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.
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:
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
16 * - Redistributions of source code must retain the above copyright notice,
17 * this list of conditions and the following disclaimer.
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.
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.
35 * Contact Information:
36 * linux-drivers@emulex.com
40 * Costa Mesa, CA 92626
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>
50 #include <linux/netdevice.h>
51 #include <net/addrconf.h>
54 #include "ocrdma_verbs.h"
55 #include "ocrdma_ah.h"
57 #include "ocrdma_hw.h"
58 #include "ocrdma_stats.h"
59 #include <rdma/ocrdma-abi.h>
61 MODULE_DESCRIPTION(OCRDMA_ROCE_DRV_DESC " " OCRDMA_ROCE_DRV_VERSION);
62 MODULE_AUTHOR("Emulex Corporation");
63 MODULE_LICENSE("Dual BSD/GPL");
65 void ocrdma_get_guid(struct ocrdma_dev *dev, u8 *guid)
69 memcpy(&mac_addr[0], &dev->nic_info.mac_addr[0], ETH_ALEN);
70 guid[0] = mac_addr[0] ^ 2;
71 guid[1] = mac_addr[1];
72 guid[2] = mac_addr[2];
75 guid[5] = mac_addr[3];
76 guid[6] = mac_addr[4];
77 guid[7] = mac_addr[5];
79 static enum rdma_link_layer ocrdma_link_layer(struct ib_device *device,
82 return IB_LINK_LAYER_ETHERNET;
85 static int ocrdma_port_immutable(struct ib_device *ibdev, u8 port_num,
86 struct ib_port_immutable *immutable)
88 struct ib_port_attr attr;
89 struct ocrdma_dev *dev;
92 dev = get_ocrdma_dev(ibdev);
93 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
94 if (ocrdma_is_udp_encap_supported(dev))
95 immutable->core_cap_flags |= RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
97 err = ib_query_port(ibdev, port_num, &attr);
101 immutable->pkey_tbl_len = attr.pkey_tbl_len;
102 immutable->gid_tbl_len = attr.gid_tbl_len;
103 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
108 static void get_dev_fw_str(struct ib_device *device, char *str)
110 struct ocrdma_dev *dev = get_ocrdma_dev(device);
112 snprintf(str, IB_FW_VERSION_NAME_MAX, "%s", &dev->attr.fw_ver[0]);
115 /* OCRDMA sysfs interface */
116 static ssize_t hw_rev_show(struct device *device,
117 struct device_attribute *attr, char *buf)
119 struct ocrdma_dev *dev =
120 rdma_device_to_drv_device(device, struct ocrdma_dev, ibdev);
122 return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->nic_info.pdev->vendor);
124 static DEVICE_ATTR_RO(hw_rev);
126 static ssize_t hca_type_show(struct device *device,
127 struct device_attribute *attr, char *buf)
129 struct ocrdma_dev *dev =
130 rdma_device_to_drv_device(device, struct ocrdma_dev, ibdev);
132 return scnprintf(buf, PAGE_SIZE, "%s\n", &dev->model_number[0]);
134 static DEVICE_ATTR_RO(hca_type);
136 static struct attribute *ocrdma_attributes[] = {
137 &dev_attr_hw_rev.attr,
138 &dev_attr_hca_type.attr,
142 static const struct attribute_group ocrdma_attr_group = {
143 .attrs = ocrdma_attributes,
146 static const struct ib_device_ops ocrdma_dev_ops = {
147 .alloc_mr = ocrdma_alloc_mr,
148 .alloc_pd = ocrdma_alloc_pd,
149 .alloc_ucontext = ocrdma_alloc_ucontext,
150 .create_ah = ocrdma_create_ah,
151 .create_cq = ocrdma_create_cq,
152 .create_qp = ocrdma_create_qp,
153 .dealloc_pd = ocrdma_dealloc_pd,
154 .dealloc_ucontext = ocrdma_dealloc_ucontext,
155 .dereg_mr = ocrdma_dereg_mr,
156 .destroy_ah = ocrdma_destroy_ah,
157 .destroy_cq = ocrdma_destroy_cq,
158 .destroy_qp = ocrdma_destroy_qp,
159 .get_dev_fw_str = get_dev_fw_str,
160 .get_dma_mr = ocrdma_get_dma_mr,
161 .get_link_layer = ocrdma_link_layer,
