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 */
* 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);
+ if (atomic_dec_and_test(&net->ibn_npeers))
+ wake_up_var(&net->ibn_npeers);
}
struct kib_peer_ni *
static void
kiblnd_del_peer_locked(struct kib_peer_ni *peer_ni)
{
- struct list_head *ctmp;
- struct list_head *cnxt;
+ 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, struct kib_conn, 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
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
rwlock_t *glock = &kiblnd_data.kib_global_lock;
struct kib_net *net = peer_ni->ibp_ni->ni_data;
struct kib_dev *dev;
- struct ib_qp_init_attr *init_qp_attr;
+ 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 = {};
*/
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;
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;
- /*
- * 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);
+ 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;
+
+ 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 {
+ /* 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);
- rc = rdma_create_qp(cmid, conn->ibc_hdev->ibh_pd, init_qp_attr);
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;
}
conn->ibc_sched = sched;
- if (conn->ibc_queue_depth != peer_ni->ibp_queue_depth)
+ 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));
kiblnd_map_rx_descs(conn);
- LIBCFS_FREE(init_qp_attr, sizeof(*init_qp_attr));
-
/* 1 ref for caller and each rxmsg */
atomic_set(&conn->ibc_refcount, 1 + IBLND_RX_MSGS(conn));
conn->ibc_nrx = IBLND_RX_MSGS(conn);
failed_2:
kiblnd_destroy_conn(conn);
LIBCFS_FREE(conn, sizeof(*conn));
- failed_1:
- LIBCFS_FREE(init_qp_attr, sizeof(*init_qp_attr));
failed_0:
return NULL;
}
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(struct kib_rx));
- }
+ 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));
kiblnd_close_peer_conns_locked(struct kib_peer_ni *peer_ni, int why)
{
struct kib_conn *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, struct kib_conn, ibc_list);
+ 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),
int version, __u64 incarnation)
{
struct kib_conn *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, struct kib_conn, ibc_list);
+ 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;
#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);
+ rd->rd_nfrags, PAGE_SIZE);
#endif
- if (unlikely(n != tx->tx_nfrags)) {
+ if (unlikely(n != rd->rd_nfrags)) {
CERROR("Failed to map mr %d/%d "
- "elements\n", n, tx->tx_nfrags);
+ "elements\n", n, rd->rd_nfrags);
return n < 0 ? n : -EINVAL;
}
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;
}
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);
/* 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);
schedule_timeout_interruptible(interval);
if (tpo->tpo_tx_descs == NULL)
goto out;
- for (i = 0; i < pool->po_size; 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,
+ 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]));
- }
+ rd_frags[IBLND_MAX_RDMA_FRAGS]));
+ }
- LIBCFS_FREE(tpo->tpo_tx_descs,
- pool->po_size * sizeof(struct kib_tx));
+ CFS_FREE_PTR_ARRAY(tpo->tpo_tx_descs, pool->po_size);
out:
- kiblnd_fini_pool(pool);
+ kiblnd_fini_pool(pool);
CFS_FREE_PTR(tpo);
}
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:
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));
- schedule_timeout_uninterruptible(cfs_time_seconds(1));
- }
-
- /* fall through */
+ 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 */
case IBLND_INIT_NOTHING:
break;
}
- if (kiblnd_data.kib_peers != NULL) {
- LIBCFS_FREE(kiblnd_data.kib_peers,
- sizeof(struct list_head) *
- kiblnd_data.kib_peer_hash_size);
- }
+ if (kiblnd_data.kib_peers)
+ CFS_FREE_PTR_ARRAY(kiblnd_data.kib_peers,
+ kiblnd_data.kib_peer_hash_size);
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);
{
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;
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));
- schedule_timeout_uninterruptible(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);
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);
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;
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);
+ CFS_ALLOC_PTR_ARRAY(kiblnd_data.kib_peers,
+ kiblnd_data.kib_peer_hash_size);
if (kiblnd_data.kib_peers == NULL)
goto failed;
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);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff