[PATCH 1/7] staging: lustre: Support different ko2iblnd configs between systems

From: James Simmons
Date: Wed Mar 02 2016 - 18:53:45 EST

From: Jeremy Filizetti <jeremy.filizetti@xxxxxxxxx>

This patch adds suppoort for ko2iblnd to have different values for
peer_credits and map_on_demand between systems.

Signed-off-by: Jeremy Filizetti <jeremy.filizetti@xxxxxxxxx>
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-3322
Reviewed-on: http://review.whamcloud.com/11794
Reviewed-by: Amir Shehata <amir.shehata@xxxxxxxxx>
Reviewed-by: James Simmons <uja.ornl@xxxxxxxxx>
Reviewed-by: Oleg Drokin <oleg.drokin@xxxxxxxxx>
---
.../staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c | 51 ++++---
.../staging/lustre/lnet/klnds/o2iblnd/o2iblnd.h | 36 +++--
.../staging/lustre/lnet/klnds/o2iblnd/o2iblnd_cb.c | 156 ++++++++++++--------
3 files changed, 146 insertions(+), 97 deletions(-)

diff --git a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c
index 1dc18d7..0b1ffbe 100644
--- a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c
+++ b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.c
@@ -631,7 +631,7 @@ static int kiblnd_get_completion_vector(kib_conn_t *conn, int cpt)
}

kib_conn_t *kiblnd_create_conn(kib_peer_t *peer, struct rdma_cm_id *cmid,
- int state, int version)
+ int state, int version, kib_connparams_t *cp)
{
/*
* CAVEAT EMPTOR:
@@ -686,6 +686,14 @@ kib_conn_t *kiblnd_create_conn(kib_peer_t *peer, struct rdma_cm_id *cmid,
cmid->context = conn; /* for future CM callbacks */
conn->ibc_cmid = cmid;

+ if (!cp) {
+ conn->ibc_max_frags = IBLND_CFG_RDMA_FRAGS;
+ conn->ibc_queue_depth = *kiblnd_tunables.kib_peertxcredits;
+ } else {
+ conn->ibc_max_frags = cp->ibcp_max_frags;
+ conn->ibc_queue_depth = cp->ibcp_queue_depth;
+ }
+
INIT_LIST_HEAD(&conn->ibc_early_rxs);
INIT_LIST_HEAD(&conn->ibc_tx_noops);
INIT_LIST_HEAD(&conn->ibc_tx_queue);
@@ -730,27 +738,27 @@ kib_conn_t *kiblnd_create_conn(kib_peer_t *peer, struct rdma_cm_id *cmid,
write_unlock_irqrestore(glock, flags);

LIBCFS_CPT_ALLOC(conn->ibc_rxs, lnet_cpt_table(), cpt,
- IBLND_RX_MSGS(version) * sizeof(kib_rx_t));
+ IBLND_RX_MSGS(conn) * sizeof(kib_rx_t));
if (!conn->ibc_rxs) {
CERROR("Cannot allocate RX buffers\n");
goto failed_2;
}

rc = kiblnd_alloc_pages(&conn->ibc_rx_pages, cpt,
- IBLND_RX_MSG_PAGES(version));
+ IBLND_RX_MSG_PAGES(conn));
if (rc)
goto failed_2;

kiblnd_map_rx_descs(conn);

- cq_attr.cqe = IBLND_CQ_ENTRIES(version);
+ cq_attr.cqe = IBLND_CQ_ENTRIES(conn);
cq_attr.comp_vector = kiblnd_get_completion_vector(conn, cpt);
cq = ib_create_cq(cmid->device,
kiblnd_cq_completion, kiblnd_cq_event, conn,
&cq_attr);
if (IS_ERR(cq)) {
- CERROR("Can't create CQ: %ld, cqe: %d\n",
- PTR_ERR(cq), IBLND_CQ_ENTRIES(version));
+ CERROR("Failed to create CQ with %d CQEs: %ld\n",
+ IBLND_CQ_ENTRIES(conn), PTR_ERR(cq));
goto failed_2;
}

