 0; - } + read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags); + return 0; } read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags); @@ -457,20 +451,17 @@ kiblnd_get_peer_info(struct lnet_ni *ni, int index, } static void -kiblnd_del_peer_locked (kib_peer_ni_t *peer_ni) +kiblnd_del_peer_locked(struct kib_peer_ni *peer_ni) { - struct list_head *ctmp; - struct list_head *cnxt; - kib_conn_t *conn; + struct kib_conn *cnxt; + struct kib_conn *conn; if (list_empty(&peer_ni->ibp_conns)) { kiblnd_unlink_peer_locked(peer_ni); } else { - list_for_each_safe(ctmp, cnxt, &peer_ni->ibp_conns) { - conn = list_entry(ctmp, kib_conn_t, ibc_list); - + list_for_each_entry_safe(conn, cnxt, &peer_ni->ibp_conns, + ibc_list) kiblnd_close_conn_locked(conn, 0); - } /* NB closing peer_ni's last conn unlinked it. */ } /* NB peer_ni now unlinked; might even be freed if the peer_ni table had the @@ -480,28 +471,28 @@ kiblnd_del_peer_locked (kib_peer_ni_t *peer_ni) static int kiblnd_del_peer(struct lnet_ni *ni, lnet_nid_t nid) { - struct list_head zombies = LIST_HEAD_INIT(zombies); - struct list_head *ptmp; - struct list_head *pnxt; - kib_peer_ni_t *peer_ni; - int lo; - int hi; - int i; - unsigned long flags; - int rc = -ENOENT; + LIST_HEAD(zombies); + struct hlist_node *pnxt; + struct kib_peer_ni *peer_ni; + int lo; + int hi; + int i; + unsigned long flags; + int rc = -ENOENT; write_lock_irqsave(&kiblnd_data.kib_global_lock, flags); - if (nid != LNET_NID_ANY) { - lo = hi = kiblnd_nid2peerlist(nid) - kiblnd_data.kib_peers; - } else { - lo = 0; - hi = kiblnd_data.kib_peer_hash_size - 1; - } + if (nid != LNET_NID_ANY) { + lo = hash_min(nid, HASH_BITS(kiblnd_data.kib_peers)); + hi = lo; + } else { + lo = 0; + hi = HASH_SIZE(kiblnd_data.kib_peers) - 1; + } for (i = lo; i <= hi; i++) { - list_for_each_safe(ptmp, pnxt, &kiblnd_data.kib_peers[i]) { - peer_ni = list_entry(ptmp, kib_peer_ni_t, ibp_list); + hlist_for_each_entry_safe(peer_ni, pnxt, + &kiblnd_data.kib_peers[i], ibp_list) { LASSERT(!kiblnd_peer_idle(peer_ni)); if (peer_ni->ibp_ni != ni) @@ -524,71 +515,66 @@ kiblnd_del_peer(struct lnet_ni *ni, lnet_nid_t nid) write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags); - kiblnd_txlist_done(&zombies, -EIO); + kiblnd_txlist_done(&zombies, -EIO, LNET_MSG_STATUS_LOCAL_ERROR); return rc; } -static kib_conn_t * +static struct kib_conn * kiblnd_get_conn_by_idx(struct lnet_ni *ni, int index) { - kib_peer_ni_t *peer_ni; - struct list_head *ptmp; - kib_conn_t *conn; - struct list_head *ctmp; - int i; - unsigned long flags; + struct kib_peer_ni *peer_ni; + struct kib_conn *conn; + struct list_head *ctmp; + int i; + unsigned long flags; read_lock_irqsave(&kiblnd_data.kib_global_lock, flags); - for (i = 0; i < kiblnd_data.kib_peer_hash_size; i++) { - list_for_each(ptmp, &kiblnd_data.kib_peers[i]) { + hash_for_each(kiblnd_data.kib_peers, i, peer_ni, ibp_list) { + LASSERT(!kiblnd_peer_idle(peer_ni)); - peer_ni = list_entry(ptmp, kib_peer_ni_t, ibp_list); - LASSERT(!kiblnd_peer_idle(peer_ni)); + if (peer_ni->ibp_ni != ni) + continue; - if (peer_ni->ibp_ni != ni) + list_for_each(ctmp, &peer_ni->ibp_conns) { + if (index-- > 0) continue; - list_for_each(ctmp, &peer_ni->ibp_conns) { - if (index-- > 0) - continue; - - conn = list_entry(ctmp, kib_conn_t, ibc_list); - kiblnd_conn_addref(conn); - read_unlock_irqrestore(&kiblnd_data.kib_global_lock, - flags); - return conn; - } + conn = list_entry(ctmp, struct kib_conn, ibc_list); + kiblnd_conn_addref(conn); + read_unlock_irqrestore(&kiblnd_data.kib_global_lock, + flags); + return conn; } } read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags); - return NULL; + return NULL; } static void -kiblnd_debug_rx (kib_rx_t *rx) +kiblnd_debug_rx(struct kib_rx *rx) { - CDEBUG(D_CONSOLE, " %p status %d msg_type %x cred %d\n", - rx, rx->rx_status, rx->rx_msg->ibm_type, - rx->rx_msg->ibm_credits); + CDEBUG(D_CONSOLE, " %p msg_type %x cred %d\n", + rx, rx->rx_msg->ibm_type, + rx->rx_msg->ibm_credits); } static void -kiblnd_debug_tx (kib_tx_t *tx) +kiblnd_debug_tx(struct kib_tx *tx) { - CDEBUG(D_CONSOLE, " %p snd %d q %d w %d rc %d dl %lx " + CDEBUG(D_CONSOLE, " %p snd %d q %d w %d rc %d dl %lld " "cookie %#llx msg %s%s type %x cred %d\n", tx, tx->tx_sending, tx->tx_queued, tx->tx_waiting, - tx->tx_status, tx->tx_deadline, tx->tx_cookie, + tx->tx_status, ktime_to_ns(tx->tx_deadline), tx->tx_cookie, tx->tx_lntmsg[0] == NULL ? "-" : "!", tx->tx_lntmsg[1] == NULL ? "-" : "!", tx->tx_msg->ibm_type, tx->tx_msg->ibm_credits); } void -kiblnd_debug_conn (kib_conn_t *conn) +kiblnd_debug_conn(struct kib_conn *conn) { struct list_head *tmp; int i; @@ -606,27 +592,27 @@ kiblnd_debug_conn (kib_conn_t *conn) CDEBUG(D_CONSOLE, " early_rxs:\n"); list_for_each(tmp, &conn->ibc_early_rxs) - kiblnd_debug_rx(list_entry(tmp, kib_rx_t, rx_list)); + kiblnd_debug_rx(list_entry(tmp, struct kib_rx, rx_list)); CDEBUG(D_CONSOLE, " tx_noops:\n"); list_for_each(tmp, &conn->ibc_tx_noops) - kiblnd_debug_tx(list_entry(tmp, kib_tx_t, tx_list)); + kiblnd_debug_tx(list_entry(tmp, struct kib_tx, tx_list)); CDEBUG(D_CONSOLE, " tx_queue_nocred:\n"); list_for_each(tmp, &conn->ibc_tx_queue_nocred) - kiblnd_debug_tx(list_entry(tmp, kib_tx_t, tx_list)); + kiblnd_debug_tx(list_entry(tmp, struct kib_tx, tx_list)); CDEBUG(D_CONSOLE, " tx_queue_rsrvd:\n"); list_for_each(tmp, &conn->ibc_tx_queue_rsrvd) - kiblnd_debug_tx(list_entry(tmp, kib_tx_t, tx_list)); + kiblnd_debug_tx(list_entry(tmp, struct kib_tx, tx_list)); CDEBUG(D_CONSOLE, " tx_queue:\n"); list_for_each(tmp, &conn->ibc_tx_queue) - kiblnd_debug_tx(list_entry(tmp, kib_tx_t, tx_list)); + kiblnd_debug_tx(list_entry(tmp, struct kib_tx, tx_list)); CDEBUG(D_CONSOLE, " active_txs:\n"); list_for_each(tmp, &conn->ibc_active_txs) - kiblnd_debug_tx(list_entry(tmp, kib_tx_t, tx_list)); + kiblnd_debug_tx(list_entry(tmp, struct kib_tx, tx_list)); CDEBUG(D_CONSOLE, " rxs:\n"); for (i = 0; i < IBLND_RX_MSGS(conn); i++) @@ -635,46 +621,22 @@ kiblnd_debug_conn (kib_conn_t *conn) spin_unlock(&conn->ibc_lock); } -int -kiblnd_translate_mtu(int value) -{ - switch (value) { - default: - return -1; - case 0: - return 0; - case 256: - return IB_MTU_256; - case 512: - return IB_MTU_512; - case 1024: - return IB_MTU_1024; - case 2048: - return IB_MTU_2048; - case 4096: - return IB_MTU_4096; - } -} - static void kiblnd_setup_mtu_locked(struct rdma_cm_id *cmid) { - int mtu; - /* XXX There is no path record for iWARP, set by netdev->change_mtu? */ if (cmid->route.path_rec == NULL) return; - mtu = kiblnd_translate_mtu(*kiblnd_tunables.kib_ib_mtu); - LASSERT (mtu >= 0); - if (mtu != 0) - cmid->route.path_rec->mtu = mtu; + if (*kiblnd_tunables.kib_ib_mtu) + cmid->route.path_rec->mtu = + ib_mtu_int_to_enum(*kiblnd_tunables.kib_ib_mtu); } static int -kiblnd_get_completion_vector(kib_conn_t *conn, int cpt) +kiblnd_get_completion_vector(struct kib_conn *conn, int cpt) { - cpumask_t *mask; + cpumask_var_t *mask; int vectors; int off; int i; @@ -688,8 +650,8 @@ kiblnd_get_completion_vector(kib_conn_t *conn, int cpt) /* hash NID to CPU id in this partition... */ ibp_nid = conn->ibc_peer->ibp_nid; - off = do_div(ibp_nid, cpumask_weight(mask)); - for_each_cpu(i, mask) { + off = do_div(ibp_nid, cpumask_weight(*mask)); + for_each_cpu(i, *mask) { if (off-- == 0) return i % vectors; } @@ -734,20 +696,32 @@ static unsigned int kiblnd_send_wrs(struct kib_conn *conn) * One WR for the LNet message * And ibc_max_frags for the transfer WRs */ - unsigned int ret = 1 + conn->ibc_max_frags; - __u32 dev_caps = conn->ibc_hdev->ibh_dev->ibd_dev_caps; + int ret; + int multiplier = 1 + conn->ibc_max_frags; + enum kib_dev_caps dev_caps = conn->ibc_hdev->ibh_dev->ibd_dev_caps; /* FastReg needs two extra WRs for map and invalidate */ if (dev_caps & IBLND_DEV_CAPS_FASTREG_ENABLED) - ret += 2; + multiplier += 2; /* account for a maximum of ibc_queue_depth in-flight transfers */ - ret *= conn->ibc_queue_depth; - return ret; + ret = multiplier * conn->ibc_queue_depth; + + if (ret > conn->ibc_hdev->ibh_max_qp_wr) { + CDEBUG(D_NET, "peer_credits %u will result in send work " + "request size %d larger than maximum %d device " + "can handle\n", conn->ibc_queue_depth, ret, + conn->ibc_hdev->ibh_max_qp_wr); + conn->ibc_queue_depth = + conn->ibc_hdev->ibh_max_qp_wr / multiplier; + } + + /* don't go beyond the maximum the device can handle */ + return min(ret, conn->ibc_hdev->ibh_max_qp_wr); } -kib_conn_t * -kiblnd_create_conn(kib_peer_ni_t *peer_ni, struct rdma_cm_id *cmid, +struct kib_conn * +kiblnd_create_conn(struct kib_peer_ni *peer_ni, struct rdma_cm_id *cmid, int state, int version) { /* CAVEAT EMPTOR: @@ -758,14 +732,14 @@ kiblnd_create_conn(kib_peer_ni_t *peer_ni, struct rdma_cm_id *cmid, * to destroy 'cmid' here since I'm called from the CM which still has * its ref on 'cmid'). */ rwlock_t *glock = &kiblnd_data.kib_global_lock; - kib_net_t *net = peer_ni->ibp_ni->ni_data; - kib_dev_t *dev; - struct ib_qp_init_attr *init_qp_attr; + struct kib_net *net = peer_ni->ibp_ni->ni_data; + struct kib_dev *dev; + struct ib_qp_init_attr init_qp_attr = {}; struct kib_sched_info *sched; #ifdef HAVE_IB_CQ_INIT_ATTR struct ib_cq_init_attr cq_attr = {}; #endif - kib_conn_t *conn; + struct kib_conn *conn; struct ib_cq *cq; unsigned long flags; int cpt; @@ -791,19 +765,11 @@ kiblnd_create_conn(kib_peer_ni_t *peer_ni, struct rdma_cm_id *cmid, */ cpt = sched->ibs_cpt; - LIBCFS_CPT_ALLOC(init_qp_attr, lnet_cpt_table(), cpt, - sizeof(*init_qp_attr)); - if (init_qp_attr == NULL) { - CERROR("Can't allocate qp_attr for %s\n", - libcfs_nid2str(peer_ni->ibp_nid)); - goto failed_0; - } - LIBCFS_CPT_ALLOC(conn, lnet_cpt_table(), cpt, sizeof(*conn)); if (conn == NULL) { CERROR("Can't allocate