xref: /openbmc/linux/drivers/infiniband/core/sysfs.c (revision 176f011b)
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 #include <linux/ethtool.h>
42 
43 #include <rdma/ib_mad.h>
44 #include <rdma/ib_pma.h>
45 #include <rdma/ib_cache.h>
46 
47 struct ib_port;
48 
49 struct gid_attr_group {
50 	struct ib_port		*port;
51 	struct kobject		kobj;
52 	struct attribute_group	ndev;
53 	struct attribute_group	type;
54 };
55 struct ib_port {
56 	struct kobject         kobj;
57 	struct ib_device      *ibdev;
58 	struct gid_attr_group *gid_attr_group;
59 	struct attribute_group gid_group;
60 	struct attribute_group pkey_group;
61 	struct attribute_group *pma_table;
62 	struct attribute_group *hw_stats_ag;
63 	struct rdma_hw_stats   *hw_stats;
64 	u8                     port_num;
65 };
66 
67 struct port_attribute {
68 	struct attribute attr;
69 	ssize_t (*show)(struct ib_port *, struct port_attribute *, char *buf);
70 	ssize_t (*store)(struct ib_port *, struct port_attribute *,
71 			 const char *buf, size_t count);
72 };
73 
74 #define PORT_ATTR(_name, _mode, _show, _store) \
75 struct port_attribute port_attr_##_name = __ATTR(_name, _mode, _show, _store)
76 
77 #define PORT_ATTR_RO(_name) \
78 struct port_attribute port_attr_##_name = __ATTR_RO(_name)
79 
80 struct port_table_attribute {
81 	struct port_attribute	attr;
82 	char			name[8];
83 	int			index;
84 	__be16			attr_id;
85 };
86 
87 struct hw_stats_attribute {
88 	struct attribute	attr;
89 	ssize_t			(*show)(struct kobject *kobj,
90 					struct attribute *attr, char *buf);
91 	ssize_t			(*store)(struct kobject *kobj,
92 					 struct attribute *attr,
93 					 const char *buf,
94 					 size_t count);
95 	int			index;
96 	u8			port_num;
97 };
98 
99 static ssize_t port_attr_show(struct kobject *kobj,
100 			      struct attribute *attr, char *buf)
101 {
102 	struct port_attribute *port_attr =
103 		container_of(attr, struct port_attribute, attr);
104 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
105 
106 	if (!port_attr->show)
107 		return -EIO;
108 
109 	return port_attr->show(p, port_attr, buf);
110 }
111 
112 static ssize_t port_attr_store(struct kobject *kobj,
113 			       struct attribute *attr,
114 			       const char *buf, size_t count)
115 {
116 	struct port_attribute *port_attr =
117 		container_of(attr, struct port_attribute, attr);
118 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
119 
120 	if (!port_attr->store)
121 		return -EIO;
122 	return port_attr->store(p, port_attr, buf, count);
123 }
124 
125 static const struct sysfs_ops port_sysfs_ops = {
126 	.show	= port_attr_show,
127 	.store	= port_attr_store
128 };
129 
130 static ssize_t gid_attr_show(struct kobject *kobj,
131 			     struct attribute *attr, char *buf)
132 {
133 	struct port_attribute *port_attr =
134 		container_of(attr, struct port_attribute, attr);
135 	struct ib_port *p = container_of(kobj, struct gid_attr_group,
136 					 kobj)->port;
137 
138 	if (!port_attr->show)
139 		return -EIO;
140 
141 	return port_attr->show(p, port_attr, buf);
142 }
143 
144 static const struct sysfs_ops gid_attr_sysfs_ops = {
145 	.show = gid_attr_show
146 };
147 
148 static ssize_t state_show(struct ib_port *p, struct port_attribute *unused,
149 			  char *buf)
150 {
151 	struct ib_port_attr attr;
152 	ssize_t ret;
153 
154 	static const char *state_name[] = {
155 		[IB_PORT_NOP]		= "NOP",
156 		[IB_PORT_DOWN]		= "DOWN",
157 		[IB_PORT_INIT]		= "INIT",
158 		[IB_PORT_ARMED]		= "ARMED",
159 		[IB_PORT_ACTIVE]	= "ACTIVE",
160 		[IB_PORT_ACTIVE_DEFER]	= "ACTIVE_DEFER"
161 	};
162 
163 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
164 	if (ret)
165 		return ret;
166 
167 	return sprintf(buf, "%d: %s\n", attr.state,
168 		       attr.state >= 0 && attr.state < ARRAY_SIZE(state_name) ?
