RDS/IB: Infiniband transport
Andy Grover [Tue, 24 Feb 2009 15:30:30 +0000 (15:30 +0000)]
Registers as an RDS transport and an IB client, and uses IB CM
API to allocate ids, queue pairs, and the rest of that fun stuff.

Signed-off-by: Andy Grover <andy.grover@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/rds/ib.c [new file with mode: 0644]
net/rds/ib.h [new file with mode: 0644]
net/rds/ib_cm.c [new file with mode: 0644]

diff --git a/net/rds/ib.c b/net/rds/ib.c
new file mode 100644 (file)
index 0000000..06a7b79
--- /dev/null
@@ -0,0 +1,323 @@
+/*
+ * Copyright (c) 2006 Oracle.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/kernel.h>
+#include <linux/in.h>
+#include <linux/if.h>
+#include <linux/netdevice.h>
+#include <linux/inetdevice.h>
+#include <linux/if_arp.h>
+#include <linux/delay.h>
+
+#include "rds.h"
+#include "ib.h"
+
+unsigned int fmr_pool_size = RDS_FMR_POOL_SIZE;
+unsigned int fmr_message_size = RDS_FMR_SIZE + 1; /* +1 allows for unaligned MRs */
+
+module_param(fmr_pool_size, int, 0444);
+MODULE_PARM_DESC(fmr_pool_size, " Max number of fmr per HCA");
+module_param(fmr_message_size, int, 0444);
+MODULE_PARM_DESC(fmr_message_size, " Max size of a RDMA transfer");
+
+struct list_head rds_ib_devices;
+
+DEFINE_SPINLOCK(ib_nodev_conns_lock);
+LIST_HEAD(ib_nodev_conns);
+
+void rds_ib_add_one(struct ib_device *device)
+{
+       struct rds_ib_device *rds_ibdev;
+       struct ib_device_attr *dev_attr;
+
+       /* Only handle IB (no iWARP) devices */
+       if (device->node_type != RDMA_NODE_IB_CA)
+               return;
+
+       dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL);
+       if (!dev_attr)
+               return;
+
+       if (ib_query_device(device, dev_attr)) {
+               rdsdebug("Query device failed for %s\n", device->name);
+               goto free_attr;
+       }
+
+       rds_ibdev = kmalloc(sizeof *rds_ibdev, GFP_KERNEL);
+       if (!rds_ibdev)
+               goto free_attr;
+
+       spin_lock_init(&rds_ibdev->spinlock);
+
+       rds_ibdev->max_wrs = dev_attr->max_qp_wr;
+       rds_ibdev->max_sge = min(dev_attr->max_sge, RDS_IB_MAX_SGE);
+
+       rds_ibdev->fmr_page_shift = max(9, ffs(dev_attr->page_size_cap) - 1);
+       rds_ibdev->fmr_page_size  = 1 << rds_ibdev->fmr_page_shift;
+       rds_ibdev->fmr_page_mask  = ~((u64) rds_ibdev->fmr_page_size - 1);
+       rds_ibdev->fmr_max_remaps = dev_attr->max_map_per_fmr?: 32;
+       rds_ibdev->max_fmrs = dev_attr->max_fmr ?
+                       min_t(unsigned int, dev_attr->max_fmr, fmr_pool_size) :
+                       fmr_pool_size;
+
+       rds_ibdev->dev = device;
+       rds_ibdev->pd = ib_alloc_pd(device);
+       if (IS_ERR(rds_ibdev->pd))
+               goto free_dev;
+
+       rds_ibdev->mr = ib_get_dma_mr(rds_ibdev->pd,
+                                     IB_ACCESS_LOCAL_WRITE);
+       if (IS_ERR(rds_ibdev->mr))
+               goto err_pd;
+
+       rds_ibdev->mr_pool = rds_ib_create_mr_pool(rds_ibdev);
+       if (IS_ERR(rds_ibdev->mr_pool)) {
+               rds_ibdev->mr_pool = NULL;
+               goto err_mr;
+       }
+
+       INIT_LIST_HEAD(&rds_ibdev->ipaddr_list);
+       INIT_LIST_HEAD(&rds_ibdev->conn_list);
+       list_add_tail(&rds_ibdev->list, &rds_ib_devices);
+
+       ib_set_client_data(device, &rds_ib_client, rds_ibdev);
+
+       goto free_attr;
+
+err_mr:
+       ib_dereg_mr(rds_ibdev->mr);
+err_pd:
+       ib_dealloc_pd(rds_ibdev->pd);
+free_dev:
+       kfree(rds_ibdev);
+free_attr:
+       kfree(dev_attr);
+}
+
+void rds_ib_remove_one(struct ib_device *device)
+{
+       struct rds_ib_device *rds_ibdev;
+       struct rds_ib_ipaddr *i_ipaddr, *i_next;
+
+       rds_ibdev = ib_get_client_data(device, &rds_ib_client);
+       if (!rds_ibdev)
+               return;
+
+       list_for_each_entry_safe(i_ipaddr, i_next, &rds_ibdev->ipaddr_list, list) {
+               list_del(&i_ipaddr->list);
+               kfree(i_ipaddr);
+       }
+
+       rds_ib_remove_conns(rds_ibdev);
+
+       if (rds_ibdev->mr_pool)
+               rds_ib_destroy_mr_pool(rds_ibdev->mr_pool);
+
+       ib_dereg_mr(rds_ibdev->mr);
+
+       while (ib_dealloc_pd(rds_ibdev->pd)) {
+               rdsdebug("Failed to dealloc pd %p\n", rds_ibdev->pd);
+               msleep(1);
+       }
+
+       list_del(&rds_ibdev->list);
+       kfree(rds_ibdev);
+}
+
+struct ib_client rds_ib_client = {
+       .name   = "rds_ib",
+       .add    = rds_ib_add_one,
+       .remove = rds_ib_remove_one
+};
+
+static int rds_ib_conn_info_visitor(struct rds_connection *conn,
+                                   void *buffer)
+{
+       struct rds_info_rdma_connection *iinfo = buffer;
+       struct rds_ib_connection *ic;
+
+       /* We will only ever look at IB transports */
+       if (conn->c_trans != &rds_ib_transport)
+               return 0;
+
+       iinfo->src_addr = conn->c_laddr;
+       iinfo->dst_addr = conn->c_faddr;
+
+       memset(&iinfo->src_gid, 0, sizeof(iinfo->src_gid));
+       memset(&iinfo->dst_gid, 0, sizeof(iinfo->dst_gid));
+       if (rds_conn_state(conn) == RDS_CONN_UP) {
+               struct rds_ib_device *rds_ibdev;
+               struct rdma_dev_addr *dev_addr;
+
+               ic = conn->c_transport_data;
+               dev_addr = &ic->i_cm_id->route.addr.dev_addr;
+
+               ib_addr_get_sgid(dev_addr, (union ib_gid *) &iinfo->src_gid);
+               ib_addr_get_dgid(dev_addr, (union ib_gid *) &iinfo->dst_gid);
+
+               rds_ibdev = ib_get_client_data(ic->i_cm_id->device, &rds_ib_client);
+               iinfo->max_send_wr = ic->i_send_ring.w_nr;
+               iinfo->max_recv_wr = ic->i_recv_ring.w_nr;
+               iinfo->max_send_sge = rds_ibdev->max_sge;
+               rds_ib_get_mr_info(rds_ibdev, iinfo);
+       }
+       return 1;
+}
+
+static void rds_ib_ic_info(struct socket *sock, unsigned int len,
+                          struct rds_info_iterator *iter,
+                          struct rds_info_lengths *lens)
+{
+       rds_for_each_conn_info(sock, len, iter, lens,
+                               rds_ib_conn_info_visitor,
+                               sizeof(struct rds_info_rdma_connection));
+}
+
+
+/*
+ * Early RDS/IB was built to only bind to an address if there is an IPoIB
+ * device with that address set.
