net/rds/ib.c

   1 /*
   2  * Copyright (c) 2006 Oracle.  All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  *
  32  */
  33 #include <linux/kernel.h>
  34 #include <linux/in.h>
  35 #include <linux/if.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/inetdevice.h>
  38 #include <linux/if_arp.h>
  39 #include <linux/delay.h>
  40
  41 #include "rds.h"
  42 #include "ib.h"
  43
  44 unsigned int fmr_pool_size = RDS_FMR_POOL_SIZE;
  45 unsigned int fmr_message_size = RDS_FMR_SIZE + 1; /* +1 allows for unaligned MRs */
  46
  47 module_param(fmr_pool_size, int, 0444);
  48 MODULE_PARM_DESC(fmr_pool_size, " Max number of fmr per HCA");
  49 module_param(fmr_message_size, int, 0444);
  50 MODULE_PARM_DESC(fmr_message_size, " Max size of a RDMA transfer");
  51
  52 struct list_head rds_ib_devices;
  53
  54 DEFINE_SPINLOCK(ib_nodev_conns_lock);
  55 LIST_HEAD(ib_nodev_conns);
  56
  57 void rds_ib_add_one(struct ib_device *device)
  58 {
  59         struct rds_ib_device *rds_ibdev;
  60         struct ib_device_attr *dev_attr;
  61
  62         /* Only handle IB (no iWARP) devices */
  63         if (device->node_type != RDMA_NODE_IB_CA)
  64                 return;
  65
  66         dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL);
  67         if (!dev_attr)
  68                 return;
  69
  70         if (ib_query_device(device, dev_attr)) {
  71                 rdsdebug("Query device failed for %s\n", device->name);
  72                 goto free_attr;
  73         }
  74
  75         rds_ibdev = kmalloc(sizeof *rds_ibdev, GFP_KERNEL);
  76         if (!rds_ibdev)
  77                 goto free_attr;
  78
  79         spin_lock_init(&rds_ibdev->spinlock);
  80
  81         rds_ibdev->max_wrs = dev_attr->max_qp_wr;
  82         rds_ibdev->max_sge = min(dev_attr->max_sge, RDS_IB_MAX_SGE);
  83
  84         rds_ibdev->fmr_page_shift = max(9, ffs(dev_attr->page_size_cap) - 1);
  85         rds_ibdev->fmr_page_size  = 1 << rds_ibdev->fmr_page_shift;
  86         rds_ibdev->fmr_page_mask  = ~((u64) rds_ibdev->fmr_page_size - 1);
  87         rds_ibdev->fmr_max_remaps = dev_attr->max_map_per_fmr?: 32;
  88         rds_ibdev->max_fmrs = dev_attr->max_fmr ?
  89                         min_t(unsigned int, dev_attr->max_fmr, fmr_pool_size) :
  90                         fmr_pool_size;
  91
  92         rds_ibdev->dev = device;
  93         rds_ibdev->pd = ib_alloc_pd(device);
  94         if (IS_ERR(rds_ibdev->pd))
  95                 goto free_dev;
  96
  97         rds_ibdev->mr = ib_get_dma_mr(rds_ibdev->pd,
  98                                       IB_ACCESS_LOCAL_WRITE);
  99         if (IS_ERR(rds_ibdev->mr))
 100                 goto err_pd;
 101
 102         rds_ibdev->mr_pool = rds_ib_create_mr_pool(rds_ibdev);
 103         if (IS_ERR(rds_ibdev->mr_pool)) {
 104                 rds_ibdev->mr_pool = NULL;
 105                 goto err_mr;
 106         }
 107
 108         INIT_LIST_HEAD(&rds_ibdev->ipaddr_list);
 109         INIT_LIST_HEAD(&rds_ibdev->conn_list);
 110         list_add_tail(&rds_ibdev->list, &rds_ib_devices);
 111
 112         ib_set_client_data(device, &rds_ib_client, rds_ibdev);
 113
 114         goto free_attr;
 115
 116 err_mr:
 117         ib_dereg_mr(rds_ibdev->mr);
 118 err_pd:
 119         ib_dealloc_pd(rds_ibdev->pd);
 120 free_dev:
 121         kfree(rds_ibdev);
 122 free_attr:
 123         kfree(dev_attr);
 124 }
 125
 126 void rds_ib_remove_one(struct ib_device *device)