@@ -764,8 +772,8 @@ kib_conn_t *kiblnd_create_conn(kib_peer_t *peer, struct rdma_cm_id *cmid,

init_qp_attr->event_handler = kiblnd_qp_event;
init_qp_attr->qp_context = conn;
- init_qp_attr->cap.max_send_wr = IBLND_SEND_WRS(version);
- init_qp_attr->cap.max_recv_wr = IBLND_RECV_WRS(version);
+ init_qp_attr->cap.max_send_wr = IBLND_SEND_WRS(conn);
+ init_qp_attr->cap.max_recv_wr = IBLND_RECV_WRS(conn);
init_qp_attr->cap.max_send_sge = 1;
init_qp_attr->cap.max_recv_sge = 1;
init_qp_attr->sq_sig_type = IB_SIGNAL_REQ_WR;
@@ -786,11 +794,11 @@ kib_conn_t *kiblnd_create_conn(kib_peer_t *peer, struct rdma_cm_id *cmid,
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(version));
- conn->ibc_nrx = IBLND_RX_MSGS(version);
+ atomic_set(&conn->ibc_refcount, 1 + IBLND_RX_MSGS(conn));
+ conn->ibc_nrx = IBLND_RX_MSGS(conn);

/* post receives */
- for (i = 0; i < IBLND_RX_MSGS(version); i++) {
+ for (i = 0; i < IBLND_RX_MSGS(conn); i++) {
rc = kiblnd_post_rx(&conn->ibc_rxs[i],
IBLND_POSTRX_NO_CREDIT);
if (rc) {
@@ -804,7 +812,7 @@ kib_conn_t *kiblnd_create_conn(kib_peer_t *peer, struct rdma_cm_id *cmid,
* NB locking needed now I'm racing with completion
*/
spin_lock_irqsave(&sched->ibs_lock, flags);
- conn->ibc_nrx -= IBLND_RX_MSGS(version) - i;
+ conn->ibc_nrx -= IBLND_RX_MSGS(conn) - i;
spin_unlock_irqrestore(&sched->ibs_lock, flags);

/*
@@ -816,7 +824,7 @@ kib_conn_t *kiblnd_create_conn(kib_peer_t *peer, struct rdma_cm_id *cmid,
conn->ibc_cmid = NULL;

/* Drop my own and unused rxbuffer refcounts */
- while (i++ <= IBLND_RX_MSGS(version))
+ while (i++ <= IBLND_RX_MSGS(conn))
kiblnd_conn_decref(conn);

return NULL;
@@ -886,8 +894,7 @@ void kiblnd_destroy_conn(kib_conn_t *conn)

if (conn->ibc_rxs) {
LIBCFS_FREE(conn->ibc_rxs,
- IBLND_RX_MSGS(conn->ibc_version)
- * sizeof(kib_rx_t));
+ IBLND_RX_MSGS(conn) * sizeof(kib_rx_t));
}

if (conn->ibc_connvars)
@@ -1143,7 +1150,7 @@ void kiblnd_unmap_rx_descs(kib_conn_t *conn)
LASSERT(conn->ibc_rxs);
LASSERT(conn->ibc_hdev);

- for (i = 0; i < IBLND_RX_MSGS(conn->ibc_version); i++) {
+ for (i = 0; i < IBLND_RX_MSGS(conn); i++) {
rx = &conn->ibc_rxs[i];

LASSERT(rx->rx_nob >= 0); /* not posted */
@@ -1167,7 +1174,7 @@ void kiblnd_map_rx_descs(kib_conn_t *conn)
int ipg;
int i;

- for (pg_off = ipg = i = 0; i < IBLND_RX_MSGS(conn->ibc_version); i++) {
+ for (pg_off = ipg = i = 0; i < IBLND_RX_MSGS(conn); i++) {
pg = conn->ibc_rx_pages->ibp_pages[ipg];
rx = &conn->ibc_rxs[i];

@@ -1192,7 +1199,7 @@ void kiblnd_map_rx_descs(kib_conn_t *conn)
if (pg_off == PAGE_SIZE) {
pg_off = 0;
ipg++;
- LASSERT(ipg <= IBLND_RX_MSG_PAGES(conn->ibc_version));
+ LASSERT(ipg <= IBLND_RX_MSG_PAGES(conn));
}
}
}
@@ -1296,12 +1303,16 @@ static void kiblnd_map_tx_pool(kib_tx_pool_t *tpo)
}
}

-struct ib_mr *kiblnd_find_rd_dma_mr(kib_hca_dev_t *hdev, kib_rdma_desc_t *rd)
+struct ib_mr *kiblnd_find_rd_dma_mr(kib_hca_dev_t *hdev, kib_rdma_desc_t *rd,
+ int negotiated_nfrags)
{
+ __u16 nfrags = (negotiated_nfrags != -1) ?
+ negotiated_nfrags : *kiblnd_tunables.kib_map_on_demand;
+
LASSERT(hdev->ibh_mrs);

if (*kiblnd_tunables.kib_map_on_demand > 0 &&
- *kiblnd_tunables.kib_map_on_demand <= rd->rd_nfrags)
+ nfrags <= rd->rd_nfrags)
return NULL;

return hdev->ibh_mrs;
diff --git a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.h b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.h
index 0c88e8b..59a26c4 100644
--- a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.h
+++ b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.h
@@ -162,18 +162,17 @@ kiblnd_concurrent_sends_v1(void)
#define IBLND_FMR_POOL 256
#define IBLND_FMR_POOL_FLUSH 192

-/* TX messages (shared by all connections) */
-#define IBLND_TX_MSGS() (*kiblnd_tunables.kib_ntx)
-
-/* RX messages (per connection) */
-#define IBLND_RX_MSGS(v) (IBLND_MSG_QUEUE_SIZE(v) * 2 + IBLND_OOB_MSGS(v))
-#define IBLND_RX_MSG_BYTES(v) (IBLND_RX_MSGS(v) * IBLND_MSG_SIZE)
-#define IBLND_RX_MSG_PAGES(v) ((IBLND_RX_MSG_BYTES(v) + PAGE_SIZE - 1) / PAGE_SIZE)
+#define IBLND_RX_MSGS(c) \
+ ((c->ibc_queue_depth) * 2 + IBLND_OOB_MSGS(c->ibc_version))
+#define IBLND_RX_MSG_BYTES(c) (IBLND_RX_MSGS(c) * IBLND_MSG_SIZE)
+#define IBLND_RX_MSG_PAGES(c) \
+ ((IBLND_RX_MSG_BYTES(c) + PAGE_SIZE - 1) / PAGE_SIZE)

/* WRs and CQEs (per connection) */
-#define IBLND_RECV_WRS(v) IBLND_RX_MSGS(v)
-#define IBLND_SEND_WRS(v) ((IBLND_RDMA_FRAGS(v) + 1) * IBLND_CONCURRENT_SENDS(v))
-#define IBLND_CQ_ENTRIES(v) (IBLND_RECV_WRS(v) + IBLND_SEND_WRS(v))
+#define IBLND_RECV_WRS(c) IBLND_RX_MSGS(c)
+#define IBLND_SEND_WRS(c) \
+ ((c->ibc_max_frags + 1) * IBLND_CONCURRENT_SENDS(c->ibc_version))
+#define IBLND_CQ_ENTRIES(c) (IBLND_RECV_WRS(c) + IBLND_SEND_WRS(c))

struct kib_hca_dev;

