1 /* This file is part of the Emulex RoCE Device Driver for
2 * RoCE (RDMA over Converged Ethernet) adapters.
3 * Copyright (C) 2012-2015 Emulex. All rights reserved.
4 * EMULEX and SLI are trademarks of Emulex.
5 * www.emulex.com
6 *
7 * This software is available to you under a choice of one of two licenses.
8 * You may choose to be licensed under the terms of the GNU General Public
9 * License (GPL) Version 2, available from the file COPYING in the main
10 * directory of this source tree, or the BSD license below:
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 *
16 * - Redistributions of source code must retain the above copyright notice,
17 * this list of conditions and the following disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in
21 * the documentation and/or other materials provided with the distribution.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
24 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
27 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
31 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
32 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
33 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 *
35 * Contact Information:
36 * linux-drivers@emulex.com
37 *
38 * Emulex
39 * 3333 Susan Street
40 * Costa Mesa, CA 92626
41 */
42
43 #include <linux/module.h>
44 #include <linux/idr.h>
45 #include <rdma/ib_verbs.h>
46 #include <rdma/ib_user_verbs.h>
47 #include <rdma/ib_addr.h>
48 #include <rdma/ib_mad.h>
49
50 #include <linux/netdevice.h>
51 #include <net/addrconf.h>
52
53 #include "ocrdma.h"
54 #include "ocrdma_verbs.h"
55 #include "ocrdma_ah.h"
56 #include "be_roce.h"
57 #include "ocrdma_hw.h"
58 #include "ocrdma_stats.h"
59 #include <rdma/ocrdma-abi.h>
60
61 MODULE_DESCRIPTION(OCRDMA_ROCE_DRV_DESC " " OCRDMA_ROCE_DRV_VERSION);
62 MODULE_AUTHOR("Emulex Corporation");
63 MODULE_LICENSE("Dual BSD/GPL");
64
65 static DEFINE_IDR(ocrdma_dev_id);
66
ocrdma_get_guid(struct ocrdma_dev * dev,u8 * guid)67 void ocrdma_get_guid(struct ocrdma_dev *dev, u8 *guid)
68 {
69 u8 mac_addr[6];
70
71 memcpy(&mac_addr[0], &dev->nic_info.mac_addr[0], ETH_ALEN);
72 guid[0] = mac_addr[0] ^ 2;
73 guid[1] = mac_addr[1];
74 guid[2] = mac_addr[2];
75 guid[3] = 0xff;
76 guid[4] = 0xfe;
77 guid[5] = mac_addr[3];
78 guid[6] = mac_addr[4];
79 guid[7] = mac_addr[5];
80 }
ocrdma_link_layer(struct ib_device * device,u8 port_num)81 static enum rdma_link_layer ocrdma_link_layer(struct ib_device *device,
82 u8 port_num)
83 {
84 return IB_LINK_LAYER_ETHERNET;
85 }
86
ocrdma_port_immutable(struct ib_device * ibdev,u8 port_num,struct ib_port_immutable * immutable)87 static int ocrdma_port_immutable(struct ib_device *ibdev, u8 port_num,
88 struct ib_port_immutable *immutable)
89 {
90 struct ib_port_attr attr;
91 struct ocrdma_dev *dev;
92 int err;
93
94 dev = get_ocrdma_dev(ibdev);
95 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
96 if (ocrdma_is_udp_encap_supported(dev))
97 immutable->core_cap_flags |= RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
