1 /*
2  * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  *
32  * Author: Upinder Malhi <umalhi@cisco.com>
33  * Author: Anant Deepak <anadeepa@cisco.com>
34  * Author: Cesare Cantu' <cantuc@cisco.com>
35  * Author: Jeff Squyres <jsquyres@cisco.com>
36  * Author: Kiran Thirumalai <kithirum@cisco.com>
37  * Author: Xuyang Wang <xuywang@cisco.com>
38  * Author: Reese Faucette <rfaucett@cisco.com>
39  *
40  */
41 
42 #include <linux/module.h>
43 #include <linux/inetdevice.h>
44 #include <linux/init.h>
45 #include <linux/slab.h>
46 #include <linux/errno.h>
47 #include <linux/pci.h>
48 #include <linux/netdevice.h>
49 
50 #include <rdma/ib_user_verbs.h>
51 #include <rdma/ib_addr.h>
52 
53 #include "usnic_abi.h"
54 #include "usnic_common_util.h"
55 #include "usnic_ib.h"
56 #include "usnic_ib_qp_grp.h"
57 #include "usnic_log.h"
58 #include "usnic_fwd.h"
59 #include "usnic_debugfs.h"
60 #include "usnic_ib_verbs.h"
61 #include "usnic_transport.h"
62 #include "usnic_uiom.h"
63 #include "usnic_ib_sysfs.h"
64 
65 unsigned int usnic_log_lvl = USNIC_LOG_LVL_ERR;
66 unsigned int usnic_ib_share_vf = 1;
67 
68 static const char usnic_version[] =
69 	DRV_NAME ": Cisco VIC (USNIC) Verbs Driver v"
70 	DRV_VERSION " (" DRV_RELDATE ")\n";
71 
72 static DEFINE_MUTEX(usnic_ib_ibdev_list_lock);
73 static LIST_HEAD(usnic_ib_ibdev_list);
74 
75 /* Callback dump funcs */
76 static int usnic_ib_dump_vf_hdr(void *obj, char *buf, int buf_sz)
77 {
78 	struct usnic_ib_vf *vf = obj;
79 	return scnprintf(buf, buf_sz, "PF: %s ", vf->pf->ib_dev.name);
80 }
81 /* End callback dump funcs */
82 
83 static void usnic_ib_dump_vf(struct usnic_ib_vf *vf, char *buf, int buf_sz)
84 {
85 	usnic_vnic_dump(vf->vnic, buf, buf_sz, vf,
86 			usnic_ib_dump_vf_hdr,
87 			usnic_ib_qp_grp_dump_hdr, usnic_ib_qp_grp_dump_rows);
88 }
89 
90 void usnic_ib_log_vf(struct usnic_ib_vf *vf)
91 {
92 	char buf[1000];
93 	usnic_ib_dump_vf(vf, buf, sizeof(buf));
94 	usnic_dbg("%s\n", buf);
95 }
96 
97 /* Start of netdev section */
98 static inline const char *usnic_ib_netdev_event_to_string(unsigned long event)
99 {
100 	const char *event2str[] = {"NETDEV_NONE", "NETDEV_UP", "NETDEV_DOWN",
101 		"NETDEV_REBOOT", "NETDEV_CHANGE",
102 		"NETDEV_REGISTER", "NETDEV_UNREGISTER", "NETDEV_CHANGEMTU",
103 		"NETDEV_CHANGEADDR", "NETDEV_GOING_DOWN", "NETDEV_FEAT_CHANGE",
104 		"NETDEV_BONDING_FAILOVER", "NETDEV_PRE_UP",
105 		"NETDEV_PRE_TYPE_CHANGE", "NETDEV_POST_TYPE_CHANGE",
106 		"NETDEV_POST_INT", "NETDEV_UNREGISTER_FINAL", "NETDEV_RELEASE",
107 		"NETDEV_NOTIFY_PEERS", "NETDEV_JOIN"
108 	};
109 
110 	if (event >= ARRAY_SIZE(event2str))
111 		return "UNKNOWN_NETDEV_EVENT";
112 	else
113 		return event2str[event];
114 }
115 
116 static void usnic_ib_qp_grp_modify_active_to_err(struct usnic_ib_dev *us_ibdev)
117 {
118 	struct usnic_ib_ucontext *ctx;
119 	struct usnic_ib_qp_grp *qp_grp;
