xref: /openbmc/linux/drivers/infiniband/core/sysfs.c (revision 9125f19b)
1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Mellanox Technologies Ltd.  All rights reserved.
4  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #include "core_priv.h"
36 
37 #include <linux/slab.h>
38 #include <linux/stat.h>
39 #include <linux/string.h>
40 #include <linux/netdevice.h>
41 
42 #include <rdma/ib_mad.h>
43 #include <rdma/ib_pma.h>
44 
45 struct ib_port;
46 
47 struct gid_attr_group {
48 	struct ib_port		*port;
49 	struct kobject		kobj;
50 	struct attribute_group	ndev;
51 	struct attribute_group	type;
52 };
53 struct ib_port {
54 	struct kobject         kobj;
55 	struct ib_device      *ibdev;
56 	struct gid_attr_group *gid_attr_group;
57 	struct attribute_group gid_group;
58 	struct attribute_group pkey_group;
59 	u8                     port_num;
60 	struct attribute_group *pma_table;
61 };
62 
63 struct port_attribute {
64 	struct attribute attr;
65 	ssize_t (*show)(struct ib_port *, struct port_attribute *, char *buf);
66 	ssize_t (*store)(struct ib_port *, struct port_attribute *,
67 			 const char *buf, size_t count);
68 };
69 
70 #define PORT_ATTR(_name, _mode, _show, _store) \
71 struct port_attribute port_attr_##_name = __ATTR(_name, _mode, _show, _store)
72 
73 #define PORT_ATTR_RO(_name) \
74 struct port_attribute port_attr_##_name = __ATTR_RO(_name)
75 
76 struct port_table_attribute {
77 	struct port_attribute	attr;
78 	char			name[8];
79 	int			index;
80 	__be16			attr_id;
81 };
82 
83 static ssize_t port_attr_show(struct kobject *kobj,
84 			      struct attribute *attr, char *buf)
85 {
86 	struct port_attribute *port_attr =
87 		container_of(attr, struct port_attribute, attr);
88 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
89 
90 	if (!port_attr->show)
91 		return -EIO;
92 
93 	return port_attr->show(p, port_attr, buf);
94 }
95 
96 static const struct sysfs_ops port_sysfs_ops = {
97 	.show = port_attr_show
98 };
99 
100 static ssize_t gid_attr_show(struct kobject *kobj,
101 			     struct attribute *attr, char *buf)
102 {
103 	struct port_attribute *port_attr =
104 		container_of(attr, struct port_attribute, attr);
105 	struct ib_port *p = container_of(kobj, struct gid_attr_group,
106 					 kobj)->port;
107 
108 	if (!port_attr->show)
109 		return -EIO;
110 
111 	return port_attr->show(p, port_attr, buf);
112 }
113 
114 static const struct sysfs_ops gid_attr_sysfs_ops = {
115 	.show = gid_attr_show
116 };
117 
118 static ssize_t state_show(struct ib_port *p, struct port_attribute *unused,
119 			  char *buf)
120 {
121 	struct ib_port_attr attr;
122 	ssize_t ret;
123 
124 	static const char *state_name[] = {
125 		[IB_PORT_NOP]		= "NOP",
126 		[IB_PORT_DOWN]		= "DOWN",
127 		[IB_PORT_INIT]		= "INIT",
128 		[IB_PORT_ARMED]		= "ARMED",
129 		[IB_PORT_ACTIVE]	= "ACTIVE",
130 		[IB_PORT_ACTIVE_DEFER]	= "ACTIVE_DEFER"
131 	};
132 
133 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
134 	if (ret)
135 		return ret;
136 
137 	return sprintf(buf, "%d: %s\n", attr.state,
138 		       attr.state >= 0 && attr.state < ARRAY_SIZE(state_name) ?
139 		       state_name[attr.state] : "UNKNOWN");
140 }
141 
142 static ssize_t lid_show(struct ib_port *p, struct port_attribute *unused,
143 			char *buf)
144 {
145 	struct ib_port_attr attr;
146 	ssize_t ret;
147 
148 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
149 	if (ret)
150 		return ret;
151 
152 	return sprintf(buf, "0x%x\n", attr.lid);
153 }
154 
155 static ssize_t lid_mask_count_show(struct ib_port *p,
156 				   struct port_attribute *unused,
157 				   char *buf)
158 {
159 	struct ib_port_attr attr;
160 	ssize_t ret;
161 
162 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
163 	if (ret)
164 		return ret;
165 
166 	return sprintf(buf, "%d\n", attr.lmc);
167 }
168 
169 static ssize_t sm_lid_show(struct ib_port *p, struct port_attribute *unused,
170 			   char *buf)
171 {
172 	struct ib_port_attr attr;
173 	ssize_t ret;
174 
175 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
176 	if (ret)
177 		return ret;
178 
179 	return sprintf(buf, "0x%x\n", attr.sm_lid);
180 }
181 
182 static ssize_t sm_sl_show(struct ib_port *p, struct port_attribute *unused,
183 			  char *buf)
184 {
185 	struct ib_port_attr attr;
186 	ssize_t ret;
187 
188 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
189 	if (ret)
190 		return ret;
191 
192 	return sprintf(buf, "%d\n", attr.sm_sl);
