xref: /linux/drivers/infiniband/hw/mlx4/mad.c (revision b43ab901d671e3e3cad425ea5e9a3c74e266dcdd)
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 and P_Key table sets, so we can
151  * synthesize LID 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_event event;
157 
158 	if ((mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
159 	     mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
160 	    mad->mad_hdr.method == IB_MGMT_METHOD_SET) {
161 		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO) {
162 			struct ib_port_info *pinfo =
163 				(struct ib_port_info *) ((struct ib_smp *) mad)->data;
164 			u16 lid = be16_to_cpu(pinfo->lid);
165 
166 			update_sm_ah(to_mdev(ibdev), port_num,
167 				     be16_to_cpu(pinfo->sm_lid),
168 				     pinfo->neighbormtu_mastersmsl & 0xf);
169 
170 			event.device	       = ibdev;
171 			event.element.port_num = port_num;
172 
173 			if (pinfo->clientrereg_resv_subnetto & 0x80) {
174 				event.event    = IB_EVENT_CLIENT_REREGISTER;
175 				ib_dispatch_event(&event);
176 			}
177 
178 			if (prev_lid != lid) {
179 				event.event    = IB_EVENT_LID_CHANGE;
180 				ib_dispatch_event(&event);
181 			}
182 		}
183 
184 		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PKEY_TABLE) {
185 			event.device	       = ibdev;
186 			event.event	       = IB_EVENT_PKEY_CHANGE;
187 			event.element.port_num = port_num;
188 			ib_dispatch_event(&event);
189 		}
190 	}
191 }
192 
193 static void node_desc_override(struct ib_device *dev,
194 			       struct ib_mad *mad)
195 {
196 	if ((mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
197 	     mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
198 	    mad->mad_hdr.method == IB_MGMT_METHOD_GET_RESP &&
199 	    mad->mad_hdr.attr_id == IB_SMP_ATTR_NODE_DESC) {
200 		spin_lock(&to_mdev(dev)->sm_lock);
201 		memcpy(((struct ib_smp *) mad)->data, dev->node_desc, 64);
202 		spin_unlock(&to_mdev(dev)->sm_lock);
203 	}
204 }
205 
206 static void forward_trap(struct mlx4_ib_dev *dev, u8 port_num, struct ib_mad *mad)
207 {
208 	int qpn = mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_SUBN_LID_ROUTED;
209 	struct ib_mad_send_buf *send_buf;
210 	struct ib_mad_agent *agent = dev->send_agent[port_num - 1][qpn];
211 	int ret;
212 
213 	if (agent) {
214 		send_buf = ib_create_send_mad(agent, qpn, 0, 0, IB_MGMT_MAD_HDR,
215 					      IB_MGMT_MAD_DATA, GFP_ATOMIC);
216 		if (IS_ERR(send_buf))
217 			return;
218 		/*
219 		 * We rely here on the fact that MLX QPs don't use the
220 		 * address handle after the send is posted (this is
221 		 * wrong following the IB spec strictly, but we know
222 		 * it's OK for our devices).
223 		 */
224 		spin_lock(&dev->sm_lock);
225 		memcpy(send_buf->mad, mad, sizeof *mad);
226 		if ((send_buf->ah = dev->sm_ah[port_num - 1]))
227 			ret = ib_post_send_mad(send_buf, NULL);
228 		else
229 			ret = -EINVAL;
230 		spin_unlock(&dev->sm_lock);
231 
232 		if (ret)
233 			ib_free_send_mad(send_buf);
234 	}
235 }
236 
237 static int ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
238 			struct ib_wc *in_wc, struct ib_grh *in_grh,
239 			struct ib_mad *in_mad, struct ib_mad *out_mad)
240 {
241 	u16 slid, prev_lid = 0;
242 	int err;
243 	struct ib_port_attr pattr;
244 
245 	slid = in_wc ? in_wc->slid : be16_to_cpu(IB_LID_PERMISSIVE);
246 
247 	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP && slid == 0) {
248 		forward_trap(to_mdev(ibdev), port_num, in_mad);
249 		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
250 	}
251 
252 	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
253 	    in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
254 		if (in_mad->mad_hdr.method   != IB_MGMT_METHOD_GET &&
255 		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_SET &&
256 		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_TRAP_REPRESS)
257 			return IB_MAD_RESULT_SUCCESS;
258 
259 		/*
260 		 * Don't process SMInfo queries or vendor-specific
261 		 * MADs -- the SMA can't handle them.