169 		       state_name[attr.state] : "UNKNOWN");
170 }
171 
172 static ssize_t lid_show(struct ib_port *p, struct port_attribute *unused,
173 			char *buf)
174 {
175 	struct ib_port_attr attr;
176 	ssize_t ret;
177 
178 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
179 	if (ret)
180 		return ret;
181 
182 	return sprintf(buf, "0x%x\n", attr.lid);
183 }
184 
185 static ssize_t lid_mask_count_show(struct ib_port *p,
186 				   struct port_attribute *unused,
187 				   char *buf)
188 {
189 	struct ib_port_attr attr;
190 	ssize_t ret;
191 
192 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
193 	if (ret)
194 		return ret;
195 
196 	return sprintf(buf, "%d\n", attr.lmc);
197 }
198 
199 static ssize_t sm_lid_show(struct ib_port *p, struct port_attribute *unused,
200 			   char *buf)
201 {
202 	struct ib_port_attr attr;
203 	ssize_t ret;
204 
205 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
206 	if (ret)
207 		return ret;
208 
209 	return sprintf(buf, "0x%x\n", attr.sm_lid);
210 }
211 
212 static ssize_t sm_sl_show(struct ib_port *p, struct port_attribute *unused,
213 			  char *buf)
214 {
215 	struct ib_port_attr attr;
216 	ssize_t ret;
217 
218 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
219 	if (ret)
220 		return ret;
221 
222 	return sprintf(buf, "%d\n", attr.sm_sl);
223 }
224 
225 static ssize_t cap_mask_show(struct ib_port *p, struct port_attribute *unused,
226 			     char *buf)
227 {
228 	struct ib_port_attr attr;
229 	ssize_t ret;
230 
231 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
232 	if (ret)
233 		return ret;
234 
235 	return sprintf(buf, "0x%08x\n", attr.port_cap_flags);
236 }
237 
238 static ssize_t rate_show(struct ib_port *p, struct port_attribute *unused,
239 			 char *buf)
240 {
241 	struct ib_port_attr attr;
242 	char *speed = "";
243 	int rate;		/* in deci-Gb/sec */
244 	ssize_t ret;
245 
246 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
247 	if (ret)
248 		return ret;
249 
250 	switch (attr.active_speed) {
251 	case IB_SPEED_DDR:
252 		speed = " DDR";
253 		rate = 50;
254 		break;
255 	case IB_SPEED_QDR:
256 		speed = " QDR";
257 		rate = 100;
258 		break;
259 	case IB_SPEED_FDR10:
260 		speed = " FDR10";
261 		rate = 100;
262 		break;
263 	case IB_SPEED_FDR:
264 		speed = " FDR";
265 		rate = 140;
266 		break;
267 	case IB_SPEED_EDR:
268 		speed = " EDR";
269 		rate = 250;
270 		break;
271 	case IB_SPEED_HDR:
272 		speed = " HDR";
273 		rate = 500;
274 		break;
275 	case IB_SPEED_SDR:
276 	default:		/* default to SDR for invalid rates */
277 		speed = " SDR";
278 		rate = 25;
279 		break;
280 	}
281 
282 	rate *= ib_width_enum_to_int(attr.active_width);
283 	if (rate < 0)
284 		return -EINVAL;
285 
286 	return sprintf(buf, "%d%s Gb/sec (%dX%s)\n",
287 		       rate / 10, rate % 10 ? ".5" : "",
288 		       ib_width_enum_to_int(attr.active_width), speed);
289 }
290 
291 static ssize_t phys_state_show(struct ib_port *p, struct port_attribute *unused,
292 			       char *buf)
293 {
294 	struct ib_port_attr attr;
295 
296 	ssize_t ret;
297 
298 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
299 	if (ret)
300 		return ret;
301 
302 	switch (attr.phys_state) {
303 	case 1:  return sprintf(buf, "1: Sleep\n");
304 	case 2:  return sprintf(buf, "2: Polling\n");
305 	case 3:  return sprintf(buf, "3: Disabled\n");
306 	case 4:  return sprintf(buf, "4: PortConfigurationTraining\n");
307 	case 5:  return sprintf(buf, "5: LinkUp\n");
308 	case 6:  return sprintf(buf, "6: LinkErrorRecovery\n");
309 	case 7:  return sprintf(buf, "7: Phy Test\n");
310 	default: return sprintf(buf, "%d: <unknown>\n", attr.phys_state);
311 	}
312 }
313 
314 static ssize_t link_layer_show(struct ib_port *p, struct port_attribute *unused,
315 			       char *buf)
316 {
317 	switch (rdma_port_get_link_layer(p->ibdev, p->port_num)) {
318 	case IB_LINK_LAYER_INFINIBAND:
319 		return sprintf(buf, "%s\n", "InfiniBand");
320 	case IB_LINK_LAYER_ETHERNET:
321 		return sprintf(buf, "%s\n", "Ethernet");
322 	default:
323 		return sprintf(buf, "%s\n", "Unknown");
324 	}
325 }
326 
327 static PORT_ATTR_RO(state);
328 static PORT_ATTR_RO(lid);
329 static PORT_ATTR_RO(lid_mask_count);
330 static PORT_ATTR_RO(sm_lid);
331 static PORT_ATTR_RO(sm_sl);
332 static PORT_ATTR_RO(cap_mask);
333 static PORT_ATTR_RO(rate);
334 static PORT_ATTR_RO(phys_state);
335 static PORT_ATTR_RO(link_layer);
336 
337 static struct attribute *port_default_attrs[] = {
338 	&port_attr_state.attr,
339 	&port_attr_lid.attr,
340 	&port_attr_lid_mask_count.attr,
341 	&port_attr_sm_lid.attr,
342 	&port_attr_sm_sl.attr,
343 	&port_attr_cap_mask.attr,
344 	&port_attr_rate.attr,
345 	&port_attr_phys_state.attr,
346 	&port_attr_link_layer.attr,
347 	NULL
348 };
349 
350 static size_t print_ndev(const struct ib_gid_attr *gid_attr, char *buf)
351 {
352 	if (!gid_attr->ndev)
353 		return -EINVAL;
354 
355 	return sprintf(buf, "%s\n", gid_attr->ndev->name);
356 }
357 
358 static size_t print_gid_type(const struct ib_gid_attr *gid_attr, char *buf)
359 {
360 	return sprintf(buf, "%s\n", ib_cache_gid_type_str(gid_attr->gid_type));
361 }
362 
363 static ssize_t _show_port_gid_attr(
364 	struct ib_port *p, struct port_attribute *attr, char *buf,
365 	size_t (*print)(const struct ib_gid_attr *gid_attr, char *buf))
366 {
367 	struct port_table_attribute *tab_attr =
368 		container_of(attr, struct port_table_attribute, attr);
369 	const struct ib_gid_attr *gid_attr;
370 	ssize_t ret;
371 
372 	gid_attr = rdma_get_gid_attr(p->ibdev, p->port_num, tab_attr->index);
373 	if (IS_ERR(gid_attr))
374 		return PTR_ERR(gid_attr);
375 
376 	ret = print(gid_attr, buf);
377 	rdma_put_gid_attr(gid_attr);
378 	return ret;
379 }
380 
381 static ssize_t show_port_gid(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 	const struct ib_gid_attr *gid_attr;
387 	ssize_t ret;
388 
389 	gid_attr = rdma_get_gid_attr(p->ibdev, p->port_num, tab_attr->index);
390 	if (IS_ERR(gid_attr)) {
391 		const union ib_gid zgid = {};
392 
393 		/* If reading GID fails, it is likely due to GID entry being
394 		 * empty (invalid) or reserved GID in the table.  User space
395 		 * expects to read GID table entries as long as it given index
396 		 * is within GID table size.  Administrative/debugging tool
397 		 * fails to query rest of the GID entries if it hits error
398 		 * while querying a GID of the given index.  To avoid user
399 		 * space throwing such error on fail to read gid, return zero
400 		 * GID as before. This maintains backward compatibility.