@@ -464,10 +463,10 @@ typedef struct {
#define IBLND_REJECT_FATAL 3 /* Anything else */
#define IBLND_REJECT_CONN_UNCOMPAT 4 /* incompatible version peer */
#define IBLND_REJECT_CONN_STALE 5 /* stale peer */
-#define IBLND_REJECT_RDMA_FRAGS 6 /* Fatal: peer's rdma frags can't match */
- /* mine */
-#define IBLND_REJECT_MSG_QUEUE_SIZE 7 /* Fatal: peer's msg queue size can't */
- /* match mine */
+/* peer's rdma frags doesn't match mine */
+#define IBLND_REJECT_RDMA_FRAGS 6
+/* peer's msg queue size doesn't match mine */
+#define IBLND_REJECT_MSG_QUEUE_SIZE 7

/***********************************************************************/

@@ -535,6 +534,10 @@ typedef struct kib_conn {
int ibc_outstanding_credits; /* # credits to return */
int ibc_reserved_credits; /* # ACK/DONE msg credits */
int ibc_comms_error; /* set on comms error */
+ /* connections queue depth */
+ __u16 ibc_queue_depth;
+ /* connections max frags */
+ __u16 ibc_max_frags;
unsigned int ibc_nrx:16; /* receive buffers owned */
unsigned int ibc_scheduled:1; /* scheduled for attention */
unsigned int ibc_ready:1; /* CQ callback fired */
@@ -907,7 +910,8 @@ static inline unsigned int kiblnd_sg_dma_len(struct ib_device *dev,
#define KIBLND_CONN_PARAM_LEN(e) ((e)->param.conn.private_data_len)

struct ib_mr *kiblnd_find_rd_dma_mr(kib_hca_dev_t *hdev,
- kib_rdma_desc_t *rd);
+ kib_rdma_desc_t *rd,
+ int negotiated_nfrags);
void kiblnd_map_rx_descs(kib_conn_t *conn);
void kiblnd_unmap_rx_descs(kib_conn_t *conn);
void kiblnd_pool_free_node(kib_pool_t *pool, struct list_head *node);
@@ -942,7 +946,7 @@ int kiblnd_close_stale_conns_locked(kib_peer_t *peer,
int kiblnd_close_peer_conns_locked(kib_peer_t *peer, int why);

kib_conn_t *kiblnd_create_conn(kib_peer_t *peer, struct rdma_cm_id *cmid,
- int state, int version);
+ int state, int version, kib_connparams_t *cp);
void kiblnd_destroy_conn(kib_conn_t *conn);
void kiblnd_close_conn(kib_conn_t *conn, int error);
void kiblnd_close_conn_locked(kib_conn_t *conn, int error);
diff --git a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd_cb.c b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd_cb.c
index 0bd6120..3937735 100644
--- a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd_cb.c
+++ b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd_cb.c
@@ -328,14 +328,13 @@ kiblnd_handle_rx(kib_rx_t *rx)
spin_lock(&conn->ibc_lock);

if (conn->ibc_credits + credits >
- IBLND_MSG_QUEUE_SIZE(conn->ibc_version)) {
+ conn->ibc_queue_depth) {
rc2 = conn->ibc_credits;
spin_unlock(&conn->ibc_lock);

CERROR("Bad credits from %s: %d + %d > %d\n",
libcfs_nid2str(conn->ibc_peer->ibp_nid),
- rc2, credits,
- IBLND_MSG_QUEUE_SIZE(conn->ibc_version));
+ rc2, credits, conn->ibc_queue_depth);

kiblnd_close_conn(conn, -EPROTO);
kiblnd_post_rx(rx, IBLND_POSTRX_NO_CREDIT);
@@ -653,8 +652,8 @@ static int kiblnd_map_tx(lnet_ni_t *ni, kib_tx_t *tx, kib_rdma_desc_t *rd,
nob += rd->rd_frags[i].rf_nob;
}