98
99 err = ib_query_port(ibdev, port_num, &attr);
100 if (err)
101 return err;
102
103 immutable->pkey_tbl_len = attr.pkey_tbl_len;
104 immutable->gid_tbl_len = attr.gid_tbl_len;
105 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
106
107 return 0;
108 }
109
get_dev_fw_str(struct ib_device * device,char * str)110 static void get_dev_fw_str(struct ib_device *device, char *str)
111 {
112 struct ocrdma_dev *dev = get_ocrdma_dev(device);
113
114 snprintf(str, IB_FW_VERSION_NAME_MAX, "%s", &dev->attr.fw_ver[0]);
115 }
116
ocrdma_register_device(struct ocrdma_dev * dev)117 static int ocrdma_register_device(struct ocrdma_dev *dev)
118 {
119 strlcpy(dev->ibdev.name, "ocrdma%d", IB_DEVICE_NAME_MAX);
120 ocrdma_get_guid(dev, (u8 *)&dev->ibdev.node_guid);
121 BUILD_BUG_ON(sizeof(OCRDMA_NODE_DESC) > IB_DEVICE_NODE_DESC_MAX);
122 memcpy(dev->ibdev.node_desc, OCRDMA_NODE_DESC,
123 sizeof(OCRDMA_NODE_DESC));
124 dev->ibdev.owner = THIS_MODULE;
125 dev->ibdev.uverbs_abi_ver = OCRDMA_ABI_VERSION;
126 dev->ibdev.uverbs_cmd_mask =
127 OCRDMA_UVERBS(GET_CONTEXT) |
128 OCRDMA_UVERBS(QUERY_DEVICE) |
129 OCRDMA_UVERBS(QUERY_PORT) |
130 OCRDMA_UVERBS(ALLOC_PD) |
131 OCRDMA_UVERBS(DEALLOC_PD) |
132 OCRDMA_UVERBS(REG_MR) |
133 OCRDMA_UVERBS(DEREG_MR) |
134 OCRDMA_UVERBS(CREATE_COMP_CHANNEL) |
135 OCRDMA_UVERBS(CREATE_CQ) |
136 OCRDMA_UVERBS(RESIZE_CQ) |
137 OCRDMA_UVERBS(DESTROY_CQ) |
138 OCRDMA_UVERBS(REQ_NOTIFY_CQ) |
139 OCRDMA_UVERBS(CREATE_QP) |
140 OCRDMA_UVERBS(MODIFY_QP) |
141 OCRDMA_UVERBS(QUERY_QP) |
142 OCRDMA_UVERBS(DESTROY_QP) |
143 OCRDMA_UVERBS(POLL_CQ) |
144 OCRDMA_UVERBS(POST_SEND) |
145 OCRDMA_UVERBS(POST_RECV);
146
147 dev->ibdev.uverbs_cmd_mask |=
148 OCRDMA_UVERBS(CREATE_AH) |
149 OCRDMA_UVERBS(MODIFY_AH) |
150 OCRDMA_UVERBS(QUERY_AH) |
151 OCRDMA_UVERBS(DESTROY_AH);
152
153 dev->ibdev.node_type = RDMA_NODE_IB_CA;
154 dev->ibdev.phys_port_cnt = 1;
155 dev->ibdev.num_comp_vectors = dev->eq_cnt;
156
157 /* mandatory verbs. */
158 dev->ibdev.query_device = ocrdma_query_device;
159 dev->ibdev.query_port = ocrdma_query_port;
160 dev->ibdev.modify_port = ocrdma_modify_port;
161 dev->ibdev.get_netdev = ocrdma_get_netdev;
162 dev->ibdev.get_link_layer = ocrdma_link_layer;
163 dev->ibdev.alloc_pd = ocrdma_alloc_pd;
164 dev->ibdev.dealloc_pd = ocrdma_dealloc_pd;
165
166 dev->ibdev.create_cq = ocrdma_create_cq;
167 dev->ibdev.destroy_cq = ocrdma_destroy_cq;
168 dev->ibdev.resize_cq = ocrdma_resize_cq;
169
170 dev->ibdev.create_qp = ocrdma_create_qp;
171 dev->ibdev.modify_qp = ocrdma_modify_qp;
172 dev->ibdev.query_qp = ocrdma_query_qp;
173 dev->ibdev.destroy_qp = ocrdma_destroy_qp;
174
175 dev->ibdev.query_pkey = ocrdma_query_pkey;
176 dev->ibdev.create_ah = ocrdma_create_ah;
177 dev->ibdev.destroy_ah = ocrdma_destroy_ah;
178 dev->ibdev.query_ah = ocrdma_query_ah;
179
180 dev->ibdev.poll_cq = ocrdma_poll_cq;
181 dev->ibdev.post_send = ocrdma_post_send;
182 dev->ibdev.post_recv = ocrdma_post_recv;
183 dev->ibdev.req_notify_cq = ocrdma_arm_cq;
184
185 dev->ibdev.get_dma_mr = ocrdma_get_dma_mr;
