xref: /linux/drivers/infiniband/hw/mlx4/mad.c (revision 25aee3debe0464f6c680173041fa3de30ec9ff54)
1 /*
2  * Copyright (c) 2007 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  * OpenIB.org 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 
33 #include <rdma/ib_mad.h>
34 #include <rdma/ib_smi.h>
35 
36 #include <linux/mlx4/cmd.h>
37 #include <linux/gfp.h>
38 #include <rdma/ib_pma.h>
39 
40 #include "mlx4_ib.h"
41 
42 enum {
43 	MLX4_IB_VENDOR_CLASS1 = 0x9,
44 	MLX4_IB_VENDOR_CLASS2 = 0xa
45 };
46 
47 int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey,
48 		 int port, struct ib_wc *in_wc, struct ib_grh *in_grh,
49 		 void *in_mad, void *response_mad)
50 {
51 	struct mlx4_cmd_mailbox *inmailbox, *outmailbox;
52 	void *inbox;
53 	int err;
54 	u32 in_modifier = port;
55 	u8 op_modifier = 0;
56 
57 	inmailbox = mlx4_alloc_cmd_mailbox(dev->dev);
58 	if (IS_ERR(inmailbox))
59 		return PTR_ERR(inmailbox);
60 	inbox = inmailbox->buf;
61 
62 	outmailbox = mlx4_alloc_cmd_mailbox(dev->dev);
63 	if (IS_ERR(outmailbox)) {
64 		mlx4_free_cmd_mailbox(dev->dev, inmailbox);
65 		return PTR_ERR(outmailbox);
66 	}
67 
68 	memcpy(inbox, in_mad, 256);
69 
70 	/*
71 	 * Key check traps can't be generated unless we have in_wc to
72 	 * tell us where to send the trap.
73 	 */
74 	if (ignore_mkey || !in_wc)
75 		op_modifier |= 0x1;
76 	if (ignore_bkey || !in_wc)
77 		op_modifier |= 0x2;
78 
79 	if (in_wc) {
80 		struct {
81 			__be32		my_qpn;
82 			u32		reserved1;
83 			__be32		rqpn;
84 			u8		sl;
85 			u8		g_path;
86 			u16		reserved2[2];
87 			__be16		pkey;
88 			u32		reserved3[11];
89 			u8		grh[40];
90 		} *ext_info;
91 
92 		memset(inbox + 256, 0, 256);
93 		ext_info = inbox + 256;
94 
95 		ext_info->my_qpn = cpu_to_be32(in_wc->qp->qp_num);
96 		ext_info->rqpn   = cpu_to_be32(in_wc->src_qp);
97 		ext_info->sl     = in_wc->sl << 4;
98 		ext_info->g_path = in_wc->dlid_path_bits |
99 			(in_wc->wc_flags & IB_WC_GRH ? 0x80 : 0);
100 		ext_info->pkey   = cpu_to_be16(in_wc->pkey_index);
101 
102 		if (in_grh)
103 			memcpy(ext_info->grh, in_grh, 40);
104 
105 		op_modifier |= 0x4;
106 
107 		in_modifier |= in_wc->slid << 16;
108 	}
109 
110 	err = mlx4_cmd_box(dev->dev, inmailbox->dma, outmailbox->dma,
111 			   in_modifier, op_modifier,
112 			   MLX4_CMD_MAD_IFC, MLX4_CMD_TIME_CLASS_C,
113 			   MLX4_CMD_NATIVE);
114 
115 	if (!err)
116 		memcpy(response_mad, outmailbox->buf, 256);
117 
118 	mlx4_free_cmd_mailbox(dev->dev, inmailbox);
119 	mlx4_free_cmd_mailbox(dev->dev, outmailbox);
120 
121 	return err;
122 }
123 
124 static void update_sm_ah(struct mlx4_ib_dev *dev, u8 port_num, u16 lid, u8 sl)
125 {
126 	struct ib_ah *new_ah;
127 	struct ib_ah_attr ah_attr;
128 
129 	if (!dev->send_agent[port_num - 1][0])
130 		return;
131 
132 	memset(&ah_attr, 0, sizeof ah_attr);
133 	ah_attr.dlid     = lid;
134 	ah_attr.sl       = sl;
135 	ah_attr.port_num = port_num;
136 
137 	new_ah = ib_create_ah(dev->send_agent[port_num - 1][0]->qp->pd,
138 			      &ah_attr);
139 	if (IS_ERR(new_ah))
140 		return;
141 
142 	spin_lock(&dev->sm_lock);
143 	if (dev->sm_ah[port_num - 1])
144 		ib_destroy_ah(dev->sm_ah[port_num - 1]);
145 	dev->sm_ah[port_num - 1] = new_ah;
146 	spin_unlock(&dev->sm_lock);
147 }
148 
149 /*
150  * Snoop SM MADs for port info, GUID info, and  P_Key table sets, so we can
151  * synthesize LID change, Client-Rereg, GID change, and P_Key change events.