401 		 */
402 		return sprintf(buf, "%pI6\n", zgid.raw);
403 	}
404 
405 	ret = sprintf(buf, "%pI6\n", gid_attr->gid.raw);
406 	rdma_put_gid_attr(gid_attr);
407 	return ret;
408 }
409 
410 static ssize_t show_port_gid_attr_ndev(struct ib_port *p,
411 				       struct port_attribute *attr, char *buf)
412 {
413 	return _show_port_gid_attr(p, attr, buf, print_ndev);
414 }
415 
416 static ssize_t show_port_gid_attr_gid_type(struct ib_port *p,
417 					   struct port_attribute *attr,
418 					   char *buf)
419 {
420 	return _show_port_gid_attr(p, attr, buf, print_gid_type);
421 }
422 
423 static ssize_t show_port_pkey(struct ib_port *p, struct port_attribute *attr,
424 			      char *buf)
425 {
426 	struct port_table_attribute *tab_attr =
427 		container_of(attr, struct port_table_attribute, attr);
428 	u16 pkey;
429 	ssize_t ret;
430 
431 	ret = ib_query_pkey(p->ibdev, p->port_num, tab_attr->index, &pkey);
432 	if (ret)
433 		return ret;
434 
435 	return sprintf(buf, "0x%04x\n", pkey);
436 }
437 
438 #define PORT_PMA_ATTR(_name, _counter, _width, _offset)			\
439 struct port_table_attribute port_pma_attr_##_name = {			\
440 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
441 	.index = (_offset) | ((_width) << 16) | ((_counter) << 24),	\
442 	.attr_id = IB_PMA_PORT_COUNTERS ,				\
443 }
444 
445 #define PORT_PMA_ATTR_EXT(_name, _width, _offset)			\
446 struct port_table_attribute port_pma_attr_ext_##_name = {		\
447 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
448 	.index = (_offset) | ((_width) << 16),				\
449 	.attr_id = IB_PMA_PORT_COUNTERS_EXT ,				\
450 }
451 
452 /*
453  * Get a Perfmgmt MAD block of data.
454  * Returns error code or the number of bytes retrieved.
455  */
456 static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
457 		void *data, int offset, size_t size)
458 {
459 	struct ib_mad *in_mad;
460 	struct ib_mad *out_mad;
461 	size_t mad_size = sizeof(*out_mad);
462 	u16 out_mad_pkey_index = 0;
463 	ssize_t ret;
464 
465 	if (!dev->ops.process_mad)
466 		return -ENOSYS;
467 
468 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
469 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
470 	if (!in_mad || !out_mad) {
471 		ret = -ENOMEM;
472 		goto out;
473 	}
474 
475 	in_mad->mad_hdr.base_version  = 1;
476 	in_mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_PERF_MGMT;
477 	in_mad->mad_hdr.class_version = 1;
478 	in_mad->mad_hdr.method        = IB_MGMT_METHOD_GET;
479 	in_mad->mad_hdr.attr_id       = attr;
480 
481 	if (attr != IB_PMA_CLASS_PORT_INFO)
482 		in_mad->data[41] = port_num;	/* PortSelect field */
483 
484 	if ((dev->ops.process_mad(dev, IB_MAD_IGNORE_MKEY,
485 				  port_num, NULL, NULL,
486 				  (const struct ib_mad_hdr *)in_mad, mad_size,
487 				  (struct ib_mad_hdr *)out_mad, &mad_size,
488 				  &out_mad_pkey_index) &
489 	     (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
490 	    (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
491 		ret = -EINVAL;
492 		goto out;
493 	}
494 	memcpy(data, out_mad->data + offset, size);
495 	ret = size;
496 out:
497 	kfree(in_mad);
498 	kfree(out_mad);
499 	return ret;
500 }
501 
502 static ssize_t show_pma_counter(struct ib_port *p, struct port_attribute *attr,
503 				char *buf)
504 {
505 	struct port_table_attribute *tab_attr =
506 		container_of(attr, struct port_table_attribute, attr);
507 	int offset = tab_attr->index & 0xffff;
508 	int width  = (tab_attr->index >> 16) & 0xff;
509 	ssize_t ret;
510 	u8 data[8];
511 
512 	ret = get_perf_mad(p->ibdev, p->port_num, tab_attr->attr_id, &data,
513 			40 + offset / 8, sizeof(data));
514 	if (ret < 0)
515 		return ret;
516 
517 	switch (width) {
518 	case 4:
519 		ret = sprintf(buf, "%u\n", (*data >>
520 					    (4 - (offset % 8))) & 0xf);
521 		break;
522 	case 8:
523 		ret = sprintf(buf, "%u\n", *data);
524 		break;
525 	case 16:
526 		ret = sprintf(buf, "%u\n",
527 			      be16_to_cpup((__be16 *)data));
528 		break;
529 	case 32:
530 		ret = sprintf(buf, "%u\n",
531 			      be32_to_cpup((__be32 *)data));
532 		break;
533 	case 64:
534 		ret = sprintf(buf, "%llu\n",
535 				be64_to_cpup((__be64 *)data));
536 		break;
537 
538 	default:
539 		ret = 0;
540 	}
541 
542 	return ret;
543 }
544 
545 static PORT_PMA_ATTR(symbol_error		    ,  0, 16,  32);
546 static PORT_PMA_ATTR(link_error_recovery	    ,  1,  8,  48);
