1 /*
2 * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33
34 #include <rdma/rdma_netlink.h>
35 #include <net/addrconf.h>
36 #include "rxe.h"
37 #include "rxe_loc.h"
38
39 MODULE_AUTHOR("Bob Pearson, Frank Zago, John Groves, Kamal Heib");
40 MODULE_DESCRIPTION("Soft RDMA transport");
41 MODULE_LICENSE("Dual BSD/GPL");
42
43 /* free resources for all ports on a device */
rxe_cleanup_ports(struct rxe_dev * rxe)44 static void rxe_cleanup_ports(struct rxe_dev *rxe)
45 {
46 kfree(rxe->port.pkey_tbl);
47 rxe->port.pkey_tbl = NULL;
48
49 }
50
51 /* free resources for a rxe device all objects created for this device must
52 * have been destroyed
53 */
rxe_dealloc(struct ib_device * ib_dev)54 void rxe_dealloc(struct ib_device *ib_dev)
55 {
56 struct rxe_dev *rxe = container_of(ib_dev, struct rxe_dev, ib_dev);
57
58 rxe_pool_cleanup(&rxe->uc_pool);
59 rxe_pool_cleanup(&rxe->pd_pool);
60 rxe_pool_cleanup(&rxe->ah_pool);
61 rxe_pool_cleanup(&rxe->srq_pool);
62 rxe_pool_cleanup(&rxe->qp_pool);
63 rxe_pool_cleanup(&rxe->cq_pool);
64 rxe_pool_cleanup(&rxe->mr_pool);
65 rxe_pool_cleanup(&rxe->mw_pool);
66 rxe_pool_cleanup(&rxe->mc_grp_pool);
67 rxe_pool_cleanup(&rxe->mc_elem_pool);
68
69 rxe_cleanup_ports(rxe);
70
71 if (rxe->tfm)
72 crypto_free_shash(rxe->tfm);
73 }
74
75 /* initialize rxe device parameters */
rxe_init_device_param(struct rxe_dev * rxe)76 static void rxe_init_device_param(struct rxe_dev *rxe)
77 {
78 rxe->max_inline_data = RXE_MAX_INLINE_DATA;
79
80 rxe->attr.fw_ver = RXE_FW_VER;
81 rxe->attr.max_mr_size = RXE_MAX_MR_SIZE;
82 rxe->attr.page_size_cap = RXE_PAGE_SIZE_CAP;
83 rxe->attr.vendor_id = RXE_VENDOR_ID;
84 rxe->attr.vendor_part_id = RXE_VENDOR_PART_ID;
85 rxe->attr.hw_ver = RXE_HW_VER;
86 rxe->attr.max_qp = RXE_MAX_QP;
87 rxe->attr.max_qp_wr = RXE_MAX_QP_WR;
88 rxe->attr.device_cap_flags = RXE_DEVICE_CAP_FLAGS;
89 rxe->attr.max_send_sge = RXE_MAX_SGE;
90 rxe->attr.max_recv_sge = RXE_MAX_SGE;
91 rxe->attr.max_sge_rd = RXE_MAX_SGE_RD;
92 rxe->attr.max_cq = RXE_MAX_CQ;
93 rxe->attr.max_cqe = (1 << RXE_MAX_LOG_CQE) - 1;
94 rxe->attr.max_mr = RXE_MAX_MR;
95 rxe->attr.max_pd = RXE_MAX_PD;
96 rxe->attr.max_qp_rd_atom = RXE_MAX_QP_RD_ATOM;
97 rxe->attr.max_ee_rd_atom = RXE_MAX_EE_RD_ATOM;
98 rxe->attr.max_res_rd_atom = RXE_MAX_RES_RD_ATOM;
99 rxe->attr.max_qp_init_rd_atom = RXE_MAX_QP_INIT_RD_ATOM;
100 rxe->attr.max_ee_init_rd_atom = RXE_MAX_EE_INIT_RD_ATOM;
101 rxe->attr.atomic_cap = IB_ATOMIC_HCA;
102 rxe->attr.max_ee = RXE_MAX_EE;
103 rxe->attr.max_rdd = RXE_MAX_RDD;
104 rxe->attr.max_mw = RXE_MAX_MW;
105 rxe->attr.max_raw_ipv6_qp = RXE_MAX_RAW_IPV6_QP;
106 rxe->attr.max_raw_ethy_qp = RXE_MAX_RAW_ETHY_QP;
107 rxe->attr.max_mcast_grp = RXE_MAX_MCAST_GRP;
108 rxe->attr.max_mcast_qp_attach = RXE_MAX_MCAST_QP_ATTACH;
109 rxe->attr.max_total_mcast_qp_attach = RXE_MAX_TOT_MCAST_QP_ATTACH;
110 rxe->attr.max_ah = RXE_MAX_AH;
111 rxe->attr.max_fmr = RXE_MAX_FMR;
112 rxe->attr.max_map_per_fmr = RXE_MAX_MAP_PER_FMR;
113 rxe->attr.max_srq = RXE_MAX_SRQ;
114 rxe->attr.max_srq_wr = RXE_MAX_SRQ_WR;
115 rxe->attr.max_srq_sge = RXE_MAX_SRQ_SGE;
116 rxe->attr.max_fast_reg_page_list_len = RXE_MAX_FMR_PAGE_LIST_LEN;
117 rxe->attr.max_pkeys = RXE_MAX_PKEYS;
118 rxe->attr.local_ca_ack_delay = RXE_LOCAL_CA_ACK_DELAY;
119
120 rxe->max_ucontext = RXE_MAX_UCONTEXT;
121 }
