1 /*
2  * Copyright(c) 2017 Intel Corporation.
3  * Copyright(c) 2021 Cornelis Networks.
4  *
5  * This file is provided under a dual BSD/GPLv2 license.  When using or
6  * redistributing this file, you may do so under either license.
7  *
8  * GPL LICENSE SUMMARY
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of version 2 of the GNU General Public License as
12  * published by the Free Software Foundation.
13  *
14  * This program is distributed in the hope that it will be useful, but
15  * WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  *
19  * BSD LICENSE
20  *
21  * Redistribution and use in source and binary forms, with or without
22  * modification, are permitted provided that the following conditions
23  * are met:
24  *
25  *  - Redistributions of source code must retain the above copyright
26  *    notice, this list of conditions and the following disclaimer.
27  *  - Redistributions in binary form must reproduce the above copyright
28  *    notice, this list of conditions and the following disclaimer in
29  *    the documentation and/or other materials provided with the
30  *    distribution.
31  *  - Neither the name of Intel Corporation nor the names of its
32  *    contributors may be used to endorse or promote products derived
33  *    from this software without specific prior written permission.
34  *
35  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
36  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
37  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
38  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
39  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
40  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
41  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
42  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
43  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
44  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
45  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
46  *
47  */
48 
49 /*
50  * This file contains OPX Virtual Network Interface Controller (VNIC)
51  * Ethernet Management Agent (EMA) driver
52  */
53 
54 #include <linux/module.h>
55 #include <linux/xarray.h>
56 #include <rdma/ib_addr.h>
57 #include <rdma/ib_verbs.h>
58 #include <rdma/opa_smi.h>
59 #include <rdma/opa_port_info.h>
60 
61 #include "opa_vnic_internal.h"
62 
63 char opa_vnic_driver_name[] = "opa_vnic";
64 
65 /*
66  * The trap service level is kept in bits 3 to 7 in the trap_sl_rsvd
67  * field in the class port info MAD.
68  */
69 #define GET_TRAP_SL_FROM_CLASS_PORT_INFO(x)  (((x) >> 3) & 0x1f)
70 
71 /* Cap trap bursts to a reasonable limit good for normal cases */
72 #define OPA_VNIC_TRAP_BURST_LIMIT 4
73 
74 /*
75  * VNIC trap limit timeout.
76  * Inverse of cap2_mask response time out (1.0737 secs) = 0.9
77  * secs approx IB spec 13.4.6.2.1 PortInfoSubnetTimeout and
78  * 13.4.9 Traps.
79  */
80 #define OPA_VNIC_TRAP_TIMEOUT  ((4096 * (1UL << 18)) / 1000)
81 
82 #define OPA_VNIC_UNSUP_ATTR  \
83 		cpu_to_be16(IB_MGMT_MAD_STATUS_UNSUPPORTED_METHOD_ATTRIB)
84 
85 #define OPA_VNIC_INVAL_ATTR  \
86 		cpu_to_be16(IB_MGMT_MAD_STATUS_INVALID_ATTRIB_VALUE)
87 
88 #define OPA_VNIC_CLASS_CAP_TRAP   0x1
89 
90 /* Maximum number of VNIC ports supported */
91 #define OPA_VNIC_MAX_NUM_VPORT    255
92 
93 /**
94  * struct opa_vnic_vema_port -- VNIC VEMA port details
95  * @cport: pointer to port
96  * @mad_agent: pointer to mad agent for port
97  * @class_port_info: Class port info information.
98  * @tid: Transaction id
99  * @port_num: OPA port number
100  * @vports: vnic ports
101  * @event_handler: ib event handler
102  * @lock: adapter interface lock
103  */
104 struct opa_vnic_vema_port {
105 	struct opa_vnic_ctrl_port      *cport;
106 	struct ib_mad_agent            *mad_agent;
107 	struct opa_class_port_info      class_port_info;
108 	u64                             tid;
109 	u8                              port_num;
110 	struct xarray                   vports;
111 	struct ib_event_handler         event_handler;
112 
113 	/* Lock to query/update network adapter */
114 	struct mutex                    lock;
115 };
116 
117 static int opa_vnic_vema_add_one(struct ib_device *device);
118 static void opa_vnic_vema_rem_one(struct ib_device *device,
119 				  void *client_data);
120 
121 static struct ib_client opa_vnic_client = {
122 	.name   = opa_vnic_driver_name,
123 	.add    = opa_vnic_vema_add_one,
124 	.remove = opa_vnic_vema_rem_one,
125 };
126 
127 /**
128  * vema_get_vport_num -- Get the vnic from the mad
129  * @recvd_mad:  Received mad
130  *
131  * Return: returns value of the vnic port number
132  */
133 static inline u8 vema_get_vport_num(struct opa_vnic_vema_mad *recvd_mad)
134 {
135 	return be32_to_cpu(recvd_mad->mad_hdr.attr_mod) & 0xff;
136 }
137 
138 /**
139  * vema_get_vport_adapter -- Get vnic port adapter from recvd mad
140  * @recvd_mad: received mad
141  * @port: ptr to port struct on which MAD was recvd
142  *
143  * Return: vnic adapter
144  */
145 static inline struct opa_vnic_adapter *
146 vema_get_vport_adapter(struct opa_vnic_vema_mad *recvd_mad,
147 		       struct opa_vnic_vema_port *port)
148 {
149 	u8 vport_num = vema_get_vport_num(recvd_mad);
150 
151 	return xa_load(&port->vports, vport_num);
152 }
153 
154 /**
155  * vema_mac_tbl_req_ok -- Check if mac request has correct values
156  * @mac_tbl: mac table
157  *
158  * This function checks for the validity of the offset and number of
159  * entries required.