152  */
153 static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad,
154 		      u16 prev_lid)
155 {
156 	struct ib_port_info *pinfo;
157 	u16 lid;
158 
159 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
160 	if ((mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
161 	     mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
162 	    mad->mad_hdr.method == IB_MGMT_METHOD_SET)
163 		switch (mad->mad_hdr.attr_id) {
164 		case IB_SMP_ATTR_PORT_INFO:
165 			pinfo = (struct ib_port_info *) ((struct ib_smp *) mad)->data;
166 			lid = be16_to_cpu(pinfo->lid);
167 
168 			update_sm_ah(dev, port_num,
169 				     be16_to_cpu(pinfo->sm_lid),
170 				     pinfo->neighbormtu_mastersmsl & 0xf);
171 
172 			if (pinfo->clientrereg_resv_subnetto & 0x80)
173 				mlx4_ib_dispatch_event(dev, port_num,
174 						       IB_EVENT_CLIENT_REREGISTER);
175 
176 			if (prev_lid != lid)
177 				mlx4_ib_dispatch_event(dev, port_num,
178 						       IB_EVENT_LID_CHANGE);
179 			break;
180 
181 		case IB_SMP_ATTR_PKEY_TABLE:
182 			mlx4_ib_dispatch_event(dev, port_num,
183 					       IB_EVENT_PKEY_CHANGE);
184 			break;
185 
186 		case IB_SMP_ATTR_GUID_INFO:
187 			/* paravirtualized master's guid is guid 0 -- does not change */
188 			if (!mlx4_is_master(dev->dev))
189 				mlx4_ib_dispatch_event(dev, port_num,
190 						       IB_EVENT_GID_CHANGE);
191 			break;
192 		default:
193 			break;
194 		}
195 }
196 
197 static void node_desc_override(struct ib_device *dev,
198 			       struct ib_mad *mad)
199 {
200 	if ((mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
201 	     mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
202 	    mad->mad_hdr.method == IB_MGMT_METHOD_GET_RESP &&
203 	    mad->mad_hdr.attr_id == IB_SMP_ATTR_NODE_DESC) {
204 		spin_lock(&to_mdev(dev)->sm_lock);
205 		memcpy(((struct ib_smp *) mad)->data, dev->node_desc, 64);
206 		spin_unlock(&to_mdev(dev)->sm_lock);
207 	}
208 }
209 
210 static void forward_trap(struct mlx4_ib_dev *dev, u8 port_num, struct ib_mad *mad)
211 {
212 	int qpn = mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_SUBN_LID_ROUTED;
213 	struct ib_mad_send_buf *send_buf;
214 	struct ib_mad_agent *agent = dev->send_agent[port_num - 1][qpn];
215 	int ret;
216 
217 	if (agent) {
218 		send_buf = ib_create_send_mad(agent, qpn, 0, 0, IB_MGMT_MAD_HDR,
219 					      IB_MGMT_MAD_DATA, GFP_ATOMIC);
220 		if (IS_ERR(send_buf))
221 			return;
222 		/*
223 		 * We rely here on the fact that MLX QPs don't use the
224 		 * address handle after the send is posted (this is
225 		 * wrong following the IB spec strictly, but we know
226 		 * it's OK for our devices).