193 }
194 
195 static ssize_t cap_mask_show(struct ib_port *p, struct port_attribute *unused,
196 			     char *buf)
197 {
198 	struct ib_port_attr attr;
199 	ssize_t ret;
200 
201 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
202 	if (ret)
203 		return ret;
204 
205 	return sprintf(buf, "0x%08x\n", attr.port_cap_flags);
206 }
207 
208 static ssize_t rate_show(struct ib_port *p, struct port_attribute *unused,
209 			 char *buf)
210 {
211 	struct ib_port_attr attr;
212 	char *speed = "";
213 	int rate;		/* in deci-Gb/sec */
214 	ssize_t ret;
215 
216 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
217 	if (ret)
218 		return ret;
219 
220 	switch (attr.active_speed) {
221 	case IB_SPEED_DDR:
222 		speed = " DDR";
223 		rate = 50;
224 		break;
225 	case IB_SPEED_QDR:
226 		speed = " QDR";
227 		rate = 100;
228 		break;
229 	case IB_SPEED_FDR10:
230 		speed = " FDR10";
231 		rate = 100;
232 		break;
233 	case IB_SPEED_FDR:
234 		speed = " FDR";
235 		rate = 140;
236 		break;
237 	case IB_SPEED_EDR:
238 		speed = " EDR";
239 		rate = 250;
240 		break;
241 	case IB_SPEED_SDR:
242 	default:		/* default to SDR for invalid rates */
243 		rate = 25;
244 		break;
245 	}
246 
247 	rate *= ib_width_enum_to_int(attr.active_width);
248 	if (rate < 0)
249 		return -EINVAL;
250 
251 	return sprintf(buf, "%d%s Gb/sec (%dX%s)\n",
252 		       rate / 10, rate % 10 ? ".5" : "",
253 		       ib_width_enum_to_int(attr.active_width), speed);
254 }
255 
256 static ssize_t phys_state_show(struct ib_port *p, struct port_attribute *unused,
257 			       char *buf)
258 {
259 	struct ib_port_attr attr;
260 
261 	ssize_t ret;
262 
263 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
264 	if (ret)
265 		return ret;
266 
267 	switch (attr.phys_state) {
268 	case 1:  return sprintf(buf, "1: Sleep\n");
269 	case 2:  return sprintf(buf, "2: Polling\n");
270 	case 3:  return sprintf(buf, "3: Disabled\n");
271 	case 4:  return sprintf(buf, "4: PortConfigurationTraining\n");
272 	case 5:  return sprintf(buf, "5: LinkUp\n");
273 	case 6:  return sprintf(buf, "6: LinkErrorRecovery\n");
274 	case 7:  return sprintf(buf, "7: Phy Test\n");
275 	default: return sprintf(buf, "%d: <unknown>\n", attr.phys_state);
276 	}
277 }
278 
279 static ssize_t link_layer_show(struct ib_port *p, struct port_attribute *unused,
280 			       char *buf)
281 {
282 	switch (rdma_port_get_link_layer(p->ibdev, p->port_num)) {
283 	case IB_LINK_LAYER_INFINIBAND:
284 		return sprintf(buf, "%s\n", "InfiniBand");
285 	case IB_LINK_LAYER_ETHERNET:
286 		return sprintf(buf, "%s\n", "Ethernet");
287 	default:
288 		return sprintf(buf, "%s\n", "Unknown");
289 	}
290 }
291 
292 static PORT_ATTR_RO(state);
293 static PORT_ATTR_RO(lid);
294 static PORT_ATTR_RO(lid_mask_count);
295 static PORT_ATTR_RO(sm_lid);
296 static PORT_ATTR_RO(sm_sl);
297 static PORT_ATTR_RO(cap_mask);
298 static PORT_ATTR_RO(rate);
299 static PORT_ATTR_RO(phys_state);
300 static PORT_ATTR_RO(link_layer);
301 
302 static struct attribute *port_default_attrs[] = {
303 	&port_attr_state.attr,
304 	&port_attr_lid.attr,
305 	&port_attr_lid_mask_count.attr,
306 	&port_attr_sm_lid.attr,
307 	&port_attr_sm_sl.attr,
308 	&port_attr_cap_mask.attr,
309 	&port_attr_rate.attr,
310 	&port_attr_phys_state.attr,
311 	&port_attr_link_layer.attr,
312 	NULL
313 };
314 
315 static size_t print_ndev(struct ib_gid_attr *gid_attr, char *buf)
316 {
317 	if (!gid_attr->ndev)
318 		return -EINVAL;
319 
320 	return sprintf(buf, "%s\n", gid_attr->ndev->name);
321 }
322 
323 static size_t print_gid_type(struct ib_gid_attr *gid_attr, char *buf)
324 {
325 	return sprintf(buf, "%s\n", ib_cache_gid_type_str(gid_attr->gid_type));
326 }
327 
328 static ssize_t _show_port_gid_attr(struct ib_port *p,
329 				   struct port_attribute *attr,
330 				   char *buf,
331 				   size_t (*print)(struct ib_gid_attr *gid_attr,
332 						   char *buf))
333 {
334 	struct port_table_attribute *tab_attr =
335 		container_of(attr, struct port_table_attribute, attr);
336 	union ib_gid gid;
337 	struct ib_gid_attr gid_attr = {};
338 	ssize_t ret;
339 
340 	ret = ib_query_gid(p->ibdev, p->port_num, tab_attr->index, &gid,
341 			   &gid_attr);
342 	if (ret)
343 		goto err;
344 
345 	ret = print(&gid_attr, buf);
346 
347 err:
348 	if (gid_attr.ndev)
349 		dev_put(gid_attr.ndev);
350 	return ret;
351 }
352 
353 static ssize_t show_port_gid(struct ib_port *p, struct port_attribute *attr,