connection for %s\n", libcfs_nid2str(peer_ni->ibp_nid)); - goto failed_1; + goto failed_0; } conn->ibc_state = IBLND_CONN_INIT; @@ -813,6 +779,8 @@ kiblnd_create_conn(kib_peer_ni_t *peer_ni, struct rdma_cm_id *cmid, conn->ibc_cmid = cmid; conn->ibc_max_frags = peer_ni->ibp_max_frags; conn->ibc_queue_depth = peer_ni->ibp_queue_depth; + conn->ibc_rxs = NULL; + conn->ibc_rx_pages = NULL; INIT_LIST_HEAD(&conn->ibc_early_rxs); INIT_LIST_HEAD(&conn->ibc_tx_noops); @@ -820,6 +788,7 @@ kiblnd_create_conn(kib_peer_ni_t *peer_ni, struct rdma_cm_id *cmid, INIT_LIST_HEAD(&conn->ibc_tx_queue_rsrvd); INIT_LIST_HEAD(&conn->ibc_tx_queue_nocred); INIT_LIST_HEAD(&conn->ibc_active_txs); + INIT_LIST_HEAD(&conn->ibc_zombie_txs); spin_lock_init(&conn->ibc_lock); LIBCFS_CPT_ALLOC(conn->ibc_connvars, lnet_cpt_table(), cpt, @@ -857,20 +826,6 @@ kiblnd_create_conn(kib_peer_ni_t *peer_ni, struct rdma_cm_id *cmid, write_unlock_irqrestore(glock, flags); - LIBCFS_CPT_ALLOC(conn->ibc_rxs, lnet_cpt_table(), cpt, - IBLND_RX_MSGS(conn) * sizeof(kib_rx_t)); - if (conn->ibc_rxs == NULL) { - CERROR("Cannot allocate RX buffers\n"); - goto failed_2; - } - - rc = kiblnd_alloc_pages(&conn->ibc_rx_pages, cpt, - IBLND_RX_MSG_PAGES(conn)); - if (rc != 0) - goto failed_2; - - kiblnd_map_rx_descs(conn); - #ifdef HAVE_IB_CQ_INIT_ATTR cq_attr.cqe = IBLND_CQ_ENTRIES(conn); cq_attr.comp_vector = kiblnd_get_completion_vector(conn, cpt); @@ -903,46 +858,71 @@ kiblnd_create_conn(kib_peer_ni_t *peer_ni, struct rdma_cm_id *cmid, goto failed_2; } - init_qp_attr->event_handler = kiblnd_qp_event; - init_qp_attr->qp_context = conn; - init_qp_attr->cap.max_send_sge = *kiblnd_tunables.kib_wrq_sge; - init_qp_attr->cap.max_recv_sge = 1; - init_qp_attr->sq_sig_type = IB_SIGNAL_REQ_WR; - init_qp_attr->qp_type = IB_QPT_RC; - init_qp_attr->send_cq = cq; - init_qp_attr->recv_cq = cq; + init_qp_attr.event_handler = kiblnd_qp_event; + init_qp_attr.qp_context = conn; + init_qp_attr.cap.max_send_sge = *kiblnd_tunables.kib_wrq_sge; + init_qp_attr.cap.max_recv_sge = 1; + init_qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR; + init_qp_attr.qp_type = IB_QPT_RC; + init_qp_attr.send_cq = cq; + init_qp_attr.recv_cq = cq; - conn->ibc_sched = sched; + if (peer_ni->ibp_queue_depth_mod && + peer_ni->ibp_queue_depth_mod < peer_ni->ibp_queue_depth) { + conn->ibc_queue_depth = peer_ni->ibp_queue_depth_mod; + CDEBUG(D_NET, "Use reduced queue depth %u (from %u)\n", + peer_ni->ibp_queue_depth_mod, + peer_ni->ibp_queue_depth); + } do { - init_qp_attr->cap.max_send_wr = kiblnd_send_wrs(conn); - init_qp_attr->cap.max_recv_wr = IBLND_RECV_WRS(conn); - - rc = rdma_create_qp(cmid, conn->ibc_hdev->ibh_pd, init_qp_attr); - if (!rc || conn->ibc_queue_depth < 2) + /* kiblnd_send_wrs() can change the connection's queue depth if + * the maximum work requests for the device is maxed out + */ + init_qp_attr.cap.max_send_wr = kiblnd_send_wrs(conn); + init_qp_attr.cap.max_recv_wr = IBLND_RECV_WRS(conn); + rc = rdma_create_qp(cmid, conn->ibc_hdev->ibh_pd, + &init_qp_attr); + if (rc != -ENOMEM || conn->ibc_queue_depth < 2) break; - conn->ibc_queue_depth--; } while (rc); if (rc) { CERROR("Can't create QP: %d, send_wr: %d, recv_wr: %d, " "send_sge: %d, recv_sge: %d\n", - rc, init_qp_attr->cap.max_send_wr, - init_qp_attr->cap.max_recv_wr, - init_qp_attr->cap.max_send_sge, - init_qp_attr->cap.max_recv_sge); + rc, init_qp_attr.cap.max_send_wr, + init_qp_attr.cap.max_recv_wr, + init_qp_attr.cap.max_send_sge, + init_qp_attr.cap.max_recv_sge); goto failed_2; } - if (conn->ibc_queue_depth != peer_ni->ibp_queue_depth) + conn->ibc_sched = sched; + + if (!peer_ni->ibp_queue_depth_mod && + conn->ibc_queue_depth != peer_ni->ibp_queue_depth) { CWARN("peer %s - queue depth reduced from %u to %u" " to allow for qp creation\n", libcfs_nid2str(peer_ni->ibp_nid), peer_ni->ibp_queue_depth, conn->ibc_queue_depth); + peer_ni->ibp_queue_depth_mod = conn->ibc_queue_depth; + } + + LIBCFS_CPT_ALLOC(conn->ibc_rxs, lnet_cpt_table(), cpt, + IBLND_RX_MSGS(conn) * sizeof(struct kib_rx)); + if (conn->ibc_rxs == NULL) { + CERROR("Cannot allocate RX buffers\n"); + goto failed_2; + } + + rc = kiblnd_alloc_pages(&conn->ibc_rx_pages, cpt, + IBLND_RX_MSG_PAGES(conn)); + if (rc != 0) + goto failed_2; - LIBCFS_FREE(init_qp_attr, sizeof(*init_qp_attr)); + kiblnd_map_rx_descs(conn); /* 1 ref for caller and each rxmsg */ atomic_set(&conn->ibc_refcount, 1 + IBLND_RX_MSGS(conn)); @@ -987,19 +967,17 @@ kiblnd_create_conn(kib_peer_ni_t *peer_ni, struct rdma_cm_id *cmid, return conn; failed_2: - kiblnd_destroy_conn(conn, true); - failed_1: - LIBCFS_FREE(init_qp_attr, sizeof(*init_qp_attr)); + kiblnd_destroy_conn(conn); + LIBCFS_FREE(conn, sizeof(*conn)); failed_0: return NULL; } void -kiblnd_destroy_conn(kib_conn_t *conn, bool free_conn) +kiblnd_destroy_conn(struct kib_conn *conn) { struct rdma_cm_id *cmid = conn->ibc_cmid; - kib_peer_ni_t *peer_ni = conn->ibc_peer; - int rc; + struct kib_peer_ni *peer_ni = conn->ibc_peer; LASSERT (!in_interrupt()); LASSERT (atomic_read(&conn->ibc_refcount) == 0); @@ -1030,19 +1008,17 @@ kiblnd_destroy_conn(kib_conn_t *conn, bool free_conn) if (cmid != NULL && cmid->qp != NULL) rdma_destroy_qp(cmid); - if (conn->ibc_cq != NULL) { - rc = ib_destroy_cq(conn->ibc_cq); - if (rc != 0) - CWARN("Error destroying CQ: %d\n", rc); - } + if (conn->ibc_cq) + ib_destroy_cq(conn->ibc_cq); + + kiblnd_txlist_done(&conn->ibc_zombie_txs, -ECONNABORTED, + LNET_MSG_STATUS_OK); if (conn->ibc_rx_pages != NULL) kiblnd_unmap_rx_descs(conn); - if (conn->ibc_rxs != NULL) { - LIBCFS_FREE(conn->ibc_rxs, - IBLND_RX_MSGS(conn) * sizeof(kib_rx_t)); - } + if (conn->ibc_rxs != NULL) + CFS_FREE_PTR_ARRAY(conn->ibc_rxs, IBLND_RX_MSGS(conn)); if (conn->ibc_connvars != NULL) LIBCFS_FREE(conn->ibc_connvars, sizeof(*conn->ibc_connvars)); @@ -1052,28 +1028,23 @@ kiblnd_destroy_conn(kib_conn_t *conn, bool free_conn) /* See CAVEAT EMPTOR above in kiblnd_create_conn */ if (conn->ibc_state != IBLND_CONN_INIT) { - kib_net_t *net = peer_ni->ibp_ni->ni_data; + struct kib_net *net = peer_ni->ibp_ni->ni_data; kiblnd_peer_decref(peer_ni); rdma_destroy_id(cmid); atomic_dec(&net->ibn_nconns); } - - if (free_conn) - LIBCFS_FREE(conn, sizeof(*conn)); } int -kiblnd_close_peer_conns_locked(kib_peer_ni_t *peer_ni, int why) +kiblnd_close_peer_conns_locked(struct kib_peer_ni *peer_ni, int why) { - kib_conn_t *conn; - struct list_head *ctmp; - struct list_head *cnxt; - int count = 0; - - list_for_each_safe(ctmp, cnxt, &peer_ni->ibp_conns) { - conn = list_entry(ctmp, kib_conn_t, ibc_list); + struct kib_conn *conn; + struct kib_conn *cnxt; + int count = 0; + list_for_each_entry_safe(conn, cnxt, &peer_ni->ibp_conns, + ibc_list) { CDEBUG(D_NET, "Closing conn -> %s, " "version: %x, reason: %d\n", libcfs_nid2str(peer_ni->ibp_nid), @@ -1087,17 +1058,15 @@ kiblnd_close_peer_conns_locked(kib_peer_ni_t *peer_ni, int why) } int -kiblnd_close_stale_conns_locked(kib_peer_ni_t *peer_ni, +kiblnd_close_stale_conns_locked(struct kib_peer_ni *peer_ni, int version, __u64 incarnation) { - kib_conn_t *conn; - struct list_head *ctmp; - struct list_head *cnxt; - int count = 0; - - list_for_each_safe(ctmp, cnxt, &peer_ni->ibp_conns) { - conn = list_entry(ctmp, kib_conn_t, ibc_list); + struct kib_conn *conn; + struct kib_conn *cnxt; + int count = 0; + list_for_each_entry_safe(conn, cnxt, &peer_ni->ibp_conns, + ibc_list) { if (conn->ibc_version == version && conn->ibc_incarnation == incarnation) continue; @@ -1118,28 +1087,27 @@ kiblnd_close_stale_conns_locked(kib_peer_ni_t *peer_ni, static int kiblnd_close_matching_conns(struct lnet_ni *ni, lnet_nid_t nid) { - kib_peer_ni_t *peer_ni; - struct list_head *ptmp; - struct list_head *pnxt; - int lo; - int hi; - int i; - unsigned long flags; - int count = 0; + struct kib_peer_ni *peer_ni; + struct hlist_node *pnxt; + int lo; + int hi; + int i; + unsigned long flags; + int count = 0; write_lock_irqsave(&kiblnd_data.kib_global_lock, flags); - if (nid != LNET_NID_ANY) - lo = hi = kiblnd_nid2peerlist(nid) - kiblnd_data.kib_peers; - else { + if (nid != LNET_NID_ANY) { + lo = hash_min(nid, HASH_BITS(kiblnd_data.kib_peers)); + hi = lo; + } else { lo = 0; - hi = kiblnd_data.kib_peer_hash_size - 1; + hi = HASH_SIZE(kiblnd_data.kib_peers) - 1; } for (i = lo; i <= hi; i++) { - list_for_each_safe(ptmp, pnxt, &kiblnd_data.kib_peers[i]) { - - peer_ni = list_entry(ptmp, kib_peer_ni_t, ibp_list); + hlist_for_each_entry_safe(peer_ni, pnxt, + &kiblnd_data.kib_peers[i], ibp_list) { LASSERT(!kiblnd_peer_idle(peer_ni)); if (peer_ni->ibp_ni != ni) @@ -1184,7 +1152,7 @@ kiblnd_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg) break; } case IOC_LIBCFS_GET_CONN: { - kib_conn_t *conn; + struct kib_conn *conn; rc = 0; conn = kiblnd_get_conn_by_idx(ni, data->ioc_count); @@ -1216,38 +1184,7 @@ kiblnd_ctl(struct lnet_ni *ni, unsigned int cmd, void *arg) } static void -kiblnd_query(struct lnet_ni *ni, lnet_nid_t nid, cfs_time_t *when) -{ - cfs_time_t last_alive = 0; - cfs_time_t now = cfs_time_current(); - rwlock_t *glock = &kiblnd_data.kib_global_lock; - kib_peer_ni_t *peer_ni; - unsigned long flags; - - read_lock_irqsave(glock, flags); - - peer_ni = kiblnd_find_peer_locked(ni, nid); - if (peer_ni != NULL) - last_alive = peer_ni->ibp_last_alive; - - read_unlock_irqrestore(glock, flags); - - if (last_alive != 0) - *when = last_alive; - - /* peer_ni is not persistent in hash, trigger peer_ni creation - * and connection establishment with a NULL tx */ - if (peer_ni == NULL) - kiblnd_launch_tx(ni, NULL, nid); - - CDEBUG(D_NET, "peer_ni %s %p, alive %ld secs ago\n", - libcfs_nid2str(nid), peer_ni, - last_alive ? cfs_duration_sec(now - last_alive) : -1); - return; -} - -static void -kiblnd_free_pages(kib_pages_t *p) +kiblnd_free_pages(struct kib_pages *p) { int npages = p->ibp_npages; int i; @@ -1257,23 +1194,23 @@ kiblnd_free_pages(kib_pages_t *p) __free_page(p->ibp_pages[i]); } - LIBCFS_FREE(p, offsetof(kib_pages_t, ibp_pages[npages])); + LIBCFS_FREE(p, offsetof(struct kib_pages, ibp_pages[npages])); } int -kiblnd_alloc_pages(kib_pages_t **pp, int cpt, int npages) +kiblnd_alloc_pages(struct kib_pages **pp, int cpt, int npages) { - kib_pages_t *p; - int i; + struct kib_pages *p; + int i; LIBCFS_CPT_ALLOC(p, lnet_cpt_table(), cpt, - offsetof(kib_pages_t, ibp_pages[npages])); + offsetof(struct kib_pages, ibp_pages[npages])); if (p == NULL) { CERROR("Can't allocate descriptor for %d pages\n", npages); return -ENOMEM; } - memset(p, 0, offsetof(kib_pages_t, ibp_pages[npages])); + memset(p, 0, offsetof(struct kib_pages, ibp_pages[npages])); p->ibp_npages = npages; for (i = 0; i < npages; i++) { @@ -1291,9 +1228,9 @@ kiblnd_alloc_pages(kib_pages_t **pp, int cpt, int npages) } void -kiblnd_unmap_rx_descs(kib_conn_t *conn) +kiblnd_unmap_rx_descs(struct kib_conn *conn) { - kib_rx_t *rx; + struct kib_rx *rx; int i; LASSERT (conn->ibc_rxs != NULL); @@ -1316,9 +1253,9 @@ kiblnd_unmap_rx_descs(kib_conn_t *conn) } void -kiblnd_map_rx_descs(kib_conn_t *conn) +kiblnd_map_rx_descs(struct kib_conn *conn) { - kib_rx_t *rx; + struct kib_rx *rx; struct page *pg; int pg_off; int ipg; @@ -1329,7 +1266,7 @@ kiblnd_map_rx_descs(kib_conn_t *conn) rx = &conn->ibc_rxs[i]; rx->rx_conn = conn; - rx->rx_msg = (kib_msg_t *)(((char *)page_address(pg)) + pg_off); + rx->rx_msg = (struct kib_msg *)(((char *)page_address(pg)) + pg_off); rx->rx_msgaddr = kiblnd_dma_map_single(conn->ibc_hdev->ibh_ibdev, @@ -1355,11 +1292,11 @@ kiblnd_map_rx_descs(kib_conn_t *conn) } static void -kiblnd_unmap_tx_pool(kib_tx_pool_t *tpo) +kiblnd_unmap_tx_pool(struct kib_tx_pool *tpo) { - kib_hca_dev_t *hdev = tpo->tpo_hdev; - kib_tx_t *tx; - int i; + struct kib_hca_dev *hdev = tpo->tpo_hdev; + struct kib_tx *tx; + int i; LASSERT (tpo->tpo_pool.po_allocated == 0); @@ -1378,10 +1315,10 @@ kiblnd_unmap_tx_pool(kib_tx_pool_t *tpo) tpo->tpo_hdev = NULL; } -static kib_hca_dev_t * -kiblnd_current_hdev(kib_dev_t *dev) +static struct kib_hca_dev * +kiblnd_current_hdev(struct kib_dev *dev) { - kib_hca_dev_t *hdev; + struct kib_hca_dev *hdev; unsigned long flags; int i = 0; @@ -1391,8 +1328,7 @@ kiblnd_current_hdev(kib_dev_t *dev) if (i++ % 50 == 0) CDEBUG(D_NET, "%s: Wait for failover\n", dev->ibd_ifname); - set_current_state(TASK_INTERRUPTIBLE); - schedule_timeout(cfs_time_seconds(1) / 100); + schedule_timeout_interruptible(cfs_time_seconds(1) / 100); read_lock_irqsave(&kiblnd_data.kib_global_lock, flags); } @@ -1406,14 +1342,14 @@ kiblnd_current_hdev(kib_dev_t *dev) } static void -kiblnd_map_tx_pool(kib_tx_pool_t *tpo) -{ - kib_pages_t *txpgs = tpo->tpo_tx_pages; - kib_pool_t *pool = &tpo->tpo_pool; - kib_net_t *net = pool->po_owner->ps_net; - kib_dev_t *dev; - struct page *page; - kib_tx_t *tx; +kiblnd_map_tx_pool(struct kib_tx_pool *tpo) +{ + struct kib_pages *txpgs = tpo->tpo_tx_pages; + struct kib_pool *pool = &tpo->tpo_pool; + struct kib_net *net = pool->po_owner->ps_net; + struct kib_dev *dev; + struct page *page; + struct kib_tx *tx; int page_offset; int ipage; int i; @@ -1422,11 +1358,11 @@ kiblnd_map_tx_pool(kib_tx_pool_t *tpo) dev = net->ibn_dev; - /* pre-mapped messages are not bigger than 1 page */ - CLASSERT (IBLND_MSG_SIZE <= PAGE_SIZE); + /* pre-mapped messages are not bigger than 1 page */ + BUILD_BUG_ON(IBLND_MSG_SIZE > PAGE_SIZE); - /* No fancy arithmetic when we do the buffer calculations */ - CLASSERT (PAGE_SIZE % IBLND_MSG_SIZE == 0); + /* No fancy arithmetic when we do the buffer calculations */ + BUILD_BUG_ON(PAGE_SIZE % IBLND_MSG_SIZE != 0); tpo->tpo_hdev = kiblnd_current_hdev(dev); @@ -1434,8 +1370,8 @@ kiblnd_map_tx_pool(kib_tx_pool_t *tpo) page = txpgs->ibp_pages[ipage]; tx = &tpo->tpo_tx_descs[i]; - tx->tx_msg = (kib_msg_t *)(((char *)page_address(page)) + - page_offset); + tx->tx_msg = (struct kib_msg *)(((char *)page_address(page)) + + page_offset); tx->tx_msgaddr = kiblnd_dma_map_single(tpo->tpo_hdev->ibh_ibdev, tx->tx_msg, @@ -1459,13 +1395,16 @@ kiblnd_map_tx_pool(kib_tx_pool_t *tpo) } static void -kiblnd_destroy_fmr_pool(kib_fmr_pool_t *fpo) +kiblnd_destroy_fmr_pool(struct kib_fmr_pool *fpo) { LASSERT(fpo->fpo_map_count == 0); +#ifdef HAVE_FMR_POOL_API if (fpo->fpo_is_fmr && fpo->fmr.fpo_fmr_pool) { ib_destroy_fmr_pool(fpo->fmr.fpo_fmr_pool); - } else { + } else +#endif /* HAVE_FMR_POOL_API */ + { struct kib_fast_reg_descriptor *frd, *tmp; int i = 0; @@ -1493,7 +1432,7 @@ kiblnd_destroy_fmr_pool(kib_fmr_pool_t *fpo) static void kiblnd_destroy_fmr_pool_list(struct list_head *head) { - kib_fmr_pool_t *fpo, *tmp; + struct kib_fmr_pool *fpo, *tmp; list_for_each_entry_safe(fpo, tmp, head, fpo_list) { list_del(&fpo->fpo_list); @@ -1519,10 +1458,12 @@ kiblnd_fmr_flush_trigger(struct lnet_ioctl_config_o2iblnd_tunables *tunables, return max(IBLND_FMR_POOL_FLUSH, size); } -static int kiblnd_alloc_fmr_pool(kib_fmr_poolset_t *fps, kib_fmr_pool_t *fpo) +#ifdef HAVE_FMR_POOL_API +static int kiblnd_alloc_fmr_pool(struct kib_fmr_poolset *fps, + struct kib_fmr_pool *fpo) { struct ib_fmr_pool_param param = { - .max_pages_per_fmr = LNET_MAX_PAYLOAD/PAGE_SIZE, + .max_pages_per_fmr = LNET_MAX_IOV, .page_shift = PAGE_SHIFT, .access = (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE), @@ -1546,14 +1487,18 @@ static int kiblnd_alloc_fmr_pool(kib_fmr_poolset_t *fps, kib_fmr_pool_t *fpo) return rc; } +#endif /* HAVE_FMR_POOL_API */ -static int kiblnd_alloc_freg_pool(kib_fmr_poolset_t *fps, kib_fmr_pool_t *fpo, - __u32 dev_caps) +static int kiblnd_alloc_freg_pool(struct kib_fmr_poolset *fps, + struct kib_fmr_pool *fpo, + enum kib_dev_caps dev_caps) { struct kib_fast_reg_descriptor *frd, *tmp; int i, rc; +#ifdef HAVE_FMR_POOL_API fpo->fpo_is_fmr = false; +#endif INIT_LIST_HEAD(&fpo->fast_reg.fpo_pool_list); fpo->fast_reg.fpo_pool_size = 0; @@ -1569,7 +1514,7 @@ static int kiblnd_alloc_freg_pool(kib_fmr_poolset_t *fps, kib_fmr_pool_t *fpo, #ifndef HAVE_IB_MAP_MR_SG frd->frd_frpl = ib_alloc_fast_reg_page_list(fpo->fpo_hdev->ibh_ibdev, - LNET_MAX_PAYLOAD/PAGE_SIZE); + LNET_MAX_IOV); if (IS_ERR(frd->frd_frpl)) { rc = PTR_ERR(frd->frd_frpl); CERROR("Failed to allocate ib_fast_reg_page_list: %d\n", @@ -1581,7 +1526,7 @@ static int kiblnd_alloc_freg_pool(kib_fmr_poolset_t *fps, kib_fmr_pool_t *fpo, #ifdef HAVE_IB_ALLOC_FAST_REG_MR frd->frd_mr = ib_alloc_fast_reg_mr(fpo->fpo_hdev->ibh_pd, - LNET_MAX_PAYLOAD/PAGE_SIZE); + LNET_MAX_IOV); #else /* * it is expected to get here if this is an MLX-5 card. @@ -1591,11 +1536,18 @@ static int kiblnd_alloc_freg_pool(kib_fmr_poolset_t *fps, kib_fmr_pool_t *fpo, * gaps. So we will need to track them here. */ frd->frd_mr = ib_alloc_mr(fpo->fpo_hdev->ibh_pd, - (dev_caps & - IBLND_DEV_CAPS_FASTREG_GAPS_SUPPORT) ? +#ifdef IB_MR_TYPE_SG_GAPS + ((*kiblnd_tunables.kib_use_fastreg_gaps == 1) && + (dev_caps & IBLND_DEV_CAPS_FASTREG_GAPS_SUPPORT)) ? IB_MR_TYPE_SG_GAPS : IB_MR_TYPE_MEM_REG, - LNET_MAX_PAYLOAD/PAGE_SIZE); +#else + IB_MR_TYPE_MEM_REG, +#endif + LNET_MAX_IOV); + if ((*kiblnd_tunables.kib_use_fastreg_gaps == 1) && + (dev_caps & IBLND_DEV_CAPS_FASTREG_GAPS_SUPPORT)) + CWARN("using IB_MR_TYPE_SG_GAPS, expect a performance drop\n"); #endif if (IS_ERR(frd->frd_mr)) { rc = PTR_ERR(frd->frd_mr); @@ -1638,11 +1590,11 @@ out: return rc; } -static int -kiblnd_create_fmr_pool(kib_fmr_poolset_t *fps, kib_fmr_pool_t **pp_fpo) +static int kiblnd_create_fmr_pool(struct kib_fmr_poolset *fps, + struct kib_fmr_pool **pp_fpo) { - kib_dev_t *dev = fps->fps_net->ibn_dev; - kib_fmr_pool_t *fpo; + struct kib_dev *dev = fps->fps_net->ibn_dev; + struct kib_fmr_pool *fpo; int rc; LIBCFS_CPT_ALLOC(fpo, lnet_cpt_table(), fps->fps_cpt, sizeof(*fpo)); @@ -1653,15 +1605,17 @@ kiblnd_create_fmr_pool(kib_fmr_poolset_t *fps, kib_fmr_pool_t **pp_fpo) fpo->fpo_hdev = kiblnd_current_hdev(dev); +#ifdef HAVE_FMR_POOL_API if (dev->ibd_dev_caps & IBLND_DEV_CAPS_FMR_ENABLED) rc = kiblnd_alloc_fmr_pool(fps, fpo); else +#endif /* HAVE_FMR_POOL_API */ rc = kiblnd_alloc_freg_pool(fps, fpo, dev->ibd_dev_caps); if (rc) goto out_fpo; - fpo->fpo_deadline = cfs_time_shift(IBLND_POOL_DEADLINE); - fpo->fpo_owner = fps; + fpo->fpo_deadline = ktime_get_seconds() + IBLND_POOL_DEADLINE; + fpo->fpo_owner = fps; *pp_fpo = fpo; return 0; @@ -1673,29 +1627,30 @@ out_fpo: } static void -kiblnd_fail_fmr_poolset(kib_fmr_poolset_t *fps, struct list_head *zombies) +kiblnd_fail_fmr_poolset(struct kib_fmr_poolset *fps, struct list_head *zombies) { + struct kib_fmr_pool *fpo; + if (fps->fps_net == NULL) /* intialized? */ return; spin_lock(&fps->fps_lock); - while (!list_empty(&fps->fps_pool_list)) { - kib_fmr_pool_t *fpo = list_entry(fps->fps_pool_list.next, - kib_fmr_pool_t, fpo_list); + while ((fpo = list_first_entry_or_null(&fps->fps_pool_list, + struct kib_fmr_pool, + fpo_list)) != NULL) { fpo->fpo_failed = 1; - list_del(&fpo->fpo_list); if (fpo->fpo_map_count == 0) - list_add(&fpo->fpo_list, zombies); + list_move(&fpo->fpo_list, zombies); else - list_add(&fpo->fpo_list, &fps->fps_failed_pool_list); + list_move(&fpo->fpo_list, &fps->fps_failed_pool_list); } spin_unlock(&fps->fps_lock); } static void -kiblnd_fini_fmr_poolset(kib_fmr_poolset_t *fps) +kiblnd_fini_fmr_poolset(struct kib_fmr_poolset *fps) { if (fps->fps_net != NULL) { /* initialized? */ kiblnd_destroy_fmr_pool_list(&fps->fps_failed_pool_list); @@ -1704,14 +1659,14 @@ kiblnd_fini_fmr_poolset(kib_fmr_poolset_t *fps) } static int -kiblnd_init_fmr_poolset(kib_fmr_poolset_t *fps, int cpt, int ncpts, - kib_net_t *net, +kiblnd_init_fmr_poolset(struct