160  *
161  * Return: true if offset and num_entries are valid
162  */
163 static inline bool vema_mac_tbl_req_ok(struct opa_veswport_mactable *mac_tbl)
164 {
165 	u16 offset, num_entries;
166 	u16 req_entries = ((OPA_VNIC_EMA_DATA - sizeof(*mac_tbl)) /
167 			   sizeof(mac_tbl->tbl_entries[0]));
168 
169 	offset = be16_to_cpu(mac_tbl->offset);
170 	num_entries = be16_to_cpu(mac_tbl->num_entries);
171 
172 	return ((num_entries <= req_entries) &&
173 		(offset + num_entries <= OPA_VNIC_MAC_TBL_MAX_ENTRIES));
174 }
175 
176 /*
177  * Return the power on default values in the port info structure
178  * in big endian format as required by MAD.
179  */
180 static inline void vema_get_pod_values(struct opa_veswport_info *port_info)
181 {
182 	memset(port_info, 0, sizeof(*port_info));
183 	port_info->vport.max_mac_tbl_ent =
184 		cpu_to_be16(OPA_VNIC_MAC_TBL_MAX_ENTRIES);
185 	port_info->vport.max_smac_ent =
186 		cpu_to_be16(OPA_VNIC_MAX_SMAC_LIMIT);
187 	port_info->vport.oper_state = OPA_VNIC_STATE_DROP_ALL;
188 	port_info->vport.config_state = OPA_VNIC_STATE_DROP_ALL;
189 	port_info->vesw.eth_mtu = cpu_to_be16(ETH_DATA_LEN);
190 }
191 
192 /**
193  * vema_add_vport -- Add a new vnic port
194  * @port: ptr to opa_vnic_vema_port struct
195  * @vport_num: vnic port number (to be added)
196  *
197  * Return a pointer to the vnic adapter structure
198  */
199 static struct opa_vnic_adapter *vema_add_vport(struct opa_vnic_vema_port *port,
200 					       u8 vport_num)
201 {
202 	struct opa_vnic_ctrl_port *cport = port->cport;
203 	struct opa_vnic_adapter *adapter;
204 
205 	adapter = opa_vnic_add_netdev(cport->ibdev, port->port_num, vport_num);
206 	if (!IS_ERR(adapter)) {
207 		int rc;
208 
209 		adapter->cport = cport;
210 		rc = xa_insert(&port->vports, vport_num, adapter, GFP_KERNEL);
211 		if (rc < 0) {
212 			opa_vnic_rem_netdev(adapter);
213 			adapter = ERR_PTR(rc);
214 		}
215 	}
216 
217 	return adapter;
218 }
219 
220 /**
221  * vema_get_class_port_info -- Get class info for port
222  * @port:  Port on whic MAD was received
223  * @recvd_mad: pointer to the received mad
224  * @rsp_mad:   pointer to respose mad
225  *
226  * This function copies the latest class port info value set for the
227  * port and stores it for generating traps
228  */
229 static void vema_get_class_port_info(struct opa_vnic_vema_port *port,
230 				     struct opa_vnic_vema_mad *recvd_mad,
231 				     struct opa_vnic_vema_mad *rsp_mad)
232 {
233 	struct opa_class_port_info *port_info;
234 
235 	port_info = (struct opa_class_port_info *)rsp_mad->data;
236 	memcpy(port_info, &port->class_port_info, sizeof(*port_info));
237 	port_info->base_version = OPA_MGMT_BASE_VERSION;
238 	port_info->class_version = OPA_EMA_CLASS_VERSION;
239 
240 	/*
241 	 * Set capability mask bit indicating agent generates traps,
242 	 * and set the maximum number of VNIC ports supported.
243 	 */
244 	port_info->cap_mask = cpu_to_be16((OPA_VNIC_CLASS_CAP_TRAP |
245 					   (OPA_VNIC_MAX_NUM_VPORT << 8)));
246 
247 	/*
248 	 * Since a get routine is always sent by the EM first we
249 	 * set the expected response time to
250 	 * 4.096 usec * 2^18 == 1.0737 sec here.