547 static PORT_PMA_ATTR(link_downed		    ,  2,  8,  56);
548 static PORT_PMA_ATTR(port_rcv_errors		    ,  3, 16,  64);
549 static PORT_PMA_ATTR(port_rcv_remote_physical_errors,  4, 16,  80);
550 static PORT_PMA_ATTR(port_rcv_switch_relay_errors   ,  5, 16,  96);
551 static PORT_PMA_ATTR(port_xmit_discards		    ,  6, 16, 112);
552 static PORT_PMA_ATTR(port_xmit_constraint_errors    ,  7,  8, 128);
553 static PORT_PMA_ATTR(port_rcv_constraint_errors	    ,  8,  8, 136);
554 static PORT_PMA_ATTR(local_link_integrity_errors    ,  9,  4, 152);
555 static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10,  4, 156);
556 static PORT_PMA_ATTR(VL15_dropped		    , 11, 16, 176);
557 static PORT_PMA_ATTR(port_xmit_data		    , 12, 32, 192);
558 static PORT_PMA_ATTR(port_rcv_data		    , 13, 32, 224);
559 static PORT_PMA_ATTR(port_xmit_packets		    , 14, 32, 256);
560 static PORT_PMA_ATTR(port_rcv_packets		    , 15, 32, 288);
561 static PORT_PMA_ATTR(port_xmit_wait		    ,  0, 32, 320);
562 
563 /*
564  * Counters added by extended set
565  */
566 static PORT_PMA_ATTR_EXT(port_xmit_data		    , 64,  64);
567 static PORT_PMA_ATTR_EXT(port_rcv_data		    , 64, 128);
568 static PORT_PMA_ATTR_EXT(port_xmit_packets	    , 64, 192);
569 static PORT_PMA_ATTR_EXT(port_rcv_packets	    , 64, 256);
570 static PORT_PMA_ATTR_EXT(unicast_xmit_packets	    , 64, 320);
571 static PORT_PMA_ATTR_EXT(unicast_rcv_packets	    , 64, 384);
572 static PORT_PMA_ATTR_EXT(multicast_xmit_packets	    , 64, 448);
573 static PORT_PMA_ATTR_EXT(multicast_rcv_packets	    , 64, 512);
574 
575 static struct attribute *pma_attrs[] = {
576 	&port_pma_attr_symbol_error.attr.attr,
577 	&port_pma_attr_link_error_recovery.attr.attr,
578 	&port_pma_attr_link_downed.attr.attr,
579 	&port_pma_attr_port_rcv_errors.attr.attr,
580 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
581 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
582 	&port_pma_attr_port_xmit_discards.attr.attr,
583 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
584 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
585 	&port_pma_attr_local_link_integrity_errors.attr.attr,
586 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
587 	&port_pma_attr_VL15_dropped.attr.attr,
588 	&port_pma_attr_port_xmit_data.attr.attr,
589 	&port_pma_attr_port_rcv_data.attr.attr,
590 	&port_pma_attr_port_xmit_packets.attr.attr,
591 	&port_pma_attr_port_rcv_packets.attr.attr,
592 	&port_pma_attr_port_xmit_wait.attr.attr,
593 	NULL
594 };
595 
596 static struct attribute *pma_attrs_ext[] = {
597 	&port_pma_attr_symbol_error.attr.attr,
598 	&port_pma_attr_link_error_recovery.attr.attr,
599 	&port_pma_attr_link_downed.attr.attr,
600 	&port_pma_attr_port_rcv_errors.attr.attr,
601 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
602 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
603 	&port_pma_attr_port_xmit_discards.attr.attr,
604 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
605 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
606 	&port_pma_attr_local_link_integrity_errors.attr.attr,
607 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
608 	&port_pma_attr_VL15_dropped.attr.attr,
609 	&port_pma_attr_ext_port_xmit_data.attr.attr,
610 	&port_pma_attr_ext_port_rcv_data.attr.attr,
611 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
612 	&port_pma_attr_port_xmit_wait.attr.attr,
613 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
614 	&port_pma_attr_ext_unicast_rcv_packets.attr.attr,
615 	&port_pma_attr_ext_unicast_xmit_packets.attr.attr,
616 	&port_pma_attr_ext_multicast_rcv_packets.attr.attr,
617 	&port_pma_attr_ext_multicast_xmit_packets.attr.attr,
618 	NULL
619 };
620 
621 static struct attribute *pma_attrs_noietf[] = {
622 	&port_pma_attr_symbol_error.attr.attr,
623 	&port_pma_attr_link_error_recovery.attr.attr,
624 	&port_pma_attr_link_downed.attr.attr,
625 	&port_pma_attr_port_rcv_errors.attr.attr,
626 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
627 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
628 	&port_pma_attr_port_xmit_discards.attr.attr,
629 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
630 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
631 	&port_pma_attr_local_link_integrity_errors.attr.attr,
632 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
633 	&port_pma_attr_VL15_dropped.attr.attr,