354 			     char *buf)
355 {
356 	struct port_table_attribute *tab_attr =
357 		container_of(attr, struct port_table_attribute, attr);
358 	union ib_gid gid;
359 	ssize_t ret;
360 
361 	ret = ib_query_gid(p->ibdev, p->port_num, tab_attr->index, &gid, NULL);
362 	if (ret)
363 		return ret;
364 
365 	return sprintf(buf, "%pI6\n", gid.raw);
366 }
367 
368 static ssize_t show_port_gid_attr_ndev(struct ib_port *p,
369 				       struct port_attribute *attr, char *buf)
370 {
371 	return _show_port_gid_attr(p, attr, buf, print_ndev);
372 }
373 
374 static ssize_t show_port_gid_attr_gid_type(struct ib_port *p,
375 					   struct port_attribute *attr,
376 					   char *buf)
377 {
378 	return _show_port_gid_attr(p, attr, buf, print_gid_type);
379 }
380 
381 static ssize_t show_port_pkey(struct ib_port *p, struct port_attribute *attr,
382 			      char *buf)
383 {
384 	struct port_table_attribute *tab_attr =
385 		container_of(attr, struct port_table_attribute, attr);
386 	u16 pkey;
387 	ssize_t ret;
388 
389 	ret = ib_query_pkey(p->ibdev, p->port_num, tab_attr->index, &pkey);
390 	if (ret)
391 		return ret;
392 
393 	return sprintf(buf, "0x%04x\n", pkey);
394 }
395 
396 #define PORT_PMA_ATTR(_name, _counter, _width, _offset)			\
397 struct port_table_attribute port_pma_attr_##_name = {			\
398 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
399 	.index = (_offset) | ((_width) << 16) | ((_counter) << 24),	\
400 	.attr_id = IB_PMA_PORT_COUNTERS ,				\
401 }
402 
403 #define PORT_PMA_ATTR_EXT(_name, _width, _offset)			\
404 struct port_table_attribute port_pma_attr_ext_##_name = {		\
405 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
406 	.index = (_offset) | ((_width) << 16),				\
407 	.attr_id = IB_PMA_PORT_COUNTERS_EXT ,				\
408 }
409 
410 /*
411  * Get a Perfmgmt MAD block of data.
412  * Returns error code or the number of bytes retrieved.
413  */
414 static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
415 		void *data, int offset, size_t size)
416 {
417 	struct ib_mad *in_mad;
418 	struct ib_mad *out_mad;
419 	size_t mad_size = sizeof(*out_mad);
420 	u16 out_mad_pkey_index = 0;
421 	ssize_t ret;
422 
423 	if (!dev->process_mad)
424 		return -ENOSYS;
425 
426 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
427 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
428 	if (!in_mad || !out_mad) {
429 		ret = -ENOMEM;
430 		goto out;
431 	}
432 
433 	in_mad->mad_hdr.base_version  = 1;
434 	in_mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_PERF_MGMT;
435 	in_mad->mad_hdr.class_version = 1;
436 	in_mad->mad_hdr.method        = IB_MGMT_METHOD_GET;
437 	in_mad->mad_hdr.attr_id       = attr;
438 
439 	if (attr != IB_PMA_CLASS_PORT_INFO)
440 		in_mad->data[41] = port_num;	/* PortSelect field */
441 
442 	if ((dev->process_mad(dev, IB_MAD_IGNORE_MKEY,
443 		 port_num, NULL, NULL,
444 		 (const struct ib_mad_hdr *)in_mad, mad_size,
445 		 (struct ib_mad_hdr *)out_mad, &mad_size,
446 		 &out_mad_pkey_index) &
447 	     (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
448 	    (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
449 		ret = -EINVAL;
450 		goto out;
451 	}
452 	memcpy(data, out_mad->data + offset, size);
453 	ret = size;
454 out:
455 	kfree(in_mad);
456 	kfree(out_mad);
457 	return ret;
458 }
459 
460 static ssize_t show_pma_counter(struct ib_port *p, struct port_attribute *attr,
461 				char *buf)
462 {
463 	struct port_table_attribute *tab_attr =
464 		container_of(attr, struct port_table_attribute, attr);
465 	int offset = tab_attr->index & 0xffff;
466 	int width  = (tab_attr->index >> 16) & 0xff;
467 	ssize_t ret;
468 	u8 data[8];
469 
470 	ret = get_perf_mad(p->ibdev, p->port_num, tab_attr->attr_id, &data,
471 			40 + offset / 8, sizeof(data));
472 	if (ret < 0)
473 		return sprintf(buf, "N/A (no PMA)\n");
474 
475 	switch (width) {
476 	case 4:
477 		ret = sprintf(buf, "%u\n", (*data >>
478 					    (4 - (offset % 8))) & 0xf);
479 		break;
480 	case 8:
481 		ret = sprintf(buf, "%u\n", *data);
482 		break;
483 	case 16:
484 		ret = sprintf(buf, "%u\n",
485 			      be16_to_cpup((__be16 *)data));
486 		break;
487 	case 32:
488 		ret = sprintf(buf, "%u\n",
489 			      be32_to_cpup((__be32 *)data));
490 		break;
491 	case 64:
492 		ret = sprintf(buf, "%llu\n",
493 				be64_to_cpup((__be64 *)data));
494 		break;
495 
496 	default:
497 		ret = 0;
498 	}
499 
500 	return ret;
501 }
502 
