From 1f0dba1e51cfc93bf4545811839a84c879086fd4 Mon Sep 17 00:00:00 2001 From: Chien Tung Date: Mon, 27 Apr 2009 13:36:03 -0700 Subject: RDMA/nes: Fix unused variable compile warning when INFINIBAND_NES_DEBUG=n Remove the NES_DEBUG that is causing the compile warning about an unused variable when INFINIBAND_NES_DEBUG is not enabled. Signed-off-by: Chien Tung Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_cm.c | 4 ---- 1 file changed, 4 deletions(-) (limited to 'drivers/infiniband/hw/nes/nes_cm.c') diff --git a/drivers/infiniband/hw/nes/nes_cm.c b/drivers/infiniband/hw/nes/nes_cm.c index dbd9a75474e..7da5437d985 100644 --- a/drivers/infiniband/hw/nes/nes_cm.c +++ b/drivers/infiniband/hw/nes/nes_cm.c @@ -854,7 +854,6 @@ static struct nes_cm_listener *find_listener(struct nes_cm_core *cm_core, { unsigned long flags; struct nes_cm_listener *listen_node; - __be32 tmp_addr = cpu_to_be32(dst_addr); /* walk list and find cm_node associated with this session ID */ spin_lock_irqsave(&cm_core->listen_list_lock, flags); @@ -871,9 +870,6 @@ static struct nes_cm_listener *find_listener(struct nes_cm_core *cm_core, } spin_unlock_irqrestore(&cm_core->listen_list_lock, flags); - nes_debug(NES_DBG_CM, "Unable to find listener for %pI4:%x\n", - &tmp_addr, dst_port); - /* no listener */ return NULL; } -- cgit v1.2.3 From 53094c388f11d79f742eaf743c9fd740a881f2c0 Mon Sep 17 00:00:00 2001 From: Faisal Latif Date: Mon, 27 Apr 2009 13:37:34 -0700 Subject: RDMA/nes: Do not set apbvt entry for loopback When a connect request comes, apbvt should only be set for non-loopback connections. Signed-off-by: Faisal Latif Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_cm.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) (limited to 'drivers/infiniband/hw/nes/nes_cm.c') diff --git a/drivers/infiniband/hw/nes/nes_cm.c b/drivers/infiniband/hw/nes/nes_cm.c index 7da5437d985..1efe0beca06 100644 --- a/drivers/infiniband/hw/nes/nes_cm.c +++ b/drivers/infiniband/hw/nes/nes_cm.c @@ -2955,6 +2955,7 @@ int nes_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) struct nes_device *nesdev; struct nes_cm_node *cm_node; struct nes_cm_info cm_info; + int apbvt_set = 0; ibqp = nes_get_qp(cm_id->device, conn_param->qpn); if (!ibqp) @@ -2992,9 +2993,11 @@ int nes_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) conn_param->private_data_len); if (cm_id->local_addr.sin_addr.s_addr != - cm_id->remote_addr.sin_addr.s_addr) + cm_id->remote_addr.sin_addr.s_addr) { nes_manage_apbvt(nesvnic, ntohs(cm_id->local_addr.sin_port), PCI_FUNC(nesdev->pcidev->devfn), NES_MANAGE_APBVT_ADD); + apbvt_set = 1; + } /* set up the connection params for the node */ cm_info.loc_addr = htonl(cm_id->local_addr.sin_addr.s_addr); @@ -3011,8 +3014,7 @@ int nes_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) conn_param->private_data_len, (void *)conn_param->private_data, &cm_info); if (!cm_node) { - if (cm_id->local_addr.sin_addr.s_addr != - cm_id->remote_addr.sin_addr.s_addr) + if (apbvt_set) nes_manage_apbvt(nesvnic, ntohs(cm_id->local_addr.sin_port), PCI_FUNC(nesdev->pcidev->devfn), NES_MANAGE_APBVT_DEL); @@ -3021,7 +3023,7 @@ int nes_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) return -ENOMEM; } - cm_node->apbvt_set = 1; + cm_node->apbvt_set = apbvt_set; nesqp->cm_node = cm_node; cm_node->nesqp = nesqp; nes_add_ref(&nesqp->ibqp); -- cgit v1.2.3 From c11470f9f4d4490cd7e9563f604c4c7868caf6de Mon Sep 17 00:00:00 2001 From: Faisal Latif Date: Mon, 27 Apr 2009 13:38:31 -0700 Subject: RDMA/nes: Check for sequence number wrap-around check_seq() was not checking if the seq#s have wrapped. Fix it. Signed-off-by: Faisal Latif Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_cm.