634 	&port_pma_attr_ext_port_xmit_data.attr.attr,
635 	&port_pma_attr_ext_port_rcv_data.attr.attr,
636 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
637 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
638 	&port_pma_attr_port_xmit_wait.attr.attr,
639 	NULL
640 };
641 
642 static struct attribute_group pma_group = {
643 	.name  = "counters",
644 	.attrs  = pma_attrs
645 };
646 
647 static struct attribute_group pma_group_ext = {
648 	.name  = "counters",
649 	.attrs  = pma_attrs_ext
650 };
651 
652 static struct attribute_group pma_group_noietf = {
653 	.name  = "counters",
654 	.attrs  = pma_attrs_noietf
655 };
656 
657 static void ib_port_release(struct kobject *kobj)
658 {
659 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
660 	struct attribute *a;
661 	int i;
662 
663 	if (p->gid_group.attrs) {
664 		for (i = 0; (a = p->gid_group.attrs[i]); ++i)
665 			kfree(a);
666 
667 		kfree(p->gid_group.attrs);
668 	}
669 
670 	if (p->pkey_group.attrs) {
671 		for (i = 0; (a = p->pkey_group.attrs[i]); ++i)
672 			kfree(a);
673 
674 		kfree(p->pkey_group.attrs);
675 	}
676 
677 	kfree(p);
678 }
679 
680 static void ib_port_gid_attr_release(struct kobject *kobj)
681 {
682 	struct gid_attr_group *g = container_of(kobj, struct gid_attr_group,
683 						kobj);
684 	struct attribute *a;
685 	int i;
686 
687 	if (g->ndev.attrs) {
688 		for (i = 0; (a = g->ndev.attrs[i]); ++i)
689 			kfree(a);
690 
691 		kfree(g->ndev.attrs);
692 	}
693 
694 	if (g->type.attrs) {
695 		for (i = 0; (a = g->type.attrs[i]); ++i)
696 			kfree(a);
697 
698 		kfree(g->type.attrs);
699 	}
700 
701 	kfree(g);
702 }
703 
704 static struct kobj_type port_type = {
705 	.release       = ib_port_release,
706 	.sysfs_ops     = &port_sysfs_ops,
707 	.default_attrs = port_default_attrs
708 };
709 
710 static struct kobj_type gid_attr_type = {
711 	.sysfs_ops      = &gid_attr_sysfs_ops,
712 	.release        = ib_port_gid_attr_release
713 };
714 
715 static struct attribute **
716 alloc_group_attrs(ssize_t (*show)(struct ib_port *,
717 				  struct port_attribute *, char *buf),
718 		  int len)
719 {
720 	struct attribute **tab_attr;
721 	struct port_table_attribute *element;
722 	int i;
723 
724 	tab_attr = kcalloc(1 + len, sizeof(struct attribute *), GFP_KERNEL);
725 	if (!tab_attr)
726 		return NULL;
727 
728 	for (i = 0; i < len; i++) {
729 		element = kzalloc(sizeof(struct port_table_attribute),
730 				  GFP_KERNEL);
731 		if (!element)
732 			goto err;
733 
734 		if (snprintf(element->name, sizeof(element->name),
735 			     "%d", i) >= sizeof(element->name)) {
736 			kfree(element);
737 			goto err;
738 		}
739 
740 		element->attr.attr.name  = element->name;
741 		element->attr.attr.mode  = S_IRUGO;
742 		element->attr.show       = show;
743 		element->index		 = i;
744 		sysfs_attr_init(&element->attr.attr);
745 
746 		tab_attr[i] = &element->attr.attr;
747 	}
748 
749 	return tab_attr;
750 
751 err:
752 	while (--i >= 0)
753 		kfree(tab_attr[i]);
754 	kfree(tab_attr);
755 	return NULL;
756 }
757 
758 /*
759  * Figure out which counter table to use depending on
760  * the device capabilities.
761  */
762 static struct attribute_group *get_counter_table(struct ib_device *dev,
763 						 int port_num)
764 {
765 	struct ib_class_port_info cpi;
766 
767 	if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
768 				&cpi, 40, sizeof(cpi)) >= 0) {
769 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
770 			/* We have extended counters */
771 			return &pma_group_ext;
772 
773 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
774 			/* But not the IETF ones */
775 			return &pma_group_noietf;
776 	}
777 
778 	/* Fall back to normal counters */
779 	return &pma_group;
780 }
781 
782 static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
783 			   u8 port_num, int index)
784 {
785 	int ret;
786 
787 	if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
788 		return 0;
789 	ret = dev->ops.get_hw_stats(dev, stats, port_num, index);
790 	if (ret < 0)
791 		return ret;
792 	if (ret == stats->num_counters)
793 		stats->timestamp = jiffies;
794 
795 	return 0;
796 }
797 
798 static ssize_t print_hw_stat(struct rdma_hw_stats *stats, int index, char *buf)
799 {
800 	return sprintf(buf, "%llu\n", stats->value[index]);
801 }
802 
803 static ssize_t show_hw_stats(struct kobject *kobj, struct attribute *attr,