503 static PORT_PMA_ATTR(symbol_error		    ,  0, 16,  32);
504 static PORT_PMA_ATTR(link_error_recovery	    ,  1,  8,  48);
505 static PORT_PMA_ATTR(link_downed		    ,  2,  8,  56);
506 static PORT_PMA_ATTR(port_rcv_errors		    ,  3, 16,  64);
507 static PORT_PMA_ATTR(port_rcv_remote_physical_errors,  4, 16,  80);
508 static PORT_PMA_ATTR(port_rcv_switch_relay_errors   ,  5, 16,  96);
509 static PORT_PMA_ATTR(port_xmit_discards		    ,  6, 16, 112);
510 static PORT_PMA_ATTR(port_xmit_constraint_errors    ,  7,  8, 128);
511 static PORT_PMA_ATTR(port_rcv_constraint_errors	    ,  8,  8, 136);
512 static PORT_PMA_ATTR(local_link_integrity_errors    ,  9,  4, 152);
513 static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10,  4, 156);
514 static PORT_PMA_ATTR(VL15_dropped		    , 11, 16, 176);
515 static PORT_PMA_ATTR(port_xmit_data		    , 12, 32, 192);
516 static PORT_PMA_ATTR(port_rcv_data		    , 13, 32, 224);
517 static PORT_PMA_ATTR(port_xmit_packets		    , 14, 32, 256);
518 static PORT_PMA_ATTR(port_rcv_packets		    , 15, 32, 288);
519 
520 /*
521  * Counters added by extended set
522  */
523 static PORT_PMA_ATTR_EXT(port_xmit_data		    , 64,  64);
524 static PORT_PMA_ATTR_EXT(port_rcv_data		    , 64, 128);
525 static PORT_PMA_ATTR_EXT(port_xmit_packets	    , 64, 192);
526 static PORT_PMA_ATTR_EXT(port_rcv_packets	    , 64, 256);
527 static PORT_PMA_ATTR_EXT(unicast_xmit_packets	    , 64, 320);
528 static PORT_PMA_ATTR_EXT(unicast_rcv_packets	    , 64, 384);
529 static PORT_PMA_ATTR_EXT(multicast_xmit_packets	    , 64, 448);
530 static PORT_PMA_ATTR_EXT(multicast_rcv_packets	    , 64, 512);
531 
532 static struct attribute *pma_attrs[] = {
533 	&port_pma_attr_symbol_error.attr.attr,
534 	&port_pma_attr_link_error_recovery.attr.attr,
535 	&port_pma_attr_link_downed.attr.attr,
536 	&port_pma_attr_port_rcv_errors.attr.attr,
537 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
538 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
539 	&port_pma_attr_port_xmit_discards.attr.attr,
540 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
541 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
542 	&port_pma_attr_local_link_integrity_errors.attr.attr,
543 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
544 	&port_pma_attr_VL15_dropped.attr.attr,
545 	&port_pma_attr_port_xmit_data.attr.attr,
546 	&port_pma_attr_port_rcv_data.attr.attr,
547 	&port_pma_attr_port_xmit_packets.attr.attr,
548 	&port_pma_attr_port_rcv_packets.attr.attr,
549 	NULL
550 };
551 
552 static struct attribute *pma_attrs_ext[] = {
553 	&port_pma_attr_symbol_error.attr.attr,
554 	&port_pma_attr_link_error_recovery.attr.attr,
555 	&port_pma_attr_link_downed.attr.attr,
556 	&port_pma_attr_port_rcv_errors.attr.attr,
557 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
558 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
559 	&port_pma_attr_port_xmit_discards.attr.attr,
560 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
561 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
562 	&port_pma_attr_local_link_integrity_errors.attr.attr,
563 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
564 	&port_pma_attr_VL15_dropped.attr.attr,
565 	&port_pma_attr_ext_port_xmit_data.attr.attr,
566 	&port_pma_attr_ext_port_rcv_data.attr.attr,
567 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
568 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
569 	&port_pma_attr_ext_unicast_rcv_packets.attr.attr,
570 	&port_pma_attr_ext_unicast_xmit_packets.attr.attr,
571 	&port_pma_attr_ext_multicast_rcv_packets.attr.attr,
572 	&port_pma_attr_ext_multicast_xmit_packets.attr.attr,
573 	NULL
574 };
575 
576 static struct attribute *pma_attrs_noietf[] = {
577 	&port_pma_attr_symbol_error.attr.attr,
578 	&port_pma_attr_link_error_recovery.attr.attr,
579 	&port_pma_attr_link_downed.attr.attr,
580 	&port_pma_attr_port_rcv_errors.attr.attr,
581 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
582 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
583 	&port_pma_attr_port_xmit_discards.attr.attr,
584 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
585 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
586 	&port_pma_attr_local_link_integrity_errors.attr.attr,
587 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
588 	&port_pma_attr_VL15_dropped.attr.attr,
589 	&port_pma_attr_ext_port_xmit_data.attr.attr,
590 	&port_pma_attr_ext_port_rcv_data.attr.attr,