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/infiniband/hw/nes/nes_cm.c') diff --git a/drivers/infiniband/hw/nes/nes_cm.c b/drivers/infiniband/hw/nes/nes_cm.c index 1efe0beca06..1f7a659e6e9 100644 --- a/drivers/infiniband/hw/nes/nes_cm.c +++ b/drivers/infiniband/hw/nes/nes_cm.c @@ -56,6 +56,7 @@ #include #include #include +#include #include "nes.h" @@ -1514,7 +1515,7 @@ static int check_seq(struct nes_cm_node *cm_node, struct tcphdr *tcph, rcv_wnd = cm_node->tcp_cntxt.rcv_wnd; if (ack_seq != loc_seq_num) err = 1; - else if ((seq + rcv_wnd) < rcv_nxt) + else if (!between(seq, rcv_nxt, (rcv_nxt+rcv_wnd))) err = 1; if (err) { nes_debug(NES_DBG_CM, "%s[%u] create abort for cm_node=%p " -- cgit v1.2.3 From 4e9c390036196f89208cf9574dfd19daae146776 Mon Sep 17 00:00:00 2001 From: Faisal Latif Date: Mon, 27 Apr 2009 13:39:36 -0700 Subject: RDMA/nes: Increase rexmit timeout interval Under heavy load with large cluster testing, it may take longer to receive a response to MPA requests. Change the driver to wait longer after each rexmit to max time value. Signed-off-by: Faisal Latif Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_cm.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'drivers/infiniband/hw/nes/nes_cm.c') diff --git a/drivers/infiniband/hw/nes/nes_cm.c b/drivers/infiniband/hw/nes/nes_cm.c index 1f7a659e6e9..4969c386785 100644 --- a/drivers/infiniband/hw/nes/nes_cm.c +++ b/drivers/infiniband/hw/nes/nes_cm.c @@ -541,6 +541,7 @@ static void nes_cm_timer_tick(unsigned long pass) struct list_head *list_node; struct nes_cm_core *cm_core = g_cm_core; u32 settimer = 0; + unsigned long timetosend; int ret = NETDEV_TX_OK; struct list_head timer_list; @@ -645,8 +646,11 @@ static void nes_cm_timer_tick(unsigned long pass) send_entry->retrycount); if (send_entry->send_retrans) { send_entry->retranscount--; + timetosend = (NES_RETRY_TIMEOUT << + (NES_DEFAULT_RETRANS - send_entry->retranscount)); + send_entry->timetosend = jiffies + - NES_RETRY_TIMEOUT; + min(timetosend, NES_MAX_TIMEOUT); if (nexttimeout > send_entry->timetosend || !settimer) { nexttimeout = send_entry->timetosend; -- cgit v1.2.3 From 109d67e4f12b828113ca8ccf4a735972dd984f40 Mon Sep 17 00:00:00 2001 From: Faisal Latif Date: Mon, 27 Apr 2009 13:41:06 -0700 Subject: RDMA/nes: Fix hang issues for large cluster dynamic connections Running large cluster setup, we are hanging after many hours of testing. Fixing this required going over the code and making sure the rexmit entry was properly removed based on the cm_node's state and packet received. Also when receiving a FIN packet, check seq# and make sure there were no errors before calling handle_fin(). Following are the changes done in nes_cm.c: * handle_ack_pkt() needs to return error value, so in case of error, handle_fin() is not called. Some cleanup done while going over the code. * handle_rst_pkt(), handling of cm_node's NES_CM_STATE_LAST_ACK is missing. * process_packet(), in case of FIN only packet is received, call check_seq() before processing. * in handle_fin_pkt(), we are calling cleanup_retrans_entry() for all conditions, even if the packets need to be dropped. Signed-off-by: Faisal Latif Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_cm.c | 56 +++++++++++++++++--------------------- 1 file changed, 25 insertions(+), 31 deletions(-) (limited to 'drivers/infiniband/hw/nes/nes_cm.c') diff --git a/drivers/infiniband/hw/nes/nes_cm.c b/drivers/infiniband/hw/nes/nes_cm.c index 4969c386785..2c90b38daef 100644 --- a/drivers/infiniband/hw/nes/nes_cm.c +++ b/drivers/infiniband/hw/nes/nes_cm.c @@ -1326,18 +1326,20 @@ static void handle_fin_pkt(struct nes_cm_node *cm_node) nes_debug(NES_DBG_CM, "Received FIN, cm_node = %p, state = %u. " "refcnt=%d\n", cm_node, cm_node->state, atomic_read(&cm_node->ref_count)); - cm_node->tcp_cntxt.rcv_nxt++; - cleanup_retrans_entry(cm_node); switch (cm_node->state) { case NES_CM_STATE_SYN_RCVD: case