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 "iw.h"
43
44 unsigned int fastreg_pool_size = RDS_FASTREG_POOL_SIZE;
45 unsigned int fastreg_message_size = RDS_FASTREG_SIZE + 1; /* +1 allows for unaligned MRs */
46
47 module_param(fastreg_pool_size, int, 0444);
48 MODULE_PARM_DESC(fastreg_pool_size, " Max number of fastreg MRs per device");
49 module_param(fastreg_message_size, int, 0444);
50 MODULE_PARM_DESC(fastreg_message_size, " Max size of a RDMA transfer (fastreg MRs)");
51
52 struct list_head rds_iw_devices;
53
54 /* NOTE: if also grabbing iwdev lock, grab this first */
55 DEFINE_SPINLOCK(iw_nodev_conns_lock);
56 LIST_HEAD(iw_nodev_conns);
57
58 void rds_iw_add_one(struct ib_device *device)
59 {
60 struct rds_iw_device *rds_iwdev;
61 struct ib_device_attr *dev_attr;
62
63 /* Only handle iwarp devices */
64 if (device->node_type != RDMA_NODE_RNIC)
65 return;
66
67 dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL);
68 if (!dev_attr)
69 return;
70
71 if (ib_query_device(device, dev_attr)) {
72 rdsdebug("Query device failed for %s\n", device->name);
73 goto free_attr;
74 }
75
76 rds_iwdev = kmalloc(sizeof *rds_iwdev, GFP_KERNEL);
77 if (!rds_iwdev)
78 goto free_attr;
79
80 spin_lock_init(&rds_iwdev->spinlock);
81
82 rds_iwdev->dma_local_lkey = !!(dev_attr->device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY);
83 rds_iwdev->max_wrs = dev_attr->max_qp_wr;
84 rds_iwdev->max_sge = min(dev_attr->max_sge, RDS_IW_MAX_SGE);
85
86 rds_iwdev->page_shift = max(PAGE_SHIFT, ffs(dev_attr->page_size_cap) - 1);
87
88 rds_iwdev->dev = device;
89 rds_iwdev->pd = ib_alloc_pd(device);
90 if (IS_ERR(rds_iwdev->pd))
91 goto free_dev;
92
93 if (!rds_iwdev->dma_local_lkey) {
94 if (device->node_type != RDMA_NODE_RNIC) {
95 rds_iwdev->mr = ib_get_dma_mr(rds_iwdev->pd,
96 IB_ACCESS_LOCAL_WRITE);
97 } else {
98 rds_iwdev->mr = ib_get_dma_mr(rds_iwdev->pd,
99 IB_ACCESS_REMOTE_READ |
100 IB_ACCESS_REMOTE_WRITE |
101 IB_ACCESS_LOCAL_WRITE);
102 }
103 if (IS_ERR(rds_iwdev->mr))
104 goto err_pd;
105 } else
106 rds_iwdev->mr = NULL;
107
108 rds_iwdev->mr_pool = rds_iw_create_mr_pool(rds_iwdev);
109 if (IS_ERR(rds_iwdev->mr_pool)) {
110 rds_iwdev->mr_pool = NULL;
111 goto err_mr;
112 }
113
114 INIT_LIST_HEAD(&rds_iwdev->cm_id_list);
115 INIT_LIST_HEAD(&rds_iwdev->conn_list);
116 list_add_tail(&rds_iwdev->list, &rds_iw_devices);
117
118 ib_set_client_data(device, &rds_iw_client, rds_iwdev);
119
120 goto free_attr;
121
122 err_mr:
123 if (rds_iwdev->mr)
124 ib_dereg_mr(rds_iwdev->mr);
125 err_pd:
126 ib_dealloc_pd(rds_iwdev->pd);
127 free_dev:
128 kfree(rds_iwdev);
129 free_attr:
130 kfree(dev_attr);
131 }
132
133 void rds_iw_remove_one(struct ib_device *device)
134 {
135 struct rds_iw_device *rds_iwdev;
136 struct rds_iw_cm_id *i_cm_id, *next;
137
138 rds_iwdev = ib_get_client_data(device, &rds_iw_client);
