1 /*
2  * Copyright(c) 2017 Intel Corporation.
3  *
4  * This file is provided under a dual BSD/GPLv2 license.  When using or
5  * redistributing this file, you may do so under either license.
6  *
7  * GPL LICENSE SUMMARY
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of version 2 of the GNU General Public License as
11  * published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful, but
14  * WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * General Public License for more details.
17  *
18  * BSD LICENSE
19  *
20  * Redistribution and use in source and binary forms, with or without
21  * modification, are permitted provided that the following conditions
22  * are met:
23  *
24  *  - Redistributions of source code must retain the above copyright
25  *    notice, this list of conditions and the following disclaimer.
26  *  - Redistributions in binary form must reproduce the above copyright
27  *    notice, this list of conditions and the following disclaimer in
28  *    the documentation and/or other materials provided with the
29  *    distribution.
30  *  - Neither the name of Intel Corporation nor the names of its
31  *    contributors may be used to endorse or promote products derived
32  *    from this software without specific prior written permission.
33  *
34  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
35  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
36  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
37  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
38  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
39  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
40  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
41  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
42  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
43  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
44  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
45  *
46  */
47 
48 /*
49  * This file contains OPA Virtual Network Interface Controller (VNIC)
50  * Ethernet Management Agent (EMA) driver
51  */
52 
53 #include <linux/module.h>
54 #include <linux/xarray.h>
55 #include <rdma/ib_addr.h>
56 #include <rdma/ib_verbs.h>
57 #include <rdma/opa_smi.h>
58 #include <rdma/opa_port_info.h>
59 
60 #include "opa_vnic_internal.h"
61 
62 char opa_vnic_driver_name[] = "opa_vnic";
63 
64 /*
65  * The trap service level is kept in bits 3 to 7 in the trap_sl_rsvd
66  * field in the class port info MAD.
67  */
68 #define GET_TRAP_SL_FROM_CLASS_PORT_INFO(x)  (((x) >> 3) & 0x1f)
69 
70 /* Cap trap bursts to a reasonable limit good for normal cases */
71 #define OPA_VNIC_TRAP_BURST_LIMIT 4
72 
73 /*
74  * VNIC trap limit timeout.
75  * Inverse of cap2_mask response time out (1.0737 secs) = 0.9
76  * secs approx IB spec 13.4.6.2.1 PortInfoSubnetTimeout and
77  * 13.4.9 Traps.
78  */
79 #define OPA_VNIC_TRAP_TIMEOUT  ((4096 * (1UL << 18)) / 1000)
80 
81 #define OPA_VNIC_UNSUP_ATTR  \
82 		cpu_to_be16(IB_MGMT_MAD_STATUS_UNSUPPORTED_METHOD_ATTRIB)
83 
84 #define OPA_VNIC_INVAL_ATTR  \
85 		cpu_to_be16(IB_MGMT_MAD_STATUS_INVALID_ATTRIB_VALUE)
86 
87 #define OPA_VNIC_CLASS_CAP_TRAP   0x1
88 
89 /* Maximum number of VNIC ports supported */
90 #define OPA_VNIC_MAX_NUM_VPORT    255
91 
92 /**
93  * struct opa_vnic_vema_port -- VNIC VEMA port details
94  * @cport: pointer to port
95  * @mad_agent: pointer to mad agent for port
96  * @class_port_info: Class port info information.
97  * @tid: Transaction id
98  * @port_num: OPA port number
99  * @vports: vnic ports
100  * @event_handler: ib event handler
101  * @lock: adapter interface lock
102  */
103 struct opa_vnic_vema_port {
104 	struct opa_vnic_ctrl_port      *cport;
105 	struct ib_mad_agent            *mad_agent;
106 	struct opa_class_port_info      class_port_info;
107 	u64                             tid;
108 	u8                              port_num;
109 	struct xarray                   vports;
110 	struct ib_event_handler         event_handler;
111 
112 	/* Lock to query/update network adapter */
113 	struct mutex                    lock;
114 };
115 
116 static int opa_vnic_vema_add_one(struct ib_device *device);
117 static void opa_vnic_vema_rem_one(struct ib_device *device,
118 				  void *client_data);
119 
120 static struct ib_client opa_vnic_client = {
121 	.name   = opa_vnic_driver_name,
122 	.add    = opa_vnic_vema_add_one,
123 	.remove = opa_vnic_vema_rem_one,
124 };
125 
126 /**
127  * vema_get_vport_num -- Get the vnic from the mad
128  * @recvd_mad:  Received mad
129  *
130  * Return: returns value of the vnic port number
131  */
132 static inline u8 vema_get_vport_num(struct opa_vnic_vema_mad *recvd_mad)
133 {
134 	return be32_to_cpu(recvd_mad->mad_hdr.attr_mod) & 0xff;
135 }
136 
137 /**
138  * vema_get_vport_adapter -- Get vnic port adapter from recvd mad
139  * @recvd_mad: received mad
140  * @port: ptr to port struct on which MAD was recvd
141  *
142  * Return: vnic adapter
143  */
144 static inline struct opa_vnic_adapter *
145 vema_get_vport_adapter(struct opa_vnic_vema_mad *recvd_mad,
146 		       struct opa_vnic_vema_port *port)
147 {
148 	u8 vport_num = vema_get_vport_num(recvd_mad);
149 
150 	return xa_load(&port->vports, vport_num);
151 }
152 
153 /**
154  * vema_mac_tbl_req_ok -- Check if mac request has correct values
155  * @mac_tbl: mac table
156  *
157  * This function checks for the validity of the offset and number of
158  * entries required.