227 		 */
228 		spin_lock(&dev->sm_lock);
229 		memcpy(send_buf->mad, mad, sizeof *mad);
230 		if ((send_buf->ah = dev->sm_ah[port_num - 1]))
231 			ret = ib_post_send_mad(send_buf, NULL);
232 		else
233 			ret = -EINVAL;
234 		spin_unlock(&dev->sm_lock);
235 
236 		if (ret)
237 			ib_free_send_mad(send_buf);
238 	}
239 }
240 
241 static int ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
242 			struct ib_wc *in_wc, struct ib_grh *in_grh,
243 			struct ib_mad *in_mad, struct ib_mad *out_mad)
244 {
245 	u16 slid, prev_lid = 0;
246 	int err;
247 	struct ib_port_attr pattr;
248 
249 	if (in_wc && in_wc->qp->qp_num) {
250 		pr_debug("received MAD: slid:%d sqpn:%d "
251 			"dlid_bits:%d dqpn:%d wc_flags:0x%x, cls %x, mtd %x, atr %x\n",
252 			in_wc->slid, in_wc->src_qp,
253 			in_wc->dlid_path_bits,
254 			in_wc->qp->qp_num,
255 			in_wc->wc_flags,
256 			in_mad->mad_hdr.mgmt_class, in_mad->mad_hdr.method,
257 			be16_to_cpu(in_mad->mad_hdr.attr_id));
258 		if (in_wc->wc_flags & IB_WC_GRH) {
259 			pr_debug("sgid_hi:0x%016llx sgid_lo:0x%016llx\n",
260 				 be64_to_cpu(in_grh->sgid.global.subnet_prefix),
261 				 be64_to_cpu(in_grh->sgid.global.interface_id));
262 			pr_debug("dgid_hi:0x%016llx dgid_lo:0x%016llx\n",
263 				 be64_to_cpu(in_grh->dgid.global.subnet_prefix),
264 				 be64_to_cpu(in_grh->dgid.global.interface_id));
265 		}
266 	}
267 
268 	slid = in_wc ? in_wc->slid : be16_to_cpu(IB_LID_PERMISSIVE);
269 
270 	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP && slid == 0) {
271 		forward_trap(to_mdev(ibdev), port_num, in_mad);
272 		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
273 	}
274 
275 	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
276 	    in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
277 		if (in_mad->mad_hdr.method   != IB_MGMT_METHOD_GET &&
278 		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_SET &&
279 		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_TRAP_REPRESS)
280 			return IB_MAD_RESULT_SUCCESS;
281 
282 		/*
283 		 * Don't process SMInfo queries -- the SMA can't handle them.
284 		 */
285 		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_SM_INFO)
286 			return IB_MAD_RESULT_SUCCESS;
287 	} else if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_PERF_MGMT ||
288 		   in_mad->mad_hdr.mgmt_class == MLX4_IB_VENDOR_CLASS1   ||
289 		   in_mad->mad_hdr.mgmt_class == MLX4_IB_VENDOR_CLASS2   ||
290 		   in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_CONG_MGMT) {
291 		if (in_mad->mad_hdr.method  != IB_MGMT_METHOD_GET &&
292 		    in_mad->mad_hdr.method  != IB_MGMT_METHOD_SET)
293 			return IB_MAD_RESULT_SUCCESS;
294 	} else
295 		return IB_MAD_RESULT_SUCCESS;
296 
297 	if ((in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
298 	     in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
299 	    in_mad->mad_hdr.method == IB_MGMT_METHOD_SET &&
300 	    in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
301 	    !ib_query_port(ibdev, port_num, &pattr))
302 		prev_lid = pattr.lid;
303 
304 	err = mlx4_MAD_IFC(to_mdev(ibdev),
305 			   mad_flags & IB_MAD_IGNORE_MKEY,
306 			   mad_flags & IB_MAD_IGNORE_BKEY,
307 			   port_num, in_wc, in_grh, in_mad, out_mad);
308 	if (err)
309 		return IB_MAD_RESULT_FAILURE;
310 
311 	if (!out_mad->mad_hdr.status) {