139 if (!rds_iwdev)
140 return;
141
142 spin_lock_irq(&rds_iwdev->spinlock);
143 list_for_each_entry_safe(i_cm_id, next, &rds_iwdev->cm_id_list, list) {
144 list_del(&i_cm_id->list);
145 kfree(i_cm_id);
146 }
147 spin_unlock_irq(&rds_iwdev->spinlock);
148
149 rds_iw_destroy_conns(rds_iwdev);
150
151 if (rds_iwdev->mr_pool)
152 rds_iw_destroy_mr_pool(rds_iwdev->mr_pool);
153
154 if (rds_iwdev->mr)
155 ib_dereg_mr(rds_iwdev->mr);
156
157 while (ib_dealloc_pd(rds_iwdev->pd)) {
158 rdsdebug("Failed to dealloc pd %p\n", rds_iwdev->pd);
159 msleep(1);
160 }
161
162 list_del(&rds_iwdev->list);
163 kfree(rds_iwdev);
164 }
165
166 struct ib_client rds_iw_client = {
167 .name = "rds_iw",
168 .add = rds_iw_add_one,
169 .remove = rds_iw_remove_one
170 };
171
172 static int rds_iw_conn_info_visitor(struct rds_connection *conn,
173 void *buffer)
174 {
175 struct rds_info_rdma_connection *iinfo = buffer;
176 struct rds_iw_connection *ic;
177
178 /* We will only ever look at IB transports */
179 if (conn->c_trans != &rds_iw_transport)
180 return 0;
181
182 iinfo->src_addr = conn->c_laddr;
183 iinfo->dst_addr = conn->c_faddr;
184
185 memset(&iinfo->src_gid, 0, sizeof(iinfo->src_gid));
186 memset(&iinfo->dst_gid, 0, sizeof(iinfo->dst_gid));
187 if (rds_conn_state(conn) == RDS_CONN_UP) {
188 struct rds_iw_device *rds_iwdev;
189 struct rdma_dev_addr *dev_addr;
190
191 ic = conn->c_transport_data;
192 dev_addr = &ic->i_cm_id->route.addr.dev_addr;
193
194 ib_addr_get_sgid(dev_addr, (union ib_gid *) &iinfo->src_gid);
195 ib_addr_get_dgid(dev_addr, (union ib_gid *) &iinfo->dst_gid);
196
197 rds_iwdev = ib_get_client_data(ic->i_cm_id->device, &rds_iw_client);
198 iinfo->max_send_wr = ic->i_send_ring.w_nr;
199 iinfo->max_recv_wr = ic->i_recv_ring.w_nr;
200 iinfo->max_send_sge = rds_iwdev->max_sge;
201 rds_iw_get_mr_info(rds_iwdev, iinfo);
202 }
203 return 1;
204 }
205
206 static void rds_iw_ic_info(struct socket *sock, unsigned int len,
207 struct rds_info_iterator *iter,
208 struct rds_info_lengths *lens)
209 {
210 rds_for_each_conn_info(sock, len, iter, lens,
211 rds_iw_conn_info_visitor,
212 sizeof(struct rds_info_rdma_connection));
213 }
214
215
216 /*
217 * Early RDS/IB was built to only bind to an address if there is an IPoIB
218 * device with that address set.
219 *
220 * If it were me, I'd advocate for something more flexible. Sending and
221 * receiving should be device-agnostic. Transports would try and maintain
222 * connections between peers who have messages queued. Userspace would be
223 * allowed to influence which paths have priority. We could call userspace
224 * asserting this policy "routing".
225 */
226 static int rds_iw_laddr_check(__be32 addr)
227 {
228 int ret;
229 struct rdma_cm_id *cm_id;
230 struct sockaddr_in sin;
231
232 /* Create a CMA ID and try to bind it. This catches both
233 * IB and iWARP capable NICs.