- /* looking for pre-mapping MR */
- mr = kiblnd_find_rd_dma_mr(hdev, rd);
+ mr = kiblnd_find_rd_dma_mr(hdev, rd, tx->tx_conn ?
+ tx->tx_conn->ibc_max_frags : -1);
if (mr) {
/* found pre-mapping MR */
rd->rd_key = (rd != tx->tx_rd) ? mr->rkey : mr->lkey;
@@ -774,13 +773,13 @@ kiblnd_post_tx_locked(kib_conn_t *conn, kib_tx_t *tx, int credit)
LASSERT(tx->tx_queued);
/* We rely on this for QP sizing */
LASSERT(tx->tx_nwrq > 0);
- LASSERT(tx->tx_nwrq <= 1 + IBLND_RDMA_FRAGS(ver));
+ LASSERT(tx->tx_nwrq <= 1 + conn->ibc_max_frags);

LASSERT(!credit || credit == 1);
LASSERT(conn->ibc_outstanding_credits >= 0);
- LASSERT(conn->ibc_outstanding_credits <= IBLND_MSG_QUEUE_SIZE(ver));
+ LASSERT(conn->ibc_outstanding_credits <= conn->ibc_queue_depth);
LASSERT(conn->ibc_credits >= 0);
- LASSERT(conn->ibc_credits <= IBLND_MSG_QUEUE_SIZE(ver));
+ LASSERT(conn->ibc_credits <= conn->ibc_queue_depth);

if (conn->ibc_nsends_posted == IBLND_CONCURRENT_SENDS(ver)) {
/* tx completions outstanding... */
@@ -1089,10 +1088,10 @@ kiblnd_init_rdma(kib_conn_t *conn, kib_tx_t *tx, int type,
break;
}

- if (tx->tx_nwrq == IBLND_RDMA_FRAGS(conn->ibc_version)) {
- CERROR("RDMA too fragmented for %s (%d): %d/%d src %d/%d dst frags\n",
+ if (tx->tx_nwrq >= conn->ibc_max_frags) {
+ CERROR("RDMA has too many fragments for peer %s (%d), src idx/frags: %d/%d dst idx/frags: %d/%d\n",
libcfs_nid2str(conn->ibc_peer->ibp_nid),
- IBLND_RDMA_FRAGS(conn->ibc_version),
+ conn->ibc_max_frags,
srcidx, srcrd->rd_nfrags,
dstidx, dstrd->rd_nfrags);
rc = -EMSGSIZE;
@@ -2243,7 +2242,7 @@ kiblnd_passive_connect(struct rdma_cm_id *cmid, void *priv, int priv_nob)
if (!ni || /* no matching net */
ni->ni_nid != reqmsg->ibm_dstnid || /* right NET, wrong NID! */
net->ibn_dev != ibdev) { /* wrong device */
- CERROR("Can't accept %s on %s (%s:%d:%pI4h): bad dst nid %s\n",
+ CERROR("Can't accept conn from %s on %s (%s:%d:%pI4h): bad dst nid %s\n",
libcfs_nid2str(nid),
!ni ? "NA" : libcfs_nid2str(ni->ni_nid),
ibdev->ibd_ifname, ibdev->ibd_nnets,
@@ -2270,10 +2269,11 @@ kiblnd_passive_connect(struct rdma_cm_id *cmid, void *priv, int priv_nob)
goto failed;
}

- if (reqmsg->ibm_u.connparams.ibcp_queue_depth !=
+ if (reqmsg->ibm_u.connparams.ibcp_queue_depth >
IBLND_MSG_QUEUE_SIZE(version)) {
- CERROR("Can't accept %s: incompatible queue depth %d (%d wanted)\n",
- libcfs_nid2str(nid), reqmsg->ibm_u.connparams.ibcp_queue_depth,
+ CERROR("Can't accept conn from %s, queue depth too large: %d (<=%d wanted)\n",
+ libcfs_nid2str(nid),
+ reqmsg->ibm_u.connparams.ibcp_queue_depth,
IBLND_MSG_QUEUE_SIZE(version));

if (version == IBLND_MSG_VERSION)
@@ -2282,14 +2282,25 @@ kiblnd_passive_connect(struct rdma_cm_id *cmid, void *priv, int priv_nob)
goto failed;
}