+ *
+ * If it were me, I'd advocate for something more flexible.  Sending and
+ * receiving should be device-agnostic.  Transports would try and maintain
+ * connections between peers who have messages queued.  Userspace would be
+ * allowed to influence which paths have priority.  We could call userspace
+ * asserting this policy "routing".
+ */
+static int rds_ib_laddr_check(__be32 addr)
+{
+       int ret;
+       struct rdma_cm_id *cm_id;
+       struct sockaddr_in sin;
+
+       /* Create a CMA ID and try to bind it. This catches both
+        * IB and iWARP capable NICs.
+        */
+       cm_id = rdma_create_id(NULL, NULL, RDMA_PS_TCP);
+       if (!cm_id)
+               return -EADDRNOTAVAIL;
+
+       memset(&sin, 0, sizeof(sin));
+       sin.sin_family = AF_INET;
+       sin.sin_addr.s_addr = addr;
+
+       /* rdma_bind_addr will only succeed for IB & iWARP devices */
+       ret = rdma_bind_addr(cm_id, (struct sockaddr *)&sin);
+       /* due to this, we will claim to support iWARP devices unless we
+          check node_type. */
+       if (ret || cm_id->device->node_type != RDMA_NODE_IB_CA)
+               ret = -EADDRNOTAVAIL;
+
+       rdsdebug("addr %pI4 ret %d node type %d\n",
+               &addr, ret,
+               cm_id->device ? cm_id->device->node_type : -1);
+
+       rdma_destroy_id(cm_id);
+
+       return ret;
+}
+
+void rds_ib_exit(void)
+{
+       rds_info_deregister_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info);
+       rds_ib_remove_nodev_conns();
+       ib_unregister_client(&rds_ib_client);
+       rds_ib_sysctl_exit();
+       rds_ib_recv_exit();
+       rds_trans_unregister(&rds_ib_transport);
+}
+
+struct rds_transport rds_ib_transport = {
+       .laddr_check            = rds_ib_laddr_check,
+       .xmit_complete          = rds_ib_xmit_complete,
+       .xmit                   = rds_ib_xmit,
+       .xmit_cong_map          = NULL,
+       .xmit_rdma              = rds_ib_xmit_rdma,
+       .recv                   = rds_ib_recv,
+       .conn_alloc             = rds_ib_conn_alloc,
+       .conn_free              = rds_ib_conn_free,
+       .conn_connect           = rds_ib_conn_connect,
+       .conn_shutdown          = rds_ib_conn_shutdown,
+       .inc_copy_to_user       = rds_ib_inc_copy_to_user,
+       .inc_purge              = rds_ib_inc_purge,
+       .inc_free               = rds_ib_inc_free,
+       .cm_initiate_connect    = rds_ib_cm_initiate_connect,
+       .cm_handle_connect      = rds_ib_cm_handle_connect,
+       .cm_connect_complete    = rds_ib_cm_connect_complete,
+       .stats_info_copy        = rds_ib_stats_info_copy,
+       .exit                   = rds_ib_exit,
+       .get_mr                 = rds_ib_get_mr,
+       .sync_mr                = rds_ib_sync_mr,
+       .free_mr                = rds_ib_free_mr,
+       .flush_mrs              = rds_ib_flush_mrs,
+       .t_owner                = THIS_MODULE,
+       .t_name                 = "infiniband",
+};
+
+int __init rds_ib_init(void)
+{
+       int ret;
+
+       INIT_LIST_HEAD(&rds_ib_devices);
+
+       ret = ib_register_client(&rds_ib_client);
+       if (ret)
+               goto out;
+
+       ret = rds_ib_sysctl_init();
+       if (ret)
+               goto out_ibreg;
+
+       ret = rds_ib_recv_init();
+       if (ret)
+               goto out_sysctl;
+
+       ret = rds_trans_register(&rds_ib_transport);
+       if (ret)
+               goto out_recv;
+
+       rds_info_register_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info);
+
+       goto out;
+
+out_recv:
+       rds_ib_recv_exit();
+out_sysctl:
+       rds_ib_sysctl_exit();
+out_ibreg:
+       ib_unregister_client(&rds_ib_client);
+out:
+       return ret;
+}
+
+MODULE_LICENSE("GPL");
+
diff --git a/net/rds/ib.h b/net/rds/ib.h
new file mode 100644 (file)
index 0000000..8be563a
--- /dev/null
@@ -0,0 +1,367 @@
+#ifndef _RDS_IB_H
+#define _RDS_IB_H
+
+#include <rdma/ib_verbs.h>
+#include <rdma/rdma_cm.h>
+#include "rds.h"
+#include "rdma_transport.h"
+
+#define RDS_FMR_SIZE                   256
+#define RDS_FMR_POOL_SIZE              4096
+
+#define RDS_IB_MAX_SGE                 8
+#define RDS_IB_RECV_SGE                2
+
+#define RDS_IB_DEFAULT_RECV_WR         1024
+#define RDS_IB_DEFAULT_SEND_WR         256
+
+#define RDS_IB_SUPPORTED_PROTOCOLS     0x00000003      /* minor versions supported */
+
+extern struct list_head rds_ib_devices;
+
+/*
+ * IB posts RDS_FRAG_SIZE fragments of pages to the receive queues to
+ * try and minimize the amount of memory tied up both the device and
+ * socket receive queues.