159  *
160  * Return: true if offset and num_entries are valid
161  */
162 static inline bool vema_mac_tbl_req_ok(struct opa_veswport_mactable *mac_tbl)
163 {
164 	u16 offset, num_entries;
165 	u16 req_entries = ((OPA_VNIC_EMA_DATA - sizeof(*mac_tbl)) /
166 			   sizeof(mac_tbl->tbl_entries[0]));
167 
168 	offset = be16_to_cpu(mac_tbl->offset);
169 	num_entries = be16_to_cpu(mac_tbl->num_entries);
170 
171 	return ((num_entries <= req_entries) &&
172 		(offset + num_entries <= OPA_VNIC_MAC_TBL_MAX_ENTRIES));
173 }
174 
175 /*
176  * Return the power on default values in the port info structure
177  * in big endian format as required by MAD.
178  */
179 static inline void vema_get_pod_values(struct opa_veswport_info *port_info)
180 {
181 	memset(port_info, 0, sizeof(*port_info));
182 	port_info->vport.max_mac_tbl_ent =
183 		cpu_to_be16(OPA_VNIC_MAC_TBL_MAX_ENTRIES);
184 	port_info->vport.max_smac_ent =
185 		cpu_to_be16(OPA_VNIC_MAX_SMAC_LIMIT);
186 	port_info->vport.oper_state = OPA_VNIC_STATE_DROP_ALL;
187 	port_info->vport.config_state = OPA_VNIC_STATE_DROP_ALL;
188 	port_info->vesw.eth_mtu = cpu_to_be16(ETH_DATA_LEN);
189 }
190 
191 /**
192  * vema_add_vport -- Add a new vnic port
193  * @port: ptr to opa_vnic_vema_port struct
194  * @vport_num: vnic port number (to be added)
195  *
196  * Return a pointer to the vnic adapter structure
197  */
198 static struct opa_vnic_adapter *vema_add_vport(struct opa_vnic_vema_port *port,
199 					       u8 vport_num)
200 {
201 	struct opa_vnic_ctrl_port *cport = port->cport;
202 	struct opa_vnic_adapter *adapter;
203 
204 	adapter = opa_vnic_add_netdev(cport->ibdev, port->port_num, vport_num);
205 	if (!IS_ERR(adapter)) {
206 		int rc;
207 
208 		adapter->cport = cport;
209 		rc = xa_insert(&port->vports, vport_num, adapter, GFP_KERNEL);
210 		if (rc < 0) {
211 			opa_vnic_rem_netdev(adapter);
212 			adapter = ERR_PTR(rc);
213 		}
214 	}
215 
216 	return adapter;
217 }
218 
219 /**
220  * vema_get_class_port_info -- Get class info for port
221  * @port:  Port on whic MAD was received
222  * @recvd_mad: pointer to the received mad
223  * @rsp_mad:   pointer to respose mad
224  *
225  * This function copies the latest class port info value set for the
226  * port and stores it for generating traps
227  */
228 static void vema_get_class_port_info(struct opa_vnic_vema_port *port,
229 				     struct opa_vnic_vema_mad *recvd_mad,
230 				     struct opa_vnic_vema_mad *rsp_mad)
231 {
232 	struct opa_class_port_info *port_info;
233 
234 	port_info = (struct opa_class_port_info *)rsp_mad->data;
235 	memcpy(port_info, &port->class_port_info, sizeof(*port_info));
236 	port_info->base_version = OPA_MGMT_BASE_VERSION,
237 	port_info->class_version = OPA_EMA_CLASS_VERSION;
238 
239 	/*
240 	 * Set capability mask bit indicating agent generates traps,
241 	 * and set the maximum number of VNIC ports supported.