- if (reqmsg->ibm_u.connparams.ibcp_max_frags !=
+ if (reqmsg->ibm_u.connparams.ibcp_max_frags >
IBLND_RDMA_FRAGS(version)) {
- CERROR("Can't accept %s(version %x): incompatible max_frags %d (%d wanted)\n",
- libcfs_nid2str(nid), version,
- reqmsg->ibm_u.connparams.ibcp_max_frags,
- IBLND_RDMA_FRAGS(version));
+ CWARN("Can't accept conn from %s (version %x): max_frags %d too large (%d wanted)\n",
+ libcfs_nid2str(nid), version,
+ reqmsg->ibm_u.connparams.ibcp_max_frags,
+ IBLND_RDMA_FRAGS(version));

- if (version == IBLND_MSG_VERSION)
+ if (version >= IBLND_MSG_VERSION)
+ rej.ibr_why = IBLND_REJECT_RDMA_FRAGS;
+
+ goto failed;
+ } else if (reqmsg->ibm_u.connparams.ibcp_max_frags <
+ IBLND_RDMA_FRAGS(version) && !net->ibn_fmr_ps) {
+ CWARN("Can't accept conn from %s (version %x): max_frags %d incompatible without FMR pool (%d wanted)\n",
+ libcfs_nid2str(nid), version,
+ reqmsg->ibm_u.connparams.ibcp_max_frags,
+ IBLND_RDMA_FRAGS(version));
+
+ if (version >= IBLND_MSG_VERSION)
rej.ibr_why = IBLND_REJECT_RDMA_FRAGS;

goto failed;
@@ -2371,7 +2382,8 @@ kiblnd_passive_connect(struct rdma_cm_id *cmid, void *priv, int priv_nob)
write_unlock_irqrestore(g_lock, flags);
}

- conn = kiblnd_create_conn(peer, cmid, IBLND_CONN_PASSIVE_WAIT, version);
+ conn = kiblnd_create_conn(peer, cmid, IBLND_CONN_PASSIVE_WAIT, version,
+ &reqmsg->ibm_u.connparams);
if (!conn) {
kiblnd_peer_connect_failed(peer, 0, -ENOMEM);
kiblnd_peer_decref(peer);
@@ -2384,19 +2396,21 @@ kiblnd_passive_connect(struct rdma_cm_id *cmid, void *priv, int priv_nob)
* CM callback doesn't destroy cmid.
*/
conn->ibc_incarnation = reqmsg->ibm_srcstamp;
- conn->ibc_credits = IBLND_MSG_QUEUE_SIZE(version);
- conn->ibc_reserved_credits = IBLND_MSG_QUEUE_SIZE(version);
- LASSERT(conn->ibc_credits + conn->ibc_reserved_credits + IBLND_OOB_MSGS(version)
- <= IBLND_RX_MSGS(version));
+ conn->ibc_credits = reqmsg->ibm_u.connparams.ibcp_queue_depth;
+ conn->ibc_reserved_credits = reqmsg->ibm_u.connparams.ibcp_queue_depth;
+ LASSERT(conn->ibc_credits + conn->ibc_reserved_credits +
+ IBLND_OOB_MSGS(version) <= IBLND_RX_MSGS(conn));

ackmsg = &conn->ibc_connvars->cv_msg;
memset(ackmsg, 0, sizeof(*ackmsg));

kiblnd_init_msg(ackmsg, IBLND_MSG_CONNACK,
sizeof(ackmsg->ibm_u.connparams));
- ackmsg->ibm_u.connparams.ibcp_queue_depth = IBLND_MSG_QUEUE_SIZE(version);
+ ackmsg->ibm_u.connparams.ibcp_queue_depth =
+ reqmsg->ibm_u.connparams.ibcp_queue_depth;
+ ackmsg->ibm_u.connparams.ibcp_max_frags =
+ reqmsg->ibm_u.connparams.ibcp_max_frags;
ackmsg->ibm_u.connparams.ibcp_max_msg_size = IBLND_MSG_SIZE;
- ackmsg->ibm_u.connparams.ibcp_max_frags = IBLND_RDMA_FRAGS(version);

kiblnd_pack_msg(ni, ackmsg, version, 0, nid, reqmsg->ibm_srcstamp);