591 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
592 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
593 	NULL
594 };
595 
596 static struct attribute_group pma_group = {
597 	.name  = "counters",
598 	.attrs  = pma_attrs
599 };
600 
601 static struct attribute_group pma_group_ext = {
602 	.name  = "counters",
603 	.attrs  = pma_attrs_ext
604 };
605 
606 static struct attribute_group pma_group_noietf = {
607 	.name  = "counters",
608 	.attrs  = pma_attrs_noietf
609 };
610 
611 static void ib_port_release(struct kobject *kobj)
612 {
613 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
614 	struct attribute *a;
615 	int i;
616 
617 	if (p->gid_group.attrs) {
618 		for (i = 0; (a = p->gid_group.attrs[i]); ++i)
619 			kfree(a);
620 
621 		kfree(p->gid_group.attrs);
622 	}
623 
624 	if (p->pkey_group.attrs) {
625 		for (i = 0; (a = p->pkey_group.attrs[i]); ++i)
626 			kfree(a);
627 
628 		kfree(p->pkey_group.attrs);
629 	}
630 
631 	kfree(p);
632 }
633 
634 static void ib_port_gid_attr_release(struct kobject *kobj)
635 {
636 	struct gid_attr_group *g = container_of(kobj, struct gid_attr_group,
637 						kobj);
638 	struct attribute *a;
639 	int i;
640 
641 	if (g->ndev.attrs) {
642 		for (i = 0; (a = g->ndev.attrs[i]); ++i)
643 			kfree(a);
644 
645 		kfree(g->ndev.attrs);
646 	}
647 
648 	if (g->type.attrs) {
649 		for (i = 0; (a = g->type.attrs[i]); ++i)
650 			kfree(a);
651 
652 		kfree(g->type.attrs);
653 	}
654 
655 	kfree(g);
656 }
657 
658 static struct kobj_type port_type = {
659 	.release       = ib_port_release,
660 	.sysfs_ops     = &port_sysfs_ops,
661 	.default_attrs = port_default_attrs
662 };
663 
664 static struct kobj_type gid_attr_type = {
665 	.sysfs_ops      = &gid_attr_sysfs_ops,
666 	.release        = ib_port_gid_attr_release
667 };
668 
669 static struct attribute **
670 alloc_group_attrs(ssize_t (*show)(struct ib_port *,
671 				  struct port_attribute *, char *buf),
672 		  int len)
673 {
674 	struct attribute **tab_attr;
675 	struct port_table_attribute *element;
676 	int i;
677 
678 	tab_attr = kcalloc(1 + len, sizeof(struct attribute *), GFP_KERNEL);
679 	if (!tab_attr)
680 		return NULL;
681 
682 	for (i = 0; i < len; i++) {
683 		element = kzalloc(sizeof(struct port_table_attribute),
684 				  GFP_KERNEL);
685 		if (!element)
686 			goto err;
687 
688 		if (snprintf(element->name, sizeof(element->name),
689 			     "%d", i) >= sizeof(element->name)) {
690 			kfree(element);
691 			goto err;
692 		}
693 
694 		element->attr.attr.name  = element->name;
695 		element->attr.attr.mode  = S_IRUGO;
696 		element->attr.show       = show;
697 		element->index		 = i;
698 		sysfs_attr_init(&element->attr.attr);
699 
700 		tab_attr[i] = &element->attr.attr;
701 	}
702 
703 	return tab_attr;
704 
705 err:
706 	while (--i >= 0)
707 		kfree(tab_attr[i]);
708 	kfree(tab_attr);
709 	return NULL;
710 }
711 
712 /*
713  * Figure out which counter table to use depending on
714  * the device capabilities.
715  */
716 static struct attribute_group *get_counter_table(struct ib_device *dev,
717 						 int port_num)
718 {
719 	struct ib_class_port_info cpi;
720 
721 	if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
722 				&cpi, 40, sizeof(cpi)) >= 0) {
723 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
724 			/* We have extended counters */
725 			return &pma_group_ext;
726 
727 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
728 			/* But not the IETF ones */
729 			return &pma_group_noietf;
730 	}
731 
732 	/* Fall back to normal counters */
733 	return &pma_group;
734 }
735 
736 static int add_port(struct ib_device *device, int port_num,
737 		    int (*port_callback)(struct ib_device *,
738 					 u8, struct kobject *))
739 {
740 	struct ib_port *p;
741 	struct ib_port_attr attr;
742 	int i;
743 	int ret;
744 
745 	ret = ib_query_port(device, port_num, &attr);
746 	if (ret)
747 		return ret;
748 
749 	p = kzalloc(sizeof *p, GFP_KERNEL);
750 	if (!p)
751 		return -ENOMEM;
752 
753 	p->ibdev      = device;
754 	p->port_num   = port_num;
755 
756 	ret = kobject_init_and_add(&p->kobj, &port_type,
757 				   device->ports_parent,
758 				   "%d", port_num);
759 	if (ret) {
760 		kfree(p);
761 		return ret;
762 	}
763 
764 	p->gid_attr_group = kzalloc(sizeof(*p->gid_attr_group), GFP_KERNEL);
765 	if (!p->gid_attr_group) {
766 		ret = -ENOMEM;
767 		goto err_put;
768 	}
769 
770 	p->gid_attr_group->port = p;
