xref: /openbmc/linux/drivers/infiniband/hw/mlx4/mad.c (revision 81d67439)
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 
114 	if (!err)
115 		memcpy(response_mad, outmailbox->buf, 256);
116 
117 	mlx4_free_cmd_mailbox(dev->dev, inmailbox);
118 	mlx4_free_cmd_mailbox(dev->dev, outmailbox);
119 
120 	return err;
121 }
122 
123 static void update_sm_ah(struct mlx4_ib_dev *dev, u8 port_num, u16 lid, u8 sl)
124 {
125 	struct ib_ah *new_ah;
126 	struct ib_ah_attr ah_attr;
127 
128 	if (!dev->send_agent[port_num - 1][0])
129 		return;
130 
131 	memset(&ah_attr, 0, sizeof ah_attr);
132 	ah_attr.dlid     = lid;
133 	ah_attr.sl       = sl;
134 	ah_attr.port_num = port_num;
135 
136 	new_ah = ib_create_ah(dev->send_agent[port_num - 1][0]->qp->pd,
137 			      &ah_attr);
138 	if (IS_ERR(new_ah))
139 		return;
140 
141 	spin_lock(&dev->sm_lock);
142 	if (dev->sm_ah[port_num - 1])
143 		ib_destroy_ah(dev->sm_ah[port_num - 1]);
144 	dev->sm_ah[port_num - 1] = new_ah;
145 	spin_unlock(&dev->sm_lock);
146 }
147 
148 /*
149  * Snoop SM MADs for port info and P_Key table sets, so we can
150  * synthesize LID change and P_Key change events.
151  */
152 static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad,
153 				u16 prev_lid)
154 {
155 	struct ib_event event;
156 
157 	if ((mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
158 	     mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
159 	    mad->mad_hdr.method == IB_MGMT_METHOD_SET) {
160 		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO) {
161 			struct ib_port_info *pinfo =
162 				(struct ib_port_info *) ((struct ib_smp *) mad)->data;
163 			u16 lid = be16_to_cpu(pinfo->lid);
164 
165 			update_sm_ah(to_mdev(ibdev), port_num,
166 				     be16_to_cpu(pinfo->sm_lid),
167 				     pinfo->neighbormtu_mastersmsl & 0xf);
168 
169 			event.device	       = ibdev;
170 			event.element.port_num = port_num;
171 
172 			if (pinfo->clientrereg_resv_subnetto & 0x80) {
173 				event.event    = IB_EVENT_CLIENT_REREGISTER;
174 				ib_dispatch_event(&event);
175 			}
176 
177 			if (prev_lid != lid) {
178 				event.event    = IB_EVENT_LID_CHANGE;
179 				ib_dispatch_event(&event);
180 			}
181 		}
182 
183 		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PKEY_TABLE) {
184 			event.device	       = ibdev;
185 			event.event	       = IB_EVENT_PKEY_CHANGE;
186 			event.element.port_num = port_num;
187 			ib_dispatch_event(&event);
188 		}
189 	}
190 }
191 
192 static void node_desc_override(struct ib_device *dev,
193 			       struct ib_mad *mad)
194 {
195 	if ((mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
196 	     mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
197 	    mad->mad_hdr.method == IB_MGMT_METHOD_GET_RESP &&
198 	    mad->mad_hdr.attr_id == IB_SMP_ATTR_NODE_DESC) {
199 		spin_lock(&to_mdev(dev)->sm_lock);
200 		memcpy(((struct ib_smp *) mad)->data, dev->node_desc, 64);
201 		spin_unlock(&to_mdev(dev)->sm_lock);
202 	}
203 }
204 
205 static void forward_trap(struct mlx4_ib_dev *dev, u8 port_num, struct ib_mad *mad)
206 {
207 	int qpn = mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_SUBN_LID_ROUTED;
208 	struct ib_mad_send_buf *send_buf;
209 	struct ib_mad_agent *agent = dev->send_agent[port_num - 1][qpn];
210 	int ret;
211 
212 	if (agent) {
213 		send_buf = ib_create_send_mad(agent, qpn, 0, 0, IB_MGMT_MAD_HDR,
214 					      IB_MGMT_MAD_DATA, GFP_ATOMIC);
215 		if (IS_ERR(send_buf))
216 			return;
217 		/*
218 		 * We rely here on the fact that MLX QPs don't use the
219 		 * address handle after the send is posted (this is
220 		 * wrong following the IB spec strictly, but we know
221 		 * it's OK for our devices).