+ */
+/* page offset of the final full frag that fits in the page */
+#define RDS_PAGE_LAST_OFF (((PAGE_SIZE  / RDS_FRAG_SIZE) - 1) * RDS_FRAG_SIZE)
+struct rds_page_frag {
+       struct list_head        f_item;
+       struct page             *f_page;
+       unsigned long           f_offset;
+       dma_addr_t              f_mapped;
+};
+
+struct rds_ib_incoming {
+       struct list_head        ii_frags;
+       struct rds_incoming     ii_inc;
+};
+
+struct rds_ib_connect_private {
+       /* Add new fields at the end, and don't permute existing fields. */
+       __be32                  dp_saddr;
+       __be32                  dp_daddr;
+       u8                      dp_protocol_major;
+       u8                      dp_protocol_minor;
+       __be16                  dp_protocol_minor_mask; /* bitmask */
+       __be32                  dp_reserved1;
+       __be64                  dp_ack_seq;
+       __be32                  dp_credit;              /* non-zero enables flow ctl */
+};
+
+struct rds_ib_send_work {
+       struct rds_message      *s_rm;
+       struct rds_rdma_op      *s_op;
+       struct ib_send_wr       s_wr;
+       struct ib_sge           s_sge[RDS_IB_MAX_SGE];
+       unsigned long           s_queued;
+};
+
+struct rds_ib_recv_work {
+       struct rds_ib_incoming  *r_ibinc;
+       struct rds_page_frag    *r_frag;
+       struct ib_recv_wr       r_wr;
+       struct ib_sge           r_sge[2];
+};
+
+struct rds_ib_work_ring {
+       u32             w_nr;
+       u32             w_alloc_ptr;
+       u32             w_alloc_ctr;
+       u32             w_free_ptr;
+       atomic_t        w_free_ctr;
+};
+
+struct rds_ib_device;
+
+struct rds_ib_connection {
+
+       struct list_head        ib_node;
+       struct rds_ib_device    *rds_ibdev;
+       struct rds_connection   *conn;
+
+       /* alphabet soup, IBTA style */
+       struct rdma_cm_id       *i_cm_id;
+       struct ib_pd            *i_pd;
+       struct ib_mr            *i_mr;
+       struct ib_cq            *i_send_cq;
+       struct ib_cq            *i_recv_cq;
+
+       /* tx */
+       struct rds_ib_work_ring i_send_ring;
+       struct rds_message      *i_rm;
+       struct rds_header       *i_send_hdrs;
+       u64                     i_send_hdrs_dma;
+       struct rds_ib_send_work *i_sends;
+
+       /* rx */
+       struct mutex            i_recv_mutex;
+       struct rds_ib_work_ring i_recv_ring;
+       struct rds_ib_incoming  *i_ibinc;
+       u32                     i_recv_data_rem;
+       struct rds_header       *i_recv_hdrs;
+       u64                     i_recv_hdrs_dma;
+       struct rds_ib_recv_work *i_recvs;
+       struct rds_page_frag    i_frag;
+       u64                     i_ack_recv;     /* last ACK received */
+
+       /* sending acks */
+       unsigned long           i_ack_flags;
+       u64                     i_ack_next;     /* next ACK to send */
+       struct rds_header       *i_ack;
+       struct ib_send_wr       i_ack_wr;
+       struct ib_sge           i_ack_sge;
+       u64                     i_ack_dma;
+       unsigned long           i_ack_queued;
+
+       /* Flow control related information
+        *
+        * Our algorithm uses a pair variables that we need to access
+        * atomically - one for the send credits, and one posted
+        * recv credits we need to transfer to remote.
+        * Rather than protect them using a slow spinlock, we put both into
+        * a single atomic_t and update it using cmpxchg
+        */
+       atomic_t                i_credits;
+
+       /* Protocol version specific information */
+       unsigned int            i_flowctl:1;    /* enable/disable flow ctl */
+
+       /* Batched completions */
+       unsigned int            i_unsignaled_wrs;
+       long                    i_unsignaled_bytes;
+};
+
+/* This assumes that atomic_t is at least 32 bits */
+#define IB_GET_SEND_CREDITS(v) ((v) & 0xffff)
+#define IB_GET_POST_CREDITS(v) ((v) >> 16)
+#define IB_SET_SEND_CREDITS(v) ((v) & 0xffff)
+#define IB_SET_POST_CREDITS(v) ((v) << 16)
+
+struct rds_ib_ipaddr {
+       struct list_head        list;
+       __be32                  ipaddr;
+};
+
+struct rds_ib_device {
+       struct list_head        list;
+       struct list_head        ipaddr_list;
+       struct list_head        conn_list;
+       struct ib_device        *dev;
+       struct ib_pd            *pd;
+       struct ib_mr            *mr;
+       struct rds_ib_mr_pool   *mr_pool;
+       int                     fmr_page_shift;
+       int                     fmr_page_size;
+       u64                     fmr_page_mask;
+       unsigned int            fmr_max_remaps;
+       unsigned int            max_fmrs;
+       int                     max_sge;
+       unsigned int            max_wrs;
+       spinlock_t              spinlock;       /* protect the above */
+};
+
+/* bits for i_ack_flags */
+#define IB_ACK_IN_FLIGHT       0
+#define IB_ACK_REQUESTED       1
+
+/* Magic WR_ID for ACKs */
+#define RDS_IB_ACK_WR_ID       (~(u64) 0)
+
+struct rds_ib_statistics {
+       uint64_t        s_ib_connect_raced;
+       uint64_t        s_ib_listen_closed_stale;
+       uint64_t        s_ib_tx_cq_call;
+       uint64_t        s_ib_tx_cq_event;
+       uint64_t        s_ib_tx_ring_full;
+       uint64_t        s_ib_tx_throttle;
+       uint64_t        s_ib_tx_sg_mapping_failure;
+       uint64_t        s_ib_tx_stalled;
+       uint64_t        s_ib_tx_credit_updates;
+       uint64_t        s_ib_rx_cq_call;
+       uint64_t        s_ib_rx_cq_event;
+       uint64_t        s_ib_rx_ring_empty;
+       uint64_t        s_ib_rx_refill_from_cq;
+       uint64_t        s_ib_rx_refill_from_thread;
+       uint64_t        s_ib_rx_alloc_limit;
+       uint64_t        s_ib_rx_credit_updates;
+       uint64_t        s_ib_ack_sent;
+       uint64_t        s_ib_ack_send_failure;
+       uint64_t        s_ib_ack_send_delayed;
+       uint64_t        s_ib_ack_send_piggybacked;
+       uint64_t        s_ib_ack_received;
+       uint64_t        s_ib_rdma_mr_alloc;
+       uint64_t        s_ib_rdma_mr_free;
+       uint64_t        s_ib_rdma_mr_used;
+       uint64_t        s_ib_rdma_mr_pool_flush;
+       uint64_t        s_ib_rdma_mr_pool_wait;
+       uint64_t        s_ib_rdma_mr_pool_depleted;
+};
+
+extern struct workqueue_struct *rds_ib_wq;
+
+/*
+ * Fake ib_dma_sync_sg_for_{cpu,device} as long as ib_verbs.h
+ * doesn't define it.