771 	ret = kobject_init_and_add(&p->gid_attr_group->kobj, &gid_attr_type,
772 				   &p->kobj, "gid_attrs");
773 	if (ret) {
774 		kfree(p->gid_attr_group);
775 		goto err_put;
776 	}
777 
778 	p->pma_table = get_counter_table(device, port_num);
779 	ret = sysfs_create_group(&p->kobj, p->pma_table);
780 	if (ret)
781 		goto err_put_gid_attrs;
782 
783 	p->gid_group.name  = "gids";
784 	p->gid_group.attrs = alloc_group_attrs(show_port_gid, attr.gid_tbl_len);
785 	if (!p->gid_group.attrs) {
786 		ret = -ENOMEM;
787 		goto err_remove_pma;
788 	}
789 
790 	ret = sysfs_create_group(&p->kobj, &p->gid_group);
791 	if (ret)
792 		goto err_free_gid;
793 
794 	p->gid_attr_group->ndev.name = "ndevs";
795 	p->gid_attr_group->ndev.attrs = alloc_group_attrs(show_port_gid_attr_ndev,
796 							  attr.gid_tbl_len);
797 	if (!p->gid_attr_group->ndev.attrs) {
798 		ret = -ENOMEM;
799 		goto err_remove_gid;
800 	}
801 
802 	ret = sysfs_create_group(&p->gid_attr_group->kobj,
803 				 &p->gid_attr_group->ndev);
804 	if (ret)
805 		goto err_free_gid_ndev;
806 
807 	p->gid_attr_group->type.name = "types";
808 	p->gid_attr_group->type.attrs = alloc_group_attrs(show_port_gid_attr_gid_type,
809 							  attr.gid_tbl_len);
810 	if (!p->gid_attr_group->type.attrs) {
811 		ret = -ENOMEM;
812 		goto err_remove_gid_ndev;
813 	}
814 
815 	ret = sysfs_create_group(&p->gid_attr_group->kobj,
816 				 &p->gid_attr_group->type);
817 	if (ret)
818 		goto err_free_gid_type;
819 
820 	p->pkey_group.name  = "pkeys";
821 	p->pkey_group.attrs = alloc_group_attrs(show_port_pkey,
822 						attr.pkey_tbl_len);
823 	if (!p->pkey_group.attrs) {
824 		ret = -ENOMEM;
825 		goto err_remove_gid_type;
826 	}
827 
828 	ret = sysfs_create_group(&p->kobj, &p->pkey_group);
829 	if (ret)
830 		goto err_free_pkey;
831 
832 	if (port_callback) {
833 		ret = port_callback(device, port_num, &p->kobj);
834 		if (ret)
835 			goto err_remove_pkey;
836 	}
837 
838 	list_add_tail(&p->kobj.entry, &device->port_list);
839 
840 	kobject_uevent(&p->kobj, KOBJ_ADD);
841 	return 0;
842 
843 err_remove_pkey:
844 	sysfs_remove_group(&p->kobj, &p->pkey_group);
845 
846 err_free_pkey:
847 	for (i = 0; i < attr.pkey_tbl_len; ++i)
848 		kfree(p->pkey_group.attrs[i]);
849 
850 	kfree(p->pkey_group.attrs);
851 	p->pkey_group.attrs = NULL;
852 
853 err_remove_gid_type:
854 	sysfs_remove_group(&p->gid_attr_group->kobj,
855 			   &p->gid_attr_group->type);
856 
857 err_free_gid_type:
858 	for (i = 0; i < attr.gid_tbl_len; ++i)
859 		kfree(p->gid_attr_group->type.attrs[i]);
860 
861 	kfree(p->gid_attr_group->type.attrs);
862 	p->gid_attr_group->type.attrs = NULL;
863 
864 err_remove_gid_ndev:
865 	sysfs_remove_group(&p->gid_attr_group->kobj,
866 			   &p->gid_attr_group->ndev);
867 
868 err_free_gid_ndev:
869 	for (i = 0; i < attr.gid_tbl_len; ++i)
870 		kfree(p->gid_attr_group->ndev.attrs[i]);
871 
872 	kfree(p->gid_attr_group->ndev.attrs);
873 	p->gid_attr_group->ndev.attrs = NULL;
874 
875 err_remove_gid:
876 	sysfs_remove_group(&p->kobj, &p->gid_group);
877 
878 err_free_gid:
879 	for (i = 0; i < attr.gid_tbl_len; ++i)
880 		kfree(p->gid_group.attrs[i]);
881 
882 	kfree(p->gid_group.attrs);
883 	p->gid_group.attrs = NULL;
884 
885 err_remove_pma:
886 	sysfs_remove_group(&p->kobj, p->pma_table);
887 
888 err_put_gid_attrs:
889 	kobject_put(&p->gid_attr_group->kobj);
890 
891 err_put:
892 	kobject_put(&p->kobj);
893 	return ret;
894 }
895 
896 static ssize_t show_node_type(struct device *device,
897 			      struct device_attribute *attr, char *buf)
898 {
899 	struct ib_device *dev = container_of(device, struct ib_device, dev);
900 
901 	switch (dev->node_type) {
902 	case RDMA_NODE_IB_CA:	  return sprintf(buf, "%d: CA\n", dev->node_type);
903 	case RDMA_NODE_RNIC:	  return sprintf(buf, "%d: RNIC\n", dev->node_type);
904 	case RDMA_NODE_USNIC:	  return sprintf(buf, "%d: usNIC\n", dev->node_type);
905 	case RDMA_NODE_USNIC_UDP: return sprintf(buf, "%d: usNIC UDP\n", dev->node_type);
906 	case RDMA_NODE_IB_SWITCH: return sprintf(buf, "%d: switch\n", dev->node_type);
907 	case RDMA_NODE_IB_ROUTER: return sprintf(buf, "%d: router\n", dev->node_type);
908 	default:		  return sprintf(buf, "%d: <unknown>\n", dev->node_type);
909 	}
910 }
911 
912 static ssize_t show_sys_image_guid(struct device *device,
913 				   struct device_attribute *dev_attr, char *buf)
914 {
915 	struct ib_device *dev = container_of(device, struct ib_device, dev);
916 