242 	 */
243 	port_info->cap_mask = cpu_to_be16((OPA_VNIC_CLASS_CAP_TRAP |
244 					   (OPA_VNIC_MAX_NUM_VPORT << 8)));
245 
246 	/*
247 	 * Since a get routine is always sent by the EM first we
248 	 * set the expected response time to
249 	 * 4.096 usec * 2^18 == 1.0737 sec here.
250 	 */
251 	port_info->cap_mask2_resp_time = cpu_to_be32(18);
252 }
253 
254 /**
255  * vema_set_class_port_info -- Get class info for port
256  * @port:  Port on whic MAD was received
257  * @recvd_mad: pointer to the received mad
258  * @rsp_mad:   pointer to respose mad
259  *
260  * This function updates the port class info for the specific vnic
261  * and sets up the response mad data
262  */
263 static void vema_set_class_port_info(struct opa_vnic_vema_port *port,
264 				     struct opa_vnic_vema_mad *recvd_mad,
265 				     struct opa_vnic_vema_mad *rsp_mad)
266 {
267 	memcpy(&port->class_port_info, recvd_mad->data,
268 	       sizeof(port->class_port_info));
269 
270 	vema_get_class_port_info(port, recvd_mad, rsp_mad);
271 }
272 
273 /**
274  * vema_get_veswport_info -- Get veswport info
275  * @port:      source port on which MAD was received
276  * @recvd_mad: pointer to the received mad
277  * @rsp_mad:   pointer to respose mad
278  */
279 static void vema_get_veswport_info(struct opa_vnic_vema_port *port,
280 				   struct opa_vnic_vema_mad *recvd_mad,
281 				   struct opa_vnic_vema_mad *rsp_mad)
282 {
283 	struct opa_veswport_info *port_info =
284 				  (struct opa_veswport_info *)rsp_mad->data;
285 	struct opa_vnic_adapter *adapter;
286 
287 	adapter = vema_get_vport_adapter(recvd_mad, port);
288 	if (adapter) {
289 		memset(port_info, 0, sizeof(*port_info));
290 		opa_vnic_get_vesw_info(adapter, &port_info->vesw);
291 		opa_vnic_get_per_veswport_info(adapter,
292 					       &port_info->vport);
293 	} else {
294 		vema_get_pod_values(port_info);
295 	}
296 }
297 
298 /**
299  * vema_set_veswport_info -- Set veswport info
300  * @port:      source port on which MAD was received
301  * @recvd_mad: pointer to the received mad
302  * @rsp_mad:   pointer to respose mad
303  *
304  * This function gets the port class infor for vnic
305  */
306 static void vema_set_veswport_info(struct opa_vnic_vema_port *port,
307 				   struct opa_vnic_vema_mad *recvd_mad,
308 				   struct opa_vnic_vema_mad *rsp_mad)
309 {
310 	struct opa_vnic_ctrl_port *cport = port->cport;
311 	struct opa_veswport_info *port_info;
312 	struct opa_vnic_adapter *adapter;
313 	u8 vport_num;
314 
315 	vport_num = vema_get_vport_num(recvd_mad);
316 
317 	adapter = vema_get_vport_adapter(recvd_mad, port);
318 	if (!adapter) {
319 		adapter = vema_add_vport(port, vport_num);
320 		if (IS_ERR(adapter)) {
321 			c_err("failed to add vport %d: %ld\n",
322 			      vport_num, PTR_ERR(adapter));
323 			goto err_exit;
324 		}
325 	}
326 
327 	port_info = (struct opa_veswport_info *)recvd_mad->data;
328 	opa_vnic_set_vesw_info(adapter, &port_info->vesw);
329 	opa_vnic_set_per_veswport_info(adapter, &port_info->vport);
330 
331 	/* Process the new config settings */
332 	opa_vnic_process_vema_config(adapter);
333 
334 	vema_get_veswport_info(port, recvd_mad, rsp_mad);
335 	return;
336 
337 err_exit:
338 	rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
339 }
340 
341 /**
342  * vema_get_mac_entries -- Get MAC entries in VNIC MAC table
343  * @port:      source port on which MAD was received
344  * @recvd_mad: pointer to the received mad
345  * @rsp_mad:   pointer to respose mad
346  *
347  * This function gets the MAC entries that are programmed into
348  * the VNIC MAC forwarding table. It checks for the validity of
349  * the index into the MAC table and the number of entries that
350  * are to be retrieved.