804 			     char *buf)
805 {
806 	struct ib_device *dev;
807 	struct ib_port *port;
808 	struct hw_stats_attribute *hsa;
809 	struct rdma_hw_stats *stats;
810 	int ret;
811 
812 	hsa = container_of(attr, struct hw_stats_attribute, attr);
813 	if (!hsa->port_num) {
814 		dev = container_of((struct device *)kobj,
815 				   struct ib_device, dev);
816 		stats = dev->hw_stats;
817 	} else {
818 		port = container_of(kobj, struct ib_port, kobj);
819 		dev = port->ibdev;
820 		stats = port->hw_stats;
821 	}
822 	mutex_lock(&stats->lock);
823 	ret = update_hw_stats(dev, stats, hsa->port_num, hsa->index);
824 	if (ret)
825 		goto unlock;
826 	ret = print_hw_stat(stats, hsa->index, buf);
827 unlock:
828 	mutex_unlock(&stats->lock);
829 
830 	return ret;
831 }
832 
833 static ssize_t show_stats_lifespan(struct kobject *kobj,
834 				   struct attribute *attr,
835 				   char *buf)
836 {
837 	struct hw_stats_attribute *hsa;
838 	struct rdma_hw_stats *stats;
839 	int msecs;
840 
841 	hsa = container_of(attr, struct hw_stats_attribute, attr);
842 	if (!hsa->port_num) {
843 		struct ib_device *dev = container_of((struct device *)kobj,
844 						     struct ib_device, dev);
845 
846 		stats = dev->hw_stats;
847 	} else {
848 		struct ib_port *p = container_of(kobj, struct ib_port, kobj);
849 
850 		stats = p->hw_stats;
851 	}
852 
853 	mutex_lock(&stats->lock);
854 	msecs = jiffies_to_msecs(stats->lifespan);
855 	mutex_unlock(&stats->lock);
856 
857 	return sprintf(buf, "%d\n", msecs);
858 }
859 
860 static ssize_t set_stats_lifespan(struct kobject *kobj,
861 				  struct attribute *attr,
862 				  const char *buf, size_t count)
863 {
864 	struct hw_stats_attribute *hsa;
865 	struct rdma_hw_stats *stats;
866 	int msecs;
867 	int jiffies;
868 	int ret;
869 
870 	ret = kstrtoint(buf, 10, &msecs);
871 	if (ret)
872 		return ret;
873 	if (msecs < 0 || msecs > 10000)
874 		return -EINVAL;
875 	jiffies = msecs_to_jiffies(msecs);
876 	hsa = container_of(attr, struct hw_stats_attribute, attr);
877 	if (!hsa->port_num) {
878 		struct ib_device *dev = container_of((struct device *)kobj,
879 						     struct ib_device, dev);
880 
881 		stats = dev->hw_stats;
882 	} else {
883 		struct ib_port *p = container_of(kobj, struct ib_port, kobj);
884 
885 		stats = p->hw_stats;
886 	}
887 
888 	mutex_lock(&stats->lock);
889 	stats->lifespan = jiffies;
890 	mutex_unlock(&stats->lock);
891 
892 	return count;
893 }
894 
895 static void free_hsag(struct kobject *kobj, struct attribute_group *attr_group)
896 {
897 	struct attribute **attr;
898 
899 	sysfs_remove_group(kobj, attr_group);
900 
901 	for (attr = attr_group->attrs; *attr; attr++)
902 		kfree(*attr);
903 	kfree(attr_group);
904 }
905 
906 static struct attribute *alloc_hsa(int index, u8 port_num, const char *name)
907 {
908 	struct hw_stats_attribute *hsa;
909 
910 	hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
911 	if (!hsa)
912 		return NULL;
913 
914 	hsa->attr.name = (char *)name;
915 	hsa->attr.mode = S_IRUGO;
916 	hsa->show = show_hw_stats;
917 	hsa->store = NULL;
918 	hsa->index = index;
919 	hsa->port_num = port_num;
920 
921 	return &hsa->attr;
922 }
923 
924 static struct attribute *alloc_hsa_lifespan(char *name, u8 port_num)
925 {
926 	struct hw_stats_attribute *hsa;
927 
928 	hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
929 	if (!hsa)
930 		return NULL;
931 
932 	hsa->attr.name = name;
933 	hsa->attr.mode = S_IWUSR | S_IRUGO;
934 	hsa->show = show_stats_lifespan;
935 	hsa->store = set_stats_lifespan;
936 	hsa->index = 0;
937 	hsa->port_num = port_num;
938 
939 	return &hsa->attr;
940 }
941 
942 static void setup_hw_stats(struct ib_device *device, struct ib_port *port,
943 			   u8 port_num)
944 {
945 	struct attribute_group *hsag;
946 	struct rdma_hw_stats *stats;
947 	int i, ret;
948 
949 	stats = device->ops.alloc_hw_stats(device, port_num);
950 
951 	if (!stats)
952 		return;
953 
954 	if (!stats->names || stats->num_counters <= 0)
955 		goto err_free_stats;
956 
957 	/*
958 	 * Two extra attribue elements here, one for the lifespan entry and
959 	 * one to NULL terminate the list for the sysfs core code
960 	 */
961 	hsag = kzalloc(sizeof(*hsag) +
962 		       sizeof(void *) * (stats->num_counters + 2),
963 		       GFP_KERNEL);
964 	if (!hsag)
965 		goto err_free_stats;
966 
967 	ret = device->ops.get_hw_stats(device, stats, port_num,