 127 {
 128         struct rds_ib_device *rds_ibdev;
 129         struct rds_ib_ipaddr *i_ipaddr, *i_next;
 130
 131         rds_ibdev = ib_get_client_data(device, &rds_ib_client);
 132         if (!rds_ibdev)
 133                 return;
 134
 135         list_for_each_entry_safe(i_ipaddr, i_next, &rds_ibdev->ipaddr_list, list) {
 136                 list_del(&i_ipaddr->list);
 137                 kfree(i_ipaddr);
 138         }
 139
 140         rds_ib_remove_conns(rds_ibdev);
 141
 142         if (rds_ibdev->mr_pool)
 143                 rds_ib_destroy_mr_pool(rds_ibdev->mr_pool);
 144
 145         ib_dereg_mr(rds_ibdev->mr);
 146
 147         while (ib_dealloc_pd(rds_ibdev->pd)) {
 148                 rdsdebug("Failed to dealloc pd %p\n", rds_ibdev->pd);
 149                 msleep(1);
 150         }
 151
 152         list_del(&rds_ibdev->list);
 153         kfree(rds_ibdev);
 154 }
 155
 156 struct ib_client rds_ib_client = {
 157         .name   = "rds_ib",
 158         .add    = rds_ib_add_one,
 159         .remove = rds_ib_remove_one
 160 };
 161
 162 static int rds_ib_conn_info_visitor(struct rds_connection *conn,
 163                                     void *buffer)
 164 {
 165         struct rds_info_rdma_connection *iinfo = buffer;
 166         struct rds_ib_connection *ic;
 167
 168         /* We will only ever look at IB transports */
 169         if (conn->c_trans != &rds_ib_transport)
 170                 return 0;
 171
 172         iinfo->src_addr = conn->c_laddr;
 173         iinfo->dst_addr = conn->c_faddr;
 174
 175         memset(&iinfo->src_gid, 0, sizeof(iinfo->src_gid));
 176         memset(&iinfo->dst_gid, 0, sizeof(iinfo->dst_gid));
 177         if (rds_conn_state(conn) == RDS_CONN_UP) {
 178                 struct rds_ib_device *rds_ibdev;
 179                 struct rdma_dev_addr *dev_addr;
 180
 181                 ic = conn->c_transport_data;
 182                 dev_addr = &ic->i_cm_id->route.addr.dev_addr;
 183
 184                 ib_addr_get_sgid(dev_addr, (union ib_gid *) &iinfo->src_gid);
 185                 ib_addr_get_dgid(dev_addr, (union ib_gid *) &iinfo->dst_gid);
 186
 187                 rds_ibdev = ib_get_client_data(ic->i_cm_id->device, &rds_ib_client);
 188                 iinfo->max_send_wr = ic->i_send_ring.w_nr;
 189                 iinfo->max_recv_wr = ic->i_recv_ring.w_nr;
 190                 iinfo->max_send_sge = rds_ibdev->max_sge;
 191                 rds_ib_get_mr_info(rds_ibdev, iinfo);
 192         }
 193         return 1;
 194 }
 195
 196 static void rds_ib_ic_info(struct socket *sock, unsigned int len,
 197                            struct rds_info_iterator *iter,
 198                            struct rds_info_lengths *lens)
 199 {
 200         rds_for_each_conn_info(sock, len, iter, lens,
 201                                 rds_ib_conn_info_visitor,
 202                                 sizeof(struct rds_info_rdma_connection));
 203 }
 204
 205
 206 /*
 207  * Early RDS/IB was built to only bind to an address if there is an IPoIB
 208  * device with that address set.
 209  *
 210  * If it were me, I'd advocate for something more flexible.  Sending and
 211  * receiving should be device-agnostic.  Transports would try and maintain
 212  * connections between peers who have messages queued.  Userspace would be
 213  * allowed to influence which paths have priority.  We could call userspace
 214  * asserting this policy "routing".