222 		 */
223 		spin_lock(&dev->sm_lock);
224 		memcpy(send_buf->mad, mad, sizeof *mad);
225 		if ((send_buf->ah = dev->sm_ah[port_num - 1]))
226 			ret = ib_post_send_mad(send_buf, NULL);
227 		else
228 			ret = -EINVAL;
229 		spin_unlock(&dev->sm_lock);
230 
231 		if (ret)
232 			ib_free_send_mad(send_buf);
233 	}
234 }
235 
236 static int ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
237 			struct ib_wc *in_wc, struct ib_grh *in_grh,
238 			struct ib_mad *in_mad, struct ib_mad *out_mad)
239 {
240 	u16 slid, prev_lid = 0;
241 	int err;
242 	struct ib_port_attr pattr;
243 
244 	slid = in_wc ? in_wc->slid : be16_to_cpu(IB_LID_PERMISSIVE);
245 
246 	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP && slid == 0) {
247 		forward_trap(to_mdev(ibdev), port_num, in_mad);
248 		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
249 	}
250 
251 	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
252 	    in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
253 		if (in_mad->mad_hdr.method   != IB_MGMT_METHOD_GET &&
254 		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_SET &&
255 		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_TRAP_REPRESS)
256 			return IB_MAD_RESULT_SUCCESS;
257 
258 		/*
259 		 * Don't process SMInfo queries or vendor-specific
260 		 * MADs -- the SMA can't handle them.
261 		 */
262 		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_SM_INFO ||
263 		    ((in_mad->mad_hdr.attr_id & IB_SMP_ATTR_VENDOR_MASK) ==
264 		     IB_SMP_ATTR_VENDOR_MASK))
265 			return IB_MAD_RESULT_SUCCESS;
266 	} else if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_PERF_MGMT ||
267 		   in_mad->mad_hdr.mgmt_class == MLX4_IB_VENDOR_CLASS1   ||
268 		   in_mad->mad_hdr.mgmt_class == MLX4_IB_VENDOR_CLASS2   ||
269 		   in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_CONG_MGMT) {
270 		if (in_mad->mad_hdr.method  != IB_MGMT_METHOD_GET &&
271 		    in_mad->mad_hdr.method  != IB_MGMT_METHOD_SET)
272 			return IB_MAD_RESULT_SUCCESS;
273 	} else
274 		return IB_MAD_RESULT_SUCCESS;
275 
276 	if ((in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
277 	     in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
278 	    in_mad->mad_hdr.method == IB_MGMT_METHOD_SET &&
279 	    in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
280 	    !ib_query_port(ibdev, port_num, &pattr))
281 		prev_lid = pattr.lid;
282 
283 	err = mlx4_MAD_IFC(to_mdev(ibdev),
284 			   mad_flags & IB_MAD_IGNORE_MKEY,
285 			   mad_flags & IB_MAD_IGNORE_BKEY,
286 			   port_num, in_wc, in_grh, in_mad, out_mad);
287 	if (err)
288 		return IB_MAD_RESULT_FAILURE;
289 
290 	if (!out_mad->mad_hdr.status) {
291 		smp_snoop(ibdev, port_num, in_mad, prev_lid);
292 		node_desc_override(ibdev, out_mad);
293 	}
294 
295 	/* set return bit in status of directed route responses */
296 	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
297 		out_mad->mad_hdr.status |= cpu_to_be16(1 << 15);
298 
299 	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP_REPRESS)
300 		/* no response for trap repress */
301 		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
302 
303 	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
304 }
305 
306 static void edit_counter(struct mlx4_counter *cnt,
307 					struct ib_pma_portcounters *pma_cnt)
308 {
309 	pma_cnt->port_xmit_data = cpu_to_be32((be64_to_cpu(cnt->tx_bytes)>>2));
310 	pma_cnt->port_rcv_data  = cpu_to_be32((be64_to_cpu(cnt->rx_bytes)>>2));