251 	 */
252 	port_info->cap_mask2_resp_time = cpu_to_be32(18);
253 }
254 
255 /**
256  * vema_set_class_port_info -- Get class info for port
257  * @port:  Port on whic MAD was received
258  * @recvd_mad: pointer to the received mad
259  * @rsp_mad:   pointer to respose mad
260  *
261  * This function updates the port class info for the specific vnic
262  * and sets up the response mad data
263  */
264 static void vema_set_class_port_info(struct opa_vnic_vema_port *port,
265 				     struct opa_vnic_vema_mad *recvd_mad,
266 				     struct opa_vnic_vema_mad *rsp_mad)
267 {
268 	memcpy(&port->class_port_info, recvd_mad->data,
269 	       sizeof(port->class_port_info));
270 
271 	vema_get_class_port_info(port, recvd_mad, rsp_mad);
272 }
273 
274 /**
275  * vema_get_veswport_info -- Get veswport info
276  * @port:      source port on which MAD was received
277  * @recvd_mad: pointer to the received mad
278  * @rsp_mad:   pointer to respose mad
279  */
280 static void vema_get_veswport_info(struct opa_vnic_vema_port *port,
281 				   struct opa_vnic_vema_mad *recvd_mad,
282 				   struct opa_vnic_vema_mad *rsp_mad)
283 {
284 	struct opa_veswport_info *port_info =
285 				  (struct opa_veswport_info *)rsp_mad->data;
286 	struct opa_vnic_adapter *adapter;
287 
288 	adapter = vema_get_vport_adapter(recvd_mad, port);
289 	if (adapter) {
290 		memset(port_info, 0, sizeof(*port_info));
291 		opa_vnic_get_vesw_info(adapter, &port_info->vesw);
292 		opa_vnic_get_per_veswport_info(adapter,
293 					       &port_info->vport);
294 	} else {
295 		vema_get_pod_values(port_info);
296 	}
297 }
298 
299 /**
300  * vema_set_veswport_info -- Set veswport info
301  * @port:      source port on which MAD was received
302  * @recvd_mad: pointer to the received mad
303  * @rsp_mad:   pointer to respose mad
304  *
305  * This function gets the port class infor for vnic
306  */
307 static void vema_set_veswport_info(struct opa_vnic_vema_port *port,
308 				   struct opa_vnic_vema_mad *recvd_mad,
309 				   struct opa_vnic_vema_mad *rsp_mad)
310 {
311 	struct opa_vnic_ctrl_port *cport = port->cport;
312 	struct opa_veswport_info *port_info;
313 	struct opa_vnic_adapter *adapter;
314 	u8 vport_num;
315 
316 	vport_num = vema_get_vport_num(recvd_mad);
317 
318 	adapter = vema_get_vport_adapter(recvd_mad, port);
319 	if (!adapter) {
320 		adapter = vema_add_vport(port, vport_num);
321 		if (IS_ERR(adapter)) {
322 			c_err("failed to add vport %d: %ld\n",
323 			      vport_num, PTR_ERR(adapter));
324 			goto err_exit;
325 		}
326 	}
327 
328 	port_info = (struct opa_veswport_info *)recvd_mad->data;
329 	opa_vnic_set_vesw_info(adapter, &port_info->vesw);
330 	opa_vnic_set_per_veswport_info(adapter, &port_info->vport);
331 
332 	/* Process the new config settings */
333 	opa_vnic_process_vema_config(adapter);
334 
335 	vema_get_veswport_info(port, recvd_mad, rsp_mad);
336 	return;
337 
338 err_exit:
339 	rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
340 }
341 
342 /**
343  * vema_get_mac_entries -- Get MAC entries in VNIC MAC table
344  * @port:      source port on which MAD was received
345  * @recvd_mad: pointer to the received mad
346  * @rsp_mad:   pointer to respose mad
347  *
348  * This function gets the MAC entries that are programmed into
349  * the VNIC MAC forwarding table. It checks for the validity of
350  * the index into the MAC table and the number of entries that
351  * are to be retrieved.
352  */
353 static void vema_get_mac_entries(struct opa_vnic_vema_port *port,
354 				 struct opa_vnic_vema_mad *recvd_mad,
355 				 struct opa_vnic_vema_mad *rsp_mad)
356 {
357 	struct opa_veswport_mactable *mac_tbl_in, *mac_tbl_out;
358 	struct opa_vnic_adapter *adapter;
359 
360 	adapter = vema_get_vport_adapter(recvd_mad, port);
361 	if (!adapter) {
362 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
363 		return;
364 	}
365 
366 	mac_tbl_in = (struct opa_veswport_mactable *)recvd_mad->data;
367 	mac_tbl_out = (struct opa_veswport_mactable *)rsp_mad->data;
368 
369 	if (vema_mac_tbl_req_ok(mac_tbl_in)) {
370 		mac_tbl_out->offset = mac_tbl_in->offset;
371 		mac_tbl_out->num_entries = mac_tbl_in->num_entries;
372 		opa_vnic_query_mac_tbl(adapter, mac_tbl_out);
373 	} else {
374 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
375 	}
376 }
377 
378 /**
379  * vema_set_mac_entries -- Set MAC entries in VNIC MAC table
380  * @port:      source port on which MAD was received
381  * @recvd_mad: pointer to the received mad
382  * @rsp_mad:   pointer to respose mad
383  *
384  * This function sets the MAC entries in the VNIC forwarding table
385  * It checks for the validity of the index and the number of forwarding
386  * table entries to be programmed.
387  */
388 static void vema_set_mac_entries(struct opa_vnic_vema_port *port,
389 				 struct opa_vnic_vema_mad *recvd_mad,
390 				 struct opa_vnic_vema_mad *rsp_mad)
391 {
392 	struct opa_veswport_mactable *mac_tbl;
393 	struct opa_vnic_adapter *adapter;
394 
395 	adapter = vema_get_vport_adapter(recvd_mad, port);
396 	if (!adapter) {
397 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
398 		return;
399 	}
400 
401 	mac_tbl = (struct opa_veswport_mactable *)recvd_mad->data;
402 	if (vema_mac_tbl_req_ok(mac_tbl)) {
403 		if (opa_vnic_update_mac_tbl(adapter, mac_tbl))
404 			rsp_mad->mad_hdr.status = OPA_VNIC_UNSUP_ATTR;
405 	} else {
406 		rsp_mad->mad_hdr.status = OPA_VNIC_UNSUP_ATTR;
407 	}
408 	vema_get_mac_entries(port, recvd_mad, rsp_mad);
409 }
410 
411 /**
412  * vema_set_delete_vesw -- Reset VESW info to POD values
413  * @port:      source port on which MAD was received
414  * @recvd_mad: pointer to the received mad
415  * @rsp_mad:   pointer to respose mad
416  *
417  * This function clears all the fields of veswport info for the requested vesw
418  * and sets them back to the power-on default values. It does not delete the
419  * vesw.
