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 
ocrdma_link_layer(struct ib_device * device,u32 port_num)65 static enum rdma_link_layer ocrdma_link_layer(struct ib_device *device,
66 					      u32 port_num)
67 {
68 	return IB_LINK_LAYER_ETHERNET;
69 }
70 
ocrdma_port_immutable(struct ib_device * ibdev,u32 port_num,struct ib_port_immutable * immutable)71 static int ocrdma_port_immutable(struct ib_device *ibdev, u32 port_num,
72 			         struct ib_port_immutable *immutable)
73 {
74 	struct ib_port_attr attr;
75 	struct ocrdma_dev *dev;
76 	int err;
77 
78 	dev = get_ocrdma_dev(ibdev);
79 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
80 	if (ocrdma_is_udp_encap_supported(dev))
81 		immutable->core_cap_flags |= RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
82 
83 	err = ib_query_port(ibdev, port_num, &attr);
84 	if (err)
85 		return err;
86 
87 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
88 	immutable->gid_tbl_len = attr.gid_tbl_len;
89 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
90 
91 	return 0;
92 }
93 
get_dev_fw_str(struct ib_device * device,char * str)94 static void get_dev_fw_str(struct ib_device *device, char *str)
95 {
96 	struct ocrdma_dev *dev = get_ocrdma_dev(device);
97 
98 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%s", &dev->attr.fw_ver[0]);
99 }
100 
101 /* OCRDMA sysfs interface */
hw_rev_show(struct device * device,struct device_attribute * attr,char * buf)102 static ssize_t hw_rev_show(struct device *device,
103 			   struct device_attribute *attr, char *buf)
104 {
105 	struct ocrdma_dev *dev =
106 		rdma_device_to_drv_device(device, struct ocrdma_dev, ibdev);
107 
108 	return sysfs_emit(buf, "0x%x\n", dev->nic_info.pdev->vendor);
109 }
110 static DEVICE_ATTR_RO(hw_rev);
111 
hca_type_show(struct device * device,struct device_attribute * attr,char * buf)112 static ssize_t hca_type_show(struct device *device,
113 			     struct device_attribute *attr, char *buf)
114 {
115 	struct ocrdma_dev *dev =
116 		rdma_device_to_drv_device(device, struct ocrdma_dev, ibdev);
117 
118 	return sysfs_emit(buf, "%s\n", &dev->model_number[0]);
119 }
120 static DEVICE_ATTR_RO(hca_type);
121 
122 static struct attribute *ocrdma_attributes[] = {
123 	&dev_attr_hw_rev.attr,
124 	&dev_attr_hca_type.attr,
125 	NULL
126 };
127 
128 static const struct attribute_group ocrdma_attr_group = {
129 	.attrs = ocrdma_attributes,
130 };
131 
132 static const struct ib_device_ops ocrdma_dev_ops = {
133 	.owner = THIS_MODULE,
134 	.driver_id = RDMA_DRIVER_OCRDMA,
135 	.uverbs_abi_ver = OCRDMA_ABI_VERSION,
136 
137 	.alloc_mr = ocrdma_alloc_mr,
138 	.alloc_pd = ocrdma_alloc_pd,
139 	.alloc_ucontext = ocrdma_alloc_ucontext,
140 	.create_ah = ocrdma_create_ah,
141 	.create_cq = ocrdma_create_cq,
142 	.create_qp = ocrdma_create_qp,
143 	.create_user_ah = ocrdma_create_ah,
144 	.dealloc_pd = ocrdma_dealloc_pd,
145 	.dealloc_ucontext = ocrdma_dealloc_ucontext,
146 	.dereg_mr = ocrdma_dereg_mr,