162 .get_port_immutable = ocrdma_port_immutable,
163 .map_mr_sg = ocrdma_map_mr_sg,
165 .modify_port = ocrdma_modify_port,
166 .modify_qp = ocrdma_modify_qp,
167 .poll_cq = ocrdma_poll_cq,
168 .post_recv = ocrdma_post_recv,
169 .post_send = ocrdma_post_send,
170 .process_mad = ocrdma_process_mad,
171 .query_ah = ocrdma_query_ah,
172 .query_device = ocrdma_query_device,
173 .query_pkey = ocrdma_query_pkey,
174 .query_port = ocrdma_query_port,
175 .query_qp = ocrdma_query_qp,
176 .reg_user_mr = ocrdma_reg_user_mr,
177 .req_notify_cq = ocrdma_arm_cq,
178 .resize_cq = ocrdma_resize_cq,
180 INIT_RDMA_OBJ_SIZE(ib_ah, ocrdma_ah, ibah),
181 INIT_RDMA_OBJ_SIZE(ib_pd, ocrdma_pd, ibpd),
182 INIT_RDMA_OBJ_SIZE(ib_ucontext, ocrdma_ucontext, ibucontext),
185 static const struct ib_device_ops ocrdma_dev_srq_ops = {
186 .create_srq = ocrdma_create_srq,
187 .destroy_srq = ocrdma_destroy_srq,
188 .modify_srq = ocrdma_modify_srq,
189 .post_srq_recv = ocrdma_post_srq_recv,
190 .query_srq = ocrdma_query_srq,
192 INIT_RDMA_OBJ_SIZE(ib_srq, ocrdma_srq, ibsrq),
195 static int ocrdma_register_device(struct ocrdma_dev *dev)
199 ocrdma_get_guid(dev, (u8 *)&dev->ibdev.node_guid);
200 BUILD_BUG_ON(sizeof(OCRDMA_NODE_DESC) > IB_DEVICE_NODE_DESC_MAX);
201 memcpy(dev->ibdev.node_desc, OCRDMA_NODE_DESC,
202 sizeof(OCRDMA_NODE_DESC));
203 dev->ibdev.owner = THIS_MODULE;
204 dev->ibdev.uverbs_abi_ver = OCRDMA_ABI_VERSION;
205 dev->ibdev.uverbs_cmd_mask =
206 OCRDMA_UVERBS(GET_CONTEXT) |
207 OCRDMA_UVERBS(QUERY_DEVICE) |
208 OCRDMA_UVERBS(QUERY_PORT) |
209 OCRDMA_UVERBS(ALLOC_PD) |
210 OCRDMA_UVERBS(DEALLOC_PD) |
211 OCRDMA_UVERBS(REG_MR) |
212 OCRDMA_UVERBS(DEREG_MR) |
213 OCRDMA_UVERBS(CREATE_COMP_CHANNEL) |
214 OCRDMA_UVERBS(CREATE_CQ) |
215 OCRDMA_UVERBS(RESIZE_CQ) |
216 OCRDMA_UVERBS(DESTROY_CQ) |
217 OCRDMA_UVERBS(REQ_NOTIFY_CQ) |
218 OCRDMA_UVERBS(CREATE_QP) |
219 OCRDMA_UVERBS(MODIFY_QP) |
220 OCRDMA_UVERBS(QUERY_QP) |
221 OCRDMA_UVERBS(DESTROY_QP) |
222 OCRDMA_UVERBS(POLL_CQ) |
223 OCRDMA_UVERBS(POST_SEND) |
224 OCRDMA_UVERBS(POST_RECV);
226 dev->ibdev.uverbs_cmd_mask |=
227 OCRDMA_UVERBS(CREATE_AH) |
228 OCRDMA_UVERBS(MODIFY_AH) |
229 OCRDMA_UVERBS(QUERY_AH) |
230 OCRDMA_UVERBS(DESTROY_AH);
232 dev->ibdev.node_type = RDMA_NODE_IB_CA;
233 dev->ibdev.phys_port_cnt = 1;
234 dev->ibdev.num_comp_vectors = dev->eq_cnt;
236 /* mandatory to support user space verbs consumer. */
237 dev->ibdev.dev.parent = &dev->nic_info.pdev->dev;
239 ib_set_device_ops(&dev->ibdev, &ocrdma_dev_ops);
241 if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
242 dev->ibdev.uverbs_cmd_mask |=
243 OCRDMA_UVERBS(CREATE_SRQ) |
244 OCRDMA_UVERBS(MODIFY_SRQ) |
245 OCRDMA_UVERBS(QUERY_SRQ) |
246 OCRDMA_UVERBS(DESTROY_SRQ) |
247 OCRDMA_UVERBS(POST_SRQ_RECV);
249 ib_set_device_ops(&dev->ibdev, &ocrdma_dev_srq_ops);
251 rdma_set_device_sysfs_group(&dev->ibdev, &ocrdma_attr_group);
252 dev->ibdev.driver_id = RDMA_DRIVER_OCRDMA;
253 ret = ib_device_set_netdev(&dev->ibdev, dev->nic_info.netdev, 1);
257 return ib_register_device(&dev->ibdev, "ocrdma%d");
260 static int ocrdma_alloc_resources(struct ocrdma_dev *dev)
262 mutex_init(&dev->dev_lock);
263 dev->cq_tbl = kcalloc(OCRDMA_MAX_CQ, sizeof(struct ocrdma_cq *),
268 if (dev->attr.max_qp) {
269 dev->qp_tbl = kcalloc(OCRDMA_MAX_QP,
270 sizeof(struct ocrdma_qp *),
276 dev->stag_arr = kcalloc(OCRDMA_MAX_STAG, sizeof(u64), GFP_KERNEL);
277 if (dev->stag_arr == NULL)
280 ocrdma_alloc_pd_pool(dev);
282 if (!ocrdma_alloc_stats_resources(dev)) {
283 pr_err("%s: stats resource allocation failed\n", __func__);
287 spin_lock_init(&dev->av_tbl.lock);