+ */
+static inline void rds_ib_dma_sync_sg_for_cpu(struct ib_device *dev,
+               struct scatterlist *sg, unsigned int sg_dma_len, int direction)
+{
+       unsigned int i;
+
+       for (i = 0; i < sg_dma_len; ++i) {
+               ib_dma_sync_single_for_cpu(dev,
+                               ib_sg_dma_address(dev, &sg[i]),
+                               ib_sg_dma_len(dev, &sg[i]),
+                               direction);
+       }
+}
+#define ib_dma_sync_sg_for_cpu rds_ib_dma_sync_sg_for_cpu
+
+static inline void rds_ib_dma_sync_sg_for_device(struct ib_device *dev,
+               struct scatterlist *sg, unsigned int sg_dma_len, int direction)
+{
+       unsigned int i;
+
+       for (i = 0; i < sg_dma_len; ++i) {
+               ib_dma_sync_single_for_device(dev,
+                               ib_sg_dma_address(dev, &sg[i]),
+                               ib_sg_dma_len(dev, &sg[i]),
+                               direction);
+       }
+}
+#define ib_dma_sync_sg_for_device      rds_ib_dma_sync_sg_for_device
+
+
+/* ib.c */
+extern struct rds_transport rds_ib_transport;
+extern void rds_ib_add_one(struct ib_device *device);
+extern void rds_ib_remove_one(struct ib_device *device);
+extern struct ib_client rds_ib_client;
+
+extern unsigned int fmr_pool_size;
+extern unsigned int fmr_message_size;
+
+extern spinlock_t ib_nodev_conns_lock;
+extern struct list_head ib_nodev_conns;
+
+/* ib_cm.c */
+int rds_ib_conn_alloc(struct rds_connection *conn, gfp_t gfp);
+void rds_ib_conn_free(void *arg);
+int rds_ib_conn_connect(struct rds_connection *conn);
+void rds_ib_conn_shutdown(struct rds_connection *conn);
+void rds_ib_state_change(struct sock *sk);
+int __init rds_ib_listen_init(void);
+void rds_ib_listen_stop(void);
+void __rds_ib_conn_error(struct rds_connection *conn, const char *, ...);
+int rds_ib_cm_handle_connect(struct rdma_cm_id *cm_id,
+                            struct rdma_cm_event *event);
+int rds_ib_cm_initiate_connect(struct rdma_cm_id *cm_id);
+void rds_ib_cm_connect_complete(struct rds_connection *conn,
+                               struct rdma_cm_event *event);
+
+
+#define rds_ib_conn_error(conn, fmt...) \
+       __rds_ib_conn_error(conn, KERN_WARNING "RDS/IB: " fmt)
+
+/* ib_rdma.c */
+int rds_ib_update_ipaddr(struct rds_ib_device *rds_ibdev, __be32 ipaddr);
+int rds_ib_add_conn(struct rds_ib_device *rds_ibdev, struct rds_connection *conn);
+void rds_ib_remove_nodev_conns(void);
+void rds_ib_remove_conns(struct rds_ib_device *rds_ibdev);
+struct rds_ib_mr_pool *rds_ib_create_mr_pool(struct rds_ib_device *);
+void rds_ib_get_mr_info(struct rds_ib_device *rds_ibdev, struct rds_info_rdma_connection *iinfo);
+void rds_ib_destroy_mr_pool(struct rds_ib_mr_pool *);
+void *rds_ib_get_mr(struct scatterlist *sg, unsigned long nents,
+                   struct rds_sock *rs, u32 *key_ret);
+void rds_ib_sync_mr(void *trans_private, int dir);
+void rds_ib_free_mr(void *trans_private, int invalidate);
+void rds_ib_flush_mrs(void);
+
+/* ib_recv.c */
+int __init rds_ib_recv_init(void);
+void rds_ib_recv_exit(void);
+int rds_ib_recv(struct rds_connection *conn);
+int rds_ib_recv_refill(struct rds_connection *conn, gfp_t kptr_gfp,
+                      gfp_t page_gfp, int prefill);
+void rds_ib_inc_purge(struct rds_incoming *inc);
+void rds_ib_inc_free(struct rds_incoming *inc);
+int rds_ib_inc_copy_to_user(struct rds_incoming *inc, struct iovec *iov,
+                            size_t size);
+void rds_ib_recv_cq_comp_handler(struct ib_cq *cq, void *context);
+void rds_ib_recv_init_ring(struct rds_ib_connection *ic);
+void rds_ib_recv_clear_ring(struct rds_ib_connection *ic);
+void rds_ib_recv_init_ack(struct rds_ib_connection *ic);
+void rds_ib_attempt_ack(struct rds_ib_connection *ic);
+void rds_ib_ack_send_complete(struct rds_ib_connection *ic);
+u64 rds_ib_piggyb_ack(struct rds_ib_connection *ic);
+
+/* ib_ring.c */
+void rds_ib_ring_init(struct rds_ib_work_ring *ring, u32 nr);
+void rds_ib_ring_resize(struct rds_ib_work_ring *ring, u32 nr);
+u32 rds_ib_ring_alloc(struct rds_ib_work_ring *ring, u32 val, u32 *pos);
+void rds_ib_ring_free(struct rds_ib_work_ring *ring, u32 val);
+void rds_ib_ring_unalloc(struct rds_ib_work_ring *ring, u32 val);
+int rds_ib_ring_empty(struct rds_ib_work_ring *ring);
+int rds_ib_ring_low(struct rds_ib_work_ring *ring);
+u32 rds_ib_ring_oldest(struct rds_ib_work_ring *ring);
+u32 rds_ib_ring_completed(struct rds_ib_work_ring *ring, u32 wr_id, u32 oldest);
+extern wait_queue_head_t rds_ib_ring_empty_wait;
+
+/* ib_send.c */
+void rds_ib_xmit_complete(struct rds_connection *conn);
+int rds_ib_xmit(struct rds_connection *conn, struct rds_message *rm,
+               unsigned int hdr_off, unsigned int sg, unsigned int off);
+void rds_ib_send_cq_comp_handler(struct ib_cq *cq, void *context);
+void rds_ib_send_init_ring(struct rds_ib_connection *ic);
+void rds_ib_send_clear_ring(struct rds_ib_connection *ic);
+int rds_ib_xmit_rdma(struct rds_connection *conn, struct rds_rdma_op *op);
+void rds_ib_send_add_credits(struct rds_connection *conn, unsigned int credits);
+void rds_ib_advertise_credits(struct rds_connection *conn, unsigned int posted);
+int rds_ib_send_grab_credits(struct rds_ib_connection *ic, u32 wanted,
+                            u32 *adv_credits, int need_posted);
+
+/* ib_stats.c */
+DECLARE_PER_CPU(struct rds_ib_statistics, rds_ib_stats);
+#define rds_ib_stats_inc(member) rds_stats_inc_which(rds_ib_stats, member)
+unsigned int rds_ib_stats_info_copy(struct rds_info_iterator *iter,
+                                   unsigned int avail);
+
+/* ib_sysctl.c */
+int __init rds_ib_sysctl_init(void);
+void rds_ib_sysctl_exit(void);
+extern unsigned long rds_ib_sysctl_max_send_wr;
+extern unsigned long rds_ib_sysctl_max_recv_wr;
+extern unsigned long rds_ib_sysctl_max_unsig_wrs;
+extern unsigned long rds_ib_sysctl_max_unsig_bytes;
+extern unsigned long rds_ib_sysctl_max_recv_allocation;
+extern unsigned int rds_ib_sysctl_flow_control;
+extern ctl_table rds_ib_sysctl_table[];
+
+/*
+ * Helper functions for getting/setting the header and data SGEs in
+ * RDS packets (not RDMA)
+ */
+static inline struct ib_sge *
+rds_ib_header_sge(struct rds_ib_connection *ic, struct ib_sge *sge)
+{
+       return &sge[0];
+}
+
+static inline struct ib_sge *
+rds_ib_data_sge(struct rds_ib_connection *ic, struct ib_sge *sge)
+{
+       return &sge[1];
+}
+
+static inline void rds_ib_set_64bit(u64 *ptr, u64 val)
+{
+#if BITS_PER_LONG == 64
+       *ptr = val;
+#else
+       set_64bit(ptr, val);
+#endif
+}
+
+#endif
diff --git a/net/rds/ib_cm.c b/net/rds/ib_cm.c
new file mode 100644 (file)
index 0000000..0532237
--- /dev/null
@@ -0,0 +1,726 @@
+/*
+ * Copyright (c) 2006 Oracle.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/kernel.h>
+#include <linux/in.h>
+#include <linux/vmalloc.h>
+
+#include "rds.h"
+#include "ib.h"
+
+/*
+ * Set the selected protocol version
+ */
+static void rds_ib_set_protocol(struct rds_connection *conn, unsigned int version)
+{
+       conn->c_version = version;
+}
+
+/*
+ * Set up flow control
+ */
+static void rds_ib_set_flow_control(struct rds_connection *conn, u32 credits)
+{
+       struct rds_ib_connection *ic = conn->c_transport_data;
+
+       if (rds_ib_sysctl_flow_control && credits != 0) {
+               /* We're doing flow control */
+               ic->i_flowctl = 1;
+               rds_ib_send_add_credits(conn, credits);
+       } else {
+               ic->i_flowctl = 0;
+       }
+}
+
+/*
+ * Tune RNR behavior. Without flow control, we use a rather
+ * low timeout, but not the absolute minimum - this should
+ * be tunable.