120 	enum ib_qp_state cur_state;
121 	int status;
122 
123 	BUG_ON(!mutex_is_locked(&us_ibdev->usdev_lock));
124 
125 	list_for_each_entry(ctx, &us_ibdev->ctx_list, link) {
126 		list_for_each_entry(qp_grp, &ctx->qp_grp_list, link) {
127 			cur_state = qp_grp->state;
128 			if (cur_state == IB_QPS_INIT ||
129 				cur_state == IB_QPS_RTR ||
130 				cur_state == IB_QPS_RTS) {
131 				status = usnic_ib_qp_grp_modify(qp_grp,
132 								IB_QPS_ERR,
133 								NULL);
134 				if (status) {
135 					usnic_err("Failed to transistion qp grp %u from %s to %s\n",
136 						qp_grp->grp_id,
137 						usnic_ib_qp_grp_state_to_string
138 						(cur_state),
139 						usnic_ib_qp_grp_state_to_string
140 						(IB_QPS_ERR));
141 				}
142 			}
143 		}
144 	}
145 }
146 
147 static void usnic_ib_handle_usdev_event(struct usnic_ib_dev *us_ibdev,
148 					unsigned long event)
149 {
150 	struct net_device *netdev;
151 	struct ib_event ib_event;
152 
153 	memset(&ib_event, 0, sizeof(ib_event));
154 
155 	mutex_lock(&us_ibdev->usdev_lock);
156 	netdev = us_ibdev->netdev;
157 	switch (event) {
158 	case NETDEV_REBOOT:
159 		usnic_info("PF Reset on %s\n", us_ibdev->ib_dev.name);
160 		usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
161 		ib_event.event = IB_EVENT_PORT_ERR;
162 		ib_event.device = &us_ibdev->ib_dev;
163 		ib_event.element.port_num = 1;
164 		ib_dispatch_event(&ib_event);
165 		break;
166 	case NETDEV_UP:
167 	case NETDEV_DOWN:
168 	case NETDEV_CHANGE:
169 		if (!us_ibdev->ufdev->link_up &&
170 				netif_carrier_ok(netdev)) {
171 			usnic_fwd_carrier_up(us_ibdev->ufdev);
172 			usnic_info("Link UP on %s\n", us_ibdev->ib_dev.name);
173 			ib_event.event = IB_EVENT_PORT_ACTIVE;
174 			ib_event.device = &us_ibdev->ib_dev;
175 			ib_event.element.port_num = 1;
176 			ib_dispatch_event(&ib_event);
177 		} else if (us_ibdev->ufdev->link_up &&
178 				!netif_carrier_ok(netdev)) {
179 			usnic_fwd_carrier_down(us_ibdev->ufdev);
180 			usnic_info("Link DOWN on %s\n", us_ibdev->ib_dev.name);
181 			usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
182 			ib_event.event = IB_EVENT_PORT_ERR;
183 			ib_event.device = &us_ibdev->ib_dev;
184 			ib_event.element.port_num = 1;
185 			ib_dispatch_event(&ib_event);
186 		} else {
187 			usnic_dbg("Ignoring %s on %s\n",
188 					usnic_ib_netdev_event_to_string(event),
189 					us_ibdev->ib_dev.name);
190 		}
191 		break;
192 	case NETDEV_CHANGEADDR:
193 		if (!memcmp(us_ibdev->ufdev->mac, netdev->dev_addr,
194 				sizeof(us_ibdev->ufdev->mac))) {
195 			usnic_dbg("Ignoring addr change on %s\n",
196 					us_ibdev->ib_dev.name);
197 		} else {
198 			usnic_info(" %s old mac: %pM new mac: %pM\n",
199 					us_ibdev->ib_dev.name,
200 					us_ibdev->ufdev->mac,
201 					netdev->dev_addr);
202 			usnic_fwd_set_mac(us_ibdev->ufdev, netdev->dev_addr);