351  */
352 static void vema_get_mac_entries(struct opa_vnic_vema_port *port,
353 				 struct opa_vnic_vema_mad *recvd_mad,
354 				 struct opa_vnic_vema_mad *rsp_mad)
355 {
356 	struct opa_veswport_mactable *mac_tbl_in, *mac_tbl_out;
357 	struct opa_vnic_adapter *adapter;
358 
359 	adapter = vema_get_vport_adapter(recvd_mad, port);
360 	if (!adapter) {
361 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
362 		return;
363 	}
364 
365 	mac_tbl_in = (struct opa_veswport_mactable *)recvd_mad->data;
366 	mac_tbl_out = (struct opa_veswport_mactable *)rsp_mad->data;
367 
368 	if (vema_mac_tbl_req_ok(mac_tbl_in)) {
369 		mac_tbl_out->offset = mac_tbl_in->offset;
370 		mac_tbl_out->num_entries = mac_tbl_in->num_entries;
371 		opa_vnic_query_mac_tbl(adapter, mac_tbl_out);
372 	} else {
373 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
374 	}
375 }
376 
377 /**
378  * vema_set_mac_entries -- Set MAC entries in VNIC MAC table
379  * @port:      source port on which MAD was received
380  * @recvd_mad: pointer to the received mad
381  * @rsp_mad:   pointer to respose mad
382  *
383  * This function sets the MAC entries in the VNIC forwarding table
384  * It checks for the validity of the index and the number of forwarding
385  * table entries to be programmed.
386  */
387 static void vema_set_mac_entries(struct opa_vnic_vema_port *port,
388 				 struct opa_vnic_vema_mad *recvd_mad,
389 				 struct opa_vnic_vema_mad *rsp_mad)
390 {
391 	struct opa_veswport_mactable *mac_tbl;
392 	struct opa_vnic_adapter *adapter;
393 
394 	adapter = vema_get_vport_adapter(recvd_mad, port);
395 	if (!adapter) {
396 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
397 		return;
398 	}
399 
400 	mac_tbl = (struct opa_veswport_mactable *)recvd_mad->data;
401 	if (vema_mac_tbl_req_ok(mac_tbl)) {
402 		if (opa_vnic_update_mac_tbl(adapter, mac_tbl))
403 			rsp_mad->mad_hdr.status = OPA_VNIC_UNSUP_ATTR;
404 	} else {
405 		rsp_mad->mad_hdr.status = OPA_VNIC_UNSUP_ATTR;
406 	}
407 	vema_get_mac_entries(port, recvd_mad, rsp_mad);
408 }
409 
410 /**
411  * vema_set_delete_vesw -- Reset VESW info to POD values
412  * @port:      source port on which MAD was received
413  * @recvd_mad: pointer to the received mad
414  * @rsp_mad:   pointer to respose mad
415  *
416  * This function clears all the fields of veswport info for the requested vesw
417  * and sets them back to the power-on default values. It does not delete the
418  * vesw.
419  */
420 static void vema_set_delete_vesw(struct opa_vnic_vema_port *port,
421 				 struct opa_vnic_vema_mad *recvd_mad,
422 				 struct opa_vnic_vema_mad *rsp_mad)
423 {
424 	struct opa_veswport_info *port_info =
425 				  (struct opa_veswport_info *)rsp_mad->data;
426 	struct opa_vnic_adapter *adapter;
427 
428 	adapter = vema_get_vport_adapter(recvd_mad, port);
429 	if (!adapter) {
430 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
431 		return;
432 	}
433 
434 	vema_get_pod_values(port_info);
435 	opa_vnic_set_vesw_info(adapter, &port_info->vesw);
436 	opa_vnic_set_per_veswport_info(adapter, &port_info->vport);
437 
438 	/* Process the new config settings */
439 	opa_vnic_process_vema_config(adapter);
440 
441 	opa_vnic_release_mac_tbl(adapter);
442 
443 	vema_get_veswport_info(port, recvd_mad, rsp_mad);
444 }
445 
446 /**
447  * vema_get_mac_list -- Get the unicast/multicast macs.