186 dev->ibdev.dereg_mr = ocrdma_dereg_mr;
187 dev->ibdev.reg_user_mr = ocrdma_reg_user_mr;
188
189 dev->ibdev.alloc_mr = ocrdma_alloc_mr;
190 dev->ibdev.map_mr_sg = ocrdma_map_mr_sg;
191
192 /* mandatory to support user space verbs consumer. */
193 dev->ibdev.alloc_ucontext = ocrdma_alloc_ucontext;
194 dev->ibdev.dealloc_ucontext = ocrdma_dealloc_ucontext;
195 dev->ibdev.mmap = ocrdma_mmap;
196 dev->ibdev.dev.parent = &dev->nic_info.pdev->dev;
197
198 dev->ibdev.process_mad = ocrdma_process_mad;
199 dev->ibdev.get_port_immutable = ocrdma_port_immutable;
200 dev->ibdev.get_dev_fw_str = get_dev_fw_str;
201
202 if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
203 dev->ibdev.uverbs_cmd_mask |=
204 OCRDMA_UVERBS(CREATE_SRQ) |
205 OCRDMA_UVERBS(MODIFY_SRQ) |
206 OCRDMA_UVERBS(QUERY_SRQ) |
207 OCRDMA_UVERBS(DESTROY_SRQ) |
208 OCRDMA_UVERBS(POST_SRQ_RECV);
209
210 dev->ibdev.create_srq = ocrdma_create_srq;
211 dev->ibdev.modify_srq = ocrdma_modify_srq;
212 dev->ibdev.query_srq = ocrdma_query_srq;
213 dev->ibdev.destroy_srq = ocrdma_destroy_srq;
214 dev->ibdev.post_srq_recv = ocrdma_post_srq_recv;
215 }
216 dev->ibdev.driver_id = RDMA_DRIVER_OCRDMA;
217 return ib_register_device(&dev->ibdev, NULL);
218 }
219
ocrdma_alloc_resources(struct ocrdma_dev * dev)220 static int ocrdma_alloc_resources(struct ocrdma_dev *dev)
221 {
222 mutex_init(&dev->dev_lock);
223 dev->cq_tbl = kcalloc(OCRDMA_MAX_CQ, sizeof(struct ocrdma_cq *),
224 GFP_KERNEL);
225 if (!dev->cq_tbl)
226 goto alloc_err;
227
228 if (dev->attr.max_qp) {
229 dev->qp_tbl = kcalloc(OCRDMA_MAX_QP,
230 sizeof(struct ocrdma_qp *),
231 GFP_KERNEL);
232 if (!dev->qp_tbl)
233 goto alloc_err;
234 }
235
236 dev->stag_arr = kcalloc(OCRDMA_MAX_STAG, sizeof(u64), GFP_KERNEL);
237 if (dev->stag_arr == NULL)
238 goto alloc_err;
239
240 ocrdma_alloc_pd_pool(dev);
241
242 if (!ocrdma_alloc_stats_resources(dev)) {
243 pr_err("%s: stats resource allocation failed\n", __func__);
244 goto alloc_err;
245 }
246
247 spin_lock_init(&dev->av_tbl.lock);
248 spin_lock_init(&dev->flush_q_lock);
249 return 0;
250 alloc_err:
251 pr_err("%s(%d) error.\n", __func__, dev->id);
252 return -ENOMEM;
253 }
254
ocrdma_free_resources(struct ocrdma_dev * dev)255 static void ocrdma_free_resources(struct ocrdma_dev *dev)
256 {
257 ocrdma_release_stats_resources(dev);
258 kfree(dev->stag_arr);
259 kfree(dev->qp_tbl);
260 kfree(dev->cq_tbl);
261 }
262
263 /* OCRDMA sysfs interface */
show_rev(struct device * device,struct device_attribute * attr,char * buf)264 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
265 char *buf)
266 {
267 struct ocrdma_dev *dev = dev_get_drvdata(device);
268
269 return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->nic_info.pdev->vendor);
270 }
271
show_hca_type(struct device * device,struct device_attribute * attr,char * buf)272 static ssize_t show_hca_type(struct device *device,
273 struct device_attribute *attr, char *buf)
274 {
275 struct ocrdma_dev *dev = dev_get_drvdata(device);
276