420  */
421 static void vema_set_delete_vesw(struct opa_vnic_vema_port *port,
422 				 struct opa_vnic_vema_mad *recvd_mad,
423 				 struct opa_vnic_vema_mad *rsp_mad)
424 {
425 	struct opa_veswport_info *port_info =
426 				  (struct opa_veswport_info *)rsp_mad->data;
427 	struct opa_vnic_adapter *adapter;
428 
429 	adapter = vema_get_vport_adapter(recvd_mad, port);
430 	if (!adapter) {
431 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
432 		return;
433 	}
434 
435 	vema_get_pod_values(port_info);
436 	opa_vnic_set_vesw_info(adapter, &port_info->vesw);
437 	opa_vnic_set_per_veswport_info(adapter, &port_info->vport);
438 
439 	/* Process the new config settings */
440 	opa_vnic_process_vema_config(adapter);
441 
442 	opa_vnic_release_mac_tbl(adapter);
443 
444 	vema_get_veswport_info(port, recvd_mad, rsp_mad);
445 }
446 
447 /**
448  * vema_get_mac_list -- Get the unicast/multicast macs.
449  * @port:      source port on which MAD was received
450  * @recvd_mad: Received mad contains fields to set vnic parameters
451  * @rsp_mad:   Response mad to be built
452  * @attr_id:   Attribute ID indicating multicast or unicast mac list
453  */
454 static void vema_get_mac_list(struct opa_vnic_vema_port *port,
455 			      struct opa_vnic_vema_mad *recvd_mad,
456 			      struct opa_vnic_vema_mad *rsp_mad,
457 			      u16 attr_id)
458 {
459 	struct opa_veswport_iface_macs *macs_in, *macs_out;
460 	int max_entries = (OPA_VNIC_EMA_DATA - sizeof(*macs_out)) / ETH_ALEN;
461 	struct opa_vnic_adapter *adapter;
462 
463 	adapter = vema_get_vport_adapter(recvd_mad, port);
464 	if (!adapter) {
465 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
466 		return;
467 	}
468 
469 	macs_in = (struct opa_veswport_iface_macs *)recvd_mad->data;
470 	macs_out = (struct opa_veswport_iface_macs *)rsp_mad->data;
471 
472 	macs_out->start_idx = macs_in->start_idx;
473 	if (macs_in->num_macs_in_msg)
474 		macs_out->num_macs_in_msg = macs_in->num_macs_in_msg;
475 	else
476 		macs_out->num_macs_in_msg = cpu_to_be16(max_entries);
477 
478 	if (attr_id == OPA_EM_ATTR_IFACE_MCAST_MACS)
479 		opa_vnic_query_mcast_macs(adapter, macs_out);
480 	else
481 		opa_vnic_query_ucast_macs(adapter, macs_out);
482 }
483 
484 /**
485  * vema_get_summary_counters -- Gets summary counters.
486  * @port:      source port on which MAD was received
487  * @recvd_mad: Received mad contains fields to set vnic parameters
488  * @rsp_mad:   Response mad to be built
489  */
490 static void vema_get_summary_counters(struct opa_vnic_vema_port *port,
491 				      struct opa_vnic_vema_mad *recvd_mad,
492 				      struct opa_vnic_vema_mad *rsp_mad)
493 {
494 	struct opa_veswport_summary_counters *cntrs;
495 	struct opa_vnic_adapter *adapter;
496 
497 	adapter = vema_get_vport_adapter(recvd_mad, port);
498 	if (adapter) {
499 		cntrs = (struct opa_veswport_summary_counters *)rsp_mad->data;
500 		opa_vnic_get_summary_counters(adapter, cntrs);
501 	} else {
502 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
503 	}
504 }
505 
506 /**
507  * vema_get_error_counters -- Gets summary counters.
