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 bool rxe_initialized;
52
53 /* free resources for a rxe device all objects created for this device must
54 * have been destroyed
55 */
rxe_dealloc(struct ib_device * ib_dev)56 void rxe_dealloc(struct ib_device *ib_dev)
57 {
58 struct rxe_dev *rxe = container_of(ib_dev, struct rxe_dev, ib_dev);
59
60 rxe_pool_cleanup(&rxe->uc_pool);
61 rxe_pool_cleanup(&rxe->pd_pool);
62 rxe_pool_cleanup(&rxe->ah_pool);
63 rxe_pool_cleanup(&rxe->srq_pool);
64 rxe_pool_cleanup(&rxe->qp_pool);
65 rxe_pool_cleanup(&rxe->cq_pool);
66 rxe_pool_cleanup(&rxe->mr_pool);
67 rxe_pool_cleanup(&rxe->mw_pool);
68 rxe_pool_cleanup(&rxe->mc_grp_pool);
69 rxe_pool_cleanup(&rxe->mc_elem_pool);
70
71 rxe_cleanup_ports(rxe);
72
73 if (rxe->tfm)
74 crypto_free_shash(rxe->tfm);
75 }
76
77 /* initialize rxe device parameters */
rxe_init_device_param(struct rxe_dev * rxe)78 static void rxe_init_device_param(struct rxe_dev *rxe)
79 {
80 rxe->max_inline_data = RXE_MAX_INLINE_DATA;
81
82 rxe->attr.fw_ver = RXE_FW_VER;
83 rxe->attr.max_mr_size = RXE_MAX_MR_SIZE;
84 rxe->attr.page_size_cap = RXE_PAGE_SIZE_CAP;
85 rxe->attr.vendor_id = RXE_VENDOR_ID;
86 rxe->attr.vendor_part_id = RXE_VENDOR_PART_ID;
87 rxe->attr.hw_ver = RXE_HW_VER;
88 rxe->attr.max_qp = RXE_MAX_QP;
89 rxe->attr.max_qp_wr = RXE_MAX_QP_WR;
90 rxe->attr.device_cap_flags = RXE_DEVICE_CAP_FLAGS;
91 rxe->attr.max_send_sge = RXE_MAX_SGE;
92 rxe->attr.max_recv_sge = RXE_MAX_SGE;
93 rxe->attr.max_sge_rd = RXE_MAX_SGE_RD;
94 rxe->attr.max_cq = RXE_MAX_CQ;
95 rxe->attr.max_cqe = (1 << RXE_MAX_LOG_CQE) - 1;
96 rxe->attr.max_mr = RXE_MAX_MR;
97 rxe->attr.max_pd = RXE_MAX_PD;
98 rxe->attr.max_qp_rd_atom = RXE_MAX_QP_RD_ATOM;
99 rxe->attr.max_ee_rd_atom = RXE_MAX_EE_RD_ATOM;
100 rxe->attr.max_res_rd_atom = RXE_MAX_RES_RD_ATOM;
101 rxe->attr.max_qp_init_rd_atom = RXE_MAX_QP_INIT_RD_ATOM;
102 rxe->attr.max_ee_init_rd_atom = RXE_MAX_EE_INIT_RD_ATOM;
103 rxe->attr.atomic_cap = IB_ATOMIC_HCA;
104 rxe->attr.max_ee = RXE_MAX_EE;
105 rxe->attr.max_rdd = RXE_MAX_RDD;
106 rxe->attr.max_mw = RXE_MAX_MW;
107 rxe->attr.max_raw_ipv6_qp = RXE_MAX_RAW_IPV6_QP;
108 rxe->attr.max_raw_ethy_qp = RXE_MAX_RAW_ETHY_QP;
109 rxe->attr.max_mcast_grp = RXE_MAX_MCAST_GRP;
110 rxe->attr.max_mcast_qp_attach = RXE_MAX_MCAST_QP_ATTACH;
111 rxe->attr.max_total_mcast_qp_attach = RXE_MAX_TOT_MCAST_QP_ATTACH;
112 rxe->attr.max_ah = RXE_MAX_AH;
113 rxe->attr.max_fmr = RXE_MAX_FMR;
114 rxe->attr.max_map_per_fmr = RXE_MAX_MAP_PER_FMR;
115 rxe->attr.max_srq = RXE_MAX_SRQ;
116 rxe->attr.max_srq_wr = RXE_MAX_SRQ_WR;
117 rxe->attr.max_srq_sge = RXE_MAX_SRQ_SGE;
118 rxe->attr.max_fast_reg_page_list_len = RXE_MAX_FMR_PAGE_LIST_LEN;
119 rxe->attr.max_pkeys = RXE_MAX_PKEYS;
120 rxe->attr.local_ca_ack_delay = RXE_LOCAL_CA_ACK_DELAY;
121 addrconf_addr_eui48((unsigned char *)&rxe->attr.sys_image_guid,
122 rxe->ndev->dev_addr);
123
124 rxe->max_ucontext = RXE_MAX_UCONTEXT;
125 }
126
127 /* initialize port attributes */
rxe_init_port_param(struct rxe_port * port)128 static int rxe_init_port_param(struct rxe_port *port)
129 {
130 port->attr.state = IB_PORT_DOWN;
131 port->attr.max_mtu = IB_MTU_4096;