@@ -2479,6 +2493,31 @@ kiblnd_reconnect(kib_conn_t *conn, int version,
reason = "Unknown";
break;

+ case IBLND_REJECT_RDMA_FRAGS:
+ if (conn->ibc_max_frags <= cp->ibcp_max_frags) {
+ CNETERR("Unsupported max frags, peer supports %d\n",
+ cp->ibcp_max_frags);
+ goto failed;
+ } else if (!*kiblnd_tunables.kib_map_on_demand) {
+ CNETERR("map_on_demand must be enabled to support map_on_demand peers\n");
+ goto failed;
+ }
+
+ conn->ibc_max_frags = cp->ibcp_max_frags;
+ reason = "rdma fragments";
+ break;
+
+ case IBLND_REJECT_MSG_QUEUE_SIZE:
+ if (conn->ibc_queue_depth <= cp->ibcp_queue_depth) {
+ CNETERR("Unsupported queue depth, peer supports %d\n",
+ cp->ibcp_queue_depth);
+ goto failed;
+ }
+
+ conn->ibc_queue_depth = cp->ibcp_queue_depth;
+ reason = "queue depth";
+ break;
+
case IBLND_REJECT_CONN_STALE:
reason = "stale";
break;
@@ -2495,11 +2534,17 @@ kiblnd_reconnect(kib_conn_t *conn, int version,
CNETERR("%s: retrying (%s), %x, %x, queue_dep: %d, max_frag: %d, msg_size: %d\n",
libcfs_nid2str(peer->ibp_nid),
reason, IBLND_MSG_VERSION, version,
- cp ? cp->ibcp_queue_depth : IBLND_MSG_QUEUE_SIZE(version),
- cp ? cp->ibcp_max_frags : IBLND_RDMA_FRAGS(version),
+ conn->ibc_queue_depth, conn->ibc_max_frags,
cp ? cp->ibcp_max_msg_size : IBLND_MSG_SIZE);

kiblnd_connect_peer(peer);
+ return;
+failed:
+ write_lock_irqsave(&kiblnd_data.kib_global_lock, flags);
+ peer->ibp_connecting--;
+ write_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);
+
+ return;
}

static void
@@ -2595,24 +2640,10 @@ kiblnd_rejected(kib_conn_t *conn, int reason, void *priv, int priv_nob)
case IBLND_REJECT_CONN_RACE:
case IBLND_REJECT_CONN_STALE:
case IBLND_REJECT_CONN_UNCOMPAT:
- kiblnd_reconnect(conn, rej->ibr_version,
- incarnation, rej->ibr_why, cp);
- break;
-
case IBLND_REJECT_MSG_QUEUE_SIZE:
- CERROR("%s rejected: incompatible message queue depth %d, %d\n",
- libcfs_nid2str(peer->ibp_nid),
- cp ? cp->ibcp_queue_depth :
- IBLND_MSG_QUEUE_SIZE(rej->ibr_version),
- IBLND_MSG_QUEUE_SIZE(conn->ibc_version));
- break;
-
case IBLND_REJECT_RDMA_FRAGS:
- CERROR("%s rejected: incompatible # of RDMA fragments %d, %d\n",
- libcfs_nid2str(peer->ibp_nid),
- cp ? cp->ibcp_max_frags :
- IBLND_RDMA_FRAGS(rej->ibr_version),
- IBLND_RDMA_FRAGS(conn->ibc_version));
+ kiblnd_reconnect(conn, rej->ibr_version,
+ incarnation, rej->ibr_why, cp);
break;

case IBLND_REJECT_NO_RESOURCES:
@@ -2676,22 +2707,22 @@ kiblnd_check_connreply(kib_conn_t *conn, void *priv, int priv_nob)
goto failed;
}