312 		if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV))
313 			smp_snoop(ibdev, port_num, in_mad, prev_lid);
314 		node_desc_override(ibdev, out_mad);
315 	}
316 
317 	/* set return bit in status of directed route responses */
318 	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
319 		out_mad->mad_hdr.status |= cpu_to_be16(1 << 15);
320 
321 	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP_REPRESS)
322 		/* no response for trap repress */
323 		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
324 
325 	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
326 }
327 
328 static void edit_counter(struct mlx4_counter *cnt,
329 					struct ib_pma_portcounters *pma_cnt)
330 {
331 	pma_cnt->port_xmit_data = cpu_to_be32((be64_to_cpu(cnt->tx_bytes)>>2));
332 	pma_cnt->port_rcv_data  = cpu_to_be32((be64_to_cpu(cnt->rx_bytes)>>2));
333 	pma_cnt->port_xmit_packets = cpu_to_be32(be64_to_cpu(cnt->tx_frames));
334 	pma_cnt->port_rcv_packets  = cpu_to_be32(be64_to_cpu(cnt->rx_frames));
335 }
336 
337 static int iboe_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
338 			struct ib_wc *in_wc, struct ib_grh *in_grh,
339 			struct ib_mad *in_mad, struct ib_mad *out_mad)
340 {
341 	struct mlx4_cmd_mailbox *mailbox;
342 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
343 	int err;
344 	u32 inmod = dev->counters[port_num - 1] & 0xffff;
345 	u8 mode;
346 
347 	if (in_mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_PERF_MGMT)
348 		return -EINVAL;
349 
350 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
351 	if (IS_ERR(mailbox))
352 		return IB_MAD_RESULT_FAILURE;
353 
354 	err = mlx4_cmd_box(dev->dev, 0, mailbox->dma, inmod, 0,
355 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
356 			   MLX4_CMD_WRAPPED);
357 	if (err)
358 		err = IB_MAD_RESULT_FAILURE;
359 	else {
360 		memset(out_mad->data, 0, sizeof out_mad->data);
361 		mode = ((struct mlx4_counter *)mailbox->buf)->counter_mode;
362 		switch (mode & 0xf) {
363 		case 0:
364 			edit_counter(mailbox->buf,
365 						(void *)(out_mad->data + 40));
366 			err = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
367 			break;
368 		default:
369 			err = IB_MAD_RESULT_FAILURE;
370 		}
371 	}
372 
373 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
374 
375 	return err;
376 }
377 
378 int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
379 			struct ib_wc *in_wc, struct ib_grh *in_grh,
380 			struct ib_mad *in_mad, struct ib_mad *out_mad)
381 {
382 	switch (rdma_port_get_link_layer(ibdev, port_num)) {
383 	case IB_LINK_LAYER_INFINIBAND:
384 		return ib_process_mad(ibdev, mad_flags, port_num, in_wc,
385 				      in_grh, in_mad, out_mad);
386 	case IB_LINK_LAYER_ETHERNET:
387 		return iboe_process_mad(ibdev, mad_flags, port_num, in_wc,
388 					  in_grh, in_mad, out_mad);
389 	default:
390 		return -EINVAL;
391 	}
392 }
393 
394 static void send_handler(struct ib_mad_agent *agent,
395 			 struct ib_mad_send_wc *mad_send_wc)
396 {
397 	ib_free_send_mad(mad_send_wc->send_buf);
398 }
399 
400 int mlx4_ib_mad_init(struct mlx4_ib_dev *dev)
401 {
402 	struct ib_mad_agent *agent;
403 	int p, q;
404 	int ret;
405 	enum rdma_link_layer ll;
406 
407 	for (p = 0; p < dev->num_ports; ++p) {
408 		ll = rdma_port_get_link_layer(&dev->ib_dev, p + 1);