917 	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
918 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[0]),
919 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[1]),
920 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[2]),
921 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[3]));
922 }
923 
924 static ssize_t show_node_guid(struct device *device,
925 			      struct device_attribute *attr, char *buf)
926 {
927 	struct ib_device *dev = container_of(device, struct ib_device, dev);
928 
929 	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
930 		       be16_to_cpu(((__be16 *) &dev->node_guid)[0]),
931 		       be16_to_cpu(((__be16 *) &dev->node_guid)[1]),
932 		       be16_to_cpu(((__be16 *) &dev->node_guid)[2]),
933 		       be16_to_cpu(((__be16 *) &dev->node_guid)[3]));
934 }
935 
936 static ssize_t show_node_desc(struct device *device,
937 			      struct device_attribute *attr, char *buf)
938 {
939 	struct ib_device *dev = container_of(device, struct ib_device, dev);
940 
941 	return sprintf(buf, "%.64s\n", dev->node_desc);
942 }
943 
944 static ssize_t set_node_desc(struct device *device,
945 			     struct device_attribute *attr,
946 			     const char *buf, size_t count)
947 {
948 	struct ib_device *dev = container_of(device, struct ib_device, dev);
949 	struct ib_device_modify desc = {};
950 	int ret;
951 
952 	if (!dev->modify_device)
953 		return -EIO;
954 
955 	memcpy(desc.node_desc, buf, min_t(int, count, 64));
956 	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
957 	if (ret)
958 		return ret;
959 
960 	return count;
961 }
962 
963 static DEVICE_ATTR(node_type, S_IRUGO, show_node_type, NULL);
964 static DEVICE_ATTR(sys_image_guid, S_IRUGO, show_sys_image_guid, NULL);
965 static DEVICE_ATTR(node_guid, S_IRUGO, show_node_guid, NULL);
966 static DEVICE_ATTR(node_desc, S_IRUGO | S_IWUSR, show_node_desc, set_node_desc);
967 
968 static struct device_attribute *ib_class_attributes[] = {
969 	&dev_attr_node_type,
970 	&dev_attr_sys_image_guid,
971 	&dev_attr_node_guid,
972 	&dev_attr_node_desc
973 };
974 
975 /* Show a given an attribute in the statistics group */
976 static ssize_t show_protocol_stat(const struct device *device,
977 			    struct device_attribute *attr, char *buf,
978 			    unsigned offset)
979 {
980 	struct ib_device *dev = container_of(device, struct ib_device, dev);
981 	union rdma_protocol_stats stats;
982 	ssize_t ret;
983 
984 	ret = dev->get_protocol_stats(dev, &stats);
985 	if (ret)
986 		return ret;
987 
988 	return sprintf(buf, "%llu\n",
989 		       (unsigned long long) ((u64 *) &stats)[offset]);
990 }
991 
992 /* generate a read-only iwarp statistics attribute */
993 #define IW_STATS_ENTRY(name)						\
994 static ssize_t show_##name(struct device *device,			\
995 			   struct device_attribute *attr, char *buf)	\
996 {									\
997 	return show_protocol_stat(device, attr, buf,			\
998 				  offsetof(struct iw_protocol_stats, name) / \
999 				  sizeof (u64));			\
1000 }									\
1001 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
1002 
1003 IW_STATS_ENTRY(ipInReceives);
1004 IW_STATS_ENTRY(ipInHdrErrors);
1005 IW_STATS_ENTRY(ipInTooBigErrors);
1006 IW_STATS_ENTRY(ipInNoRoutes);
1007 IW_STATS_ENTRY(ipInAddrErrors);
1008 IW_STATS_ENTRY(ipInUnknownProtos);
1009 IW_STATS_ENTRY(ipInTruncatedPkts);
1010 IW_STATS_ENTRY(ipInDiscards);
1011 IW_STATS_ENTRY(ipInDelivers);
1012 IW_STATS_ENTRY(ipOutForwDatagrams);
1013 IW_STATS_ENTRY(ipOutRequests);
1014 IW_STATS_ENTRY(ipOutDiscards);
1015 IW_STATS_ENTRY(ipOutNoRoutes);
1016 IW_STATS_ENTRY(ipReasmTimeout);
1017 IW_STATS_ENTRY(ipReasmReqds);
1018 IW_STATS_ENTRY(ipReasmOKs);
1019 IW_STATS_ENTRY(ipReasmFails);
1020 IW_STATS_ENTRY(ipFragOKs);
1021 IW_STATS_ENTRY(ipFragFails);
1022 IW_STATS_ENTRY(ipFragCreates);
1023 IW_STATS_ENTRY(ipInMcastPkts);
1024 IW_STATS_ENTRY(ipOutMcastPkts);
1025 IW_STATS_ENTRY(ipInBcastPkts);
1026 IW_STATS_ENTRY(ipOutBcastPkts);
1027 IW_STATS_ENTRY(tcpRtoAlgorithm);
1028 IW_STATS_ENTRY(tcpRtoMin);
1029 IW_STATS_ENTRY(tcpRtoMax);
1030 IW_STATS_ENTRY(tcpMaxConn);
1031 IW_STATS_ENTRY(tcpActiveOpens);
1032 IW_STATS_ENTRY(tcpPassiveOpens);
1033 IW_STATS_ENTRY(tcpAttemptFails);
1034 IW_STATS_ENTRY(tcpEstabResets);
1035 IW_STATS_ENTRY(tcpCurrEstab);
1036 IW_STATS_ENTRY(tcpInSegs);
1037 IW_STATS_ENTRY(tcpOutSegs);
1038 IW_STATS_ENTRY(tcpRetransSegs);
1039 IW_STATS_ENTRY(tcpInErrs);
1040 IW_STATS_ENTRY(tcpOutRsts);