508  * @port:      source port on which MAD was received
509  * @recvd_mad: Received mad contains fields to set vnic parameters
510  * @rsp_mad:   Response mad to be built
511  */
512 static void vema_get_error_counters(struct opa_vnic_vema_port *port,
513 				    struct opa_vnic_vema_mad *recvd_mad,
514 				    struct opa_vnic_vema_mad *rsp_mad)
515 {
516 	struct opa_veswport_error_counters *cntrs;
517 	struct opa_vnic_adapter *adapter;
518 
519 	adapter = vema_get_vport_adapter(recvd_mad, port);
520 	if (adapter) {
521 		cntrs = (struct opa_veswport_error_counters *)rsp_mad->data;
522 		opa_vnic_get_error_counters(adapter, cntrs);
523 	} else {
524 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
525 	}
526 }
527 
528 /**
529  * vema_get -- Process received get MAD
530  * @port:      source port on which MAD was received
531  * @recvd_mad: Received mad
532  * @rsp_mad:   Response mad to be built
533  */
534 static void vema_get(struct opa_vnic_vema_port *port,
535 		     struct opa_vnic_vema_mad *recvd_mad,
536 		     struct opa_vnic_vema_mad *rsp_mad)
537 {
538 	u16 attr_id = be16_to_cpu(recvd_mad->mad_hdr.attr_id);
539 
540 	switch (attr_id) {
541 	case OPA_EM_ATTR_CLASS_PORT_INFO:
542 		vema_get_class_port_info(port, recvd_mad, rsp_mad);
543 		break;
544 	case OPA_EM_ATTR_VESWPORT_INFO:
545 		vema_get_veswport_info(port, recvd_mad, rsp_mad);
546 		break;
547 	case OPA_EM_ATTR_VESWPORT_MAC_ENTRIES:
548 		vema_get_mac_entries(port, recvd_mad, rsp_mad);
549 		break;
550 	case OPA_EM_ATTR_IFACE_UCAST_MACS:
551 	case OPA_EM_ATTR_IFACE_MCAST_MACS:
552 		vema_get_mac_list(port, recvd_mad, rsp_mad, attr_id);
553 		break;
554 	case OPA_EM_ATTR_VESWPORT_SUMMARY_COUNTERS:
555 		vema_get_summary_counters(port, recvd_mad, rsp_mad);
556 		break;
557 	case OPA_EM_ATTR_VESWPORT_ERROR_COUNTERS:
558 		vema_get_error_counters(port, recvd_mad, rsp_mad);
559 		break;
560 	default:
561 		rsp_mad->mad_hdr.status = OPA_VNIC_UNSUP_ATTR;
562 		break;
563 	}
564 }
565 
566 /**
567  * vema_set -- Process received set MAD
568  * @port:      source port on which MAD was received
569  * @recvd_mad: Received mad contains fields to set vnic parameters
570  * @rsp_mad:   Response mad to be built
571  */
572 static void vema_set(struct opa_vnic_vema_port *port,
573 		     struct opa_vnic_vema_mad *recvd_mad,
574 		     struct opa_vnic_vema_mad *rsp_mad)
575 {
576 	u16 attr_id = be16_to_cpu(recvd_mad->mad_hdr.attr_id);
577 
578 	switch (attr_id) {
579 	case OPA_EM_ATTR_CLASS_PORT_INFO:
580 		vema_set_class_port_info(port, recvd_mad, rsp_mad);
581 		break;
582 	case OPA_EM_ATTR_VESWPORT_INFO:
583 		vema_set_veswport_info(port, recvd_mad, rsp_mad);
584 		break;
585 	case OPA_EM_ATTR_VESWPORT_MAC_ENTRIES:
586 		vema_set_mac_entries(port, recvd_mad, rsp_mad);
587 		break;
588 	case OPA_EM_ATTR_DELETE_VESW:
589 		vema_set_delete_vesw(port, recvd_mad, rsp_mad);
590 		break;
591 	default:
592 		rsp_mad->mad_hdr.status = OPA_VNIC_UNSUP_ATTR;
593 		break;
594 	}
595 }
596 
597 /**
598  * vema_send -- Send handler for VEMA MAD agent
599  * @mad_agent: pointer to the mad agent
600  * @mad_wc:    pointer to mad send work completion information
601  *
602  * Free all the data structures associated with the sent MAD
603  */
604 static void vema_send(struct ib_mad_agent *mad_agent,
605 		      struct ib_mad_send_wc *mad_wc)
606 {
607 	rdma_destroy_ah(mad_wc->send_buf->ah, RDMA_DESTROY_AH_SLEEPABLE);
608 	ib_free_send_mad(mad_wc->send_buf);
609 }
610 
611 /**
612  * vema_recv -- Recv handler for VEMA MAD agent
613  * @mad_agent: pointer to the mad agent
614  * @send_buf: Send buffer if found, else NULL
615  * @mad_wc:    pointer to mad send work completion information
616  *
617  * Handle only set and get methods and respond to other methods
618  * as unsupported. Allocate response buffer and address handle
619  * for the response MAD.
620  */
621 static void vema_recv(struct ib_mad_agent *mad_agent,
622 		      struct ib_mad_send_buf *send_buf,
623 		      struct ib_mad_recv_wc *mad_wc)
624 {
625 	struct opa_vnic_vema_port *port;
626 	struct ib_ah              *ah;
627 	struct ib_mad_send_buf    *rsp;
628 	struct opa_vnic_vema_mad  *vema_mad;
629 
630 	if (!mad_wc || !mad_wc->recv_buf.mad)
631 		return;
632 
633 	port = mad_agent->context;
634 	ah = ib_create_ah_from_wc(mad_agent->qp->pd, mad_wc->wc,