+ *
+ * We already set the RNR retry count to 7 (which is the
+ * smallest infinite number :-) above.
+ * If flow control is off, we want to change this back to 0
+ * so that we learn quickly when our credit accounting is
+ * buggy.
+ *
+ * Caller passes in a qp_attr pointer - don't waste stack spacv
+ * by allocation this twice.
+ */
+static void
+rds_ib_tune_rnr(struct rds_ib_connection *ic, struct ib_qp_attr *attr)
+{
+       int ret;
+
+       attr->min_rnr_timer = IB_RNR_TIMER_000_32;
+       ret = ib_modify_qp(ic->i_cm_id->qp, attr, IB_QP_MIN_RNR_TIMER);
+       if (ret)
+               printk(KERN_NOTICE "ib_modify_qp(IB_QP_MIN_RNR_TIMER): err=%d\n", -ret);
+}
+
+/*
+ * Connection established.
+ * We get here for both outgoing and incoming connection.
+ */
+void rds_ib_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_event *event)
+{
+       const struct rds_ib_connect_private *dp = NULL;
+       struct rds_ib_connection *ic = conn->c_transport_data;
+       struct rds_ib_device *rds_ibdev;
+       struct ib_qp_attr qp_attr;
+       int err;
+
+       if (event->param.conn.private_data_len) {
+               dp = event->param.conn.private_data;
+
+               rds_ib_set_protocol(conn,
+                               RDS_PROTOCOL(dp->dp_protocol_major,
+                                       dp->dp_protocol_minor));
+               rds_ib_set_flow_control(conn, be32_to_cpu(dp->dp_credit));
+       }
+
+       printk(KERN_NOTICE "RDS/IB: connected to %pI4 version %u.%u%s\n",
+                       &conn->c_laddr,
+                       RDS_PROTOCOL_MAJOR(conn->c_version),
+                       RDS_PROTOCOL_MINOR(conn->c_version),
+                       ic->i_flowctl ? ", flow control" : "");
+
+       /* Tune RNR behavior */
+       rds_ib_tune_rnr(ic, &qp_attr);
+
+       qp_attr.qp_state = IB_QPS_RTS;
+       err = ib_modify_qp(ic->i_cm_id->qp, &qp_attr, IB_QP_STATE);
+       if (err)
+               printk(KERN_NOTICE "ib_modify_qp(IB_QP_STATE, RTS): err=%d\n", err);
+
+       /* update ib_device with this local ipaddr & conn */
+       rds_ibdev = ib_get_client_data(ic->i_cm_id->device, &rds_ib_client);
+       err = rds_ib_update_ipaddr(rds_ibdev, conn->c_laddr);
+       if (err)
+               printk(KERN_ERR "rds_ib_update_ipaddr failed (%d)\n", err);
+       err = rds_ib_add_conn(rds_ibdev, conn);
+       if (err)
+               printk(KERN_ERR "rds_ib_add_conn failed (%d)\n", err);
+
+       /* If the peer gave us the last packet it saw, process this as if
+        * we had received a regular ACK. */
+       if (dp && dp->dp_ack_seq)
+               rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL);
+
+       rds_connect_complete(conn);
+}
+
+static void rds_ib_cm_fill_conn_param(struct rds_connection *conn,
+                       struct rdma_conn_param *conn_param,
+                       struct rds_ib_connect_private *dp,
+                       u32 protocol_version)
+{
+       memset(conn_param, 0, sizeof(struct rdma_conn_param));
+       /* XXX tune these? */
+       conn_param->responder_resources = 1;
+       conn_param->initiator_depth = 1;
+       conn_param->retry_count = 7;
+       conn_param->rnr_retry_count = 7;
+
+       if (dp) {
+               struct rds_ib_connection *ic = conn->c_transport_data;
+
+               memset(dp, 0, sizeof(*dp));
+               dp->dp_saddr = conn->c_laddr;
+               dp->dp_daddr = conn->c_faddr;
+               dp->dp_protocol_major = RDS_PROTOCOL_MAJOR(protocol_version);
+               dp->dp_protocol_minor = RDS_PROTOCOL_MINOR(protocol_version);
+               dp->dp_protocol_minor_mask = cpu_to_be16(RDS_IB_SUPPORTED_PROTOCOLS);
+               dp->dp_ack_seq = rds_ib_piggyb_ack(ic);
+
+               /* Advertise flow control */
+               if (ic->i_flowctl) {
+                       unsigned int credits;
+
+                       credits = IB_GET_POST_CREDITS(atomic_read(&ic->i_credits));
+                       dp->dp_credit = cpu_to_be32(credits);
+                       atomic_sub(IB_SET_POST_CREDITS(credits), &ic->i_credits);
+               }
+
+               conn_param->private_data = dp;
+               conn_param->private_data_len = sizeof(*dp);
+       }
+}
+
+static void rds_ib_cq_event_handler(struct ib_event *event, void *data)
+{
+       rdsdebug("event %u data %p\n", event->event, data);
+}
+
+static void rds_ib_qp_event_handler(struct ib_event *event, void *data)
+{
+       struct rds_connection *conn = data;
+       struct rds_ib_connection *ic = conn->c_transport_data;
+
+       rdsdebug("conn %p ic %p event %u\n", conn, ic, event->event);
+
+       switch (event->event) {
+       case IB_EVENT_COMM_EST:
+               rdma_notify(ic->i_cm_id, IB_EVENT_COMM_EST);
+               break;
+       default:
+               printk(KERN_WARNING "RDS/ib: unhandled QP event %u "
+                      "on connection to %pI4\n", event->event,
+                      &conn->c_faddr);
+               break;
+       }
+}
+
+/*
+ * This needs to be very careful to not leave IS_ERR pointers around for
+ * cleanup to trip over.