147 	.destroy_ah = ocrdma_destroy_ah,
148 	.destroy_cq = ocrdma_destroy_cq,
149 	.destroy_qp = ocrdma_destroy_qp,
150 	.device_group = &ocrdma_attr_group,
151 	.get_dev_fw_str = get_dev_fw_str,
152 	.get_dma_mr = ocrdma_get_dma_mr,
153 	.get_link_layer = ocrdma_link_layer,
154 	.get_port_immutable = ocrdma_port_immutable,
155 	.map_mr_sg = ocrdma_map_mr_sg,
156 	.mmap = ocrdma_mmap,
157 	.modify_qp = ocrdma_modify_qp,
158 	.poll_cq = ocrdma_poll_cq,
159 	.post_recv = ocrdma_post_recv,
160 	.post_send = ocrdma_post_send,
161 	.process_mad = ocrdma_process_mad,
162 	.query_ah = ocrdma_query_ah,
163 	.query_device = ocrdma_query_device,
164 	.query_pkey = ocrdma_query_pkey,
165 	.query_port = ocrdma_query_port,
166 	.query_qp = ocrdma_query_qp,
167 	.reg_user_mr = ocrdma_reg_user_mr,
168 	.req_notify_cq = ocrdma_arm_cq,
169 	.resize_cq = ocrdma_resize_cq,
170 
171 	INIT_RDMA_OBJ_SIZE(ib_ah, ocrdma_ah, ibah),
172 	INIT_RDMA_OBJ_SIZE(ib_cq, ocrdma_cq, ibcq),
173 	INIT_RDMA_OBJ_SIZE(ib_pd, ocrdma_pd, ibpd),
174 	INIT_RDMA_OBJ_SIZE(ib_qp, ocrdma_qp, ibqp),
175 	INIT_RDMA_OBJ_SIZE(ib_ucontext, ocrdma_ucontext, ibucontext),
176 };
177 
178 static const struct ib_device_ops ocrdma_dev_srq_ops = {
179 	.create_srq = ocrdma_create_srq,
180 	.destroy_srq = ocrdma_destroy_srq,
181 	.modify_srq = ocrdma_modify_srq,
182 	.post_srq_recv = ocrdma_post_srq_recv,
183 	.query_srq = ocrdma_query_srq,
184 
185 	INIT_RDMA_OBJ_SIZE(ib_srq, ocrdma_srq, ibsrq),
186 };
187 
ocrdma_register_device(struct ocrdma_dev * dev)188 static int ocrdma_register_device(struct ocrdma_dev *dev)
189 {
190 	int ret;
191 
192 	addrconf_addr_eui48((u8 *)&dev->ibdev.node_guid,
193 			    dev->nic_info.mac_addr);
194 	BUILD_BUG_ON(sizeof(OCRDMA_NODE_DESC) > IB_DEVICE_NODE_DESC_MAX);
195 	memcpy(dev->ibdev.node_desc, OCRDMA_NODE_DESC,
196 	       sizeof(OCRDMA_NODE_DESC));
197 
198 	dev->ibdev.node_type = RDMA_NODE_IB_CA;
199 	dev->ibdev.phys_port_cnt = 1;
200 	dev->ibdev.num_comp_vectors = dev->eq_cnt;
201 
202 	/* mandatory to support user space verbs consumer. */
203 	dev->ibdev.dev.parent = &dev->nic_info.pdev->dev;
204 
205 	ib_set_device_ops(&dev->ibdev, &ocrdma_dev_ops);
206 
207 	if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R)
208 		ib_set_device_ops(&dev->ibdev, &ocrdma_dev_srq_ops);
209 
210 	ret = ib_device_set_netdev(&dev->ibdev, dev->nic_info.netdev, 1);
211 	if (ret)
212 		return ret;
213 
214 	dma_set_max_seg_size(&dev->nic_info.pdev->dev, UINT_MAX);
215 	return ib_register_device(&dev->ibdev, "ocrdma%d",
216 				  &dev->nic_info.pdev->dev);
217 }
218 
ocrdma_alloc_resources(struct ocrdma_dev * dev)219 static int ocrdma_alloc_resources(struct ocrdma_dev *dev)
220 {
221 	mutex_init(&dev->dev_lock);
222 	dev->cq_tbl = kcalloc(OCRDMA_MAX_CQ, sizeof(struct ocrdma_cq *),
223 			      GFP_KERNEL);
224 	if (!dev->cq_tbl)
225 		goto alloc_err;
226 
227 	if (dev->attr.max_qp) {
228 		dev->qp_tbl = kcalloc(OCRDMA_MAX_QP,
229 				      sizeof(struct ocrdma_qp *),
230 				      GFP_KERNEL);