277 return scnprintf(buf, PAGE_SIZE, "%s\n", &dev->model_number[0]);
278 }
279
280 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
281 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
282
283 static struct device_attribute *ocrdma_attributes[] = {
284 &dev_attr_hw_rev,
285 &dev_attr_hca_type
286 };
287
ocrdma_remove_sysfiles(struct ocrdma_dev * dev)288 static void ocrdma_remove_sysfiles(struct ocrdma_dev *dev)
289 {
290 int i;
291
292 for (i = 0; i < ARRAY_SIZE(ocrdma_attributes); i++)
293 device_remove_file(&dev->ibdev.dev, ocrdma_attributes[i]);
294 }
295
ocrdma_add(struct be_dev_info * dev_info)296 static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info)
297 {
298 int status = 0, i;
299 u8 lstate = 0;
300 struct ocrdma_dev *dev;
301
302 dev = (struct ocrdma_dev *)ib_alloc_device(sizeof(struct ocrdma_dev));
303 if (!dev) {
304 pr_err("Unable to allocate ib device\n");
305 return NULL;
306 }
307 dev->mbx_cmd = kzalloc(sizeof(struct ocrdma_mqe_emb_cmd), GFP_KERNEL);
308 if (!dev->mbx_cmd)
309 goto idr_err;
310
311 memcpy(&dev->nic_info, dev_info, sizeof(*dev_info));
312 dev->id = idr_alloc(&ocrdma_dev_id, NULL, 0, 0, GFP_KERNEL);
313 if (dev->id < 0)
314 goto idr_err;
315
316 status = ocrdma_init_hw(dev);
317 if (status)
318 goto init_err;
319
320 status = ocrdma_alloc_resources(dev);
321 if (status)
322 goto alloc_err;
323
324 ocrdma_init_service_level(dev);
325 status = ocrdma_register_device(dev);
326 if (status)
327 goto alloc_err;
328
329 /* Query Link state and update */
330 status = ocrdma_mbx_get_link_speed(dev, NULL, &lstate);
331 if (!status)
332 ocrdma_update_link_state(dev, lstate);
333
334 for (i = 0; i < ARRAY_SIZE(ocrdma_attributes); i++)
335 if (device_create_file(&dev->ibdev.dev, ocrdma_attributes[i]))
336 goto sysfs_err;
337 /* Init stats */
338 ocrdma_add_port_stats(dev);
339 /* Interrupt Moderation */
340 INIT_DELAYED_WORK(&dev->eqd_work, ocrdma_eqd_set_task);
341 schedule_delayed_work(&dev->eqd_work, msecs_to_jiffies(1000));
342
343 pr_info("%s %s: %s \"%s\" port %d\n",
344 dev_name(&dev->nic_info.pdev->dev), hca_name(dev),
345 port_speed_string(dev), dev->model_number,
346 dev->hba_port_num);
347 pr_info("%s ocrdma%d driver loaded successfully\n",
348 dev_name(&dev->nic_info.pdev->dev), dev->id);
349 return dev;
350
351 sysfs_err:
352 ocrdma_remove_sysfiles(dev);
353 alloc_err:
354 ocrdma_free_resources(dev);
355 ocrdma_cleanup_hw(dev);
356 init_err:
357 idr_remove(&ocrdma_dev_id, dev->id);
358 idr_err:
359 kfree(dev->mbx_cmd);
360 ib_dealloc_device(&dev->ibdev);
361 pr_err("%s() leaving. ret=%d\n", __func__, status);
362 return NULL;
363 }
364
ocrdma_remove_free(struct ocrdma_dev * dev)365 static void ocrdma_remove_free(struct ocrdma_dev *dev)
366 {
367
368 idr_remove(&ocrdma_dev_id, dev->id);
369 kfree(dev->mbx_cmd);
370 ib_dealloc_device(&dev->ibdev);
371 }
372
ocrdma_remove(struct ocrdma_dev * dev)373 static void ocrdma_remove(struct ocrdma_dev *dev)
374 {
375 /* first unregister with stack to stop all the active traffic
376 * of the registered clients.