- if (msg->ibm_u.connparams.ibcp_queue_depth !=
- IBLND_MSG_QUEUE_SIZE(ver)) {
- CERROR("%s has incompatible queue depth %d(%d wanted)\n",
+ if (msg->ibm_u.connparams.ibcp_queue_depth >
+ conn->ibc_queue_depth) {
+ CERROR("%s has incompatible queue depth %d (<=%d wanted)\n",
libcfs_nid2str(peer->ibp_nid),
msg->ibm_u.connparams.ibcp_queue_depth,
- IBLND_MSG_QUEUE_SIZE(ver));
+ conn->ibc_queue_depth);
rc = -EPROTO;
goto failed;
}

- if (msg->ibm_u.connparams.ibcp_max_frags !=
- IBLND_RDMA_FRAGS(ver)) {
- CERROR("%s has incompatible max_frags %d (%d wanted)\n",
+ if (msg->ibm_u.connparams.ibcp_max_frags >
+ conn->ibc_max_frags) {
+ CERROR("%s has incompatible max_frags %d (<=%d wanted)\n",
libcfs_nid2str(peer->ibp_nid),
msg->ibm_u.connparams.ibcp_max_frags,
- IBLND_RDMA_FRAGS(ver));
+ conn->ibc_max_frags);
rc = -EPROTO;
goto failed;
}
@@ -2721,10 +2752,12 @@ kiblnd_check_connreply(kib_conn_t *conn, void *priv, int priv_nob)
}

conn->ibc_incarnation = msg->ibm_srcstamp;
- conn->ibc_credits =
- conn->ibc_reserved_credits = IBLND_MSG_QUEUE_SIZE(ver);
- LASSERT(conn->ibc_credits + conn->ibc_reserved_credits + IBLND_OOB_MSGS(ver)
- <= IBLND_RX_MSGS(ver));
+ conn->ibc_credits = msg->ibm_u.connparams.ibcp_queue_depth;
+ conn->ibc_reserved_credits = msg->ibm_u.connparams.ibcp_queue_depth;
+ conn->ibc_queue_depth = msg->ibm_u.connparams.ibcp_queue_depth;
+ conn->ibc_max_frags = msg->ibm_u.connparams.ibcp_max_frags;
+ LASSERT(conn->ibc_credits + conn->ibc_reserved_credits +
+ IBLND_OOB_MSGS(ver) <= IBLND_RX_MSGS(conn));

kiblnd_connreq_done(conn, 0);
return;
@@ -2761,7 +2794,8 @@ kiblnd_active_connect(struct rdma_cm_id *cmid)

read_unlock_irqrestore(&kiblnd_data.kib_global_lock, flags);

- conn = kiblnd_create_conn(peer, cmid, IBLND_CONN_ACTIVE_CONNECT, version);
+ conn = kiblnd_create_conn(peer, cmid, IBLND_CONN_ACTIVE_CONNECT,
+ version, NULL);
if (!conn) {
kiblnd_peer_connect_failed(peer, 1, -ENOMEM);
kiblnd_peer_decref(peer); /* lose cmid's ref */
@@ -2777,8 +2811,8 @@ kiblnd_active_connect(struct rdma_cm_id *cmid)

memset(msg, 0, sizeof(*msg));
kiblnd_init_msg(msg, IBLND_MSG_CONNREQ, sizeof(msg->ibm_u.connparams));
- msg->ibm_u.connparams.ibcp_queue_depth = IBLND_MSG_QUEUE_SIZE(version);
- msg->ibm_u.connparams.ibcp_max_frags = IBLND_RDMA_FRAGS(version);
+ msg->ibm_u.connparams.ibcp_queue_depth = conn->ibc_queue_depth;
+ msg->ibm_u.connparams.ibcp_max_frags = conn->ibc_max_frags;
msg->ibm_u.connparams.ibcp_max_msg_size = IBLND_MSG_SIZE;

kiblnd_pack_msg(peer->ibp_ni, msg, version,
--
1.7.1