122
123 /* initialize port attributes */
rxe_init_port_param(struct rxe_port * port)124 static int rxe_init_port_param(struct rxe_port *port)
125 {
126 port->attr.state = IB_PORT_DOWN;
127 port->attr.max_mtu = IB_MTU_4096;
128 port->attr.active_mtu = IB_MTU_256;
129 port->attr.gid_tbl_len = RXE_PORT_GID_TBL_LEN;
130 port->attr.port_cap_flags = RXE_PORT_PORT_CAP_FLAGS;
131 port->attr.max_msg_sz = RXE_PORT_MAX_MSG_SZ;
132 port->attr.bad_pkey_cntr = RXE_PORT_BAD_PKEY_CNTR;
133 port->attr.qkey_viol_cntr = RXE_PORT_QKEY_VIOL_CNTR;
134 port->attr.pkey_tbl_len = RXE_PORT_PKEY_TBL_LEN;
135 port->attr.lid = RXE_PORT_LID;
136 port->attr.sm_lid = RXE_PORT_SM_LID;
137 port->attr.lmc = RXE_PORT_LMC;
138 port->attr.max_vl_num = RXE_PORT_MAX_VL_NUM;
139 port->attr.sm_sl = RXE_PORT_SM_SL;
140 port->attr.subnet_timeout = RXE_PORT_SUBNET_TIMEOUT;
141 port->attr.init_type_reply = RXE_PORT_INIT_TYPE_REPLY;
142 port->attr.active_width = RXE_PORT_ACTIVE_WIDTH;
143 port->attr.active_speed = RXE_PORT_ACTIVE_SPEED;
144 port->attr.phys_state = RXE_PORT_PHYS_STATE;
145 port->mtu_cap = ib_mtu_enum_to_int(IB_MTU_256);
146 port->subnet_prefix = cpu_to_be64(RXE_PORT_SUBNET_PREFIX);
147
148 return 0;
149 }
150
151 /* initialize port state, note IB convention that HCA ports are always
152 * numbered from 1
153 */
rxe_init_ports(struct rxe_dev * rxe)154 static int rxe_init_ports(struct rxe_dev *rxe)
155 {
156 struct rxe_port *port = &rxe->port;
157
158 rxe_init_port_param(port);
159
160 if (!port->attr.pkey_tbl_len || !port->attr.gid_tbl_len)
161 return -EINVAL;
162
163 port->pkey_tbl = kcalloc(port->attr.pkey_tbl_len,
164 sizeof(*port->pkey_tbl), GFP_KERNEL);
165
166 if (!port->pkey_tbl)
167 return -ENOMEM;
168
169 port->pkey_tbl[0] = 0xffff;
170 addrconf_addr_eui48((unsigned char *)&port->port_guid,
171 rxe->ndev->dev_addr);
172
173 spin_lock_init(&port->port_lock);
174
175 return 0;
176 }
177
178 /* init pools of managed objects */
rxe_init_pools(struct rxe_dev * rxe)179 static int rxe_init_pools(struct rxe_dev *rxe)
180 {
181 int err;
182
183 err = rxe_pool_init(rxe, &rxe->uc_pool, RXE_TYPE_UC,
184 rxe->max_ucontext);
185 if (err)
186 goto err1;
187
188 err = rxe_pool_init(rxe, &rxe->pd_pool, RXE_TYPE_PD,
189 rxe->attr.max_pd);
190 if (err)
191 goto err2;
192
193 err = rxe_pool_init(rxe, &rxe->ah_pool, RXE_TYPE_AH,
194 rxe->attr.max_ah);
195 if (err)
196 goto err3;
197
198 err = rxe_pool_init(rxe, &rxe->srq_pool, RXE_TYPE_SRQ,
199 rxe->attr.max_srq);
200 if (err)
201 goto err4;
202
203 err = rxe_pool_init(rxe, &rxe->qp_pool, RXE_TYPE_QP,
204 rxe->attr.max_qp);
205 if (err)
206 goto err5;
207
208 err = rxe_pool_init(rxe, &rxe->cq_pool, RXE_TYPE_CQ,
209 rxe->attr.max_cq);
210 if (err)
211 goto err6;
212
213 err = rxe_pool_init(rxe, &rxe->mr_pool, RXE_TYPE_MR,
214 rxe->attr.max_mr);
215 if (err)
216 goto err7;
217
218 err = rxe_pool_init(rxe, &rxe->mw_pool, RXE_TYPE_MW,
219 rxe->attr.max_mw);
220 if (err)
221 goto err8;
222
223 err = rxe_pool_init(rxe, &rxe->mc_grp_pool, RXE_TYPE_MC_GRP,
224 rxe->attr.max_mcast_grp);
225 if (err)
226 goto err9;
227
228 err = rxe_pool_init(rxe, &rxe->mc_elem_pool, RXE_TYPE_MC_ELEM,
229 rxe->attr.max_total_mcast_qp_attach);
230 if (err)
231 goto err10;
232
233 return 0;
234
235 err10:
236 rxe_pool_cleanup(&rxe->mc_grp_pool);
237 err9:
238 rxe_pool_cleanup(&rxe->mw_pool);
239 err8:
240 rxe_pool_cleanup(&rxe->mr_pool);
241 err7:
242 rxe_pool_cleanup(&rxe->cq_pool);
243 err6:
244 rxe_pool_cleanup(&rxe->qp_pool);
245 err5:
246 rxe_pool_cleanup(&rxe->srq_pool);
247 err4:
248 rxe_pool_cleanup(&rxe->ah_pool);
249 err3:
250 rxe_pool_cleanup(&rxe->pd_pool);
251 err2:
252 rxe_pool_cleanup(&rxe->uc_pool);
253 err1:
254 return err;
255 }
256
257 /* initialize rxe device state */
rxe_init(struct rxe_dev * rxe)258 static int rxe_init(struct rxe_dev *rxe)
259 {
260 int err;
261
262 /* init default device parameters */
263 rxe_init_device_param(rxe);
264
265 err = rxe_init_ports(rxe);
266 if (err)
267 goto err1;
268
269 err = rxe_init_pools(rxe);
270 if (err)
271 goto err2;
272
273 /* init pending mmap list */
274 spin_lock_init(&rxe->mmap_offset_lock);
275 spin_lock_init(&rxe->pending_lock);
276 INIT_LIST_HEAD(&rxe->pending_mmaps);
277
278 mutex_init(&rxe->usdev_lock);
279
280 return 0;
281
282 err2:
283 rxe_cleanup_ports(rxe);
284 err1:
285 return err;
286 }
287
rxe_set_mtu(struct rxe_dev * rxe,unsigned int ndev_mtu)288 void rxe_set_mtu(struct rxe_dev *rxe, unsigned int ndev_mtu)
289 {
290 struct rxe_port *port = &rxe->port;
291 enum ib_mtu mtu;
292
293 mtu = eth_mtu_int_to_enum(ndev_mtu);
294
295 /* Make sure that new MTU in range */
296 mtu = mtu ? min_t(enum ib_mtu, mtu, IB_MTU_4096) : IB_MTU_256;
297
298 port->attr.active_mtu = mtu;
299 port->mtu_cap = ib_mtu_enum_to_int(mtu);
300 }
301
302 /* called by ifc layer to create new rxe device.
303 * The caller should allocate memory for rxe by calling ib_alloc_device.
304 */
rxe_add(struct rxe_dev * rxe,unsigned int mtu,const char * ibdev_name)305 int rxe_add(struct rxe_dev *rxe, unsigned int mtu, const char *ibdev_name)
306 {
307 int err;
308
309 err = rxe_init(rxe);
310 if (err)
311 return err;
312
313 rxe_set_mtu(rxe, mtu);
314
315 return rxe_register_device(rxe, ibdev_name);
316 }
317
rxe_newlink(const char * ibdev_name,struct net_device * ndev)318 static int rxe_newlink(const char *ibdev_name, struct net_device *ndev)
319 {
320 struct rxe_dev *exists;
321 int err = 0;
322
323 exists = rxe_get_dev_from_net(ndev);
324 if (exists) {
325 ib_device_put(&exists->ib_dev);
326 pr_err("already configured on %s\n", ndev->name);
327 err = -EEXIST;
328 goto err;
329 }
330
331 err = rxe_net_add(ibdev_name, ndev);
332 if (err) {
333 pr_err("failed to add %s\n", ndev->name);
334 goto err;
335 }
336 err:
337 return err;
338 }
339
340 static struct rdma_link_ops rxe_link_ops = {
341 .type = "rxe",
342 .newlink = rxe_newlink,
343 };
344
rxe_module_init(void)345 static int __init rxe_module_init(void)
346 {
347 int err;
348
349 /* initialize slab caches for managed objects */
350 err = rxe_cache_init();
351 if (err) {
352 pr_err("unable to init object pools\n");
353 return err;
354 }
355
356 err = rxe_net_init();
357 if (err)
358 return err;
359
360 rdma_link_register(&rxe_link_ops);
361 pr_info("loaded\n");
362 return 0;
363 }
364
rxe_module_exit(void)365 static void __exit rxe_module_exit(void)
366 {
367 rdma_link_unregister(&rxe_link_ops);
368 ib_unregister_driver(RDMA_DRIVER_RXE);
369 rxe_net_exit();
370 rxe_cache_exit();
371
372 pr_info("unloaded\n");
373 }
374
375 late_initcall(rxe_module_init);
376 module_exit(rxe_module_exit);
377
378 MODULE_ALIAS_RDMA_LINK("rxe");
379