203 			usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
204 			ib_event.event = IB_EVENT_GID_CHANGE;
205 			ib_event.device = &us_ibdev->ib_dev;
206 			ib_event.element.port_num = 1;
207 			ib_dispatch_event(&ib_event);
208 		}
209 
210 		break;
211 	case NETDEV_CHANGEMTU:
212 		if (us_ibdev->ufdev->mtu != netdev->mtu) {
213 			usnic_info("MTU Change on %s old: %u new: %u\n",
214 					us_ibdev->ib_dev.name,
215 					us_ibdev->ufdev->mtu, netdev->mtu);
216 			usnic_fwd_set_mtu(us_ibdev->ufdev, netdev->mtu);
217 			usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
218 		} else {
219 			usnic_dbg("Ignoring MTU change on %s\n",
220 					us_ibdev->ib_dev.name);
221 		}
222 		break;
223 	default:
224 		usnic_dbg("Ignoring event %s on %s",
225 				usnic_ib_netdev_event_to_string(event),
226 				us_ibdev->ib_dev.name);
227 	}
228 	mutex_unlock(&us_ibdev->usdev_lock);
229 }
230 
231 static int usnic_ib_netdevice_event(struct notifier_block *notifier,
232 					unsigned long event, void *ptr)
233 {
234 	struct usnic_ib_dev *us_ibdev;
235 
236 	struct net_device *netdev = netdev_notifier_info_to_dev(ptr);
237 
238 	mutex_lock(&usnic_ib_ibdev_list_lock);
239 	list_for_each_entry(us_ibdev, &usnic_ib_ibdev_list, ib_dev_link) {
240 		if (us_ibdev->netdev == netdev) {
241 			usnic_ib_handle_usdev_event(us_ibdev, event);
242 			break;
243 		}
244 	}
245 	mutex_unlock(&usnic_ib_ibdev_list_lock);
246 
247 	return NOTIFY_DONE;
248 }
249 
250 static struct notifier_block usnic_ib_netdevice_notifier = {
251 	.notifier_call = usnic_ib_netdevice_event
252 };
253 /* End of netdev section */
254 
255 /* Start of inet section */
256 static int usnic_ib_handle_inet_event(struct usnic_ib_dev *us_ibdev,
257 					unsigned long event, void *ptr)
258 {
259 	struct in_ifaddr *ifa = ptr;
260 	struct ib_event ib_event;
261 
262 	mutex_lock(&us_ibdev->usdev_lock);
263 
264 	switch (event) {
265 	case NETDEV_DOWN:
266 		usnic_info("%s via ip notifiers",
267 				usnic_ib_netdev_event_to_string(event));
268 		usnic_fwd_del_ipaddr(us_ibdev->ufdev);
269 		usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
270 		ib_event.event = IB_EVENT_GID_CHANGE;
271 		ib_event.device = &us_ibdev->ib_dev;
272 		ib_event.element.port_num = 1;
273 		ib_dispatch_event(&ib_event);
274 		break;
275 	case NETDEV_UP:
276 		usnic_fwd_add_ipaddr(us_ibdev->ufdev, ifa->ifa_address);
277 		usnic_info("%s via ip notifiers: ip %pI4",
278 				usnic_ib_netdev_event_to_string(event),
279 				&us_ibdev->ufdev->inaddr);
280 		ib_event.event = IB_EVENT_GID_CHANGE;
281 		ib_event.device = &us_ibdev->ib_dev;
282 		ib_event.element.port_num = 1;
283 		ib_dispatch_event(&ib_event);
284 		break;
285 	default:
286 		usnic_info("Ignoring event %s on %s",
287 				usnic_ib_netdev_event_to_string(event),
288 				us_ibdev->ib_dev.name);
289 	}
290 	mutex_unlock(&us_ibdev->usdev_lock);
291 
292 	return NOTIFY_DONE;