635 				  mad_wc->recv_buf.grh, mad_agent->port_num);
636 	if (IS_ERR(ah))
637 		goto free_recv_mad;
638 
639 	rsp = ib_create_send_mad(mad_agent, mad_wc->wc->src_qp,
640 				 mad_wc->wc->pkey_index, 0,
641 				 IB_MGMT_VENDOR_HDR, OPA_VNIC_EMA_DATA,
642 				 GFP_KERNEL, OPA_MGMT_BASE_VERSION);
643 	if (IS_ERR(rsp))
644 		goto err_rsp;
645 
646 	rsp->ah = ah;
647 	vema_mad = rsp->mad;
648 	memcpy(vema_mad, mad_wc->recv_buf.mad, IB_MGMT_VENDOR_HDR);
649 	vema_mad->mad_hdr.method = IB_MGMT_METHOD_GET_RESP;
650 	vema_mad->mad_hdr.status = 0;
651 
652 	/* Lock ensures network adapter is not removed */
653 	mutex_lock(&port->lock);
654 
655 	switch (mad_wc->recv_buf.mad->mad_hdr.method) {
656 	case IB_MGMT_METHOD_GET:
657 		vema_get(port, (struct opa_vnic_vema_mad *)mad_wc->recv_buf.mad,
658 			 vema_mad);
659 		break;
660 	case IB_MGMT_METHOD_SET:
661 		vema_set(port, (struct opa_vnic_vema_mad *)mad_wc->recv_buf.mad,
662 			 vema_mad);
663 		break;
664 	default:
665 		vema_mad->mad_hdr.status = OPA_VNIC_UNSUP_ATTR;
666 		break;
667 	}
668 	mutex_unlock(&port->lock);
669 
670 	if (!ib_post_send_mad(rsp, NULL)) {
671 		/*
672 		 * with post send successful ah and send mad
673 		 * will be destroyed in send handler
674 		 */
675 		goto free_recv_mad;
676 	}
677 
678 	ib_free_send_mad(rsp);
679 
680 err_rsp:
681 	rdma_destroy_ah(ah, RDMA_DESTROY_AH_SLEEPABLE);
682 free_recv_mad:
683 	ib_free_recv_mad(mad_wc);
684 }
685 
686 /**
687  * vema_get_port -- Gets the opa_vnic_vema_port
688  * @cport: pointer to control dev
689  * @port_num: Port number
690  *
691  * This function loops through the ports and returns
692  * the opa_vnic_vema port structure that is associated
693  * with the OPA port number
694  *
695  * Return: ptr to requested opa_vnic_vema_port strucure
696  *         if success, NULL if not
697  */
698 static struct opa_vnic_vema_port *
699 vema_get_port(struct opa_vnic_ctrl_port *cport, u8 port_num)
700 {
701 	struct opa_vnic_vema_port *port = (void *)cport + sizeof(*cport);
702 
703 	if (port_num > cport->num_ports)
704 		return NULL;
705 
706 	return port + (port_num - 1);
707 }
708 
709 /**
710  * opa_vnic_vema_send_trap -- This function sends a trap to the EM
711  * @adapter: pointer to vnic adapter
712  * @data: pointer to trap data filled by calling function
713  * @lid:  issuers lid (encap_slid from vesw_port_info)
714  *
715  * This function is called from the VNIC driver to send a trap if there
716  * is somethng the EM should be notified about. These events currently
717  * are
718  * 1) UNICAST INTERFACE MACADDRESS changes
719  * 2) MULTICAST INTERFACE MACADDRESS changes
720  * 3) ETHERNET LINK STATUS changes
721  * While allocating the send mad the remote site qpn used is 1
722  * as this is the well known QP.
723  *
724  */
725 void opa_vnic_vema_send_trap(struct opa_vnic_adapter *adapter,
726 			     struct __opa_veswport_trap *data, u32 lid)
727 {
728 	struct opa_vnic_ctrl_port *cport = adapter->cport;
729 	struct ib_mad_send_buf *send_buf;
730 	struct opa_vnic_vema_port *port;
731 	struct ib_device *ibp;
732 	struct opa_vnic_vema_mad_trap *trap_mad;
733 	struct opa_class_port_info *class;
734 	struct rdma_ah_attr ah_attr;
735 	struct ib_ah *ah;
736 	struct opa_veswport_trap *trap;
737 	u32 trap_lid;
738 	u16 pkey_idx;
739 
740 	if (!cport)
741 		goto err_exit;
742 	ibp = cport->ibdev;
743 	port = vema_get_port(cport, data->opaportnum);
744 	if (!port || !port->mad_agent)
745 		goto err_exit;
746 
747 	if (time_before(jiffies, adapter->trap_timeout)) {
748 		if (adapter->trap_count == OPA_VNIC_TRAP_BURST_LIMIT) {
749 			v_warn("Trap rate exceeded\n");
750 			goto err_exit;
751 		} else {
752 			adapter->trap_count++;
753 		}
754 	} else {
755 		adapter->trap_count = 0;
756 	}
757 
758 	class = &port->class_port_info;
759 	/* Set up address handle */
760 	memset(&ah_attr, 0, sizeof(ah_attr));
761 	ah_attr.type = rdma_ah_find_type(ibp, port->port_num);
762 	rdma_ah_set_sl(&ah_attr,
763 		       GET_TRAP_SL_FROM_CLASS_PORT_INFO(class->trap_sl_rsvd));
764 	rdma_ah_set_port_num(&ah_attr, port->port_num);
765 	trap_lid = be32_to_cpu(class->trap_lid);
766 	/*
767 	 * check for trap lid validity, must not be zero
768 	 * The trap sink could change after we fashion the MAD but since traps
769 	 * are not guaranteed we won't use a lock as anyway the change will take
770 	 * place even with locking.