 215  */
 216 static int rds_ib_laddr_check(__be32 addr)
 217 {
 218         int ret;
 219         struct rdma_cm_id *cm_id;
 220         struct sockaddr_in sin;
 221
 222         /* Create a CMA ID and try to bind it. This catches both
 223          * IB and iWARP capable NICs.
 224          */
 225         cm_id = rdma_create_id(NULL, NULL, RDMA_PS_TCP);
 226         if (!cm_id)
 227                 return -EADDRNOTAVAIL;
 228
 229         memset(&sin, 0, sizeof(sin));
 230         sin.sin_family = AF_INET;
 231         sin.sin_addr.s_addr = addr;
 232
 233         /* rdma_bind_addr will only succeed for IB & iWARP devices */
 234         ret = rdma_bind_addr(cm_id, (struct sockaddr *)&sin);
 235         /* due to this, we will claim to support iWARP devices unless we
 236            check node_type. */
 237         if (ret || cm_id->device->node_type != RDMA_NODE_IB_CA)
 238                 ret = -EADDRNOTAVAIL;
 239
 240         rdsdebug("addr %pI4 ret %d node type %d\n",
 241                 &addr, ret,
 242                 cm_id->device ? cm_id->device->node_type : -1);
 243
 244         rdma_destroy_id(cm_id);
 245
 246         return ret;
 247 }
 248
 249 void rds_ib_exit(void)
 250 {
 251         rds_info_deregister_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info);
 252         rds_ib_remove_nodev_conns();
 253         ib_unregister_client(&rds_ib_client);
 254         rds_ib_sysctl_exit();
 255         rds_ib_recv_exit();
 256         rds_trans_unregister(&rds_ib_transport);
 257 }
 258
 259 struct rds_transport rds_ib_transport = {
 260         .laddr_check            = rds_ib_laddr_check,
 261         .xmit_complete          = rds_ib_xmit_complete,
 262         .xmit                   = rds_ib_xmit,
 263         .xmit_cong_map          = NULL,
 264         .xmit_rdma              = rds_ib_xmit_rdma,
 265         .recv                   = rds_ib_recv,
 266         .conn_alloc             = rds_ib_conn_alloc,
 267         .conn_free              = rds_ib_conn_free,
 268         .conn_connect           = rds_ib_conn_connect,
 269         .conn_shutdown          = rds_ib_conn_shutdown,
 270         .inc_copy_to_user       = rds_ib_inc_copy_to_user,
 271         .inc_purge              = rds_ib_inc_purge,
 272         .inc_free               = rds_ib_inc_free,
 273         .cm_initiate_connect    = rds_ib_cm_initiate_connect,
 274         .cm_handle_connect      = rds_ib_cm_handle_connect,
 275         .cm_connect_complete    = rds_ib_cm_connect_complete,
 276         .stats_info_copy        = rds_ib_stats_info_copy,
 277         .exit                   = rds_ib_exit,
 278         .get_mr                 = rds_ib_get_mr,
 279         .sync_mr                = rds_ib_sync_mr,
 280         .free_mr                = rds_ib_free_mr,
 281         .flush_mrs              = rds_ib_flush_mrs,
 282         .t_owner                = THIS_MODULE,
 283         .t_name                 = "infiniband",
 284 };
 285
 286 int __init rds_ib_init(void)
 287 {
 288         int ret;
 289
 290         INIT_LIST_HEAD(&rds_ib_devices);
 291
 292         ret = ib_register_client(&rds_ib_client);
 293         if (ret)
 294                 goto out;
 295
 296         ret = rds_ib_sysctl_init();
 297         if (ret)
 298                 goto out_ibreg;
 299
 300         ret = rds_ib_recv_init();
 301         if (ret)
 302                 goto out_sysctl;
 303
 304         ret = rds_trans_register(&rds_ib_transport);
 305         if (ret)
 306                 goto out_recv;
 307
 308         rds_info_register_func(RDS_INFO_IB_CONNECTIONS, rds_ib_ic_info);
 309
 310         goto out;
 311
 312 out_recv:
 313         rds_ib_recv_exit();
 314 out_sysctl:
 315         rds_ib_sysctl_exit();
 316 out_ibreg:
 317         ib_unregister_client(&rds_ib_client);
 318 out:
 319         return ret;
 320 }
 321
 322 MODULE_LICENSE("GPL");
 323