132 port->attr.active_mtu = IB_MTU_256;
133 port->attr.gid_tbl_len = RXE_PORT_GID_TBL_LEN;
134 port->attr.port_cap_flags = RXE_PORT_PORT_CAP_FLAGS;
135 port->attr.max_msg_sz = RXE_PORT_MAX_MSG_SZ;
136 port->attr.bad_pkey_cntr = RXE_PORT_BAD_PKEY_CNTR;
137 port->attr.qkey_viol_cntr = RXE_PORT_QKEY_VIOL_CNTR;
138 port->attr.pkey_tbl_len = RXE_PORT_PKEY_TBL_LEN;
139 port->attr.lid = RXE_PORT_LID;
140 port->attr.sm_lid = RXE_PORT_SM_LID;
141 port->attr.lmc = RXE_PORT_LMC;
142 port->attr.max_vl_num = RXE_PORT_MAX_VL_NUM;
143 port->attr.sm_sl = RXE_PORT_SM_SL;
144 port->attr.subnet_timeout = RXE_PORT_SUBNET_TIMEOUT;
145 port->attr.init_type_reply = RXE_PORT_INIT_TYPE_REPLY;
146 port->attr.active_width = RXE_PORT_ACTIVE_WIDTH;
147 port->attr.active_speed = RXE_PORT_ACTIVE_SPEED;
148 port->attr.phys_state = RXE_PORT_PHYS_STATE;
149 port->mtu_cap = ib_mtu_enum_to_int(IB_MTU_256);
150 port->subnet_prefix = cpu_to_be64(RXE_PORT_SUBNET_PREFIX);
151
152 return 0;
153 }
154
155 /* initialize port state, note IB convention that HCA ports are always
156 * numbered from 1
157 */
rxe_init_ports(struct rxe_dev * rxe)158 static int rxe_init_ports(struct rxe_dev *rxe)
159 {
160 struct rxe_port *port = &rxe->port;
161
162 rxe_init_port_param(port);
163
164 port->pkey_tbl = kcalloc(port->attr.pkey_tbl_len,
165 sizeof(*port->pkey_tbl), GFP_KERNEL);
166
167 if (!port->pkey_tbl)
168 return -ENOMEM;
169
170 port->pkey_tbl[0] = 0xffff;
171 addrconf_addr_eui48((unsigned char *)&port->port_guid,
172 rxe->ndev->dev_addr);
173
174 spin_lock_init(&port->port_lock);
175
176 return 0;
177 }
178
179 /* init pools of managed objects */
rxe_init_pools(struct rxe_dev * rxe)180 static int rxe_init_pools(struct rxe_dev *rxe)
181 {
182 int err;
183
184 err = rxe_pool_init(rxe, &rxe->uc_pool, RXE_TYPE_UC,
185 rxe->max_ucontext);
186 if (err)
187 goto err1;
188
189 err = rxe_pool_init(rxe, &rxe->pd_pool, RXE_TYPE_PD,
190 rxe->attr.max_pd);
191 if (err)
192 goto err2;
193
194 err = rxe_pool_init(rxe, &rxe->ah_pool, RXE_TYPE_AH,
195 rxe->attr.max_ah);
196 if (err)
197 goto err3;
198
199 err = rxe_pool_init(rxe, &rxe->srq_pool, RXE_TYPE_SRQ,
200 rxe->attr.max_srq);
201 if (err)
202 goto err4;
203
204 err = rxe_pool_init(rxe, &rxe->qp_pool, RXE_TYPE_QP,
205 rxe->attr.max_qp);
206 if (err)
207 goto err5;
208
209 err = rxe_pool_init(rxe, &rxe->cq_pool, RXE_TYPE_CQ,
210 rxe->attr.max_cq);
211 if (err)
212 goto err6;
213
214 err = rxe_pool_init(rxe, &rxe->mr_pool, RXE_TYPE_MR,
215 rxe->attr.max_mr);
216 if (err)
217 goto err7;
218
219 err = rxe_pool_init(rxe, &rxe->mw_pool, RXE_TYPE_MW,
220 rxe->attr.max_mw);
221 if (err)
222 goto err8;
223
224 err = rxe_pool_init(rxe, &rxe->mc_grp_pool, RXE_TYPE_MC_GRP,
225 rxe->attr.max_mcast_grp);
226 if (err)
227 goto err9;
228
229 err = rxe_pool_init(rxe, &rxe->mc_elem_pool, RXE_TYPE_MC_ELEM,
230 rxe->attr.max_total_mcast_qp_attach);
231 if (err)
232 goto err10;
233
234 return 0;
235
236 err10:
237 rxe_pool_cleanup(&rxe->mc_grp_pool);
238 err9:
239 rxe_pool_cleanup(&rxe->mw_pool);
240 err8:
241 rxe_pool_cleanup(&rxe->mr_pool);
242 err7:
243 rxe_pool_cleanup(&rxe->cq_pool);
244 err6:
245 rxe_pool_cleanup(&rxe->qp_pool);
246 err5:
247 rxe_pool_cleanup(&rxe->srq_pool);
248 err4:
249 rxe_pool_cleanup(&rxe->ah_pool);
250 err3:
251 rxe_pool_cleanup(&rxe->pd_pool);
252 err2:
253 rxe_pool_cleanup(&rxe->uc_pool);
254 err1:
255 return err;
256 }
257
258 /* initialize rxe device state */
rxe_init(struct rxe_dev * rxe)259 static int rxe_init(struct rxe_dev *rxe)
260 {
261 int err;
262
263 /* init default device parameters */
264 rxe_init_device_param(rxe);
265
266 err = rxe_init_ports(rxe);
267 if (err)
268 goto err1;
269
270 err = rxe_init_pools(rxe);
271 if (err)
272 goto err2;
273
274 /* init pending mmap list */
275 spin_lock_init(&rxe->mmap_offset_lock);
276 spin_lock_init(&rxe->pending_lock);
277 INIT_LIST_HEAD(&rxe->pending_mmaps);
278
279 mutex_init(&rxe->usdev_lock);
280
281 return 0;
282
283 err2:
284 rxe_cleanup_ports(rxe);
285 err1:
286 return err;
287 }
288
rxe_set_mtu(struct rxe_dev * rxe,unsigned int ndev_mtu)289 void rxe_set_mtu(struct rxe_dev *rxe, unsigned int ndev_mtu)
290 {
291 struct rxe_port *port = &rxe->port;
292 enum ib_mtu mtu;
293
294 mtu = eth_mtu_int_to_enum(ndev_mtu);
295
296 /* Make sure that new MTU in range */
297 mtu = mtu ? min_t(enum ib_mtu, mtu, IB_MTU_4096) : IB_MTU_256;
298
299 port->attr.active_mtu = mtu;
300 port->mtu_cap = ib_mtu_enum_to_int(mtu);
301 }
302
303 /* called by ifc layer to create new rxe device.
304 * The caller should allocate memory for rxe by calling ib_alloc_device.
305 */
rxe_add(struct rxe_dev * rxe,unsigned int mtu,const char * ibdev_name)306 int rxe_add(struct rxe_dev *rxe, unsigned int mtu, const char *ibdev_name)
307 {
308 int err;
309
310 err = rxe_init(rxe);
311 if (err)
312 return err;
313
314 rxe_set_mtu(rxe, mtu);
315
316 return rxe_register_device(rxe, ibdev_name);
317 }
318
rxe_newlink(const char * ibdev_name,struct net_device * ndev)319 static int rxe_newlink(const char *ibdev_name, struct net_device *ndev)
320 {
321 struct rxe_dev *exists;
322 int err = 0;
323
324 exists = rxe_get_dev_from_net(ndev);
325 if (exists) {
326 ib_device_put(&exists->ib_dev);
327 pr_err("already configured on %s\n", ndev->name);
328 err = -EEXIST;
329 goto err;
330 }
331
332 err = rxe_net_add(ibdev_name, ndev);
333 if (err) {
334 pr_err("failed to add %s\n", ndev->name);
335 goto err;
336 }
337 err:
338 return err;
339 }
340
341 static struct rdma_link_ops rxe_link_ops = {
342 .type = "rxe",
343 .newlink = rxe_newlink,
344 };
345
rxe_module_init(void)346 static int __init rxe_module_init(void)
347 {
348 int err;
349
350 /* initialize slab caches for managed objects */
351 err = rxe_cache_init();
352 if (err) {
353 pr_err("unable to init object pools\n");
354 return err;
355 }
356
357 err = rxe_net_init();
358 if (err)
359 return err;
360
361 rdma_link_register(&rxe_link_ops);
362 rxe_initialized = true;
363 pr_info("loaded\n");
364 return 0;
365 }
366
rxe_module_exit(void)367 static void __exit rxe_module_exit(void)
368 {
369 rdma_link_unregister(&rxe_link_ops);
370 ib_unregister_driver(RDMA_DRIVER_RXE);
371 rxe_net_exit();
372 rxe_cache_exit();
373
374 rxe_initialized = false;
375 pr_info("unloaded\n");
376 }
377
378 late_initcall(rxe_module_init);
379 module_exit(rxe_module_exit);
380
381 MODULE_ALIAS_RDMA_LINK("rxe");
382