NES_CM_STATE_SYN_SENT: case NES_CM_STATE_ESTABLISHED: case NES_CM_STATE_MPAREQ_SENT: case NES_CM_STATE_MPAREJ_RCVD: + cm_node->tcp_cntxt.rcv_nxt++; + cleanup_retrans_entry(cm_node); cm_node->state = NES_CM_STATE_LAST_ACK; send_fin(cm_node, NULL); break; case NES_CM_STATE_FIN_WAIT1: + cm_node->tcp_cntxt.rcv_nxt++; + cleanup_retrans_entry(cm_node); cm_node->state = NES_CM_STATE_CLOSING; send_ack(cm_node, NULL); /* Wait for ACK as this is simultanous close.. @@ -1345,11 +1347,15 @@ static void handle_fin_pkt(struct nes_cm_node *cm_node) * Just rm the node.. Done.. */ break; case NES_CM_STATE_FIN_WAIT2: + cm_node->tcp_cntxt.rcv_nxt++; + cleanup_retrans_entry(cm_node); cm_node->state = NES_CM_STATE_TIME_WAIT; send_ack(cm_node, NULL); schedule_nes_timer(cm_node, NULL, NES_TIMER_TYPE_CLOSE, 1, 0); break; case NES_CM_STATE_TIME_WAIT: + cm_node->tcp_cntxt.rcv_nxt++; + cleanup_retrans_entry(cm_node); cm_node->state = NES_CM_STATE_CLOSED; rem_ref_cm_node(cm_node->cm_core, cm_node); break; @@ -1385,7 +1391,6 @@ static void handle_rst_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, passive_state = atomic_add_return(1, &cm_node->passive_state); if (passive_state == NES_SEND_RESET_EVENT) create_event(cm_node, NES_CM_EVENT_RESET); - cleanup_retrans_entry(cm_node); cm_node->state = NES_CM_STATE_CLOSED; dev_kfree_skb_any(skb); break; @@ -1399,17 +1404,16 @@ static void handle_rst_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, active_open_err(cm_node, skb, reset); break; case NES_CM_STATE_CLOSED: - cleanup_retrans_entry(cm_node); drop_packet(skb); break; + case NES_CM_STATE_LAST_ACK: + cm_node->cm_id->rem_ref(cm_node->cm_id); case NES_CM_STATE_TIME_WAIT: - cleanup_retrans_entry(cm_node); cm_node->state = NES_CM_STATE_CLOSED; rem_ref_cm_node(cm_node->cm_core, cm_node); drop_packet(skb); break; case NES_CM_STATE_FIN_WAIT1: - cleanup_retrans_entry(cm_node); nes_debug(NES_DBG_CM, "Bad state %s[%u]\n", __func__, __LINE__); default: drop_packet(skb); @@ -1456,6 +1460,7 @@ static void handle_rcv_mpa(struct nes_cm_node *cm_node, struct sk_buff *skb) NES_PASSIVE_STATE_INDICATED); break; case NES_CM_STATE_MPAREQ_SENT: + cleanup_retrans_entry(cm_node); if (res_type == NES_MPA_REQUEST_REJECT) { type = NES_CM_EVENT_MPA_REJECT; cm_node->state = NES_CM_STATE_MPAREJ_RCVD; @@ -1653,49 +1658,39 @@ static void handle_synack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, } } -static void handle_ack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, +static int handle_ack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, struct tcphdr *tcph) { int datasize = 0; u32 inc_sequence; u32 rem_seq_ack; u32 rem_seq; - int ret; + int ret = 0; int optionsize; optionsize = (tcph->doff << 2) - sizeof(struct tcphdr); if (check_seq(cm_node, tcph, skb)) - return; + return -EINVAL; skb_pull(skb, tcph->doff << 2); inc_sequence = ntohl(tcph->seq); rem_seq = ntohl(tcph->seq); rem_seq_ack = ntohl(tcph->ack_seq); datasize = skb->len; - cleanup_retrans_entry(cm_node); switch (cm_node->state) { case NES_CM_STATE_SYN_RCVD: /* Passive OPEN */ + cleanup_retrans_entry(cm_node); ret = handle_tcp_options(cm_node, tcph, skb, optionsize, 1); if (ret) break; cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq); - if (cm_node->tcp_cntxt.rem_ack_num != - cm_node->tcp_cntxt.loc_seq_num) { - nes_debug(NES_DBG_CM, "rem_ack_num != loc_seq_num\n"); - cleanup_retrans_entry(cm_node); - send_reset(cm_node, skb); - return; - } cm_node->state = NES_CM_STATE_ESTABLISHED; - cleanup_retrans_entry(cm_node); if (datasize) { cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize; handle_rcv_mpa(cm_node, skb); - } else { /* rcvd ACK only */ + } else /* rcvd ACK only */ dev_kfree_skb_any(skb); - cleanup_retrans_entry(cm_node); - } break; case NES_CM_STATE_ESTABLISHED: /* Passive OPEN */ @@ -1707,15 +1702,12 @@ static void handle_ack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, drop_packet(skb); break; case NES_CM_STATE_MPAREQ_SENT: - cleanup_retrans_entry(cm_node); cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq); if (datasize) { cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize; handle_rcv_mpa(cm_node, skb); - } else { /* Could be just an ack pkt.. */ - cleanup_retrans_entry(cm_node); + } else /* Could be just an ack pkt.. */ dev_kfree_skb_any(skb); - } break; case NES_CM_STATE_LISTENING: case NES_CM_STATE_CLOSED: @@ -1723,11 +1715,10 @@ static void handle_ack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, send_reset(cm_node, skb); break; case NES_CM_STATE_LAST_ACK: + case NES_CM_STATE_CLOSING: cleanup_retrans_entry(cm_node); cm_node->state = NES_CM_STATE_CLOSED; cm_node->cm_id->rem_ref(cm_node->cm_id); - case NES_CM_STATE_CLOSING: - cleanup_retrans_entry(cm_node); rem_ref_cm_node(cm_node->cm_core, cm_node); drop_packet(skb); break; @@ -1742,9 +1733,11 @@ static void handle_ack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb, case NES_CM_STATE_MPAREQ_RCVD: case NES_CM_STATE_UNKNOWN: default: + cleanup_retrans_entry(cm_node); drop_packet(skb); break; } + return ret; } @@ -1850,6 +1843,7 @@ static void process_packet(struct nes_cm_node *cm_node, struct sk_buff *skb, enum nes_tcpip_pkt_type pkt_type = NES_PKT_TYPE_UNKNOWN; struct tcphdr *tcph = tcp_hdr(skb); u32 fin_set = 0; + int ret = 0; skb_pull(skb, ip_hdr(skb)->ihl << 2); nes_debug(NES_DBG_CM, "process_packet: cm_node=%p state =%d syn=%d " @@ -1875,17 +1869,17 @@ static void process_packet(struct nes_cm_node *cm_node, struct sk_buff *skb, handle_synack_pkt(cm_node, skb, tcph); break; case NES_PKT_TYPE_ACK: - handle_ack_pkt(cm_node, skb, tcph); - if (fin_set) + ret = handle_ack_pkt(cm_node, skb, tcph); + if (fin_set && !ret) handle_fin_pkt(cm_node); break; case NES_PKT_TYPE_RST: handle_rst_pkt(cm_node, skb, tcph); break; default: - drop_packet(skb); - if (fin_set) + if ((fin_set) && (!check_seq(cm_node, tcph, skb))) handle_fin_pkt(cm_node); + drop_packet(skb); break; } } -- cgit v1.2.3 From 9256b2513074164b4555617c4a3b82d36abf03e5 Mon Sep 17 00:00:00 2001 From: Faisal Latif Date: Mon, 27 Apr 2009 13:45:19 -0700 Subject: RDMA/nes: Fix error path in nes_accept() If reg_phys_mem() fails, we need to free memory allocated for MPA frame with private data before returning the error. Also move nes_add_ref() after the reg_phys_mem() is successful. Signed-off-by: Faisal Latif Signed-off-by: Roland Dreier --- drivers/infiniband/hw/nes/nes_cm.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'drivers/infiniband/hw/nes/nes_cm.c') diff --git a/drivers/infiniband/hw/nes/nes_cm.c b/drivers/infiniband/hw/nes/nes_cm.c index 2c90b38daef..11c7d664201 100644 --- a/drivers/infiniband/hw/nes/nes_cm.c +++ b/drivers/infiniband/hw/nes/nes_cm.c @@ -2705,7 +2705,6 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) /* associate the node with the QP */ nesqp->cm_node = (void *)cm_node; cm_node->nesqp = nesqp; - nes_add_ref(&nesqp->ibqp); nes_debug(NES_DBG_CM, "QP%u, cm_node=%p, jiffies = %lu listener = %p\n", nesqp->hwqp.qp_id, cm_node, jiffies, cm_node->listener); @@ -2758,6 +2757,9 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) nes_debug(NES_DBG_CM, "Unable to register memory region" "for lSMM for cm_node = %p \n", cm_node); + pci_free_consistent(nesdev->pcidev, + nesqp->private_data_len+sizeof(struct ietf_mpa_frame), + nesqp->ietf_frame, nesqp->ietf_frame_pbase); return -ENOMEM; } @@ -2874,6 +2876,7 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) /* notify OF layer that accept event was successful */ cm_id->add_ref(cm_id); + nes_add_ref(&nesqp->ibqp); cm_event.event = IW_CM_EVENT_ESTABLISHED; cm_event.status = IW_CM_EVENT_STATUS_ACCEPTED; -- cgit v1.2.3