293 }
294 
295 static int usnic_ib_inetaddr_event(struct notifier_block *notifier,
296 					unsigned long event, void *ptr)
297 {
298 	struct usnic_ib_dev *us_ibdev;
299 	struct in_ifaddr *ifa = ptr;
300 	struct net_device *netdev = ifa->ifa_dev->dev;
301 
302 	mutex_lock(&usnic_ib_ibdev_list_lock);
303 	list_for_each_entry(us_ibdev, &usnic_ib_ibdev_list, ib_dev_link) {
304 		if (us_ibdev->netdev == netdev) {
305 			usnic_ib_handle_inet_event(us_ibdev, event, ptr);
306 			break;
307 		}
308 	}
309 	mutex_unlock(&usnic_ib_ibdev_list_lock);
310 
311 	return NOTIFY_DONE;
312 }
313 static struct notifier_block usnic_ib_inetaddr_notifier = {
314 	.notifier_call = usnic_ib_inetaddr_event
315 };
316 /* End of inet section*/
317 
318 static int usnic_port_immutable(struct ib_device *ibdev, u8 port_num,
319 			        struct ib_port_immutable *immutable)
320 {
321 	struct ib_port_attr attr;
322 	int err;
323 
324 	immutable->core_cap_flags = RDMA_CORE_PORT_USNIC;
325 
326 	err = ib_query_port(ibdev, port_num, &attr);
327 	if (err)
328 		return err;
329 
330 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
331 	immutable->gid_tbl_len = attr.gid_tbl_len;
332 
333 	return 0;
334 }
335 
336 static void usnic_get_dev_fw_str(struct ib_device *device, char *str)
337 {
338 	struct usnic_ib_dev *us_ibdev =
339 		container_of(device, struct usnic_ib_dev, ib_dev);
340 	struct ethtool_drvinfo info;
341 
342 	mutex_lock(&us_ibdev->usdev_lock);
343 	us_ibdev->netdev->ethtool_ops->get_drvinfo(us_ibdev->netdev, &info);
344 	mutex_unlock(&us_ibdev->usdev_lock);
345 
346 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%s", info.fw_version);
347 }
348 
349 /* Start of PF discovery section */
350 static void *usnic_ib_device_add(struct pci_dev *dev)
351 {
352 	struct usnic_ib_dev *us_ibdev;
353 	union ib_gid gid;
354 	struct in_device *ind;
355 	struct net_device *netdev;
356 
357 	usnic_dbg("\n");
358 	netdev = pci_get_drvdata(dev);
359 
360 	us_ibdev = (struct usnic_ib_dev *)ib_alloc_device(sizeof(*us_ibdev));
361 	if (!us_ibdev) {
362 		usnic_err("Device %s context alloc failed\n",
363 				netdev_name(pci_get_drvdata(dev)));
364 		return ERR_PTR(-EFAULT);
365 	}
366 
367 	us_ibdev->ufdev = usnic_fwd_dev_alloc(dev);
368 	if (!us_ibdev->ufdev) {
369 		usnic_err("Failed to alloc ufdev for %s\n", pci_name(dev));
370 		goto err_dealloc;
371 	}
372 
373 	mutex_init(&us_ibdev->usdev_lock);
374 	INIT_LIST_HEAD(&us_ibdev->vf_dev_list);
375 	INIT_LIST_HEAD(&us_ibdev->ctx_list);
376 
377 	us_ibdev->pdev = dev;
378 	us_ibdev->netdev = pci_get_drvdata(dev);
379 	us_ibdev->ib_dev.owner = THIS_MODULE;
380 	us_ibdev->ib_dev.node_type = RDMA_NODE_USNIC_UDP;
381 	us_ibdev->ib_dev.phys_port_cnt = USNIC_IB_PORT_CNT;
382 	us_ibdev->ib_dev.num_comp_vectors = USNIC_IB_NUM_COMP_VECTORS;
383 	us_ibdev->ib_dev.dev.parent = &dev->dev;