448  * @port:      source port on which MAD was received
449  * @recvd_mad: Received mad contains fields to set vnic parameters
450  * @rsp_mad:   Response mad to be built
451  * @attr_id:   Attribute ID indicating multicast or unicast mac list
452  */
453 static void vema_get_mac_list(struct opa_vnic_vema_port *port,
454 			      struct opa_vnic_vema_mad *recvd_mad,
455 			      struct opa_vnic_vema_mad *rsp_mad,
456 			      u16 attr_id)
457 {
458 	struct opa_veswport_iface_macs *macs_in, *macs_out;
459 	int max_entries = (OPA_VNIC_EMA_DATA - sizeof(*macs_out)) / ETH_ALEN;
460 	struct opa_vnic_adapter *adapter;
461 
462 	adapter = vema_get_vport_adapter(recvd_mad, port);
463 	if (!adapter) {
464 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
465 		return;
466 	}
467 
468 	macs_in = (struct opa_veswport_iface_macs *)recvd_mad->data;
469 	macs_out = (struct opa_veswport_iface_macs *)rsp_mad->data;
470 
471 	macs_out->start_idx = macs_in->start_idx;
472 	if (macs_in->num_macs_in_msg)
473 		macs_out->num_macs_in_msg = macs_in->num_macs_in_msg;
474 	else
475 		macs_out->num_macs_in_msg = cpu_to_be16(max_entries);
476 
477 	if (attr_id == OPA_EM_ATTR_IFACE_MCAST_MACS)
478 		opa_vnic_query_mcast_macs(adapter, macs_out);
479 	else
480 		opa_vnic_query_ucast_macs(adapter, macs_out);
481 }
482 
483 /**
484  * vema_get_summary_counters -- Gets summary counters.
485  * @port:      source port on which MAD was received
486  * @recvd_mad: Received mad contains fields to set vnic parameters
487  * @rsp_mad:   Response mad to be built
488  */
489 static void vema_get_summary_counters(struct opa_vnic_vema_port *port,
490 				      struct opa_vnic_vema_mad *recvd_mad,
491 				      struct opa_vnic_vema_mad *rsp_mad)
492 {
493 	struct opa_veswport_summary_counters *cntrs;
494 	struct opa_vnic_adapter *adapter;
495 
496 	adapter = vema_get_vport_adapter(recvd_mad, port);
497 	if (adapter) {
498 		cntrs = (struct opa_veswport_summary_counters *)rsp_mad->data;
499 		opa_vnic_get_summary_counters(adapter, cntrs);
500 	} else {
501 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
502 	}
503 }
504 
505 /**
506  * vema_get_error_counters -- Gets summary counters.
507  * @port:      source port on which MAD was received
508  * @recvd_mad: Received mad contains fields to set vnic parameters
509  * @rsp_mad:   Response mad to be built
510  */
511 static void vema_get_error_counters(struct opa_vnic_vema_port *port,
512 				    struct opa_vnic_vema_mad *recvd_mad,
513 				    struct opa_vnic_vema_mad *rsp_mad)
514 {
515 	struct opa_veswport_error_counters *cntrs;
516 	struct opa_vnic_adapter *adapter;
517 
518 	adapter = vema_get_vport_adapter(recvd_mad, port);
519 	if (adapter) {
520 		cntrs = (struct opa_veswport_error_counters *)rsp_mad->data;
521 		opa_vnic_get_error_counters(adapter, cntrs);
522 	} else {
523 		rsp_mad->mad_hdr.status = OPA_VNIC_INVAL_ATTR;
524 	}
525 }
526 
527 /**
528  * vema_get -- Process received get MAD
529  * @port:      source port on which MAD was received
530  * @recvd_mad: Received mad
531  * @rsp_mad:   Response mad to be built
532  */
533 static void vema_get(struct opa_vnic_vema_port *port,
534 		     struct opa_vnic_vema_mad *recvd_mad,
535 		     struct opa_vnic_vema_mad *rsp_mad)
536 {
537 	u16 attr_id = be16_to_cpu(recvd_mad->mad_hdr.attr_id);
538 
539 	switch (attr_id) {
540 	case OPA_EM_ATTR_CLASS_PORT_INFO:
541 		vema_get_class_port_info(port, recvd_mad, rsp_mad);
542 		break;
543 	case OPA_EM_ATTR_VESWPORT_INFO:
544 		vema_get_veswport_info(port, recvd_mad, rsp_mad);
545 		break;
546 	case OPA_EM_ATTR_VESWPORT_MAC_ENTRIES:
547 		vema_get_mac_entries(port, recvd_mad, rsp_mad);
548 		break;
549 	case OPA_EM_ATTR_IFACE_UCAST_MACS:
550 		/* fall through */
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("Intel Corporation");
1056 MODULE_DESCRIPTION("Intel OPA Virtual Network driver");
1057