1041 
1042 static struct attribute *iw_proto_stats_attrs[] = {
1043 	&dev_attr_ipInReceives.attr,
1044 	&dev_attr_ipInHdrErrors.attr,
1045 	&dev_attr_ipInTooBigErrors.attr,
1046 	&dev_attr_ipInNoRoutes.attr,
1047 	&dev_attr_ipInAddrErrors.attr,
1048 	&dev_attr_ipInUnknownProtos.attr,
1049 	&dev_attr_ipInTruncatedPkts.attr,
1050 	&dev_attr_ipInDiscards.attr,
1051 	&dev_attr_ipInDelivers.attr,
1052 	&dev_attr_ipOutForwDatagrams.attr,
1053 	&dev_attr_ipOutRequests.attr,
1054 	&dev_attr_ipOutDiscards.attr,
1055 	&dev_attr_ipOutNoRoutes.attr,
1056 	&dev_attr_ipReasmTimeout.attr,
1057 	&dev_attr_ipReasmReqds.attr,
1058 	&dev_attr_ipReasmOKs.attr,
1059 	&dev_attr_ipReasmFails.attr,
1060 	&dev_attr_ipFragOKs.attr,
1061 	&dev_attr_ipFragFails.attr,
1062 	&dev_attr_ipFragCreates.attr,
1063 	&dev_attr_ipInMcastPkts.attr,
1064 	&dev_attr_ipOutMcastPkts.attr,
1065 	&dev_attr_ipInBcastPkts.attr,
1066 	&dev_attr_ipOutBcastPkts.attr,
1067 	&dev_attr_tcpRtoAlgorithm.attr,
1068 	&dev_attr_tcpRtoMin.attr,
1069 	&dev_attr_tcpRtoMax.attr,
1070 	&dev_attr_tcpMaxConn.attr,
1071 	&dev_attr_tcpActiveOpens.attr,
1072 	&dev_attr_tcpPassiveOpens.attr,
1073 	&dev_attr_tcpAttemptFails.attr,
1074 	&dev_attr_tcpEstabResets.attr,
1075 	&dev_attr_tcpCurrEstab.attr,
1076 	&dev_attr_tcpInSegs.attr,
1077 	&dev_attr_tcpOutSegs.attr,
1078 	&dev_attr_tcpRetransSegs.attr,
1079 	&dev_attr_tcpInErrs.attr,
1080 	&dev_attr_tcpOutRsts.attr,
1081 	NULL
1082 };
1083 
1084 static struct attribute_group iw_stats_group = {
1085 	.name	= "proto_stats",
1086 	.attrs	= iw_proto_stats_attrs,
1087 };
1088 
1089 static void free_port_list_attributes(struct ib_device *device)
1090 {
1091 	struct kobject *p, *t;
1092 
1093 	list_for_each_entry_safe(p, t, &device->port_list, entry) {
1094 		struct ib_port *port = container_of(p, struct ib_port, kobj);
1095 		list_del(&p->entry);
1096 		sysfs_remove_group(p, port->pma_table);
1097 		sysfs_remove_group(p, &port->pkey_group);
1098 		sysfs_remove_group(p, &port->gid_group);
1099 		sysfs_remove_group(&port->gid_attr_group->kobj,
1100 				   &port->gid_attr_group->ndev);
1101 		sysfs_remove_group(&port->gid_attr_group->kobj,
1102 				   &port->gid_attr_group->type);
1103 		kobject_put(&port->gid_attr_group->kobj);
1104 		kobject_put(p);
1105 	}
1106 
1107 	kobject_put(device->ports_parent);
1108 }
1109 
1110 int ib_device_register_sysfs(struct ib_device *device,
1111 			     int (*port_callback)(struct ib_device *,
1112 						  u8, struct kobject *))
1113 {
1114 	struct device *class_dev = &device->dev;
1115 	int ret;
1116 	int i;
1117 
1118 	device->dev.parent = device->dma_device;
1119 	ret = dev_set_name(class_dev, "%s", device->name);
1120 	if (ret)
1121 		return ret;
1122 
1123 	ret = device_add(class_dev);
1124 	if (ret)
1125 		goto err;
1126 
1127 	for (i = 0; i < ARRAY_SIZE(ib_class_attributes); ++i) {
1128 		ret = device_create_file(class_dev, ib_class_attributes[i]);
1129 		if (ret)
1130 			goto err_unregister;
1131 	}
1132 
1133 	device->ports_parent = kobject_create_and_add("ports",
1134 						      &class_dev->kobj);
1135 	if (!device->ports_parent) {
1136 		ret = -ENOMEM;
1137 		goto err_put;
1138 	}
1139 
1140 	if (rdma_cap_ib_switch(device)) {
1141 		ret = add_port(device, 0, port_callback);
1142 		if (ret)
1143 			goto err_put;
1144 	} else {
1145 		for (i = 1; i <= device->phys_port_cnt; ++i) {
1146 			ret = add_port(device, i, port_callback);
1147 			if (ret)
1148 				goto err_put;
1149 		}
1150 	}
1151 
1152 	if (device->node_type == RDMA_NODE_RNIC && device->get_protocol_stats) {
1153 		ret = sysfs_create_group(&class_dev->kobj, &iw_stats_group);
1154 		if (ret)
1155 			goto err_put;
1156 	}
1157 
1158 	return 0;
1159 
1160 err_put:
1161 	free_port_list_attributes(device);
1162 
1163 err_unregister:
1164 	device_unregister(class_dev);
1165 
1166 err:
1167 	return ret;
1168 }
1169 
1170 void ib_device_unregister_sysfs(struct ib_device *device)
1171 {
1172 	/* Hold kobject until ib_dealloc_device() */
1173 	struct kobject *kobj_dev = kobject_get(&device->dev.kobj);
1174 	int i;
1175 
1176 	if (device->node_type == RDMA_NODE_RNIC && device->get_protocol_stats)
1177 		sysfs_remove_group(kobj_dev, &iw_stats_group);
1178 
1179 	free_port_list_attributes(device);
1180 
1181 	for (i = 0; i < ARRAY_SIZE(ib_class_attributes); ++i)
1182 		device_remove_file(&device->dev, ib_class_attributes[i]);
1183 
1184 	device_unregister(&device->dev);
1185 }
1186