968 				       stats->num_counters);
969 	if (ret != stats->num_counters)
970 		goto err_free_hsag;
971 
972 	stats->timestamp = jiffies;
973 
974 	hsag->name = "hw_counters";
975 	hsag->attrs = (void *)hsag + sizeof(*hsag);
976 
977 	for (i = 0; i < stats->num_counters; i++) {
978 		hsag->attrs[i] = alloc_hsa(i, port_num, stats->names[i]);
979 		if (!hsag->attrs[i])
980 			goto err;
981 		sysfs_attr_init(hsag->attrs[i]);
982 	}
983 
984 	mutex_init(&stats->lock);
985 	/* treat an error here as non-fatal */
986 	hsag->attrs[i] = alloc_hsa_lifespan("lifespan", port_num);
987 	if (hsag->attrs[i])
988 		sysfs_attr_init(hsag->attrs[i]);
989 
990 	if (port) {
991 		struct kobject *kobj = &port->kobj;
992 		ret = sysfs_create_group(kobj, hsag);
993 		if (ret)
994 			goto err;
995 		port->hw_stats_ag = hsag;
996 		port->hw_stats = stats;
997 	} else {
998 		struct kobject *kobj = &device->dev.kobj;
999 		ret = sysfs_create_group(kobj, hsag);
1000 		if (ret)
1001 			goto err;
1002 		device->hw_stats_ag = hsag;
1003 		device->hw_stats = stats;
1004 	}
1005 
1006 	return;
1007 
1008 err:
1009 	for (; i >= 0; i--)
1010 		kfree(hsag->attrs[i]);
1011 err_free_hsag:
1012 	kfree(hsag);
1013 err_free_stats:
1014 	kfree(stats);
1015 	return;
1016 }
1017 
1018 static int add_port(struct ib_device *device, int port_num,
1019 		    int (*port_callback)(struct ib_device *,
1020 					 u8, struct kobject *))
1021 {
1022 	struct ib_port *p;
1023 	struct ib_port_attr attr;
1024 	int i;
1025 	int ret;
1026 
1027 	ret = ib_query_port(device, port_num, &attr);
1028 	if (ret)
1029 		return ret;
1030 
1031 	p = kzalloc(sizeof *p, GFP_KERNEL);
1032 	if (!p)
1033 		return -ENOMEM;
1034 
1035 	p->ibdev      = device;
1036 	p->port_num   = port_num;
1037 
1038 	ret = kobject_init_and_add(&p->kobj, &port_type,
1039 				   device->ports_kobj,
1040 				   "%d", port_num);
1041 	if (ret) {
1042 		kfree(p);
1043 		return ret;
1044 	}
1045 
1046 	p->gid_attr_group = kzalloc(sizeof(*p->gid_attr_group), GFP_KERNEL);
1047 	if (!p->gid_attr_group) {
1048 		ret = -ENOMEM;
1049 		goto err_put;
1050 	}
1051 
1052 	p->gid_attr_group->port = p;
1053 	ret = kobject_init_and_add(&p->gid_attr_group->kobj, &gid_attr_type,
1054 				   &p->kobj, "gid_attrs");
1055 	if (ret) {
1056 		kfree(p->gid_attr_group);
1057 		goto err_put;
1058 	}
1059 
1060 	if (device->ops.process_mad) {
1061 		p->pma_table = get_counter_table(device, port_num);
1062 		ret = sysfs_create_group(&p->kobj, p->pma_table);
1063 		if (ret)
1064 			goto err_put_gid_attrs;
1065 	}
1066 
1067 	p->gid_group.name  = "gids";
1068 	p->gid_group.attrs = alloc_group_attrs(show_port_gid, attr.gid_tbl_len);
1069 	if (!p->gid_group.attrs) {
1070 		ret = -ENOMEM;
1071 		goto err_remove_pma;
1072 	}
1073 
1074 	ret = sysfs_create_group(&p->kobj, &p->gid_group);
1075 	if (ret)
1076 		goto err_free_gid;
1077 
1078 	p->gid_attr_group->ndev.name = "ndevs";
1079 	p->gid_attr_group->ndev.attrs = alloc_group_attrs(show_port_gid_attr_ndev,
1080 							  attr.gid_tbl_len);
1081 	if (!p->gid_attr_group->ndev.attrs) {
1082 		ret = -ENOMEM;
1083 		goto err_remove_gid;
1084 	}
1085 
1086 	ret = sysfs_create_group(&p->gid_attr_group->kobj,
1087 				 &p->gid_attr_group->ndev);
1088 	if (ret)
1089 		goto err_free_gid_ndev;
1090 
1091 	p->gid_attr_group->type.name = "types";
1092 	p->gid_attr_group->type.attrs = alloc_group_attrs(show_port_gid_attr_gid_type,
1093 							  attr.gid_tbl_len);
1094 	if (!p->gid_attr_group->type.attrs) {
1095 		ret = -ENOMEM;
1096 		goto err_remove_gid_ndev;
1097 	}
1098 
1099 	ret = sysfs_create_group(&p->gid_attr_group->kobj,
1100 				 &p->gid_attr_group->type);
1101 	if (ret)
1102 		goto err_free_gid_type;
1103 
1104 	p->pkey_group.name  = "pkeys";
1105 	p->pkey_group.attrs = alloc_group_attrs(show_port_pkey,
1106 						attr.pkey_tbl_len);
1107 	if (!p->pkey_group.attrs) {
1108 		ret = -ENOMEM;
1109 		goto err_remove_gid_type;
1110 	}
1111 
1112 	ret = sysfs_create_group(&p->kobj, &p->pkey_group);
1113 	if (ret)
1114 		goto err_free_pkey;
1115 
1116 	if (port_callback) {
1117 		ret = port_callback(device, port_num, &p->kobj);
1118 		if (ret)
1119 			goto err_remove_pkey;
1120 	}
1121 
1122 	/*
1123 	 * If port == 0, it means hw_counters are per device and not per
1124 	 * port, so holder should be device. Therefore skip per port conunter
1125 	 * initialization.