384 	us_ibdev->ib_dev.uverbs_abi_ver = USNIC_UVERBS_ABI_VERSION;
385 	strlcpy(us_ibdev->ib_dev.name, "usnic_%d", IB_DEVICE_NAME_MAX);
386 
387 	us_ibdev->ib_dev.uverbs_cmd_mask =
388 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
389 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
390 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
391 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
392 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
393 		(1ull << IB_USER_VERBS_CMD_REG_MR) |
394 		(1ull << IB_USER_VERBS_CMD_DEREG_MR) |
395 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
396 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
397 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
398 		(1ull << IB_USER_VERBS_CMD_CREATE_QP) |
399 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
400 		(1ull << IB_USER_VERBS_CMD_QUERY_QP) |
401 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
402 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
403 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
404 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
405 
406 	us_ibdev->ib_dev.query_device = usnic_ib_query_device;
407 	us_ibdev->ib_dev.query_port = usnic_ib_query_port;
408 	us_ibdev->ib_dev.query_pkey = usnic_ib_query_pkey;
409 	us_ibdev->ib_dev.query_gid = usnic_ib_query_gid;
410 	us_ibdev->ib_dev.get_netdev = usnic_get_netdev;
411 	us_ibdev->ib_dev.get_link_layer = usnic_ib_port_link_layer;
412 	us_ibdev->ib_dev.alloc_pd = usnic_ib_alloc_pd;
413 	us_ibdev->ib_dev.dealloc_pd = usnic_ib_dealloc_pd;
414 	us_ibdev->ib_dev.create_qp = usnic_ib_create_qp;
415 	us_ibdev->ib_dev.modify_qp = usnic_ib_modify_qp;
416 	us_ibdev->ib_dev.query_qp = usnic_ib_query_qp;
417 	us_ibdev->ib_dev.destroy_qp = usnic_ib_destroy_qp;
418 	us_ibdev->ib_dev.create_cq = usnic_ib_create_cq;
419 	us_ibdev->ib_dev.destroy_cq = usnic_ib_destroy_cq;
420 	us_ibdev->ib_dev.reg_user_mr = usnic_ib_reg_mr;
421 	us_ibdev->ib_dev.dereg_mr = usnic_ib_dereg_mr;
422 	us_ibdev->ib_dev.alloc_ucontext = usnic_ib_alloc_ucontext;
423 	us_ibdev->ib_dev.dealloc_ucontext = usnic_ib_dealloc_ucontext;
424 	us_ibdev->ib_dev.mmap = usnic_ib_mmap;
425 	us_ibdev->ib_dev.create_ah = usnic_ib_create_ah;
426 	us_ibdev->ib_dev.destroy_ah = usnic_ib_destroy_ah;
427 	us_ibdev->ib_dev.post_send = usnic_ib_post_send;
428 	us_ibdev->ib_dev.post_recv = usnic_ib_post_recv;
429 	us_ibdev->ib_dev.poll_cq = usnic_ib_poll_cq;
430 	us_ibdev->ib_dev.req_notify_cq = usnic_ib_req_notify_cq;
431 	us_ibdev->ib_dev.get_dma_mr = usnic_ib_get_dma_mr;
432 	us_ibdev->ib_dev.get_port_immutable = usnic_port_immutable;
433 	us_ibdev->ib_dev.get_dev_fw_str     = usnic_get_dev_fw_str;
434 
435 
436 	if (ib_register_device(&us_ibdev->ib_dev, NULL))
437 		goto err_fwd_dealloc;
438 
439 	usnic_fwd_set_mtu(us_ibdev->ufdev, us_ibdev->netdev->mtu);