+ */
+static int rds_ib_setup_qp(struct rds_connection *conn)
+{
+       struct rds_ib_connection *ic = conn->c_transport_data;
+       struct ib_device *dev = ic->i_cm_id->device;
+       struct ib_qp_init_attr attr;
+       struct rds_ib_device *rds_ibdev;
+       int ret;
+
+       /* rds_ib_add_one creates a rds_ib_device object per IB device,
+        * and allocates a protection domain, memory range and FMR pool
+        * for each.  If that fails for any reason, it will not register
+        * the rds_ibdev at all.
+        */
+       rds_ibdev = ib_get_client_data(dev, &rds_ib_client);
+       if (rds_ibdev == NULL) {
+               if (printk_ratelimit())
+                       printk(KERN_NOTICE "RDS/IB: No client_data for device %s\n",
+                                       dev->name);
+               return -EOPNOTSUPP;
+       }
+
+       if (rds_ibdev->max_wrs < ic->i_send_ring.w_nr + 1)
+               rds_ib_ring_resize(&ic->i_send_ring, rds_ibdev->max_wrs - 1);
+       if (rds_ibdev->max_wrs < ic->i_recv_ring.w_nr + 1)
+               rds_ib_ring_resize(&ic->i_recv_ring, rds_ibdev->max_wrs - 1);
+
+       /* Protection domain and memory range */
+       ic->i_pd = rds_ibdev->pd;
+       ic->i_mr = rds_ibdev->mr;
+
+       ic->i_send_cq = ib_create_cq(dev, rds_ib_send_cq_comp_handler,
+                                    rds_ib_cq_event_handler, conn,
+                                    ic->i_send_ring.w_nr + 1, 0);
+       if (IS_ERR(ic->i_send_cq)) {
+               ret = PTR_ERR(ic->i_send_cq);
+               ic->i_send_cq = NULL;
+               rdsdebug("ib_create_cq send failed: %d\n", ret);
+               goto out;
+       }
+
+       ic->i_recv_cq = ib_create_cq(dev, rds_ib_recv_cq_comp_handler,
+                                    rds_ib_cq_event_handler, conn,
+                                    ic->i_recv_ring.w_nr, 0);
+       if (IS_ERR(ic->i_recv_cq)) {
+               ret = PTR_ERR(ic->i_recv_cq);
+               ic->i_recv_cq = NULL;
+               rdsdebug("ib_create_cq recv failed: %d\n", ret);
+               goto out;
+       }
+
+       ret = ib_req_notify_cq(ic->i_send_cq, IB_CQ_NEXT_COMP);
+       if (ret) {
+               rdsdebug("ib_req_notify_cq send failed: %d\n", ret);
+               goto out;
+       }
+
+       ret = ib_req_notify_cq(ic->i_recv_cq, IB_CQ_SOLICITED);
+       if (ret) {
+               rdsdebug("ib_req_notify_cq recv failed: %d\n", ret);
+               goto out;
+       }
+
+       /* XXX negotiate max send/recv with remote? */
+       memset(&attr, 0, sizeof(attr));
+       attr.event_handler = rds_ib_qp_event_handler;
+       attr.qp_context = conn;
+       /* + 1 to allow for the single ack message */
+       attr.cap.max_send_wr = ic->i_send_ring.w_nr + 1;
+       attr.cap.max_recv_wr = ic->i_recv_ring.w_nr + 1;
+       attr.cap.max_send_sge = rds_ibdev->max_sge;
+       attr.cap.max_recv_sge = RDS_IB_RECV_SGE;
+       attr.sq_sig_type = IB_SIGNAL_REQ_WR;
+       attr.qp_type = IB_QPT_RC;
+       attr.send_cq = ic->i_send_cq;
+       attr.recv_cq = ic->i_recv_cq;
+
+       /*
+        * XXX this can fail if max_*_wr is too large?  Are we supposed
+        * to back off until we get a value that the hardware can support?
+        */
+       ret = rdma_create_qp(ic->i_cm_id, ic->i_pd, &attr);
+       if (ret) {
+               rdsdebug("rdma_create_qp failed: %d\n", ret);
+               goto out;
+       }
+
+       ic->i_send_hdrs = ib_dma_alloc_coherent(dev,
+                                          ic->i_send_ring.w_nr *
+                                               sizeof(struct rds_header),
+                                          &ic->i_send_hdrs_dma, GFP_KERNEL);
+       if (ic->i_send_hdrs == NULL) {
+               ret = -ENOMEM;
+               rdsdebug("ib_dma_alloc_coherent send failed\n");
+               goto out;
+       }
+
+       ic->i_recv_hdrs = ib_dma_alloc_coherent(dev,
+                                          ic->i_recv_ring.w_nr *
+                                               sizeof(struct rds_header),
+                                          &ic->i_recv_hdrs_dma, GFP_KERNEL);
+       if (ic->i_recv_hdrs == NULL) {
+               ret = -ENOMEM;
+               rdsdebug("ib_dma_alloc_coherent recv failed\n");
+               goto out;
+       }
+
+       ic->i_ack = ib_dma_alloc_coherent(dev, sizeof(struct rds_header),
+                                      &ic->i_ack_dma, GFP_KERNEL);
+       if (ic->i_ack == NULL) {
+               ret = -ENOMEM;
+               rdsdebug("ib_dma_alloc_coherent ack failed\n");
+               goto out;
+       }
+
+       ic->i_sends = vmalloc(ic->i_send_ring.w_nr * sizeof(struct rds_ib_send_work));
+       if (ic->i_sends == NULL) {
+               ret = -ENOMEM;
+               rdsdebug("send allocation failed\n");
+               goto out;
+       }
+       rds_ib_send_init_ring(ic);
+
+       ic->i_recvs = vmalloc(ic->i_recv_ring.w_nr * sizeof(struct rds_ib_recv_work));
+       if (ic->i_recvs == NULL) {
+               ret = -ENOMEM;
+               rdsdebug("recv allocation failed\n");
+               goto out;
+       }
+
+       rds_ib_recv_init_ring(ic);
+       rds_ib_recv_init_ack(ic);
+
+       /* Post receive buffers - as a side effect, this will update
+        * the posted credit count. */
+       rds_ib_recv_refill(conn, GFP_KERNEL, GFP_HIGHUSER, 1);
+
+       rdsdebug("conn %p pd %p mr %p cq %p %p\n", conn, ic->i_pd, ic->i_mr,
+                ic->i_send_cq, ic->i_recv_cq);
+
+out:
+       return ret;
+}
+
+static u32 rds_ib_protocol_compatible(const struct rds_ib_connect_private *dp)
+{
+       u16 common;
+       u32 version = 0;
+
+       /* rdma_cm private data is odd - when there is any private data in the
+        * request, we will be given a pretty large buffer without telling us the
+        * original size. The only way to tell the difference is by looking at
+        * the contents, which are initialized to zero.
+        * If the protocol version fields aren't set, this is a connection attempt
+        * from an older version. This could could be 3.0 or 2.0 - we can't tell.