234 */
235 cm_id = rdma_create_id(NULL, NULL, RDMA_PS_TCP);
236 if (IS_ERR(cm_id))
237 return PTR_ERR(cm_id);
238
239 memset(&sin, 0, sizeof(sin));
240 sin.sin_family = AF_INET;
241 sin.sin_addr.s_addr = addr;
242
243 /* rdma_bind_addr will only succeed for IB & iWARP devices */
244 ret = rdma_bind_addr(cm_id, (struct sockaddr *)&sin);
245 /* due to this, we will claim to support IB devices unless we
246 check node_type. */
247 if (ret || cm_id->device->node_type != RDMA_NODE_RNIC)
248 ret = -EADDRNOTAVAIL;
249
250 rdsdebug("addr %pI4 ret %d node type %d\n",
251 &addr, ret,
252 cm_id->device ? cm_id->device->node_type : -1);
253
254 rdma_destroy_id(cm_id);
255
256 return ret;
257 }
258
259 void rds_iw_exit(void)
260 {
261 rds_info_deregister_func(RDS_INFO_IWARP_CONNECTIONS, rds_iw_ic_info);
262 rds_iw_destroy_nodev_conns();
263 ib_unregister_client(&rds_iw_client);
264 rds_iw_sysctl_exit();
265 rds_iw_recv_exit();
266 rds_trans_unregister(&rds_iw_transport);
267 }
268
269 struct rds_transport rds_iw_transport = {
270 .laddr_check = rds_iw_laddr_check,
271 .xmit_complete = rds_iw_xmit_complete,
272 .xmit = rds_iw_xmit,
273 .xmit_cong_map = NULL,
274 .xmit_rdma = rds_iw_xmit_rdma,
275 .recv = rds_iw_recv,
276 .conn_alloc = rds_iw_conn_alloc,
277 .conn_free = rds_iw_conn_free,
278 .conn_connect = rds_iw_conn_connect,
279 .conn_shutdown = rds_iw_conn_shutdown,
280 .inc_copy_to_user = rds_iw_inc_copy_to_user,
281 .inc_purge = rds_iw_inc_purge,
282 .inc_free = rds_iw_inc_free,
283 .cm_initiate_connect = rds_iw_cm_initiate_connect,
284 .cm_handle_connect = rds_iw_cm_handle_connect,
285 .cm_connect_complete = rds_iw_cm_connect_complete,
286 .stats_info_copy = rds_iw_stats_info_copy,
287 .exit = rds_iw_exit,
288 .get_mr = rds_iw_get_mr,
289 .sync_mr = rds_iw_sync_mr,
290 .free_mr = rds_iw_free_mr,
291 .flush_mrs = rds_iw_flush_mrs,
292 .t_owner = THIS_MODULE,
293 .t_name = "iwarp",
294 .t_prefer_loopback = 1,
295 };
296
297 int __init rds_iw_init(void)
298 {
299 int ret;
300
301 INIT_LIST_HEAD(&rds_iw_devices);
302
303 ret = ib_register_client(&rds_iw_client);
304 if (ret)
305 goto out;
306
307 ret = rds_iw_sysctl_init();
308 if (ret)
309 goto out_ibreg;
310
311 ret = rds_iw_recv_init();
312 if (ret)
313 goto out_sysctl;
314
315 ret = rds_trans_register(&rds_iw_transport);
316 if (ret)
317 goto out_recv;
318
319 rds_info_register_func(RDS_INFO_IWARP_CONNECTIONS, rds_iw_ic_info);
320
321 goto out;
322
323 out_recv:
324 rds_iw_recv_exit();
325 out_sysctl:
326 rds_iw_sysctl_exit();
327 out_ibreg:
328 ib_unregister_client(&rds_iw_client);
329 out:
330 return ret;
331 }
332
333 MODULE_LICENSE("GPL");
334
335
|
This page was automatically generated by the
LXR engine.
|