262 		 */
263 		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_SM_INFO ||
264 		    ((in_mad->mad_hdr.attr_id & IB_SMP_ATTR_VENDOR_MASK) ==
265 		     IB_SMP_ATTR_VENDOR_MASK))
266 			return IB_MAD_RESULT_SUCCESS;
267 	} else if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_PERF_MGMT ||
268 		   in_mad->mad_hdr.mgmt_class == MLX4_IB_VENDOR_CLASS1   ||
269 		   in_mad->mad_hdr.mgmt_class == MLX4_IB_VENDOR_CLASS2   ||
270 		   in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_CONG_MGMT) {
271 		if (in_mad->mad_hdr.method  != IB_MGMT_METHOD_GET &&
272 		    in_mad->mad_hdr.method  != IB_MGMT_METHOD_SET)
273 			return IB_MAD_RESULT_SUCCESS;
274 	} else
275 		return IB_MAD_RESULT_SUCCESS;
276 
277 	if ((in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
278 	     in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
279 	    in_mad->mad_hdr.method == IB_MGMT_METHOD_SET &&
280 	    in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
281 	    !ib_query_port(ibdev, port_num, &pattr))
282 		prev_lid = pattr.lid;
283 
284 	err = mlx4_MAD_IFC(to_mdev(ibdev),
285 			   mad_flags & IB_MAD_IGNORE_MKEY,
286 			   mad_flags & IB_MAD_IGNORE_BKEY,
287 			   port_num, in_wc, in_grh, in_mad, out_mad);
288 	if (err)
289 		return IB_MAD_RESULT_FAILURE;
290 
291 	if (!out_mad->mad_hdr.status) {
292 		smp_snoop(ibdev, port_num, in_mad, prev_lid);
293 		node_desc_override(ibdev, out_mad);
294 	}
295 
296 	/* set return bit in status of directed route responses */
297 	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
298 		out_mad->mad_hdr.status |= cpu_to_be16(1 << 15);
299 
300 	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP_REPRESS)
301 		/* no response for trap repress */
302 		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
303 
304 	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
305 }
306 
307 static void edit_counter(struct mlx4_counter *cnt,
308 					struct ib_pma_portcounters *pma_cnt)
309 {
310 	pma_cnt->port_xmit_data = cpu_to_be32((be64_to_cpu(cnt->tx_bytes)>>2));
311 	pma_cnt->port_rcv_data  = cpu_to_be32((be64_to_cpu(cnt->rx_bytes)>>2));
312 	pma_cnt->port_xmit_packets = cpu_to_be32(be64_to_cpu(cnt->tx_frames));
313 	pma_cnt->port_rcv_packets  = cpu_to_be32(be64_to_cpu(cnt->rx_frames));
314 }
315 
316 static int iboe_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
317 			struct ib_wc *in_wc, struct ib_grh *in_grh,
318 			struct ib_mad *in_mad, struct ib_mad *out_mad)
319 {
320 	struct mlx4_cmd_mailbox *mailbox;
321 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
322 	int err;
323 	u32 inmod = dev->counters[port_num - 1] & 0xffff;
324 	u8 mode;
325 
326 	if (in_mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_PERF_MGMT)
327 		return -EINVAL;
328 
329 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
330 	if (IS_ERR(mailbox))
331 		return IB_MAD_RESULT_FAILURE;
332 
333 	err = mlx4_cmd_box(dev->dev, 0, mailbox->dma, inmod, 0,
334 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
335 			   MLX4_CMD_WRAPPED);
336 	if (err)
337 		err = IB_MAD_RESULT_FAILURE;
338 	else {
339 		memset(out_mad->data, 0, sizeof out_mad->data);
340 		mode = ((struct mlx4_counter *)mailbox->buf)->counter_mode;
341 		switch (mode & 0xf) {
342 		case 0:
343 			edit_counter(mailbox->buf,
344 						(void *)(out_mad->data + 40));
345 			err = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
346 			break;
347 		default:
348 			err = IB_MAD_RESULT_FAILURE;
349 		}
350 	}
351 
352 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
353 
354 	return err;
355 }
356 
357 int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
358 			struct ib_wc *in_wc, struct ib_grh *in_grh,
359 			struct ib_mad *in_mad, struct ib_mad *out_mad)
360 {
361 	switch (rdma_port_get_link_layer(ibdev, port_num)) {
362 	case IB_LINK_LAYER_INFINIBAND:
363 		return ib_process_mad(ibdev, mad_flags, port_num, in_wc,
364 				      in_grh, in_mad, out_mad);
365 	case IB_LINK_LAYER_ETHERNET:
366 		return iboe_process_mad(ibdev, mad_flags, port_num, in_wc,
367 					  in_grh, in_mad, out_mad);
368 	default:
369 		return -EINVAL;
370 	}
371 }
372 
373 static void send_handler(struct ib_mad_agent *agent,
374 			 struct ib_mad_send_wc *mad_send_wc)
375 {
376 	ib_free_send_mad(mad_send_wc->send_buf);
377 }
378 
379 int mlx4_ib_mad_init(struct mlx4_ib_dev *dev)
380 {
381 	struct ib_mad_agent *agent;
382 	int p, q;
383 	int ret;
384 	enum rdma_link_layer ll;
385 
386 	for (p = 0; p < dev->num_ports; ++p) {
387 		ll = rdma_port_get_link_layer(&dev->ib_dev, p + 1);
388 		for (q = 0; q <= 1; ++q) {
389 			if (ll == IB_LINK_LAYER_INFINIBAND) {
390 				agent = ib_register_mad_agent(&dev->ib_dev, p + 1,
391 							      q ? IB_QPT_GSI : IB_QPT_SMI,
392 							      NULL, 0, send_handler,
393 							      NULL, NULL);
394 				if (IS_ERR(agent)) {
395 					ret = PTR_ERR(agent);
396 					goto err;
397 				}
398 				dev->send_agent[p][q] = agent;
399 			} else
400 				dev->send_agent[p][q] = NULL;
401 		}
402 	}
403 
404 	return 0;
405 
406 err:
407 	for (p = 0; p < dev->num_ports; ++p)
408 		for (q = 0; q <= 1; ++q)
409 			if (dev->send_agent[p][q])
410 				ib_unregister_mad_agent(dev->send_agent[p][q]);
411 
412 	return ret;
413 }
414 
415 void mlx4_ib_mad_cleanup(struct mlx4_ib_dev *dev)
416 {
417 	struct ib_mad_agent *agent;
418 	int p, q;
419 
420 	for (p = 0; p < dev->num_ports; ++p) {
421 		for (q = 0; q <= 1; ++q) {
422 			agent = dev->send_agent[p][q];
423 			if (agent) {
424 				dev->send_agent[p][q] = NULL;
425 				ib_unregister_mad_agent(agent);
426 			}
427 		}
428 
429 		if (dev->sm_ah[p])
430 			ib_destroy_ah(dev->sm_ah[p]);
431 	}
432 }
433