+        * We really should have changed this for OFED 1.3 :-( */
+       if (dp->dp_protocol_major == 0)
+               return RDS_PROTOCOL_3_0;
+
+       common = be16_to_cpu(dp->dp_protocol_minor_mask) & RDS_IB_SUPPORTED_PROTOCOLS;
+       if (dp->dp_protocol_major == 3 && common) {
+               version = RDS_PROTOCOL_3_0;
+               while ((common >>= 1) != 0)
+                       version++;
+       } else if (printk_ratelimit()) {
+               printk(KERN_NOTICE "RDS: Connection from %pI4 using "
+                       "incompatible protocol version %u.%u\n",
+                       &dp->dp_saddr,
+                       dp->dp_protocol_major,
+                       dp->dp_protocol_minor);
+       }
+       return version;
+}
+
+int rds_ib_cm_handle_connect(struct rdma_cm_id *cm_id,
+                                   struct rdma_cm_event *event)
+{
+       __be64 lguid = cm_id->route.path_rec->sgid.global.interface_id;
+       __be64 fguid = cm_id->route.path_rec->dgid.global.interface_id;
+       const struct rds_ib_connect_private *dp = event->param.conn.private_data;
+       struct rds_ib_connect_private dp_rep;
+       struct rds_connection *conn = NULL;
+       struct rds_ib_connection *ic = NULL;
+       struct rdma_conn_param conn_param;
+       u32 version;
+       int err, destroy = 1;
+
+       /* Check whether the remote protocol version matches ours. */
+       version = rds_ib_protocol_compatible(dp);
+       if (!version)
+               goto out;
+
+       rdsdebug("saddr %pI4 daddr %pI4 RDSv%u.%u lguid 0x%llx fguid "
+                "0x%llx\n", &dp->dp_saddr, &dp->dp_daddr,
+                RDS_PROTOCOL_MAJOR(version), RDS_PROTOCOL_MINOR(version),
+                (unsigned long long)be64_to_cpu(lguid),
+                (unsigned long long)be64_to_cpu(fguid));
+
+       conn = rds_conn_create(dp->dp_daddr, dp->dp_saddr, &rds_ib_transport,
+                              GFP_KERNEL);
+       if (IS_ERR(conn)) {
+               rdsdebug("rds_conn_create failed (%ld)\n", PTR_ERR(conn));
+               conn = NULL;
+               goto out;
+       }
+
+       /*
+        * The connection request may occur while the
+        * previous connection exist, e.g. in case of failover.
+        * But as connections may be initiated simultaneously
+        * by both hosts, we have a random backoff mechanism -
+        * see the comment above rds_queue_reconnect()
+        */
+       mutex_lock(&conn->c_cm_lock);
+       if (!rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING)) {
+               if (rds_conn_state(conn) == RDS_CONN_UP) {
+                       rdsdebug("incoming connect while connecting\n");
+                       rds_conn_drop(conn);
+                       rds_ib_stats_inc(s_ib_listen_closed_stale);
+               } else
+               if (rds_conn_state(conn) == RDS_CONN_CONNECTING) {
+                       /* Wait and see - our connect may still be succeeding */
+                       rds_ib_stats_inc(s_ib_connect_raced);
+               }
+               mutex_unlock(&conn->c_cm_lock);
+               goto out;
+       }
+
+       ic = conn->c_transport_data;
+
+       rds_ib_set_protocol(conn, version);
+       rds_ib_set_flow_control(conn, be32_to_cpu(dp->dp_credit));
+
+       /* If the peer gave us the last packet it saw, process this as if
+        * we had received a regular ACK. */
+       if (dp->dp_ack_seq)
+               rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL);
+
+       BUG_ON(cm_id->context);
+       BUG_ON(ic->i_cm_id);
+
+       ic->i_cm_id = cm_id;
+       cm_id->context = conn;
+
+       /* We got halfway through setting up the ib_connection, if we
+        * fail now, we have to take the long route out of this mess. */
+       destroy = 0;
+
+       err = rds_ib_setup_qp(conn);
+       if (err) {
+               rds_ib_conn_error(conn, "rds_ib_setup_qp failed (%d)\n", err);
+               goto out;
+       }
+
+       rds_ib_cm_fill_conn_param(conn, &conn_param, &dp_rep, version);
+
+       /* rdma_accept() calls rdma_reject() internally if it fails */
+       err = rdma_accept(cm_id, &conn_param);
+       mutex_unlock(&conn->c_cm_lock);
+       if (err) {
+               rds_ib_conn_error(conn, "rdma_accept failed (%d)\n", err);
+               goto out;
+       }
+
+       return 0;
+
+out:
+       rdma_reject(cm_id, NULL, 0);
+       return destroy;
+}
+
+
+int rds_ib_cm_initiate_connect(struct rdma_cm_id *cm_id)
+{
+       struct rds_connection *conn = cm_id->context;
+       struct rds_ib_connection *ic = conn->c_transport_data;
+       struct rdma_conn_param conn_param;
+       struct rds_ib_connect_private dp;
+       int ret;
+
+       /* If the peer doesn't do protocol negotiation, we must
+        * default to RDSv3.0 */
+       rds_ib_set_protocol(conn, RDS_PROTOCOL_3_0);
+       ic->i_flowctl = rds_ib_sysctl_flow_control;     /* advertise flow control */
+
+       ret = rds_ib_setup_qp(conn);
+       if (ret) {
+               rds_ib_conn_error(conn, "rds_ib_setup_qp failed (%d)\n", ret);
+               goto out;
+       }
+
+       rds_ib_cm_fill_conn_param(conn, &conn_param, &dp, RDS_PROTOCOL_VERSION);
+
+       ret = rdma_connect(cm_id, &conn_param);
+       if (ret)
+               rds_ib_conn_error(conn, "rdma_connect failed (%d)\n", ret);
+
+out:
+       /* Beware - returning non-zero tells the rdma_cm to destroy
+        * the cm_id. We should certainly not do it as long as we still
+        * "own" the cm_id. */
+       if (ret) {
+               if (ic->i_cm_id == cm_id)
+                       ret = 0;
+       }
+       return ret;
+}
+
+int rds_ib_conn_connect(struct rds_connection *conn)
+{
+       struct rds_ib_connection *ic = conn->c_transport_data;
+       struct sockaddr_in src, dest;
+       int ret;
+
+       /* XXX I wonder what affect the port space has */
+       /* delegate cm event handler to rdma_transport */
+       ic->i_cm_id = rdma_create_id(rds_rdma_cm_event_handler, conn,
+                                    RDMA_PS_TCP);
+       if (IS_ERR(ic->i_cm_id)) {
+               ret = PTR_ERR(ic->i_cm_id);
+               ic->i_cm_id = NULL;
+               rdsdebug("rdma_create_id() failed: %d\n", ret);
+               goto out;
+       }
+
+       rdsdebug("created cm id %p for conn %p\n", ic->i_cm_id, conn);
+
+       src.sin_family = AF_INET;
+       src.sin_addr.s_addr = (__force u32)conn->c_laddr;
+       src.sin_port = (__force u16)htons(0);
+
+       dest.sin_family = AF_INET;
+       dest.sin_addr.s_addr = (__force u32)conn->c_faddr;
+       dest.sin_port = (__force u16)htons(RDS_PORT);
+
+       ret = rdma_resolve_addr(ic->i_cm_id, (struct sockaddr *)&src,
+                               (struct sockaddr *)&dest,
+                               RDS_RDMA_RESOLVE_TIMEOUT_MS);
+       if (ret) {
+               rdsdebug("addr resolve failed for cm id %p: %d\n", ic->i_cm_id,
+                        ret);
+               rdma_destroy_id(ic->i_cm_id);
+               ic->i_cm_id = NULL;
+       }
+
+out:
+       return ret;
+}
+
+/*
+ * This is so careful about only cleaning up resources that were built up
+ * so that it can be called at any point during startup.  In fact it
+ * can be called multiple times for a given connection.