440 	usnic_fwd_set_mac(us_ibdev->ufdev, us_ibdev->netdev->dev_addr);
441 	if (netif_carrier_ok(us_ibdev->netdev))
442 		usnic_fwd_carrier_up(us_ibdev->ufdev);
443 
444 	ind = in_dev_get(netdev);
445 	if (ind->ifa_list)
446 		usnic_fwd_add_ipaddr(us_ibdev->ufdev,
447 				     ind->ifa_list->ifa_address);
448 	in_dev_put(ind);
449 
450 	usnic_mac_ip_to_gid(us_ibdev->netdev->perm_addr,
451 				us_ibdev->ufdev->inaddr, &gid.raw[0]);
452 	memcpy(&us_ibdev->ib_dev.node_guid, &gid.global.interface_id,
453 		sizeof(gid.global.interface_id));
454 	kref_init(&us_ibdev->vf_cnt);
455 
456 	usnic_info("Added ibdev: %s netdev: %s with mac %pM Link: %u MTU: %u\n",
457 			us_ibdev->ib_dev.name, netdev_name(us_ibdev->netdev),
458 			us_ibdev->ufdev->mac, us_ibdev->ufdev->link_up,
459 			us_ibdev->ufdev->mtu);
460 	return us_ibdev;
461 
462 err_fwd_dealloc:
463 	usnic_fwd_dev_free(us_ibdev->ufdev);
464 err_dealloc:
465 	usnic_err("failed -- deallocing device\n");
466 	ib_dealloc_device(&us_ibdev->ib_dev);
467 	return NULL;
468 }
469 
470 static void usnic_ib_device_remove(struct usnic_ib_dev *us_ibdev)
471 {
472 	usnic_info("Unregistering %s\n", us_ibdev->ib_dev.name);
473 	usnic_ib_sysfs_unregister_usdev(us_ibdev);
474 	usnic_fwd_dev_free(us_ibdev->ufdev);
475 	ib_unregister_device(&us_ibdev->ib_dev);
476 	ib_dealloc_device(&us_ibdev->ib_dev);
477 }
478 
479 static void usnic_ib_undiscover_pf(struct kref *kref)
480 {
481 	struct usnic_ib_dev *us_ibdev, *tmp;
482 	struct pci_dev *dev;
483 	bool found = false;
484 
485 	dev = container_of(kref, struct usnic_ib_dev, vf_cnt)->pdev;
486 	mutex_lock(&usnic_ib_ibdev_list_lock);
487 	list_for_each_entry_safe(us_ibdev, tmp,
488 				&usnic_ib_ibdev_list, ib_dev_link) {
489 		if (us_ibdev->pdev == dev) {
490 			list_del(&us_ibdev->ib_dev_link);
491 			usnic_ib_device_remove(us_ibdev);
492 			found = true;
493 			break;
494 		}
495 	}
496 
497 	WARN(!found, "Failed to remove PF %s\n", pci_name(dev));
498 
499 	mutex_unlock(&usnic_ib_ibdev_list_lock);
500 }
501 
502 static struct usnic_ib_dev *usnic_ib_discover_pf(struct usnic_vnic *vnic)
503 {
504 	struct usnic_ib_dev *us_ibdev;
505 	struct pci_dev *parent_pci, *vf_pci;
506 	int err;
507 
508 	vf_pci = usnic_vnic_get_pdev(vnic);
509 	parent_pci = pci_physfn(vf_pci);
510 
511 	BUG_ON(!parent_pci);
512 
513 	mutex_lock(&usnic_ib_ibdev_list_lock);
514 	list_for_each_entry(us_ibdev, &usnic_ib_ibdev_list, ib_dev_link) {
515 		if (us_ibdev->pdev == parent_pci) {
516 			kref_get(&us_ibdev->vf_cnt);
517 			goto out;
518 		}
519 	}
520 
521 	us_ibdev = usnic_ib_device_add(parent_pci);
522 	if (IS_ERR_OR_NULL(us_ibdev)) {
523 		us_ibdev = us_ibdev ? us_ibdev : ERR_PTR(-EFAULT);
524 		goto out;
525 	}
526 
527 	err = usnic_ib_sysfs_register_usdev(us_ibdev);
528 	if (err) {
529 		usnic_ib_device_remove(us_ibdev);
530 		us_ibdev = ERR_PTR(err);
531 		goto out;
532 	}