377 */
378 cancel_delayed_work_sync(&dev->eqd_work);
379 ocrdma_remove_sysfiles(dev);
380 ib_unregister_device(&dev->ibdev);
381
382 ocrdma_rem_port_stats(dev);
383 ocrdma_free_resources(dev);
384 ocrdma_cleanup_hw(dev);
385 ocrdma_remove_free(dev);
386 }
387
ocrdma_dispatch_port_active(struct ocrdma_dev * dev)388 static int ocrdma_dispatch_port_active(struct ocrdma_dev *dev)
389 {
390 struct ib_event port_event;
391
392 port_event.event = IB_EVENT_PORT_ACTIVE;
393 port_event.element.port_num = 1;
394 port_event.device = &dev->ibdev;
395 ib_dispatch_event(&port_event);
396 return 0;
397 }
398
ocrdma_dispatch_port_error(struct ocrdma_dev * dev)399 static int ocrdma_dispatch_port_error(struct ocrdma_dev *dev)
400 {
401 struct ib_event err_event;
402
403 err_event.event = IB_EVENT_PORT_ERR;
404 err_event.element.port_num = 1;
405 err_event.device = &dev->ibdev;
406 ib_dispatch_event(&err_event);
407 return 0;
408 }
409
ocrdma_shutdown(struct ocrdma_dev * dev)410 static void ocrdma_shutdown(struct ocrdma_dev *dev)
411 {
412 ocrdma_dispatch_port_error(dev);
413 ocrdma_remove(dev);
414 }
415
416 /* event handling via NIC driver ensures that all the NIC specific
417 * initialization done before RoCE driver notifies
418 * event to stack.
419 */
ocrdma_event_handler(struct ocrdma_dev * dev,u32 event)420 static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event)
421 {
422 switch (event) {
423 case BE_DEV_SHUTDOWN:
424 ocrdma_shutdown(dev);
425 break;
426 default:
427 break;
428 }
429 }
430
ocrdma_update_link_state(struct ocrdma_dev * dev,u8 lstate)431 void ocrdma_update_link_state(struct ocrdma_dev *dev, u8 lstate)
432 {
433 if (!(dev->flags & OCRDMA_FLAGS_LINK_STATUS_INIT)) {
434 dev->flags |= OCRDMA_FLAGS_LINK_STATUS_INIT;
435 if (!lstate)
436 return;
437 }
438
439 if (!lstate)
440 ocrdma_dispatch_port_error(dev);
441 else
442 ocrdma_dispatch_port_active(dev);
443 }
444
445 static struct ocrdma_driver ocrdma_drv = {
446 .name = "ocrdma_driver",
447 .add = ocrdma_add,
448 .remove = ocrdma_remove,
449 .state_change_handler = ocrdma_event_handler,
450 .be_abi_version = OCRDMA_BE_ROCE_ABI_VERSION,
451 };
452
ocrdma_init_module(void)453 static int __init ocrdma_init_module(void)
454 {
455 int status;
456
457 ocrdma_init_debugfs();
458
459 status = be_roce_register_driver(&ocrdma_drv);
460 if (status)
461 goto err_be_reg;
462
463 return 0;
464
465 err_be_reg:
466
467 return status;
468 }
469
ocrdma_exit_module(void)470 static void __exit ocrdma_exit_module(void)
471 {
472 be_roce_unregister_driver(&ocrdma_drv);
473 ocrdma_rem_debugfs();
474 idr_destroy(&ocrdma_dev_id);
475 }
476
477 module_init(ocrdma_init_module);
478 module_exit(ocrdma_exit_module);
479