+ */
+void rds_ib_conn_shutdown(struct rds_connection *conn)
+{
+       struct rds_ib_connection *ic = conn->c_transport_data;
+       int err = 0;
+
+       rdsdebug("cm %p pd %p cq %p %p qp %p\n", ic->i_cm_id,
+                ic->i_pd, ic->i_send_cq, ic->i_recv_cq,
+                ic->i_cm_id ? ic->i_cm_id->qp : NULL);
+
+       if (ic->i_cm_id) {
+               struct ib_device *dev = ic->i_cm_id->device;
+
+               rdsdebug("disconnecting cm %p\n", ic->i_cm_id);
+               err = rdma_disconnect(ic->i_cm_id);
+               if (err) {
+                       /* Actually this may happen quite frequently, when
+                        * an outgoing connect raced with an incoming connect.
+                        */
+                       rdsdebug("failed to disconnect, cm: %p err %d\n",
+                               ic->i_cm_id, err);
+               }
+
+               wait_event(rds_ib_ring_empty_wait,
+                       rds_ib_ring_empty(&ic->i_send_ring) &&
+                       rds_ib_ring_empty(&ic->i_recv_ring));
+
+               if (ic->i_send_hdrs)
+                       ib_dma_free_coherent(dev,
+                                          ic->i_send_ring.w_nr *
+                                               sizeof(struct rds_header),
+                                          ic->i_send_hdrs,
+                                          ic->i_send_hdrs_dma);
+
+               if (ic->i_recv_hdrs)
+                       ib_dma_free_coherent(dev,
+                                          ic->i_recv_ring.w_nr *
+                                               sizeof(struct rds_header),
+                                          ic->i_recv_hdrs,
+                                          ic->i_recv_hdrs_dma);
+
+               if (ic->i_ack)
+                       ib_dma_free_coherent(dev, sizeof(struct rds_header),
+                                            ic->i_ack, ic->i_ack_dma);
+
+               if (ic->i_sends)
+                       rds_ib_send_clear_ring(ic);
+               if (ic->i_recvs)
+                       rds_ib_recv_clear_ring(ic);
+
+               if (ic->i_cm_id->qp)
+                       rdma_destroy_qp(ic->i_cm_id);
+               if (ic->i_send_cq)
+                       ib_destroy_cq(ic->i_send_cq);
+               if (ic->i_recv_cq)
+                       ib_destroy_cq(ic->i_recv_cq);
+               rdma_destroy_id(ic->i_cm_id);
+
+               /*
+                * Move connection back to the nodev list.
+                */
+               if (ic->rds_ibdev) {
+
+                       spin_lock_irq(&ic->rds_ibdev->spinlock);
+                       BUG_ON(list_empty(&ic->ib_node));
+                       list_del(&ic->ib_node);
+                       spin_unlock_irq(&ic->rds_ibdev->spinlock);
+
+                       spin_lock_irq(&ib_nodev_conns_lock);
+                       list_add_tail(&ic->ib_node, &ib_nodev_conns);
+                       spin_unlock_irq(&ib_nodev_conns_lock);
+                       ic->rds_ibdev = NULL;
+               }
+
+               ic->i_cm_id = NULL;
+               ic->i_pd = NULL;
+               ic->i_mr = NULL;
+               ic->i_send_cq = NULL;
+               ic->i_recv_cq = NULL;
+               ic->i_send_hdrs = NULL;
+               ic->i_recv_hdrs = NULL;
+               ic->i_ack = NULL;
+       }
+       BUG_ON(ic->rds_ibdev);
+
+       /* Clear pending transmit */
+       if (ic->i_rm) {
+               rds_message_put(ic->i_rm);
+               ic->i_rm = NULL;
+       }
+
+       /* Clear the ACK state */
+       clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags);
+       rds_ib_set_64bit(&ic->i_ack_next, 0);
+       ic->i_ack_recv = 0;
+
+       /* Clear flow control state */
+       ic->i_flowctl = 0;
+       atomic_set(&ic->i_credits, 0);
+
+       rds_ib_ring_init(&ic->i_send_ring, rds_ib_sysctl_max_send_wr);
+       rds_ib_ring_init(&ic->i_recv_ring, rds_ib_sysctl_max_recv_wr);
+
+       if (ic->i_ibinc) {
+               rds_inc_put(&ic->i_ibinc->ii_inc);
+               ic->i_ibinc = NULL;
+       }
+
+       vfree(ic->i_sends);
+       ic->i_sends = NULL;
+       vfree(ic->i_recvs);
+       ic->i_recvs = NULL;
+}
+
+int rds_ib_conn_alloc(struct rds_connection *conn, gfp_t gfp)
+{
+       struct rds_ib_connection *ic;
+       unsigned long flags;
+
+       /* XXX too lazy? */
+       ic = kzalloc(sizeof(struct rds_ib_connection), GFP_KERNEL);
+       if (ic == NULL)
+               return -ENOMEM;
+
+       INIT_LIST_HEAD(&ic->ib_node);
+       mutex_init(&ic->i_recv_mutex);
+
+       /*
+        * rds_ib_conn_shutdown() waits for these to be emptied so they
+        * must be initialized before it can be called.
+        */
+       rds_ib_ring_init(&ic->i_send_ring, rds_ib_sysctl_max_send_wr);
+       rds_ib_ring_init(&ic->i_recv_ring, rds_ib_sysctl_max_recv_wr);
+
+       ic->conn = conn;
+       conn->c_transport_data = ic;
+
+       spin_lock_irqsave(&ib_nodev_conns_lock, flags);
+       list_add_tail(&ic->ib_node, &ib_nodev_conns);
+       spin_unlock_irqrestore(&ib_nodev_conns_lock, flags);
+
+
+       rdsdebug("conn %p conn ic %p\n", conn, conn->c_transport_data);
+       return 0;
+}
+
+void rds_ib_conn_free(void *arg)
+{
+       struct rds_ib_connection *ic = arg;
+       rdsdebug("ic %p\n", ic);
+       list_del(&ic->ib_node);
+       kfree(ic);
+}
+
+
+/*
+ * An error occurred on the connection
+ */
+void
+__rds_ib_conn_error(struct rds_connection *conn, const char *fmt, ...)
+{
+       va_list ap;
+
+       rds_conn_drop(conn);
+
+       va_start(ap, fmt);
+       vprintk(fmt, ap);
+       va_end(ap);
+}