kib_fmr_poolset *fps, int cpt, int ncpts, + struct kib_net *net, struct lnet_ioctl_config_o2iblnd_tunables *tunables) { - kib_fmr_pool_t *fpo; - int rc; + struct kib_fmr_pool *fpo; + int rc; - memset(fps, 0, sizeof(kib_fmr_poolset_t)); + memset(fps, 0, sizeof(struct kib_fmr_poolset)); fps->fps_net = net; fps->fps_cpt = cpt; @@ -1732,19 +1687,20 @@ kiblnd_init_fmr_poolset(kib_fmr_poolset_t *fps, int cpt, int ncpts, } static int -kiblnd_fmr_pool_is_idle(kib_fmr_pool_t *fpo, cfs_time_t now) +kiblnd_fmr_pool_is_idle(struct kib_fmr_pool *fpo, time64_t now) { if (fpo->fpo_map_count != 0) /* still in use */ return 0; if (fpo->fpo_failed) return 1; - return cfs_time_aftereq(now, fpo->fpo_deadline); + return now >= fpo->fpo_deadline; } +#if defined(HAVE_FMR_POOL_API) || !defined(HAVE_IB_MAP_MR_SG) static int -kiblnd_map_tx_pages(kib_tx_t *tx, kib_rdma_desc_t *rd) +kiblnd_map_tx_pages(struct kib_tx *tx, struct kib_rdma_desc *rd) { - kib_hca_dev_t *hdev; + struct kib_hca_dev *hdev; __u64 *pages = tx->tx_pages; int npages; int size; @@ -1762,33 +1718,36 @@ kiblnd_map_tx_pages(kib_tx_t *tx, kib_rdma_desc_t *rd) return npages; } +#endif void -kiblnd_fmr_pool_unmap(kib_fmr_t *fmr, int status) +kiblnd_fmr_pool_unmap(struct kib_fmr *fmr, int status) { - struct list_head zombies = LIST_HEAD_INIT(zombies); - kib_fmr_pool_t *fpo = fmr->fmr_pool; - kib_fmr_poolset_t *fps; - cfs_time_t now = cfs_time_current(); - kib_fmr_pool_t *tmp; - int rc; + LIST_HEAD(zombies); + struct kib_fmr_pool *fpo = fmr->fmr_pool; + struct kib_fmr_poolset *fps; + time64_t now = ktime_get_seconds(); + struct kib_fmr_pool *tmp; if (!fpo) return; fps = fpo->fpo_owner; + +#ifdef HAVE_FMR_POOL_API if (fpo->fpo_is_fmr) { if (fmr->fmr_pfmr) { - rc = ib_fmr_pool_unmap(fmr->fmr_pfmr); - LASSERT(!rc); + ib_fmr_pool_unmap(fmr->fmr_pfmr); fmr->fmr_pfmr = NULL; } if (status) { - rc = ib_flush_fmr_pool(fpo->fmr.fpo_fmr_pool); + int rc = ib_flush_fmr_pool(fpo->fmr.fpo_fmr_pool); LASSERT(!rc); } - } else { + } else +#endif /* HAVE_FMR_POOL_API */ + { struct kib_fast_reg_descriptor *frd = fmr->fmr_frd; if (frd) { @@ -1820,25 +1779,29 @@ kiblnd_fmr_pool_unmap(kib_fmr_t *fmr, int status) kiblnd_destroy_fmr_pool_list(&zombies); } -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) +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) { - kib_fmr_pool_t *fpo; - __u64 *pages = tx->tx_pages; + struct kib_fmr_pool *fpo; __u64 version; bool is_rx = (rd != tx->tx_rd); - bool tx_pages_mapped = 0; +#ifdef HAVE_FMR_POOL_API + __u64 *pages = tx->tx_pages; + bool tx_pages_mapped = false; int npages = 0; +#endif int rc; again: spin_lock(&fps->fps_lock); version = fps->fps_version; list_for_each_entry(fpo, &fps->fps_pool_list, fpo_list) { - fpo->fpo_deadline = cfs_time_shift(IBLND_POOL_DEADLINE); + fpo->fpo_deadline = ktime_get_seconds() + IBLND_POOL_DEADLINE; fpo->fpo_map_count++; +#ifdef HAVE_FMR_POOL_API + fmr->fmr_pfmr = NULL; if (fpo->fpo_is_fmr) { struct ib_pool_fmr *pfmr; @@ -1846,21 +1809,23 @@ again: if (!tx_pages_mapped) { npages = kiblnd_map_tx_pages(tx, rd); - tx_pages_mapped = 1; + tx_pages_mapped = true; } - pfmr = ib_fmr_pool_map_phys(fpo->fmr.fpo_fmr_pool, - pages, npages, iov); + pfmr = kib_fmr_pool_map(fpo->fmr.fpo_fmr_pool, + pages, npages, iov); if (likely(!IS_ERR(pfmr))) { fmr->fmr_key = is_rx ? pfmr->fmr->rkey - : pfmr->fmr->lkey; + : pfmr->fmr->lkey; fmr->fmr_frd = NULL; fmr->fmr_pfmr = pfmr; fmr->fmr_pool = fpo; return 0; } rc = PTR_ERR(pfmr); - } else { + } else +#endif /* HAVE_FMR_POOL_API */ + { if (!list_empty(&fpo->fast_reg.fpo_pool_list)) { struct kib_fast_reg_descriptor *frd; #ifdef HAVE_IB_MAP_MR_SG @@ -1872,9 +1837,10 @@ again: #endif struct ib_mr *mr; - frd = list_first_entry(&fpo->fast_reg.fpo_pool_list, - struct kib_fast_reg_descriptor, - frd_list); + frd = list_first_entry( + &fpo->fast_reg.fpo_pool_list, + struct kib_fast_reg_descriptor, + frd_list); list_del(&frd->frd_list); spin_unlock(&fps->fps_lock); @@ -1902,14 +1868,14 @@ again: #ifdef HAVE_IB_MAP_MR_SG #ifdef HAVE_IB_MAP_MR_SG_5ARGS n = ib_map_mr_sg(mr, tx->tx_frags, - tx->tx_nfrags, NULL, PAGE_SIZE); + rd->rd_nfrags, NULL, PAGE_SIZE); #else n = ib_map_mr_sg(mr, tx->tx_frags, - tx->tx_nfrags, PAGE_SIZE); -#endif - if (unlikely(n != tx->tx_nfrags)) { - CERROR("Failed to map mr %d/%d " - "elements\n", n, tx->tx_nfrags); + rd->rd_nfrags, PAGE_SIZE); +#endif /* HAVE_IB_MAP_MR_SG_5ARGS */ + if (unlikely(n != rd->rd_nfrags)) { + CERROR("Failed to map mr %d/%d elements\n", + n, rd->rd_nfrags); return n < 0 ? n : -EINVAL; } @@ -1924,15 +1890,15 @@ again: wr->key = is_rx ? mr->rkey : mr->lkey; wr->access = (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE); -#else +#else /* HAVE_IB_MAP_MR_SG */ if (!tx_pages_mapped) { npages = kiblnd_map_tx_pages(tx, rd); - tx_pages_mapped = 1; + tx_pages_mapped = true; } LASSERT(npages <= frpl->max_page_list_len); memcpy(frpl->page_list, pages, - sizeof(*pages) * npages); + sizeof(*pages) * npages); /* Prepare FastReg WR */ wr = &frd->frd_fastreg_wr; @@ -1947,15 +1913,14 @@ again: wr->wr.wr.fast_reg.page_shift = PAGE_SHIFT; wr->wr.wr.fast_reg.length = nob; wr->wr.wr.fast_reg.rkey = - is_rx ? mr->rkey : mr->lkey; + is_rx ? mr->rkey : mr->lkey; wr->wr.wr.fast_reg.access_flags = - (IB_ACCESS_LOCAL_WRITE | - IB_ACCESS_REMOTE_WRITE); -#endif + (IB_ACCESS_LOCAL_WRITE | + IB_ACCESS_REMOTE_WRITE); +#endif /* HAVE_IB_MAP_MR_SG */ fmr->fmr_key = is_rx ? mr->rkey : mr->lkey; fmr->fmr_frd = frd; - fmr->fmr_pfmr = NULL; fmr->fmr_pool = fpo; return 0; } @@ -1981,12 +1946,12 @@ again: spin_unlock(&fps->fps_lock); CDEBUG(D_NET, "Another thread is allocating new " "FMR pool, waiting for her to complete\n"); - schedule(); + wait_var_event(fps, !fps->fps_increasing); goto again; } - if (cfs_time_before(cfs_time_current(), fps->fps_next_retry)) { + if (ktime_get_seconds() < fps->fps_next_retry) { /* someone failed recently */ spin_unlock(&fps->fps_lock); return -EAGAIN; @@ -1999,11 +1964,12 @@ again: rc = kiblnd_create_fmr_pool(fps, &fpo); spin_lock(&fps->fps_lock); fps->fps_increasing = 0; + wake_up_var(fps); if (rc == 0) { fps->fps_version++; list_add_tail(&fpo->fpo_list, &fps->fps_pool_list); } else { - fps->fps_next_retry = cfs_time_shift(IBLND_POOL_RETRY); + fps->fps_next_retry = ktime_get_seconds() + IBLND_POOL_RETRY; } spin_unlock(&fps->fps_lock); @@ -2011,7 +1977,7 @@ again: } static void -kiblnd_fini_pool(kib_pool_t *pool) +kiblnd_fini_pool(struct kib_pool *pool) { LASSERT(list_empty(&pool->po_free_list)); LASSERT(pool->po_allocated == 0); @@ -2020,24 +1986,25 @@ kiblnd_fini_pool(kib_pool_t *pool) } static void -kiblnd_init_pool(kib_poolset_t *ps, kib_pool_t *pool, int size) +kiblnd_init_pool(struct kib_poolset *ps, struct kib_pool *pool, int size) { CDEBUG(D_NET, "Initialize %s pool\n", ps->ps_name); - memset(pool, 0, sizeof(kib_pool_t)); + memset(pool, 0, sizeof(struct kib_pool)); INIT_LIST_HEAD(&pool->po_free_list); - pool->po_deadline = cfs_time_shift(IBLND_POOL_DEADLINE); - pool->po_owner = ps; - pool->po_size = size; + pool->po_deadline = ktime_get_seconds() + IBLND_POOL_DEADLINE; + pool->po_owner = ps; + pool->po_size = size; } static void kiblnd_destroy_pool_list(struct list_head *head) { - kib_pool_t *pool; + struct kib_pool *pool; - while (!list_empty(head)) { - pool = list_entry(head->next, kib_pool_t, po_list); + while ((pool = list_first_entry_or_null(head, + struct kib_pool, + po_list)) != NULL) { list_del(&pool->po_list); LASSERT(pool->po_owner != NULL); @@ -2046,27 +2013,28 @@ kiblnd_destroy_pool_list(struct list_head *head) } static void -kiblnd_fail_poolset(kib_poolset_t *ps, struct list_head *zombies) +kiblnd_fail_poolset(struct kib_poolset *ps, struct list_head *zombies) { + struct kib_pool *po; + if (ps->ps_net == NULL) /* intialized? */ return; spin_lock(&ps->ps_lock); - while (!list_empty(&ps->ps_pool_list)) { - kib_pool_t *po = list_entry(ps->ps_pool_list.next, - kib_pool_t, po_list); + while ((po = list_first_entry_or_null(&ps->ps_pool_list, + struct kib_pool, + po_list)) != NULL) { po->po_failed = 1; - list_del(&po->po_list); if (po->po_allocated == 0) - list_add(&po->po_list, zombies); + list_move(&po->po_list, zombies); else - list_add(&po->po_list, &ps->ps_failed_pool_list); + list_move(&po->po_list, &ps->ps_failed_pool_list); } spin_unlock(&ps->ps_lock); } static void -kiblnd_fini_poolset(kib_poolset_t *ps) +kiblnd_fini_poolset(struct kib_poolset *ps) { if (ps->ps_net != NULL) { /* initialized? */ kiblnd_destroy_pool_list(&ps->ps_failed_pool_list); @@ -2075,17 +2043,17 @@ kiblnd_fini_poolset(kib_poolset_t *ps) } static int -kiblnd_init_poolset(kib_poolset_t *ps, int cpt, - kib_net_t *net, char *name, int size, +kiblnd_init_poolset(struct kib_poolset *ps, int cpt, + struct kib_net *net, char *name, int size, kib_ps_pool_create_t po_create, kib_ps_pool_destroy_t po_destroy, kib_ps_node_init_t nd_init, kib_ps_node_fini_t nd_fini) { - kib_pool_t *pool; - int rc; + struct kib_pool *pool; + int rc; - memset(ps, 0, sizeof(kib_poolset_t)); + memset(ps, 0, sizeof(struct kib_poolset)); ps->ps_cpt = cpt; ps->ps_net = net; @@ -2111,22 +2079,22 @@ kiblnd_init_poolset(kib_poolset_t *ps, int cpt, } static int -kiblnd_pool_is_idle(kib_pool_t *pool, cfs_time_t now) +kiblnd_pool_is_idle(struct kib_pool *pool, time64_t now) { if (pool->po_allocated != 0) /* still in use */ return 0; if (pool->po_failed) return 1; - return cfs_time_aftereq(now, pool->po_deadline); + return now >= pool->po_deadline; } void -kiblnd_pool_free_node(kib_pool_t *pool, struct list_head *node) +kiblnd_pool_free_node(struct kib_pool *pool, struct list_head *node) { - struct list_head zombies = LIST_HEAD_INIT(zombies); - kib_poolset_t *ps = pool->po_owner; - kib_pool_t *tmp; - cfs_time_t now = cfs_time_current(); + LIST_HEAD(zombies); + struct kib_poolset *ps = pool->po_owner; + struct kib_pool *tmp; + time64_t now = ktime_get_seconds(); spin_lock(&ps->ps_lock); @@ -2152,14 +2120,14 @@ kiblnd_pool_free_node(kib_pool_t *pool, struct list_head *node) } struct list_head * -kiblnd_pool_alloc_node(kib_poolset_t *ps) +kiblnd_pool_alloc_node(struct kib_poolset *ps) { struct list_head *node; - kib_pool_t *pool; + struct kib_pool *pool; int rc; unsigned int interval = 1; - cfs_time_t time_before; - unsigned int trips = 0; + ktime_t time_before; + unsigned int trips = 0; again: spin_lock(&ps->ps_lock); @@ -2168,7 +2136,8 @@ again: continue; pool->po_allocated++; - pool->po_deadline = cfs_time_shift(IBLND_POOL_DEADLINE); + pool->po_deadline = ktime_get_seconds() + + IBLND_POOL_DEADLINE; node = pool->po_free_list.next; list_del(node); @@ -2185,20 +2154,18 @@ again: /* another thread is allocating a new pool */ spin_unlock(&ps->ps_lock); trips++; - CDEBUG(D_NET, "Another thread is allocating new " - "%s pool, waiting %d HZs for her to complete." - "trips = %d\n", + CDEBUG(D_NET, + "Another thread is allocating new %s pool, waiting %d jiffies for her to complete. trips = %d\n", ps->ps_name, interval, trips); - set_current_state(TASK_INTERRUPTIBLE); - schedule_timeout(interval); + schedule_timeout_interruptible(interval); if (interval < cfs_time_seconds(1)) interval *= 2; goto again; } - if (cfs_time_before(cfs_time_current(), ps->ps_next_retry)) { + if (ktime_get_seconds() < ps->ps_next_retry) { /* someone failed recently */ spin_unlock(&ps->ps_lock); return NULL; @@ -2208,17 +2175,17 @@ again: spin_unlock(&ps->ps_lock); CDEBUG(D_NET, "%s pool exhausted, allocate new pool\n", ps->ps_name); - time_before = cfs_time_current(); + time_before = ktime_get(); rc = ps->ps_pool_create(ps, ps->ps_pool_size, &pool); - CDEBUG(D_NET, "ps_pool_create took %lu HZ to complete", - cfs_time_current() - time_before); + CDEBUG(D_NET, "ps_pool_create took %lld ms to complete", + ktime_ms_delta(ktime_get(), time_before)); spin_lock(&ps->ps_lock); ps->ps_increasing = 0; if (rc == 0) { list_add_tail(&pool->po_list, &ps->ps_pool_list); } else { - ps->ps_next_retry = cfs_time_shift(IBLND_POOL_RETRY); + ps->ps_next_retry = ktime_get_seconds() + IBLND_POOL_RETRY; CERROR("Can't allocate new %s pool because out of memory\n", ps->ps_name); } @@ -2228,10 +2195,11 @@ again: } static void -kiblnd_destroy_tx_pool(kib_pool_t *pool) +kiblnd_destroy_tx_pool(struct kib_pool *pool) { - kib_tx_pool_t *tpo = container_of(pool, kib_tx_pool_t, tpo_pool); - int i; + struct kib_tx_pool *tpo = container_of(pool, struct kib_tx_pool, + tpo_pool); + int i; LASSERT (pool->po_allocated == 0); @@ -2243,38 +2211,33 @@ kiblnd_destroy_tx_pool(kib_pool_t *pool) if (tpo->tpo_tx_descs == NULL) goto out; - for (i = 0; i < pool->po_size; i++) { - kib_tx_t *tx = &tpo->tpo_tx_descs[i]; + for (i = 0; i < pool->po_size; i++) { + struct kib_tx *tx = &tpo->tpo_tx_descs[i]; int wrq_sge = *kiblnd_tunables.kib_wrq_sge; list_del(&tx->tx_list); - if (tx->tx_pages != NULL) - LIBCFS_FREE(tx->tx_pages, - LNET_MAX_IOV * - sizeof(*tx->tx_pages)); - if (tx->tx_frags != NULL) - LIBCFS_FREE(tx->tx_frags, - (1 + IBLND_MAX_RDMA_FRAGS) * - sizeof(*tx->tx_frags)); - if (tx->tx_wrq != NULL) - LIBCFS_FREE(tx->tx_wrq, - (1 + IBLND_MAX_RDMA_FRAGS) * - sizeof(*tx->tx_wrq)); + if (tx->tx_pages != NULL) + CFS_FREE_PTR_ARRAY(tx->tx_pages, LNET_MAX_IOV); + if (tx->tx_frags != NULL) + CFS_FREE_PTR_ARRAY(tx->tx_frags, + (1 + IBLND_MAX_RDMA_FRAGS)); + if (tx->tx_wrq != NULL) + CFS_FREE_PTR_ARRAY(tx->tx_wrq, + (1 + IBLND_MAX_RDMA_FRAGS)); if (tx->tx_sge != NULL) - LIBCFS_FREE(tx->tx_sge, - (1 + IBLND_MAX_RDMA_FRAGS) * wrq_sge * - sizeof(*tx->tx_sge)); - if (tx->tx_rd != NULL) - LIBCFS_FREE(tx->tx_rd, - offsetof(kib_rdma_desc_t, - rd_frags[IBLND_MAX_RDMA_FRAGS])); - } + CFS_FREE_PTR_ARRAY(tx->tx_sge, + (1 + IBLND_MAX_RDMA_FRAGS) * + wrq_sge); + if (tx->tx_rd != NULL) + LIBCFS_FREE(tx->tx_rd, + offsetof(struct kib_rdma_desc, + rd_frags[IBLND_MAX_RDMA_FRAGS])); + } - LIBCFS_FREE(tpo->tpo_tx_descs, - pool->po_size * sizeof(kib_tx_t)); + CFS_FREE_PTR_ARRAY(tpo->tpo_tx_descs, pool->po_size); out: - kiblnd_fini_pool(pool); - LIBCFS_FREE(tpo, sizeof(kib_tx_pool_t)); + kiblnd_fini_pool(pool); + CFS_FREE_PTR(tpo); } static int kiblnd_tx_pool_size(struct lnet_ni *ni, int ncpts) @@ -2289,12 +2252,12 @@ static int kiblnd_tx_pool_size(struct lnet_ni *ni, int ncpts) } static int -kiblnd_create_tx_pool(kib_poolset_t *ps, int size, kib_pool_t **pp_po) +kiblnd_create_tx_pool(struct kib_poolset *ps, int size, struct kib_pool **pp_po) { int i; int npg; - kib_pool_t *pool; - kib_tx_pool_t *tpo; + struct kib_pool *pool; + struct kib_tx_pool *tpo; LIBCFS_CPT_ALLOC(tpo, lnet_cpt_table(), ps->ps_cpt, sizeof(*tpo)); if (tpo == NULL) { @@ -2310,22 +2273,22 @@ kiblnd_create_tx_pool(kib_poolset_t *ps, int size, kib_pool_t **pp_po) npg = (size * IBLND_MSG_SIZE + PAGE_SIZE - 1) / PAGE_SIZE; if (kiblnd_alloc_pages(&tpo->tpo_tx_pages, ps->ps_cpt, npg) != 0) { CERROR("Can't allocate tx pages: %d\n", npg); - LIBCFS_FREE(tpo, sizeof(kib_tx_pool_t)); + CFS_FREE_PTR(tpo); return -ENOMEM; } LIBCFS_CPT_ALLOC(tpo->tpo_tx_descs, lnet_cpt_table(), ps->ps_cpt, - size * sizeof(kib_tx_t)); + size * sizeof(struct kib_tx)); if (tpo->tpo_tx_descs == NULL) { CERROR("Can't allocate %d tx descriptors\n", size); ps->ps_pool_destroy(pool); return -ENOMEM; } - memset(tpo->tpo_tx_descs, 0, size * sizeof(kib_tx_t)); + memset(tpo->tpo_tx_descs, 0, size * sizeof(struct kib_tx)); for (i = 0; i < size; i++) { - kib_tx_t *tx = &tpo->tpo_tx_descs[i]; + struct kib_tx *tx = &tpo->tpo_tx_descs[i]; int wrq_sge = *kiblnd_tunables.kib_wrq_sge; tx->tx_pool = tpo; @@ -2358,7 +2321,7 @@ kiblnd_create_tx_pool(kib_poolset_t *ps, int size, kib_pool_t **pp_po) break; LIBCFS_CPT_ALLOC(tx->tx_rd, lnet_cpt_table(), ps->ps_cpt, - offsetof(kib_rdma_desc_t, + offsetof(struct kib_rdma_desc, rd_frags[IBLND_MAX_RDMA_FRAGS])); if (tx->tx_rd == NULL) break; @@ -2375,23 +2338,24 @@ kiblnd_create_tx_pool(kib_poolset_t *ps, int size, kib_pool_t **pp_po) } static void -kiblnd_tx_init(kib_pool_t *pool, struct list_head *node) +kiblnd_tx_init(struct kib_pool *pool, struct list_head *node) { - kib_tx_poolset_t *tps = container_of(pool->po_owner, kib_tx_poolset_t, - tps_poolset); - kib_tx_t *tx = list_entry(node, kib_tx_t, tx_list); + struct kib_tx_poolset *tps = container_of(pool->po_owner, + struct kib_tx_poolset, + tps_poolset); + struct kib_tx *tx = list_entry(node, struct kib_tx, tx_list); tx->tx_cookie = tps->tps_next_tx_cookie++; } static void -kiblnd_net_fini_pools(kib_net_t *net) +kiblnd_net_fini_pools(struct kib_net *net) { int i; cfs_cpt_for_each(i, lnet_cpt_table()) { - kib_tx_poolset_t *tps; - kib_fmr_poolset_t *fps; + struct kib_tx_poolset *tps; + struct kib_fmr_poolset *fps; if (net->ibn_tx_ps != NULL) { tps = net->ibn_tx_ps[i]; @@ -2416,7 +2380,7 @@ kiblnd_net_fini_pools(kib_net_t *net) } static int -kiblnd_net_init_pools(kib_net_t *net, struct lnet_ni *ni, __u32 *cpts, +kiblnd_net_init_pools(struct kib_net *net, struct lnet_ni *ni, __u32 *cpts, int ncpts) { struct lnet_ioctl_config_o2iblnd_tunables *tunables; @@ -2461,7 +2425,7 @@ kiblnd_net_init_pools(kib_net_t *net, struct lnet_ni *ni, __u32 *cpts, * FMR pool and map-on-demand if premapping failed */ net->ibn_fmr_ps = cfs_percpt_alloc(lnet_cpt_table(), - sizeof(kib_fmr_poolset_t)); + sizeof(struct kib_fmr_poolset)); if (net->ibn_fmr_ps == NULL) { CERROR("Failed to allocate FMR pool array\n"); rc = -ENOMEM; @@ -2486,7 +2450,7 @@ kiblnd_net_init_pools(kib_net_t *net, struct lnet_ni *ni, __u32 *cpts, create_tx_pool: #endif net->ibn_tx_ps = cfs_percpt_alloc(lnet_cpt_table(), - sizeof(kib_tx_poolset_t)); + sizeof(struct kib_tx_poolset)); if (net->ibn_tx_ps == NULL) { CERROR("Failed to allocate tx pool array\n"); rc = -ENOMEM; @@ -2516,10 +2480,95 @@ kiblnd_net_init_pools(kib_net_t *net, struct lnet_ni *ni, __u32 *cpts, } static int -kiblnd_hdev_get_attr(kib_hca_dev_t *hdev) +kiblnd_port_get_attr(struct kib_hca_dev *hdev) +{ + struct ib_port_attr *port_attr; + int rc; + unsigned long flags; + rwlock_t *g_lock = &kiblnd_data.kib_global_lock; + + LIBCFS_ALLOC(port_attr, sizeof(*port_attr)); + if (port_attr == NULL) { + CDEBUG(D_NETERROR, "Out of memory\n"); + return -ENOMEM; + } + + rc = ib_query_port(hdev->ibh_ibdev, hdev->ibh_port, port_attr); + + write_lock_irqsave(g_lock, flags); + + if (rc == 0) + hdev->ibh_state = port_attr->state == IB_PORT_ACTIVE + ? IBLND_DEV_PORT_ACTIVE + : IBLND_DEV_PORT_DOWN; + + write_unlock_irqrestore(g_lock, flags); + LIBCFS_FREE(port_attr, sizeof(*port_attr)); + + if (rc != 0) { + CDEBUG(D_NETERROR, "Failed to query IB port: %d\n", rc); + return rc; + } + return 0; +} + +static inline void +kiblnd_set_ni_fatal_on(struct kib_hca_dev *hdev, int val) +{ + struct kib_net *net; + + /* for health check */ + list_for_each_entry(net, &hdev->ibh_dev->ibd_nets, ibn_list) { + if (val) + CDEBUG(D_NETERROR, "Fatal device error for NI %s\n", + libcfs_nid2str(net->ibn_ni->ni_nid)); + atomic_set(&net->ibn_ni->ni_fatal_error_on, val); + } +} + +void +kiblnd_event_handler(struct ib_event_handler *handler, struct ib_event *event) +{ + rwlock_t *g_lock = &kiblnd_data.kib_global_lock; + struct kib_hca_dev *hdev; + unsigned long flags; + + hdev = container_of(handler, struct kib_hca_dev, ibh_event_handler); + + write_lock_irqsave(g_lock, flags); + + switch (event->event) { + case IB_EVENT_DEVICE_FATAL: + CDEBUG(D_NET, "IB device fatal\n"); + hdev->ibh_state = IBLND_DEV_FATAL; + kiblnd_set_ni_fatal_on(hdev, 1); + break; + case IB_EVENT_PORT_ACTIVE: + CDEBUG(D_NET, "IB port active\n"); + if (event->element.port_num == hdev->ibh_port) { + hdev->ibh_state = IBLND_DEV_PORT_ACTIVE; + kiblnd_set_ni_fatal_on(hdev, 0); + } + break; + case IB_EVENT_PORT_ERR: + CDEBUG(D_NET, "IB port err\n"); + if (event->element.port_num == hdev->ibh_port) { + hdev->ibh_state = IBLND_DEV_PORT_DOWN; + kiblnd_set_ni_fatal_on(hdev, 1); + } + break; + default: + break; + } + write_unlock_irqrestore(g_lock, flags); +} + +static int +kiblnd_hdev_get_attr(struct kib_hca_dev *hdev) { struct ib_device_attr *dev_attr; int rc = 0; + int rc2 = 0; /* It's safe to assume a HCA can handle a page size * matching that of the native system */ @@ -2544,28 +2593,43 @@ kiblnd_hdev_get_attr(kib_hca_dev_t *hdev) #endif hdev->ibh_mr_size = dev_attr->max_mr_size; + hdev->ibh_max_qp_wr = dev_attr->max_qp_wr; /* Setup device Memory Registration capabilities */ +#ifdef HAVE_FMR_POOL_API +#ifdef HAVE_IB_DEVICE_OPS + if (hdev->ibh_ibdev->ops.alloc_fmr && + hdev->ibh_ibdev->ops.dealloc_fmr && + hdev->ibh_ibdev->ops.map_phys_fmr && + hdev->ibh_ibdev->ops.unmap_fmr) { +#else if (hdev->ibh_ibdev->alloc_fmr && hdev->ibh_ibdev->dealloc_fmr && hdev->ibh_ibdev->map_phys_fmr && hdev->ibh_ibdev->unmap_fmr) { +#endif LCONSOLE_INFO("Using FMR for registration\n"); hdev->ibh_dev->ibd_dev_caps |= IBLND_DEV_CAPS_FMR_ENABLED; - } else if (dev_attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) { + } else +#endif /* HAVE_FMR_POOL_API */ + if (dev_attr->device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) { LCONSOLE_INFO("Using FastReg for registration\n"); hdev->ibh_dev->ibd_dev_caps |= IBLND_DEV_CAPS_FASTREG_ENABLED; #ifndef HAVE_IB_ALLOC_FAST_REG_MR +#ifdef IB_DEVICE_SG_GAPS_REG if (dev_attr->device_cap_flags & IB_DEVICE_SG_GAPS_REG) hdev->ibh_dev->ibd_dev_caps |= IBLND_DEV_CAPS_FASTREG_GAPS_SUPPORT; #endif +#endif } else { rc = -ENOSYS; } - if (rc == 0 && hdev->ibh_mr_size == ~0ULL) - hdev->ibh_mr_shift = 64; - else if (rc != 0) + rc2 = kiblnd_port_get_attr(hdev); + if (rc2 != 0) + return rc2; + + if (rc != 0) rc = -EINVAL; #ifndef HAVE_IB_DEVICE_ATTRS @@ -2583,7 +2647,7 @@ out_clean_attr: #ifdef HAVE_IB_GET_DMA_MR static void -kiblnd_hdev_cleanup_mrs(kib_hca_dev_t *hdev) +kiblnd_hdev_cleanup_mrs(struct kib_hca_dev *hdev) { if (hdev->ibh_mrs == NULL) return; @@ -2595,8 +2659,11 @@ kiblnd_hdev_cleanup_mrs(kib_hca_dev_t *hdev) #endif void -kiblnd_hdev_destroy(kib_hca_dev_t *hdev) +kiblnd_hdev_destroy(struct kib_hca_dev *hdev) { + if (hdev->ibh_event_handler.device != NULL) + ib_unregister_event_handler(&hdev->ibh_event_handler); + #ifdef HAVE_IB_GET_DMA_MR kiblnd_hdev_cleanup_mrs(hdev); #endif @@ -2612,7 +2679,7 @@ kiblnd_hdev_destroy(kib_hca_dev_t *hdev) #ifdef HAVE_IB_GET_DMA_MR static int -kiblnd_hdev_setup_mrs(kib_hca_dev_t *hdev) +kiblnd_hdev_setup_mrs(struct kib_hca_dev *hdev) { struct ib_mr *mr; int acflags = IB_ACCESS_LOCAL_WRITE | @@ -2638,7 +2705,7 @@ kiblnd_dummy_callback(struct rdma_cm_id *cmid, struct rdma_cm_event *event) } static int -kiblnd_dev_need_failover(kib_dev_t *dev) +kiblnd_dev_need_failover(struct kib_dev *dev, struct net *ns) { struct rdma_cm_id *cmid; struct sockaddr_in srcaddr; @@ -2660,8 +2727,8 @@ kiblnd_dev_need_failover(kib_dev_t *dev) * * a. rdma_bind_addr(), it will conflict with listener cmid * b. rdma_resolve_addr() to zero addr */ - cmid = kiblnd_rdma_create_id(kiblnd_dummy_callback, dev, RDMA_PS_TCP, - IB_QPT_RC); + cmid = kiblnd_rdma_create_id(ns, kiblnd_dummy_callback, dev, + RDMA_PS_TCP, IB_QPT_RC); if (IS_ERR(cmid)) { rc = PTR_ERR(cmid); CERROR("Failed to create cmid for failover: %d\n", rc); @@ -2690,16 +2757,16 @@ kiblnd_dev_need_failover(kib_dev_t *dev) } int -kiblnd_dev_failover(kib_dev_t *dev) +kiblnd_dev_failover(struct kib_dev *dev, struct net *ns) { - struct list_head zombie_tpo = LIST_HEAD_INIT(zombie_tpo); - struct list_head zombie_ppo = LIST_HEAD_INIT(zombie_ppo); - struct list_head zombie_fpo = LIST_HEAD_INIT(zombie_fpo); + LIST_HEAD(zombie_tpo); + LIST_HEAD(zombie_ppo); + LIST_HEAD(zombie_fpo); struct rdma_cm_id *cmid = NULL; - kib_hca_dev_t *hdev = NULL; - kib_hca_dev_t *old; + struct kib_hca_dev *hdev = NULL; + struct kib_hca_dev *old; struct ib_pd *pd; - kib_net_t *net; + struct kib_net *net; struct sockaddr_in addr; unsigned long flags; int rc = 0; @@ -2709,7 +2776,7 @@ kiblnd_dev_failover(kib_dev_t *dev) dev->ibd_can_failover || dev->ibd_hdev == NULL); - rc = kiblnd_dev_need_failover(dev); + rc = kiblnd_dev_need_failover(dev, ns); if (rc <= 0) goto out; @@ -2730,8 +2797,8 @@ kiblnd_dev_failover(kib_dev_t *dev) rdma_destroy_id(cmid); } - cmid = kiblnd_rdma_create_id(kiblnd_cm_callback, dev, RDMA_PS_TCP, - IB_QPT_RC); + cmid = kiblnd_rdma_create_id(ns, kiblnd_cm_callback, dev, RDMA_PS_TCP, + IB_QPT_RC); if (IS_ERR(cmid)) { rc = PTR_ERR(cmid); CERROR("Failed to create cmid for failover: %d\n", rc); @@ -2765,6 +2832,7 @@ kiblnd_dev_failover(kib_dev_t *dev) hdev->ibh_dev = dev; hdev->ibh_cmid = cmid; hdev->ibh_ibdev = cmid->device; + hdev->ibh_port = cmid->port_num; #ifdef HAVE_IB_ALLOC_PD_2ARGS pd = ib_alloc_pd(cmid->device, 0); @@ -2799,6 +2867,10 @@ kiblnd_dev_failover(kib_dev_t *dev) } #endif + INIT_IB_EVENT_HANDLER(&hdev->ibh_event_handler, + hdev->ibh_ibdev, kiblnd_event_handler); + ib_register_event_handler(&hdev->ibh_event_handler); + write_lock_irqsave(&kiblnd_data.kib_global_lock, flags); old = dev->ibd_hdev; @@ -2836,9 +2908,9 @@ kiblnd_dev_failover(kib_dev_t *dev) } void -kiblnd_destroy_dev (kib_dev_t *dev) +kiblnd_destroy_dev(struct kib_dev *dev) { - LASSERT (dev->ibd_nnets == 0); + LASSERT(dev->ibd_nnets == 0); LASSERT(list_empty(&dev->ibd_nets)); list_del(&dev->ibd_fail_list); @@ -2850,80 +2922,26 @@ kiblnd_destroy_dev (kib_dev_t *dev) LIBCFS_FREE(dev, sizeof(*dev)); } -static kib_dev_t * -kiblnd_create_dev(char *ifname) -{ - struct net_device *netdev; - kib_dev_t *dev; - __u32 netmask; - __u32 ip; - int up; - int rc; - - rc = lnet_ipif_query(ifname, &up, &ip, &netmask); - if (rc != 0) { - CERROR("Can't query IPoIB interface %s: %d\n", - ifname, rc); - return NULL; - } - - if (!up) { - CERROR("Can't query IPoIB interface %s: it's down\n", ifname); - return NULL; - } - - LIBCFS_ALLOC(dev, sizeof(*dev)); - if (dev == NULL) - return NULL; - - netdev = dev_get_by_name(&init_net, ifname); - if (netdev == NULL) { - dev->ibd_can_failover = 0; - } else { - dev->ibd_can_failover = !!(netdev->flags & IFF_MASTER); - dev_put(netdev); - } - - INIT_LIST_HEAD(&dev->ibd_nets); - INIT_LIST_HEAD(&dev->ibd_list); /* not yet in kib_devs */ - INIT_LIST_HEAD(&dev->ibd_fail_list); - dev->ibd_ifip = ip; - strcpy(&dev->ibd_ifname[0], ifname); - - /* initialize the device */ - rc = kiblnd_dev_failover(dev); - if (rc != 0) { - CERROR("Can't initialize device: %d\n", rc); - LIBCFS_FREE(dev, sizeof(*dev)); - return NULL; - } - - list_add_tail(&dev->ibd_list, - &kiblnd_data.kib_devs); - return dev; -} - static void kiblnd_base_shutdown(void) { - struct kib_sched_info *sched; - int i; + struct kib_sched_info *sched; + struct kib_peer_ni *peer_ni; + int i; LASSERT(list_empty(&kiblnd_data.kib_devs)); - CDEBUG(D_MALLOC, "before LND base cleanup: kmem %d\n", - atomic_read(&libcfs_kmemory)); + CDEBUG(D_MALLOC, "before LND base cleanup: kmem %lld\n", + libcfs_kmem_read()); - switch (kiblnd_data.kib_init) { - default: - LBUG(); + switch (kiblnd_data.kib_init) { + default: + LBUG(); - case IBLND_INIT_ALL: - case IBLND_INIT_DATA: - LASSERT (kiblnd_data.kib_peers != NULL); - for (i = 0; i < kiblnd_data.kib_peer_hash_size; i++) { - LASSERT(list_empty(&kiblnd_data.kib_peers[i])); - } + case IBLND_INIT_ALL: + case IBLND_INIT_DATA: + hash_for_each(kiblnd_data.kib_peers, i, peer_ni, ibp_list) + LASSERT(0); LASSERT(list_empty(&kiblnd_data.kib_connd_zombies)); LASSERT(list_empty(&kiblnd_data.kib_connd_conns)); LASSERT(list_empty(&kiblnd_data.kib_reconn_list)); @@ -2934,41 +2952,29 @@ kiblnd_base_shutdown(void) /* NB: we really want to stop scheduler threads net by net * instead of the whole module, this should be improved - * with dynamic configuration LNet */ + * with dynamic configuration LNet. + */ cfs_percpt_for_each(sched, i, kiblnd_data.kib_scheds) wake_up_all(&sched->ibs_waitq); - wake_up_all(&kiblnd_data.kib_connd_waitq); - wake_up_all(&kiblnd_data.kib_failover_waitq); - - i = 2; - while (atomic_read(&kiblnd_data.kib_nthreads) != 0) { - i++; - /* power of 2? */ - CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, - "Waiting for %d threads to terminate\n", - atomic_read(&kiblnd_data.kib_nthreads)); - set_current_state(TASK_UNINTERRUPTIBLE); - schedule_timeout(cfs_time_seconds(1)); - } - - /* fall through */ + wake_up(&kiblnd_data.kib_connd_waitq); + wake_up(&kiblnd_data.kib_failover_waitq); - case IBLND_INIT_NOTHING: - break; - } + wait_var_event_warning(&kiblnd_data.kib_nthreads, + !atomic_read(&kiblnd_data.kib_nthreads), + "Waiting for %d threads to terminate\n", + atomic_read(&kiblnd_data.kib_nthreads)); + /* fall through */ - if (kiblnd_data.kib_peers != NULL) { - LIBCFS_FREE(kiblnd_data.kib_peers, - sizeof(struct list_head) * - kiblnd_data.kib_peer_hash_size); + case IBLND_INIT_NOTHING: + break; } if (kiblnd_data.kib_scheds != NULL) cfs_percpt_free(kiblnd_data.kib_scheds); - CDEBUG(D_MALLOC, "after LND base cleanup: kmem %d\n", - atomic_read(&libcfs_kmemory)); + CDEBUG(D_MALLOC, "after LND base cleanup: kmem %lld\n", + libcfs_kmem_read()); kiblnd_data.kib_init = IBLND_INIT_NOTHING; module_put(THIS_MODULE); @@ -2977,18 +2983,17 @@ kiblnd_base_shutdown(void) static void kiblnd_shutdown(struct lnet_ni *ni) { - kib_net_t *net = ni->ni_data; + struct kib_net *net = ni->ni_data; rwlock_t *g_lock = &kiblnd_data.kib_global_lock; - int i; - unsigned long flags; + unsigned long flags; LASSERT(kiblnd_data.kib_init == IBLND_INIT_ALL); if (net == NULL) goto