311 	pma_cnt->port_xmit_packets = cpu_to_be32(be64_to_cpu(cnt->tx_frames));
312 	pma_cnt->port_rcv_packets  = cpu_to_be32(be64_to_cpu(cnt->rx_frames));
313 }
314 
315 static int iboe_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
316 			struct ib_wc *in_wc, struct ib_grh *in_grh,
317 			struct ib_mad *in_mad, struct ib_mad *out_mad)
318 {
319 	struct mlx4_cmd_mailbox *mailbox;
320 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
321 	int err;
322 	u32 inmod = dev->counters[port_num - 1] & 0xffff;
323 	u8 mode;
324 
325 	if (in_mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_PERF_MGMT)
326 		return -EINVAL;
327 
328 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
329 	if (IS_ERR(mailbox))
330 		return IB_MAD_RESULT_FAILURE;
331 
332 	err = mlx4_cmd_box(dev->dev, 0, mailbox->dma, inmod, 0,
333 			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C);
334 	if (err)
335 		err = IB_MAD_RESULT_FAILURE;
336 	else {
337 		memset(out_mad->data, 0, sizeof out_mad->data);
338 		mode = ((struct mlx4_counter *)mailbox->buf)->counter_mode;
339 		switch (mode & 0xf) {
340 		case 0:
341 			edit_counter(mailbox->buf,
342 						(void *)(out_mad->data + 40));
343 			err = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
344 			break;
345 		default:
346 			err = IB_MAD_RESULT_FAILURE;
347 		}
348 	}
349 
350 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
351 
352 	return err;
353 }
354 
355 int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
356 			struct ib_wc *in_wc, struct ib_grh *in_grh,
357 			struct ib_mad *in_mad, struct ib_mad *out_mad)
358 {
359 	switch (rdma_port_get_link_layer(ibdev, port_num)) {
360 	case IB_LINK_LAYER_INFINIBAND:
361 		return ib_process_mad(ibdev, mad_flags, port_num, in_wc,
362 				      in_grh, in_mad, out_mad);
363 	case IB_LINK_LAYER_ETHERNET:
364 		return iboe_process_mad(ibdev, mad_flags, port_num, in_wc,
365 					  in_grh, in_mad, out_mad);
366 	default:
367 		return -EINVAL;
368 	}
369 }
370 
371 static void send_handler(struct ib_mad_agent *agent,
372 			 struct ib_mad_send_wc *mad_send_wc)
373 {
374 	ib_free_send_mad(mad_send_wc->send_buf);
375 }
376 
377 int mlx4_ib_mad_init(struct mlx4_ib_dev *dev)
378 {
379 	struct ib_mad_agent *agent;
380 	int p, q;
381 	int ret;
382 	enum rdma_link_layer ll;
383 
384 	for (p = 0; p < dev->num_ports; ++p) {
385 		ll = rdma_port_get_link_layer(&dev->ib_dev, p + 1);
386 		for (q = 0; q <= 1; ++q) {
387 			if (ll == IB_LINK_LAYER_INFINIBAND) {
388 				agent = ib_register_mad_agent(&dev->ib_dev, p + 1,
389 							      q ? IB_QPT_GSI : IB_QPT_SMI,
390 							      NULL, 0, send_handler,
391 							      NULL, NULL);
392 				if (IS_ERR(agent)) {
393 					ret = PTR_ERR(agent);
394 					goto err;
395 				}
396 				dev->send_agent[p][q] = agent;
397 			} else
398 				dev->send_agent[p][q] = NULL;
399 		}
400 	}
401 
402 	return 0;
403 
404 err:
405 	for (p = 0; p < dev->num_ports; ++p)
406 		for (q = 0; q <= 1; ++q)
407 			if (dev->send_agent[p][q])
408 				ib_unregister_mad_agent(dev->send_agent[p][q]);
409 
410 	return ret;
411 }
412 
413 void mlx4_ib_mad_cleanup(struct mlx4_ib_dev *dev)
414 {
415 	struct ib_mad_agent *agent;
416 	int p, q;
417 
418 	for (p = 0; p < dev->num_ports; ++p) {
419 		for (q = 0; q <= 1; ++q) {
420 			agent = dev->send_agent[p][q];
421 			if (agent) {
422 				dev->send_agent[p][q] = NULL;
423 				ib_unregister_mad_agent(agent);
424 			}
425 		}
426 
427 		if (dev->sm_ah[p])
428 			ib_destroy_ah(dev->sm_ah[p]);
429 	}
430 }
431