533 
534 	list_add(&us_ibdev->ib_dev_link, &usnic_ib_ibdev_list);
535 out:
536 	mutex_unlock(&usnic_ib_ibdev_list_lock);
537 	return us_ibdev;
538 }
539 /* End of PF discovery section */
540 
541 /* Start of PCI section */
542 
543 static const struct pci_device_id usnic_ib_pci_ids[] = {
544 	{PCI_DEVICE(PCI_VENDOR_ID_CISCO, PCI_DEVICE_ID_CISCO_VIC_USPACE_NIC)},
545 	{0,}
546 };
547 
548 static int usnic_ib_pci_probe(struct pci_dev *pdev,
549 				const struct pci_device_id *id)
550 {
551 	int err;
552 	struct usnic_ib_dev *pf;
553 	struct usnic_ib_vf *vf;
554 	enum usnic_vnic_res_type res_type;
555 
556 	vf = kzalloc(sizeof(*vf), GFP_KERNEL);
557 	if (!vf)
558 		return -ENOMEM;
559 
560 	err = pci_enable_device(pdev);
561 	if (err) {
562 		usnic_err("Failed to enable %s with err %d\n",
563 				pci_name(pdev), err);
564 		goto out_clean_vf;
565 	}
566 
567 	err = pci_request_regions(pdev, DRV_NAME);
568 	if (err) {
569 		usnic_err("Failed to request region for %s with err %d\n",
570 				pci_name(pdev), err);
571 		goto out_disable_device;
572 	}
573 
574 	pci_set_master(pdev);
575 	pci_set_drvdata(pdev, vf);
576 
577 	vf->vnic = usnic_vnic_alloc(pdev);
578 	if (IS_ERR_OR_NULL(vf->vnic)) {
579 		err = vf->vnic ? PTR_ERR(vf->vnic) : -ENOMEM;
580 		usnic_err("Failed to alloc vnic for %s with err %d\n",
581 				pci_name(pdev), err);
582 		goto out_release_regions;
583 	}
584 
585 	pf = usnic_ib_discover_pf(vf->vnic);
586 	if (IS_ERR_OR_NULL(pf)) {
587 		usnic_err("Failed to discover pf of vnic %s with err%ld\n",
588 				pci_name(pdev), PTR_ERR(pf));
589 		err = pf ? PTR_ERR(pf) : -EFAULT;
590 		goto out_clean_vnic;
591 	}
592 
593 	vf->pf = pf;
594 	spin_lock_init(&vf->lock);
595 	mutex_lock(&pf->usdev_lock);
596 	list_add_tail(&vf->link, &pf->vf_dev_list);
597 	/*
598 	 * Save max settings (will be same for each VF, easier to re-write than
599 	 * to say "if (!set) { set_values(); set=1; }
600 	 */
601 	for (res_type = USNIC_VNIC_RES_TYPE_EOL+1;
602 			res_type < USNIC_VNIC_RES_TYPE_MAX;
603 			res_type++) {
604 		pf->vf_res_cnt[res_type] = usnic_vnic_res_cnt(vf->vnic,
605 								res_type);
606 	}
607 
608 	mutex_unlock(&pf->usdev_lock);
609 
610 	usnic_info("Registering usnic VF %s into PF %s\n", pci_name(pdev),
611 			pf->ib_dev.name);
612 	usnic_ib_log_vf(vf);
613 	return 0;
614 
615 out_clean_vnic:
616 	usnic_vnic_free(vf->vnic);
617 out_release_regions:
618 	pci_set_drvdata(pdev, NULL);
619 	pci_clear_master(pdev);
620 	pci_release_regions(pdev);
621 out_disable_device:
622 	pci_disable_device(pdev);
623 out_clean_vf:
624 	kfree(vf);
625 	return err;
626 }
627 
628 static void usnic_ib_pci_remove(struct pci_dev *pdev)
629 {
630 	struct usnic_ib_vf *vf = pci_get_drvdata(pdev);
631 	struct usnic_ib_dev *pf = vf->pf;
632 
633 	mutex_lock(&pf->usdev_lock);
634 	list_del(&vf->link);
635 	mutex_unlock(&pf->usdev_lock);