out; - CDEBUG(D_MALLOC, "before LND net cleanup: kmem %d\n", - atomic_read(&libcfs_kmemory)); + CDEBUG(D_MALLOC, "before LND net cleanup: kmem %lld\n", + libcfs_kmem_read()); write_lock_irqsave(g_lock, flags); net->ibn_shutdown = 1; @@ -2998,22 +3003,16 @@ kiblnd_shutdown(struct lnet_ni *ni) default: LBUG(); - case IBLND_INIT_ALL: - /* nuke all existing peers within this net */ - kiblnd_del_peer(ni, LNET_NID_ANY); + case IBLND_INIT_ALL: + /* nuke all existing peers within this net */ + kiblnd_del_peer(ni, LNET_NID_ANY); /* Wait for all peer_ni state to clean up */ - i = 2; - while (atomic_read(&net->ibn_npeers) != 0) { - i++; - /* power of 2? */ - CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, - "%s: waiting for %d peers to disconnect\n", - libcfs_nid2str(ni->ni_nid), - atomic_read(&net->ibn_npeers)); - set_current_state(TASK_UNINTERRUPTIBLE); - schedule_timeout(cfs_time_seconds(1)); - } + wait_var_event_warning(&net->ibn_npeers, + atomic_read(&net->ibn_npeers) == 0, + "%s: waiting for %d peers to disconnect\n", + libcfs_nid2str(ni->ni_nid), + atomic_read(&net->ibn_npeers)); kiblnd_net_fini_pools(net); @@ -3023,7 +3022,7 @@ kiblnd_shutdown(struct lnet_ni *ni) list_del(&net->ibn_list); write_unlock_irqrestore(g_lock, flags); - /* fall through */ + /* fall through */ case IBLND_INIT_NOTHING: LASSERT (atomic_read(&net->ibn_nconns) == 0); @@ -3035,8 +3034,8 @@ kiblnd_shutdown(struct lnet_ni *ni) break; } - CDEBUG(D_MALLOC, "after LND net cleanup: kmem %d\n", - atomic_read(&libcfs_kmemory)); + CDEBUG(D_MALLOC, "after LND net cleanup: kmem %lld\n", + libcfs_kmem_read()); net->ibn_init = IBLND_INIT_NOTHING; ni->ni_data = NULL; @@ -3046,19 +3045,20 @@ kiblnd_shutdown(struct lnet_ni *ni) out: if (list_empty(&kiblnd_data.kib_devs)) kiblnd_base_shutdown(); - return; } static int -kiblnd_base_startup(void) +kiblnd_base_startup(struct net *ns) { - struct kib_sched_info *sched; - int rc; - int i; + struct kib_sched_info *sched; + int rc; + int i; LASSERT(kiblnd_data.kib_init == IBLND_INIT_NOTHING); - try_module_get(THIS_MODULE); + if (!try_module_get(THIS_MODULE)) + goto failed; + memset(&kiblnd_data, 0, sizeof(kiblnd_data)); /* zero pointers, flags etc */ rwlock_init(&kiblnd_data.kib_global_lock); @@ -3066,18 +3066,11 @@ kiblnd_base_startup(void) INIT_LIST_HEAD(&kiblnd_data.kib_devs); INIT_LIST_HEAD(&kiblnd_data.kib_failed_devs); - kiblnd_data.kib_peer_hash_size = IBLND_PEER_HASH_SIZE; - LIBCFS_ALLOC(kiblnd_data.kib_peers, - sizeof(struct list_head) * - kiblnd_data.kib_peer_hash_size); - if (kiblnd_data.kib_peers == NULL) - goto failed; - - for (i = 0; i < kiblnd_data.kib_peer_hash_size; i++) - INIT_LIST_HEAD(&kiblnd_data.kib_peers[i]); + hash_init(kiblnd_data.kib_peers); spin_lock_init(&kiblnd_data.kib_connd_lock); INIT_LIST_HEAD(&kiblnd_data.kib_connd_conns); + INIT_LIST_HEAD(&kiblnd_data.kib_connd_waits); INIT_LIST_HEAD(&kiblnd_data.kib_connd_zombies); INIT_LIST_HEAD(&kiblnd_data.kib_reconn_list); INIT_LIST_HEAD(&kiblnd_data.kib_reconn_wait); @@ -3110,36 +3103,36 @@ kiblnd_base_startup(void) sched->ibs_cpt = i; } - kiblnd_data.kib_error_qpa.qp_state = IB_QPS_ERR; + kiblnd_data.kib_error_qpa.qp_state = IB_QPS_ERR; - /* lists/ptrs/locks initialised */ - kiblnd_data.kib_init = IBLND_INIT_DATA; - /*****************************************************/ + /* lists/ptrs/locks initialised */ + kiblnd_data.kib_init = IBLND_INIT_DATA; + /*****************************************************/ rc = kiblnd_thread_start(kiblnd_connd, NULL, "kiblnd_connd"); - if (rc != 0) { - CERROR("Can't spawn o2iblnd connd: %d\n", rc); - goto failed; - } + if (rc != 0) { + CERROR("Can't spawn o2iblnd connd: %d\n", rc); + goto failed; + } if (*kiblnd_tunables.kib_dev_failover != 0) - rc = kiblnd_thread_start(kiblnd_failover_thread, NULL, + rc = kiblnd_thread_start(kiblnd_failover_thread, ns, "kiblnd_failover"); - if (rc != 0) { - CERROR("Can't spawn o2iblnd failover thread: %d\n", rc); - goto failed; - } + if (rc != 0) { + CERROR("Can't spawn o2iblnd failover thread: %d\n", rc); + goto failed; + } - /* flag everything initialised */ - kiblnd_data.kib_init = IBLND_INIT_ALL; - /*****************************************************/ + /* flag everything initialised */ + kiblnd_data.kib_init = IBLND_INIT_ALL; + /*****************************************************/ - return 0; + return 0; failed: - kiblnd_base_shutdown(); - return -ENETDOWN; + kiblnd_base_shutdown(); + return -ENETDOWN; } static int @@ -3183,8 +3176,8 @@ kiblnd_start_schedulers(struct kib_sched_info *sched) return rc; } -static int -kiblnd_dev_start_threads(kib_dev_t *dev, int newdev, __u32 *cpts, int ncpts) +static int kiblnd_dev_start_threads(struct kib_dev *dev, bool newdev, u32 *cpts, + int ncpts) { int cpt; int rc; @@ -3209,13 +3202,13 @@ kiblnd_dev_start_threads(kib_dev_t *dev, int newdev, __u32 *cpts, int ncpts) return 0; } -static kib_dev_t * +static struct kib_dev * kiblnd_dev_search(char *ifname) { - kib_dev_t *alias = NULL; - kib_dev_t *dev; - char *colon; - char *colon2; + struct kib_dev *alias = NULL; + struct kib_dev *dev; + char *colon; + char *colon2; colon = strchr(ifname, ':'); list_for_each_entry(dev, &kiblnd_data.kib_devs, ibd_list) { @@ -3245,106 +3238,304 @@ kiblnd_dev_search(char *ifname) static int kiblnd_startup(struct lnet_ni *ni) { - char *ifname; - kib_dev_t *ibdev = NULL; - kib_net_t *net; - unsigned long flags; - int rc; - int newdev; - int node_id; + char *ifname = NULL; + struct lnet_inetdev *ifaces = NULL; + struct kib_dev *ibdev = NULL; + struct kib_net *net = NULL; + unsigned long flags; + int rc; + int i; + bool newdev; - LASSERT (ni->ni_net->net_lnd == &the_o2iblnd); + LASSERT(ni->ni_net->net_lnd == &the_o2iblnd); - if (kiblnd_data.kib_init == IBLND_INIT_NOTHING) { - rc = kiblnd_base_startup(); - if (rc != 0) - return rc; - } + if (kiblnd_data.kib_init == IBLND_INIT_NOTHING) { + rc = kiblnd_base_startup(ni->ni_net_ns); + if (rc != 0) + return rc; + } - LIBCFS_ALLOC(net, sizeof(*net)); - ni->ni_data = net; - if (net == NULL) - goto failed; + LIBCFS_ALLOC(net, sizeof(*net)); + ni->ni_data = net; + if (net == NULL) { + rc = -ENOMEM; + goto failed; + } + net->ibn_ni = ni; net->ibn_incarnation = ktime_get_real_ns() / NSEC_PER_USEC; kiblnd_tunables_setup(ni); - if (ni->ni_interfaces[0] != NULL) { + /* + * Multi-Rail wants each secondary + * IP to be treated as an unique 'struct ni' interface. + */ + if (ni->ni_interface != NULL) { /* Use the IPoIB interface specified in 'networks=' */ - - CLASSERT(LNET_INTERFACES_NUM > 1); - if (ni->ni_interfaces[1] != NULL) { - CERROR("Multiple interfaces not supported\n"); - goto failed; - } - - ifname = ni->ni_interfaces[0]; + ifname = ni->ni_interface; } else { ifname = *kiblnd_tunables.kib_default_ipif; } - if (strlen(ifname) >= sizeof(ibdev->ibd_ifname)) { - CERROR("IPoIB interface name too long: %s\n", ifname); - goto failed; - } + if (strlen(ifname) >= sizeof(ibdev->ibd_ifname)) { + CERROR("IPoIB interface name too long: %s\n", ifname); + rc = -E2BIG; + goto failed; + } - ibdev = kiblnd_dev_search(ifname); + rc = lnet_inet_enumerate(&ifaces, ni->ni_net_ns); + if (rc < 0) + goto failed; + + for (i = 0; i < rc; i++) { + if (strcmp(ifname, ifaces[i].li_name) == 0) + break; + } + if (i == rc) { + CERROR("ko2iblnd: No matching interfaces\n"); + rc = -ENOENT; + goto failed; + } + + ibdev = kiblnd_dev_search(ifname); newdev = ibdev == NULL; /* hmm...create kib_dev even for alias */ - if (ibdev == NULL || strcmp(&ibdev->ibd_ifname[0], ifname) != 0) - ibdev = kiblnd_create_dev(ifname); + if (ibdev == NULL || strcmp(&ibdev->ibd_ifname[0], ifname) != 0) { + LIBCFS_ALLOC(ibdev, sizeof(*ibdev)); + if (!ibdev) { + rc = -ENOMEM; + goto failed; + } - if (ibdev == NULL) - goto failed; + ibdev->ibd_ifip = ifaces[i].li_ipaddr; + strlcpy(ibdev->ibd_ifname, ifaces[i].li_name, + sizeof(ibdev->ibd_ifname)); + ibdev->ibd_can_failover = !!(ifaces[i].li_flags & IFF_MASTER); + + INIT_LIST_HEAD(&ibdev->ibd_nets); + INIT_LIST_HEAD(&ibdev->ibd_list); /* not yet in kib_devs */ + INIT_LIST_HEAD(&ibdev->ibd_fail_list); + + /* initialize the device */ + rc = kiblnd_dev_failover(ibdev, ni->ni_net_ns); + if (rc) { + CERROR("ko2iblnd: Can't initialize device: rc = %d\n", + rc); + goto failed; + } - node_id = dev_to_node(ibdev->ibd_hdev->ibh_ibdev->dma_device); - ni->ni_dev_cpt = cfs_cpt_of_node(lnet_cpt_table(), node_id); + list_add_tail(&ibdev->ibd_list, &kiblnd_data.kib_devs); + } net->ibn_dev = ibdev; ni->ni_nid = LNET_MKNID(LNET_NIDNET(ni->ni_nid), ibdev->ibd_ifip); - rc = kiblnd_dev_start_threads(ibdev, newdev, - ni->ni_cpts, ni->ni_ncpts); + ni->ni_dev_cpt = ifaces[i].li_cpt; + + rc = kiblnd_dev_start_threads(ibdev, newdev, ni->ni_cpts, ni->ni_ncpts); if (rc != 0) goto failed; rc = kiblnd_net_init_pools(net, ni, ni->ni_cpts, ni->ni_ncpts); - if (rc != 0) { - CERROR("Failed to initialize NI pools: %d\n", rc); - goto failed; - } + if (rc != 0) { + CERROR("Failed to initialize NI pools: %d\n", rc); + goto failed; + } write_lock_irqsave(&kiblnd_data.kib_global_lock, flags); ibdev->ibd_nnets++; list_add_tail(&net->ibn_list, &ibdev->ibd_nets); + /* for health check */ + if (ibdev->ibd_hdev->ibh_state == IBLND_DEV_PORT_DOWN) + kiblnd_set_ni_fatal_on(ibdev->ibd_hdev, 1); write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags); - net->ibn_init = IBLND_INIT_ALL; + net->ibn_init = IBLND_INIT_ALL; - return 0; + return 0; failed: if (net != NULL && net->ibn_dev == NULL && ibdev != NULL) - kiblnd_destroy_dev(ibdev); + kiblnd_destroy_dev(ibdev); + + kfree(ifaces); + kiblnd_shutdown(ni); - kiblnd_shutdown(ni); + CDEBUG(D_NET, "Configuration of device %s failed: rc = %d\n", + ifname ? ifname : "", rc); - CDEBUG(D_NET, "kiblnd_startup failed\n"); - return -ENETDOWN; + return -ENETDOWN; } -static struct lnet_lnd the_o2iblnd = { +static const struct lnet_lnd the_o2iblnd = { .lnd_type = O2IBLND, .lnd_startup = kiblnd_startup, .lnd_shutdown = kiblnd_shutdown, .lnd_ctl = kiblnd_ctl, - .lnd_query = kiblnd_query, .lnd_send = kiblnd_send, .lnd_recv = kiblnd_recv, }; +static void ko2inlnd_assert_wire_constants(void) +{ + BUILD_BUG_ON(IBLND_MSG_MAGIC != 0x0be91b91); + BUILD_BUG_ON(IBLND_MSG_VERSION_1 != 0x11); + BUILD_BUG_ON(IBLND_MSG_VERSION_2 != 0x12); + BUILD_BUG_ON(IBLND_MSG_VERSION != IBLND_MSG_VERSION_2); + + BUILD_BUG_ON(IBLND_MSG_CONNREQ != 0xc0); + BUILD_BUG_ON(IBLND_MSG_CONNACK != 0xc1); + BUILD_BUG_ON(IBLND_MSG_NOOP != 0xd0); + BUILD_BUG_ON(IBLND_MSG_IMMEDIATE != 0xd1); + BUILD_BUG_ON(IBLND_MSG_PUT_REQ != 0xd2); + BUILD_BUG_ON(IBLND_MSG_PUT_NAK != 0xd3); + BUILD_BUG_ON(IBLND_MSG_PUT_ACK != 0xd4); + BUILD_BUG_ON(IBLND_MSG_PUT_DONE != 0xd5); + BUILD_BUG_ON(IBLND_MSG_GET_REQ != 0xd6); + BUILD_BUG_ON(IBLND_MSG_GET_DONE != 0xd7); + + BUILD_BUG_ON(IBLND_REJECT_CONN_RACE != 1); + BUILD_BUG_ON(IBLND_REJECT_NO_RESOURCES != 2); + BUILD_BUG_ON(IBLND_REJECT_FATAL != 3); + BUILD_BUG_ON(IBLND_REJECT_CONN_UNCOMPAT != 4); + BUILD_BUG_ON(IBLND_REJECT_CONN_STALE != 5); + BUILD_BUG_ON(IBLND_REJECT_RDMA_FRAGS != 6); + BUILD_BUG_ON(IBLND_REJECT_MSG_QUEUE_SIZE != 7); + BUILD_BUG_ON(IBLND_REJECT_INVALID_SRV_ID != 8); + + BUILD_BUG_ON((int)sizeof(struct kib_connparams) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_connparams, ibcp_queue_depth) != 0); + BUILD_BUG_ON((int)sizeof(((struct kib_connparams *)0)->ibcp_queue_depth) != 2); + BUILD_BUG_ON((int)offsetof(struct kib_connparams, ibcp_max_frags) != 2); + BUILD_BUG_ON((int)sizeof(((struct kib_connparams *)0)->ibcp_max_frags) != 2); + BUILD_BUG_ON((int)offsetof(struct kib_connparams, ibcp_max_msg_size) != 4); + BUILD_BUG_ON((int)sizeof(((struct kib_connparams *)0)->ibcp_max_msg_size) != 4); + + BUILD_BUG_ON((int)sizeof(struct kib_immediate_msg) != 72); + BUILD_BUG_ON((int)offsetof(struct kib_immediate_msg, ibim_hdr) != 0); + BUILD_BUG_ON((int)sizeof(((struct kib_immediate_msg *)0)->ibim_hdr) != 72); + BUILD_BUG_ON((int)offsetof(struct kib_immediate_msg, ibim_payload) != 72); + BUILD_BUG_ON((int)sizeof(((struct kib_immediate_msg *)0)->ibim_payload) != 0); + + BUILD_BUG_ON((int)sizeof(struct kib_rdma_frag) != 12); + BUILD_BUG_ON((int)offsetof(struct kib_rdma_frag, rf_nob) != 0); + BUILD_BUG_ON((int)sizeof(((struct kib_rdma_frag *)0)->rf_nob) != 4); + BUILD_BUG_ON((int)offsetof(struct kib_rdma_frag, rf_addr) != 4); + BUILD_BUG_ON((int)sizeof(((struct kib_rdma_frag *)0)->rf_addr) != 8); + + BUILD_BUG_ON((int)sizeof(struct kib_rdma_desc) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_rdma_desc, rd_key) != 0); + BUILD_BUG_ON((int)sizeof(((struct kib_rdma_desc *)0)->rd_key) != 4); + BUILD_BUG_ON((int)offsetof(struct kib_rdma_desc, rd_nfrags) != 4); + BUILD_BUG_ON((int)sizeof(((struct kib_rdma_desc *)0)->rd_nfrags) != 4); + BUILD_BUG_ON((int)offsetof(struct kib_rdma_desc, rd_frags) != 8); + BUILD_BUG_ON((int)sizeof(((struct kib_rdma_desc *)0)->rd_frags) != 0); + + BUILD_BUG_ON((int)sizeof(struct kib_putreq_msg) != 80); + BUILD_BUG_ON((int)offsetof(struct kib_putreq_msg, ibprm_hdr) != 0); + BUILD_BUG_ON((int)sizeof(((struct kib_putreq_msg *)0)->ibprm_hdr) != 72); + BUILD_BUG_ON((int)offsetof(struct kib_putreq_msg, ibprm_cookie) != 72); + BUILD_BUG_ON((int)sizeof(((struct kib_putreq_msg *)0)->ibprm_cookie) != 8); + + BUILD_BUG_ON((int)sizeof(struct kib_putack_msg) != 24); + BUILD_BUG_ON((int)offsetof(struct kib_putack_msg, ibpam_src_cookie) != 0); + BUILD_BUG_ON((int)sizeof(((struct kib_putack_msg *)0)->ibpam_src_cookie) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_putack_msg, ibpam_dst_cookie) != 8); + BUILD_BUG_ON((int)sizeof(((struct kib_putack_msg *)0)->ibpam_dst_cookie) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_putack_msg, ibpam_rd) != 16); + BUILD_BUG_ON((int)sizeof(((struct kib_putack_msg *)0)->ibpam_rd) != 8); + + BUILD_BUG_ON((int)sizeof(struct kib_get_msg) != 88); + BUILD_BUG_ON((int)offsetof(struct kib_get_msg, ibgm_hdr) != 0); + BUILD_BUG_ON((int)sizeof(((struct kib_get_msg *)0)->ibgm_hdr) != 72); + BUILD_BUG_ON((int)offsetof(struct kib_get_msg, ibgm_cookie) != 72); + BUILD_BUG_ON((int)sizeof(((struct kib_get_msg *)0)->ibgm_cookie) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_get_msg, ibgm_rd) != 80); + BUILD_BUG_ON((int)sizeof(((struct kib_get_msg *)0)->ibgm_rd) != 8); + + BUILD_BUG_ON((int)sizeof(struct kib_completion_msg) != 12); + BUILD_BUG_ON((int)offsetof(struct kib_completion_msg, ibcm_cookie) != 0); + BUILD_BUG_ON((int)sizeof(((struct kib_completion_msg *)0)->ibcm_cookie) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_completion_msg, ibcm_status) != 8); + BUILD_BUG_ON((int)sizeof(((struct kib_completion_msg *)0)->ibcm_status) != 4); + + /* Checks for struct kib_msg */ + //BUILD_BUG_ON((int)sizeof(struct kib_msg) != 12); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_magic) != 0); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_magic) != 4); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_version) != 4); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_version) != 2); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_type) != 6); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_type) != 1); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_credits) != 7); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_credits) != 1); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_nob) != 8); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_nob) != 4); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_cksum) != 12); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_cksum) != 4); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_srcnid) != 16); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_srcnid) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_srcstamp) != 24); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_srcstamp) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_dstnid) != 32); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_dstnid) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_dststamp) != 40); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_dststamp) != 8); + + /* Connparams */ + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.connparams.ibcp_queue_depth) != 48); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.connparams.ibcp_queue_depth) != 2); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.connparams.ibcp_max_frags) != 50); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.connparams.ibcp_max_frags) != 2); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.connparams.ibcp_max_msg_size) != 52); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.connparams.ibcp_max_msg_size) != 4); + + /* Immediate message */ + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.immediate.ibim_hdr) != 48); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.immediate.ibim_hdr) != 72); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.immediate.ibim_payload) != 120); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.immediate.ibim_payload) != 0); + + /* PUT req message */ + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.putreq.ibprm_hdr) != 48); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.putreq.ibprm_hdr) != 72); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.putreq.ibprm_cookie) != 120); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.putreq.ibprm_cookie) != 8); + + /* Put ACK */ + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.putack.ibpam_src_cookie) != 48); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.putack.ibpam_src_cookie) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.putack.ibpam_dst_cookie) != 56); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.putack.ibpam_dst_cookie) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.putack.ibpam_rd) != 64); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.putack.ibpam_rd) != 8); + + /* GET message */ + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.get.ibgm_hdr) != 48); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.get.ibgm_hdr) != 72); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.get.ibgm_cookie) != 120); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.get.ibgm_cookie) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.get.ibgm_rd) != 128); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.get.ibgm_rd) != 8); + + /* Completion message */ + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.completion.ibcm_cookie) != 48); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.completion.ibcm_cookie) != 8); + BUILD_BUG_ON((int)offsetof(struct kib_msg, ibm_u.completion.ibcm_status) != 56); + BUILD_BUG_ON((int)sizeof(((struct kib_msg *)0)->ibm_u.completion.ibcm_status) != 4); + + /* Sanity checks */ + BUILD_BUG_ON(sizeof(struct kib_msg) > IBLND_MSG_SIZE); + BUILD_BUG_ON(offsetof(struct kib_msg, + ibm_u.get.ibgm_rd.rd_frags[IBLND_MAX_RDMA_FRAGS]) > + IBLND_MSG_SIZE); + BUILD_BUG_ON(offsetof(struct kib_msg, + ibm_u.putack.ibpam_rd.rd_frags[IBLND_MAX_RDMA_FRAGS]) > + IBLND_MSG_SIZE); +} + static void __exit ko2iblnd_exit(void) { lnet_unregister_lnd(&the_o2iblnd); @@ -3354,13 +3545,7 @@ static int __init ko2iblnd_init(void) { int rc; - CLASSERT(sizeof(kib_msg_t) <= IBLND_MSG_SIZE); - CLASSERT(offsetof(kib_msg_t, - ibm_u.get.ibgm_rd.rd_frags[IBLND_MAX_RDMA_FRAGS]) <= - IBLND_MSG_SIZE); - CLASSERT(offsetof(kib_msg_t, - ibm_u.putack.ibpam_rd.rd_frags[IBLND_MAX_RDMA_FRAGS]) - <= IBLND_MSG_SIZE); + ko2inlnd_assert_wire_constants(); rc = kiblnd_tunables_init(); if (rc != 0)