231 		if (!dev->qp_tbl)
232 			goto alloc_err;
233 	}
234 
235 	dev->stag_arr = kcalloc(OCRDMA_MAX_STAG, sizeof(u64), GFP_KERNEL);
236 	if (dev->stag_arr == NULL)
237 		goto alloc_err;
238 
239 	ocrdma_alloc_pd_pool(dev);
240 
241 	if (!ocrdma_alloc_stats_resources(dev)) {
242 		pr_err("%s: stats resource allocation failed\n", __func__);
243 		goto alloc_err;
244 	}
245 
246 	spin_lock_init(&dev->av_tbl.lock);
247 	spin_lock_init(&dev->flush_q_lock);
248 	return 0;
249 alloc_err:
250 	pr_err("%s(%d) error.\n", __func__, dev->id);
251 	return -ENOMEM;
252 }
253 
ocrdma_free_resources(struct ocrdma_dev * dev)254 static void ocrdma_free_resources(struct ocrdma_dev *dev)
255 {
256 	ocrdma_release_stats_resources(dev);
257 	kfree(dev->stag_arr);
258 	kfree(dev->qp_tbl);
259 	kfree(dev->cq_tbl);
260 }
261 
ocrdma_add(struct be_dev_info * dev_info)262 static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info)
263 {
264 	int status = 0;
265 	u8 lstate = 0;
266 	struct ocrdma_dev *dev;
267 
268 	dev = ib_alloc_device(ocrdma_dev, ibdev);
269 	if (!dev) {
270 		pr_err("Unable to allocate ib device\n");
271 		return NULL;
272 	}
273 
274 	dev->mbx_cmd = kzalloc(sizeof(struct ocrdma_mqe_emb_cmd), GFP_KERNEL);
275 	if (!dev->mbx_cmd)
276 		goto init_err;
277 
278 	memcpy(&dev->nic_info, dev_info, sizeof(*dev_info));
279 	dev->id = PCI_FUNC(dev->nic_info.pdev->devfn);
280 	status = ocrdma_init_hw(dev);
281 	if (status)
282 		goto init_err;
283 
284 	status = ocrdma_alloc_resources(dev);
285 	if (status)
286 		goto alloc_err;
287 
288 	ocrdma_init_service_level(dev);
289 	status = ocrdma_register_device(dev);
290 	if (status)
291 		goto alloc_err;
292 
293 	/* Query Link state and update */
294 	status = ocrdma_mbx_get_link_speed(dev, NULL, &lstate);
295 	if (!status)
296 		ocrdma_update_link_state(dev, lstate);
297 
298 	/* Init stats */
299 	ocrdma_add_port_stats(dev);
300 	/* Interrupt Moderation */
301 	INIT_DELAYED_WORK(&dev->eqd_work, ocrdma_eqd_set_task);
302 	schedule_delayed_work(&dev->eqd_work, msecs_to_jiffies(1000));
303 
304 	pr_info("%s %s: %s \"%s\" port %d\n",
305 		dev_name(&dev->nic_info.pdev->dev), hca_name(dev),
306 		port_speed_string(dev), dev->model_number,
307 		dev->hba_port_num);
308 	pr_info("%s ocrdma%d driver loaded successfully\n",
309 		dev_name(&dev->nic_info.pdev->dev), dev->id);
310 	return dev;
311 
312 alloc_err:
313 	ocrdma_free_resources(dev);
314 	ocrdma_cleanup_hw(dev);
315 init_err:
316 	kfree(dev->mbx_cmd);
317 	ib_dealloc_device(&dev->ibdev);
318 	pr_err("%s() leaving. ret=%d\n", __func__, status);
319 	return NULL;
320 }
321 
ocrdma_remove_free(struct ocrdma_dev * dev)322 static void ocrdma_remove_free(struct ocrdma_dev *dev)
323 {
324 
325 	kfree(dev->mbx_cmd);
326 	ib_dealloc_device(&dev->ibdev);
327 }
328 
ocrdma_remove(struct ocrdma_dev * dev)329 static void ocrdma_remove(struct ocrdma_dev *dev)
330 {
331 	/* first unregister with stack to stop all the active traffic
332 	 * of the registered clients.