636 
637 	kref_put(&pf->vf_cnt, usnic_ib_undiscover_pf);
638 	usnic_vnic_free(vf->vnic);
639 	pci_set_drvdata(pdev, NULL);
640 	pci_clear_master(pdev);
641 	pci_release_regions(pdev);
642 	pci_disable_device(pdev);
643 	kfree(vf);
644 
645 	usnic_info("Removed VF %s\n", pci_name(pdev));
646 }
647 
648 /* PCI driver entry points */
649 static struct pci_driver usnic_ib_pci_driver = {
650 	.name = DRV_NAME,
651 	.id_table = usnic_ib_pci_ids,
652 	.probe = usnic_ib_pci_probe,
653 	.remove = usnic_ib_pci_remove,
654 };
655 /* End of PCI section */
656 
657 /* Start of module section */
658 static int __init usnic_ib_init(void)
659 {
660 	int err;
661 
662 	printk_once(KERN_INFO "%s", usnic_version);
663 
664 	err = usnic_uiom_init(DRV_NAME);
665 	if (err) {
666 		usnic_err("Unable to initalize umem with err %d\n", err);
667 		return err;
668 	}
669 
670 	err = pci_register_driver(&usnic_ib_pci_driver);
671 	if (err) {
672 		usnic_err("Unable to register with PCI\n");
673 		goto out_umem_fini;
674 	}
675 
676 	err = register_netdevice_notifier(&usnic_ib_netdevice_notifier);
677 	if (err) {
678 		usnic_err("Failed to register netdev notifier\n");
679 		goto out_pci_unreg;
680 	}
681 
682 	err = register_inetaddr_notifier(&usnic_ib_inetaddr_notifier);
683 	if (err) {
684 		usnic_err("Failed to register inet addr notifier\n");
685 		goto out_unreg_netdev_notifier;
686 	}
687 
688 	err = usnic_transport_init();
689 	if (err) {
690 		usnic_err("Failed to initialize transport\n");
691 		goto out_unreg_inetaddr_notifier;
692 	}
693 
694 	usnic_debugfs_init();
695 
696 	return 0;
697 
698 out_unreg_inetaddr_notifier:
699 	unregister_inetaddr_notifier(&usnic_ib_inetaddr_notifier);
700 out_unreg_netdev_notifier:
701 	unregister_netdevice_notifier(&usnic_ib_netdevice_notifier);
702 out_pci_unreg:
703 	pci_unregister_driver(&usnic_ib_pci_driver);
704 out_umem_fini:
705 	usnic_uiom_fini();
706 
707 	return err;
708 }
709 
710 static void __exit usnic_ib_destroy(void)
711 {
712 	usnic_dbg("\n");
713 	usnic_debugfs_exit();
714 	usnic_transport_fini();
715 	unregister_inetaddr_notifier(&usnic_ib_inetaddr_notifier);
716 	unregister_netdevice_notifier(&usnic_ib_netdevice_notifier);
717 	pci_unregister_driver(&usnic_ib_pci_driver);
718 	usnic_uiom_fini();
719 }
720 
721 MODULE_DESCRIPTION("Cisco VIC (usNIC) Verbs Driver");
722 MODULE_AUTHOR("Upinder Malhi <umalhi@cisco.com>");
723 MODULE_LICENSE("Dual BSD/GPL");
724 module_param(usnic_log_lvl, uint, S_IRUGO | S_IWUSR);
725 module_param(usnic_ib_share_vf, uint, S_IRUGO | S_IWUSR);
726 MODULE_PARM_DESC(usnic_log_lvl, " Off=0, Err=1, Info=2, Debug=3");
727 MODULE_PARM_DESC(usnic_ib_share_vf, "Off=0, On=1 VF sharing amongst QPs");
728 MODULE_DEVICE_TABLE(pci, usnic_ib_pci_ids);
729 
730 module_init(usnic_ib_init);
731 module_exit(usnic_ib_destroy);
732 /* End of module section */
733