409 		for (q = 0; q <= 1; ++q) {
410 			if (ll == IB_LINK_LAYER_INFINIBAND) {
411 				agent = ib_register_mad_agent(&dev->ib_dev, p + 1,
412 							      q ? IB_QPT_GSI : IB_QPT_SMI,
413 							      NULL, 0, send_handler,
414 							      NULL, NULL);
415 				if (IS_ERR(agent)) {
416 					ret = PTR_ERR(agent);
417 					goto err;
418 				}
419 				dev->send_agent[p][q] = agent;
420 			} else
421 				dev->send_agent[p][q] = NULL;
422 		}
423 	}
424 
425 	return 0;
426 
427 err:
428 	for (p = 0; p < dev->num_ports; ++p)
429 		for (q = 0; q <= 1; ++q)
430 			if (dev->send_agent[p][q])
431 				ib_unregister_mad_agent(dev->send_agent[p][q]);
432 
433 	return ret;
434 }
435 
436 void mlx4_ib_mad_cleanup(struct mlx4_ib_dev *dev)
437 {
438 	struct ib_mad_agent *agent;
439 	int p, q;
440 
441 	for (p = 0; p < dev->num_ports; ++p) {
442 		for (q = 0; q <= 1; ++q) {
443 			agent = dev->send_agent[p][q];
444 			if (agent) {
445 				dev->send_agent[p][q] = NULL;
446 				ib_unregister_mad_agent(agent);
447 			}
448 		}
449 
450 		if (dev->sm_ah[p])
451 			ib_destroy_ah(dev->sm_ah[p]);
452 	}
453 }
454 
455 void handle_port_mgmt_change_event(struct work_struct *work)
456 {
457 	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
458 	struct mlx4_ib_dev *dev = ew->ib_dev;
459 	struct mlx4_eqe *eqe = &(ew->ib_eqe);
460 	u8 port = eqe->event.port_mgmt_change.port;
461 	u32 changed_attr;
462 
463 	switch (eqe->subtype) {
464 	case MLX4_DEV_PMC_SUBTYPE_PORT_INFO:
465 		changed_attr = be32_to_cpu(eqe->event.port_mgmt_change.params.port_info.changed_attr);
466 
467 		/* Update the SM ah - This should be done before handling
468 		   the other changed attributes so that MADs can be sent to the SM */
469 		if (changed_attr & MSTR_SM_CHANGE_MASK) {
470 			u16 lid = be16_to_cpu(eqe->event.port_mgmt_change.params.port_info.mstr_sm_lid);
471 			u8 sl = eqe->event.port_mgmt_change.params.port_info.mstr_sm_sl & 0xf;
472 			update_sm_ah(dev, port, lid, sl);
473 		}
474 
475 		/* Check if it is a lid change event */
476 		if (changed_attr & MLX4_EQ_PORT_INFO_LID_CHANGE_MASK)
477 			mlx4_ib_dispatch_event(dev, port, IB_EVENT_LID_CHANGE);
478 
479 		/* Generate GUID changed event */
480 		if (changed_attr & MLX4_EQ_PORT_INFO_GID_PFX_CHANGE_MASK)
481 			mlx4_ib_dispatch_event(dev, port, IB_EVENT_GID_CHANGE);
482 
483 		if (changed_attr & MLX4_EQ_PORT_INFO_CLIENT_REREG_MASK)
484 			mlx4_ib_dispatch_event(dev, port,
485 					       IB_EVENT_CLIENT_REREGISTER);
486 		break;
487 
488 	case MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE:
489 		mlx4_ib_dispatch_event(dev, port, IB_EVENT_PKEY_CHANGE);
490 		break;
491 	case MLX4_DEV_PMC_SUBTYPE_GUID_INFO:
492 		/* paravirtualized master's guid is guid 0 -- does not change */
493 		if (!mlx4_is_master(dev->dev))
494 			mlx4_ib_dispatch_event(dev, port, IB_EVENT_GID_CHANGE);
495 		break;
496 	default:
497 		pr_warn("Unsupported subtype 0x%x for "
498 			"Port Management Change event\n", eqe->subtype);
499 	}
500 
501 	kfree(ew);
502 }
503 
504 void mlx4_ib_dispatch_event(struct mlx4_ib_dev *dev, u8 port_num,
505 			    enum ib_event_type type)
506 {
507 	struct ib_event event;
508 
509 	event.device		= &dev->ib_dev;
510 	event.element.port_num	= port_num;
511 	event.event		= type;
512 
513 	ib_dispatch_event(&event);
514 }
515