288 spin_lock_init(&dev->flush_q_lock);
291 pr_err("%s(%d) error.\n", __func__, dev->id);
295 static void ocrdma_free_resources(struct ocrdma_dev *dev)
297 ocrdma_release_stats_resources(dev);
298 kfree(dev->stag_arr);
303 static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info)
307 struct ocrdma_dev *dev;
309 dev = ib_alloc_device(ocrdma_dev, ibdev);
311 pr_err("Unable to allocate ib device\n");
315 dev->mbx_cmd = kzalloc(sizeof(struct ocrdma_mqe_emb_cmd), GFP_KERNEL);
319 memcpy(&dev->nic_info, dev_info, sizeof(*dev_info));
320 dev->id = PCI_FUNC(dev->nic_info.pdev->devfn);
321 status = ocrdma_init_hw(dev);
325 status = ocrdma_alloc_resources(dev);
329 ocrdma_init_service_level(dev);
330 status = ocrdma_register_device(dev);
334 /* Query Link state and update */
335 status = ocrdma_mbx_get_link_speed(dev, NULL, &lstate);
337 ocrdma_update_link_state(dev, lstate);
340 ocrdma_add_port_stats(dev);
341 /* Interrupt Moderation */
342 INIT_DELAYED_WORK(&dev->eqd_work, ocrdma_eqd_set_task);
343 schedule_delayed_work(&dev->eqd_work, msecs_to_jiffies(1000));
345 pr_info("%s %s: %s \"%s\" port %d\n",
346 dev_name(&dev->nic_info.pdev->dev), hca_name(dev),
347 port_speed_string(dev), dev->model_number,
349 pr_info("%s ocrdma%d driver loaded successfully\n",
350 dev_name(&dev->nic_info.pdev->dev), dev->id);
354 ocrdma_free_resources(dev);
355 ocrdma_cleanup_hw(dev);
358 ib_dealloc_device(&dev->ibdev);
359 pr_err("%s() leaving. ret=%d\n", __func__, status);
363 static void ocrdma_remove_free(struct ocrdma_dev *dev)
367 ib_dealloc_device(&dev->ibdev);
370 static void ocrdma_remove(struct ocrdma_dev *dev)
372 /* first unregister with stack to stop all the active traffic
373 * of the registered clients.
375 cancel_delayed_work_sync(&dev->eqd_work);
376 ib_unregister_device(&dev->ibdev);
378 ocrdma_rem_port_stats(dev);
379 ocrdma_free_resources(dev);
380 ocrdma_cleanup_hw(dev);
381 ocrdma_remove_free(dev);
384 static int ocrdma_dispatch_port_active(struct ocrdma_dev *dev)
386 struct ib_event port_event;
388 port_event.event = IB_EVENT_PORT_ACTIVE;
389 port_event.element.port_num = 1;
390 port_event.device = &dev->ibdev;
391 ib_dispatch_event(&port_event);
395 static int ocrdma_dispatch_port_error(struct ocrdma_dev *dev)
397 struct ib_event err_event;
399 err_event.event = IB_EVENT_PORT_ERR;
400 err_event.element.port_num = 1;
401 err_event.device = &dev->ibdev;
402 ib_dispatch_event(&err_event);
406 static void ocrdma_shutdown(struct ocrdma_dev *dev)
408 ocrdma_dispatch_port_error(dev);
412 /* event handling via NIC driver ensures that all the NIC specific
413 * initialization done before RoCE driver notifies
416 static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event)
419 case BE_DEV_SHUTDOWN:
420 ocrdma_shutdown(dev);
427 void ocrdma_update_link_state(struct ocrdma_dev *dev, u8 lstate)
429 if (!(dev->flags & OCRDMA_FLAGS_LINK_STATUS_INIT)) {
430 dev->flags |= OCRDMA_FLAGS_LINK_STATUS_INIT;
436 ocrdma_dispatch_port_error(dev);
438 ocrdma_dispatch_port_active(dev);
441 static struct ocrdma_driver ocrdma_drv = {
442 .name = "ocrdma_driver",
444 .remove = ocrdma_remove,
445 .state_change_handler = ocrdma_event_handler,
446 .be_abi_version = OCRDMA_BE_ROCE_ABI_VERSION,
449 static int __init ocrdma_init_module(void)
453 ocrdma_init_debugfs();
455 status = be_roce_register_driver(&ocrdma_drv);
466 static void __exit ocrdma_exit_module(void)
468 be_roce_unregister_driver(&ocrdma_drv);
469 ocrdma_rem_debugfs();
472 module_init(ocrdma_init_module);
473 module_exit(ocrdma_exit_module);