771 	 */
772 	if (!trap_lid) {
773 		c_err("%s: Invalid dlid\n", __func__);
774 		goto err_exit;
775 	}
776 
777 	rdma_ah_set_dlid(&ah_attr, trap_lid);
778 	ah = rdma_create_ah(port->mad_agent->qp->pd, &ah_attr, 0);
779 	if (IS_ERR(ah)) {
780 		c_err("%s:Couldn't create new AH = %p\n", __func__, ah);
781 		c_err("%s:dlid = %d, sl = %d, port = %d\n", __func__,
782 		      rdma_ah_get_dlid(&ah_attr), rdma_ah_get_sl(&ah_attr),
783 		      rdma_ah_get_port_num(&ah_attr));
784 		goto err_exit;
785 	}
786 
787 	if (ib_find_pkey(ibp, data->opaportnum, IB_DEFAULT_PKEY_FULL,
788 			 &pkey_idx) < 0) {
789 		c_err("%s:full key not found, defaulting to partial\n",
790 		      __func__);
791 		if (ib_find_pkey(ibp, data->opaportnum, IB_DEFAULT_PKEY_PARTIAL,
792 				 &pkey_idx) < 0)
793 			pkey_idx = 1;
794 	}
795 
796 	send_buf = ib_create_send_mad(port->mad_agent, 1, pkey_idx, 0,
797 				      IB_MGMT_VENDOR_HDR, IB_MGMT_MAD_DATA,
798 				      GFP_ATOMIC, OPA_MGMT_BASE_VERSION);
799 	if (IS_ERR(send_buf)) {
800 		c_err("%s:Couldn't allocate send buf\n", __func__);
801 		goto err_sndbuf;
802 	}
803 
804 	send_buf->ah = ah;
805 
806 	/* Set up common MAD hdr */
807 	trap_mad = send_buf->mad;
808 	trap_mad->mad_hdr.base_version = OPA_MGMT_BASE_VERSION;
809 	trap_mad->mad_hdr.mgmt_class = OPA_MGMT_CLASS_INTEL_EMA;
810 	trap_mad->mad_hdr.class_version = OPA_EMA_CLASS_VERSION;
811 	trap_mad->mad_hdr.method = IB_MGMT_METHOD_TRAP;
812 	port->tid++;
813 	trap_mad->mad_hdr.tid = cpu_to_be64(port->tid);
814 	trap_mad->mad_hdr.attr_id = IB_SMP_ATTR_NOTICE;
815 
816 	/* Set up vendor OUI */
817 	trap_mad->oui[0] = INTEL_OUI_1;
818 	trap_mad->oui[1] = INTEL_OUI_2;
819 	trap_mad->oui[2] = INTEL_OUI_3;
820 
821 	/* Setup notice attribute portion */
822 	trap_mad->notice.gen_type = OPA_INTEL_EMA_NOTICE_TYPE_INFO << 1;
823 	trap_mad->notice.oui_1 = INTEL_OUI_1;
824 	trap_mad->notice.oui_2 = INTEL_OUI_2;
825 	trap_mad->notice.oui_3 = INTEL_OUI_3;
826 	trap_mad->notice.issuer_lid = cpu_to_be32(lid);
827 
828 	/* copy the actual trap data */
829 	trap = (struct opa_veswport_trap *)trap_mad->notice.raw_data;
830 	trap->fabric_id = cpu_to_be16(data->fabric_id);
831 	trap->veswid = cpu_to_be16(data->veswid);
832 	trap->veswportnum = cpu_to_be32(data->veswportnum);
833 	trap->opaportnum = cpu_to_be16(data->opaportnum);
834 	trap->veswportindex = data->veswportindex;
835 	trap->opcode = data->opcode;
836 
837 	/* If successful send set up rate limit timeout else bail */
838 	if (ib_post_send_mad(send_buf, NULL)) {
839 		ib_free_send_mad(send_buf);
840 	} else {
841 		if (adapter->trap_count)
842 			return;
843 		adapter->trap_timeout = jiffies +
844 					usecs_to_jiffies(OPA_VNIC_TRAP_TIMEOUT);
845 		return;
846 	}
847 
848 err_sndbuf:
849 	rdma_destroy_ah(ah, 0);
850 err_exit:
851 	v_err("Aborting trap\n");
852 }
853 
854 static void opa_vnic_event(struct ib_event_handler *handler,
855 			   struct ib_event *record)
856 {
857 	struct opa_vnic_vema_port *port =
858 		container_of(handler, struct opa_vnic_vema_port, event_handler);
859 	struct opa_vnic_ctrl_port *cport = port->cport;
860 	struct opa_vnic_adapter *adapter;
861 	unsigned long index;
862 
863 	if (record->element.port_num != port->port_num)
864 		return;
865 
866 	c_dbg("OPA_VNIC received event %d on device %s port %d\n",
867 	      record->event, dev_name(&record->device->dev),
868 	      record->element.port_num);
869 
870 	if (record->event != IB_EVENT_PORT_ERR &&
871 	    record->event != IB_EVENT_PORT_ACTIVE)
872 		return;
873 
874 	xa_for_each(&port->vports, index, adapter) {
875 		if (record->event == IB_EVENT_PORT_ACTIVE)
876 			netif_carrier_on(adapter->netdev);
877 		else
878 			netif_carrier_off(adapter->netdev);
879 	}
880 }
881 
882 /**
883  * vema_unregister -- Unregisters agent
884  * @cport: pointer to control port
885  *
886  * This deletes the registration by VEMA for MADs
887  */
888 static void vema_unregister(struct opa_vnic_ctrl_port *cport)
889 {
890 	struct opa_vnic_adapter *adapter;
891 	unsigned long index;
892 	int i;
893 
894 	for (i = 1; i <= cport->num_ports; i++) {
895 		struct opa_vnic_vema_port *port = vema_get_port(cport, i);
896 
897 		if (!port->mad_agent)
898 			continue;
899 
900 		/* Lock ensures no MAD is being processed */
901 		mutex_lock(&port->lock);
902 		xa_for_each(&port->vports, index, adapter)
903 			opa_vnic_rem_netdev(adapter);
904 		mutex_unlock(&port->lock);
905 