333 	 */
334 	cancel_delayed_work_sync(&dev->eqd_work);
335 	ib_unregister_device(&dev->ibdev);
336 
337 	ocrdma_rem_port_stats(dev);
338 	ocrdma_free_resources(dev);
339 	ocrdma_cleanup_hw(dev);
340 	ocrdma_remove_free(dev);
341 }
342 
ocrdma_dispatch_port_active(struct ocrdma_dev * dev)343 static int ocrdma_dispatch_port_active(struct ocrdma_dev *dev)
344 {
345 	struct ib_event port_event;
346 
347 	port_event.event = IB_EVENT_PORT_ACTIVE;
348 	port_event.element.port_num = 1;
349 	port_event.device = &dev->ibdev;
350 	ib_dispatch_event(&port_event);
351 	return 0;
352 }
353 
ocrdma_dispatch_port_error(struct ocrdma_dev * dev)354 static int ocrdma_dispatch_port_error(struct ocrdma_dev *dev)
355 {
356 	struct ib_event err_event;
357 
358 	err_event.event = IB_EVENT_PORT_ERR;
359 	err_event.element.port_num = 1;
360 	err_event.device = &dev->ibdev;
361 	ib_dispatch_event(&err_event);
362 	return 0;
363 }
364 
ocrdma_shutdown(struct ocrdma_dev * dev)365 static void ocrdma_shutdown(struct ocrdma_dev *dev)
366 {
367 	ocrdma_dispatch_port_error(dev);
368 	ocrdma_remove(dev);
369 }
370 
371 /* event handling via NIC driver ensures that all the NIC specific
372  * initialization done before RoCE driver notifies
373  * event to stack.
374  */
ocrdma_event_handler(struct ocrdma_dev * dev,u32 event)375 static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event)
376 {
377 	switch (event) {
378 	case BE_DEV_SHUTDOWN:
379 		ocrdma_shutdown(dev);
380 		break;
381 	default:
382 		break;
383 	}
384 }
385 
ocrdma_update_link_state(struct ocrdma_dev * dev,u8 lstate)386 void ocrdma_update_link_state(struct ocrdma_dev *dev, u8 lstate)
387 {
388 	if (!(dev->flags & OCRDMA_FLAGS_LINK_STATUS_INIT)) {
389 		dev->flags |= OCRDMA_FLAGS_LINK_STATUS_INIT;
390 		if (!lstate)
391 			return;
392 	}
393 
394 	if (!lstate)
395 		ocrdma_dispatch_port_error(dev);
396 	else
397 		ocrdma_dispatch_port_active(dev);
398 }
399 
400 static struct ocrdma_driver ocrdma_drv = {
401 	.name			= "ocrdma_driver",
402 	.add			= ocrdma_add,
403 	.remove			= ocrdma_remove,
404 	.state_change_handler	= ocrdma_event_handler,
405 	.be_abi_version		= OCRDMA_BE_ROCE_ABI_VERSION,
406 };
407 
ocrdma_init_module(void)408 static int __init ocrdma_init_module(void)
409 {
410 	int status;
411 
412 	ocrdma_init_debugfs();
413 
414 	status = be_roce_register_driver(&ocrdma_drv);
415 	if (status)
416 		goto err_be_reg;
417 
418 	return 0;
419 
420 err_be_reg:
421 
422 	return status;
423 }
424 
ocrdma_exit_module(void)425 static void __exit ocrdma_exit_module(void)
426 {
427 	be_roce_unregister_driver(&ocrdma_drv);
428 	ocrdma_rem_debugfs();
429 }
430 
431 module_init(ocrdma_init_module);
432 module_exit(ocrdma_exit_module);
433