906 		ib_unregister_mad_agent(port->mad_agent);
907 		port->mad_agent = NULL;
908 		mutex_destroy(&port->lock);
909 		xa_destroy(&port->vports);
910 		ib_unregister_event_handler(&port->event_handler);
911 	}
912 }
913 
914 /**
915  * vema_register -- Registers agent
916  * @cport: pointer to control port
917  *
918  * This function registers the handlers for the VEMA MADs
919  *
920  * Return: returns 0 on success. non zero otherwise
921  */
922 static int vema_register(struct opa_vnic_ctrl_port *cport)
923 {
924 	struct ib_mad_reg_req reg_req = {
925 		.mgmt_class = OPA_MGMT_CLASS_INTEL_EMA,
926 		.mgmt_class_version = OPA_MGMT_BASE_VERSION,
927 		.oui = { INTEL_OUI_1, INTEL_OUI_2, INTEL_OUI_3 }
928 	};
929 	int i;
930 
931 	set_bit(IB_MGMT_METHOD_GET, reg_req.method_mask);
932 	set_bit(IB_MGMT_METHOD_SET, reg_req.method_mask);
933 
934 	/* register ib event handler and mad agent for each port on dev */
935 	for (i = 1; i <= cport->num_ports; i++) {
936 		struct opa_vnic_vema_port *port = vema_get_port(cport, i);
937 		int ret;
938 
939 		port->cport = cport;
940 		port->port_num = i;
941 
942 		INIT_IB_EVENT_HANDLER(&port->event_handler,
943 				      cport->ibdev, opa_vnic_event);
944 		ib_register_event_handler(&port->event_handler);
945 
946 		xa_init(&port->vports);
947 		mutex_init(&port->lock);
948 		port->mad_agent = ib_register_mad_agent(cport->ibdev, i,
949 							IB_QPT_GSI, &reg_req,
950 							IB_MGMT_RMPP_VERSION,
951 							vema_send, vema_recv,
952 							port, 0);
953 		if (IS_ERR(port->mad_agent)) {
954 			ret = PTR_ERR(port->mad_agent);
955 			port->mad_agent = NULL;
956 			mutex_destroy(&port->lock);
957 			vema_unregister(cport);
958 			return ret;
959 		}
960 	}
961 
962 	return 0;
963 }
964 
965 /**
966  * opa_vnic_ctrl_config_dev -- This function sends a trap to the EM
967  * by way of ib_modify_port to indicate support for ethernet on the
968  * fabric.
969  * @cport: pointer to control port
970  * @en: enable or disable ethernet on fabric support
971  */
972 static void opa_vnic_ctrl_config_dev(struct opa_vnic_ctrl_port *cport, bool en)
973 {
974 	struct ib_port_modify pm = { 0 };
975 	int i;
976 
977 	if (en)
978 		pm.set_port_cap_mask = OPA_CAP_MASK3_IsEthOnFabricSupported;
979 	else
980 		pm.clr_port_cap_mask = OPA_CAP_MASK3_IsEthOnFabricSupported;
981 
982 	for (i = 1; i <= cport->num_ports; i++)
983 		ib_modify_port(cport->ibdev, i, IB_PORT_OPA_MASK_CHG, &pm);
984 }
985 
986 /**
987  * opa_vnic_vema_add_one -- Handle new ib device
988  * @device: ib device pointer
989  *
990  * Allocate the vnic control port and initialize it.
991  */
992 static int opa_vnic_vema_add_one(struct ib_device *device)
993 {
994 	struct opa_vnic_ctrl_port *cport;
995 	int rc, size = sizeof(*cport);
996 
997 	if (!rdma_cap_opa_vnic(device))
998 		return -EOPNOTSUPP;
999 
1000 	size += device->phys_port_cnt * sizeof(struct opa_vnic_vema_port);
1001 	cport = kzalloc(size, GFP_KERNEL);
1002 	if (!cport)
1003 		return -ENOMEM;
1004 
1005 	cport->num_ports = device->phys_port_cnt;
1006 	cport->ibdev = device;
1007 
1008 	/* Initialize opa vnic management agent (vema) */
1009 	rc = vema_register(cport);
1010 	if (!rc)
1011 		c_info("VNIC client initialized\n");
1012 
1013 	ib_set_client_data(device, &opa_vnic_client, cport);
1014 	opa_vnic_ctrl_config_dev(cport, true);
1015 	return 0;
1016 }
1017 
1018 /**
1019  * opa_vnic_vema_rem_one -- Handle ib device removal
1020  * @device: ib device pointer
1021  * @client_data: ib client data
1022  *
1023  * Uninitialize and free the vnic control port.
1024  */
1025 static void opa_vnic_vema_rem_one(struct ib_device *device,
1026 				  void *client_data)
1027 {
1028 	struct opa_vnic_ctrl_port *cport = client_data;
1029 
1030 	c_info("removing VNIC client\n");
1031 	opa_vnic_ctrl_config_dev(cport, false);
1032 	vema_unregister(cport);
1033 	kfree(cport);
1034 }
1035 
1036 static int __init opa_vnic_init(void)
1037 {
1038 	int rc;
1039 
1040 	rc = ib_register_client(&opa_vnic_client);
1041 	if (rc)
1042 		pr_err("VNIC driver register failed %d\n", rc);
1043 
1044 	return rc;
1045 }
1046 module_init(opa_vnic_init);
1047 
1048 static void opa_vnic_deinit(void)
1049 {
1050 	ib_unregister_client(&opa_vnic_client);
1051 }
1052 module_exit(opa_vnic_deinit);
1053 
1054 MODULE_LICENSE("Dual BSD/GPL");
1055 MODULE_AUTHOR("Cornelis Networks");
1056 MODULE_DESCRIPTION("Cornelis OPX Virtual Network driver");
1057