1126 	 */
1127 	if (device->ops.alloc_hw_stats && port_num)
1128 		setup_hw_stats(device, p, port_num);
1129 
1130 	list_add_tail(&p->kobj.entry, &device->port_list);
1131 
1132 	kobject_uevent(&p->kobj, KOBJ_ADD);
1133 	return 0;
1134 
1135 err_remove_pkey:
1136 	sysfs_remove_group(&p->kobj, &p->pkey_group);
1137 
1138 err_free_pkey:
1139 	for (i = 0; i < attr.pkey_tbl_len; ++i)
1140 		kfree(p->pkey_group.attrs[i]);
1141 
1142 	kfree(p->pkey_group.attrs);
1143 	p->pkey_group.attrs = NULL;
1144 
1145 err_remove_gid_type:
1146 	sysfs_remove_group(&p->gid_attr_group->kobj,
1147 			   &p->gid_attr_group->type);
1148 
1149 err_free_gid_type:
1150 	for (i = 0; i < attr.gid_tbl_len; ++i)
1151 		kfree(p->gid_attr_group->type.attrs[i]);
1152 
1153 	kfree(p->gid_attr_group->type.attrs);
1154 	p->gid_attr_group->type.attrs = NULL;
1155 
1156 err_remove_gid_ndev:
1157 	sysfs_remove_group(&p->gid_attr_group->kobj,
1158 			   &p->gid_attr_group->ndev);
1159 
1160 err_free_gid_ndev:
1161 	for (i = 0; i < attr.gid_tbl_len; ++i)
1162 		kfree(p->gid_attr_group->ndev.attrs[i]);
1163 
1164 	kfree(p->gid_attr_group->ndev.attrs);
1165 	p->gid_attr_group->ndev.attrs = NULL;
1166 
1167 err_remove_gid:
1168 	sysfs_remove_group(&p->kobj, &p->gid_group);
1169 
1170 err_free_gid:
1171 	for (i = 0; i < attr.gid_tbl_len; ++i)
1172 		kfree(p->gid_group.attrs[i]);
1173 
1174 	kfree(p->gid_group.attrs);
1175 	p->gid_group.attrs = NULL;
1176 
1177 err_remove_pma:
1178 	if (p->pma_table)
1179 		sysfs_remove_group(&p->kobj, p->pma_table);
1180 
1181 err_put_gid_attrs:
1182 	kobject_put(&p->gid_attr_group->kobj);
1183 
1184 err_put:
1185 	kobject_put(&p->kobj);
1186 	return ret;
1187 }
1188 
1189 static ssize_t node_type_show(struct device *device,
1190 			      struct device_attribute *attr, char *buf)
1191 {
1192 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1193 
1194 	switch (dev->node_type) {
1195 	case RDMA_NODE_IB_CA:	  return sprintf(buf, "%d: CA\n", dev->node_type);
1196 	case RDMA_NODE_RNIC:	  return sprintf(buf, "%d: RNIC\n", dev->node_type);
1197 	case RDMA_NODE_USNIC:	  return sprintf(buf, "%d: usNIC\n", dev->node_type);
1198 	case RDMA_NODE_USNIC_UDP: return sprintf(buf, "%d: usNIC UDP\n", dev->node_type);
1199 	case RDMA_NODE_IB_SWITCH: return sprintf(buf, "%d: switch\n", dev->node_type);
1200 	case RDMA_NODE_IB_ROUTER: return sprintf(buf, "%d: router\n", dev->node_type);
1201 	default:		  return sprintf(buf, "%d: <unknown>\n", dev->node_type);
1202 	}
1203 }
1204 static DEVICE_ATTR_RO(node_type);
1205 
1206 static ssize_t sys_image_guid_show(struct device *device,
1207 				   struct device_attribute *dev_attr, char *buf)
1208 {
1209 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1210 
1211 	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1212 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[0]),
1213 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[1]),
1214 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[2]),
1215 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[3]));
1216 }
1217 static DEVICE_ATTR_RO(sys_image_guid);
1218 
1219 static ssize_t node_guid_show(struct device *device,
1220 			      struct device_attribute *attr, char *buf)
1221 {
1222 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1223 
1224 	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1225 		       be16_to_cpu(((__be16 *) &dev->node_guid)[0]),
1226 		       be16_to_cpu(((__be16 *) &dev->node_guid)[1]),
1227 		       be16_to_cpu(((__be16 *) &dev->node_guid)[2]),
1228 		       be16_to_cpu(((__be16 *) &dev->node_guid)[3]));
1229 }
1230 static DEVICE_ATTR_RO(node_guid);
1231 
1232 static ssize_t node_desc_show(struct device *device,
1233 			      struct device_attribute *attr, char *buf)
1234 {
1235 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1236 
1237 	return sprintf(buf, "%.64s\n", dev->node_desc);
1238 }
1239 
1240 static ssize_t node_desc_store(struct device *device,
1241 			       struct device_attribute *attr,
1242 			       const char *buf, size_t count)
1243 {
1244 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1245 	struct ib_device_modify desc = {};
1246 	int ret;
1247 
1248 	if (!dev->ops.modify_device)
1249 		return -EIO;
1250 
1251 	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1252 	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1253 	if (ret)
1254 		return ret;
1255 
1256 	return count;
1257 }
1258 static DEVICE_ATTR_RW(node_desc);
1259 
1260 static ssize_t fw_ver_show(struct device *device, struct device_attribute *attr,
1261 			   char *buf)
1262 {
1263 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1264 
1265 	ib_get_device_fw_str(dev, buf);
1266 	strlcat(buf, "\n", IB_FW_VERSION_NAME_MAX);
1267 	return strlen(buf);
1268 }
1269 static DEVICE_ATTR_RO(fw_ver);
1270 
1271 static struct attribute *ib_dev_attrs[] = {
1272 	&dev_attr_node_type.attr,
1273 	&dev_attr_node_guid.attr,
1274 	&dev_attr_sys_image_guid.attr,
1275 	&dev_attr_fw_ver.attr,
1276 	&dev_attr_node_desc.attr,
1277 	NULL,
1278 };
1279 
1280 static const struct attribute_group dev_attr_group = {
1281 	.attrs = ib_dev_attrs,
1282 };
1283 
1284 static void free_port_list_attributes(struct ib_device *device)
1285 {
1286 	struct kobject *p, *t;
1287 
1288 	list_for_each_entry_safe(p, t, &device->port_list, entry) {
1289 		struct ib_port *port = container_of(p, struct ib_port, kobj);
1290 		list_del(&p->entry);
1291 		if (port->hw_stats) {
1292 			kfree(port->hw_stats);
1293 			free_hsag(&port->kobj, port->hw_stats_ag);
1294 		}
1295 
1296 		if (port->pma_table)
1297 			sysfs_remove_group(p, port->pma_table);
1298 		sysfs_remove_group(p, &port->pkey_group);
1299 		sysfs_remove_group(p, &port->gid_group);
1300 		sysfs_remove_group(&port->gid_attr_group->kobj,
1301 				   &port->gid_attr_group->ndev);
1302 		sysfs_remove_group(&port->gid_attr_group->kobj,
1303 				   &port->gid_attr_group->type);
1304 		kobject_put(&port->gid_attr_group->kobj);
1305 		kobject_put(p);
1306 	}
1307 
1308 	kobject_put(device->ports_kobj);
1309 }
1310 
1311 int ib_device_register_sysfs(struct ib_device *device,
1312 			     int (*port_callback)(struct ib_device *,
1313 						  u8, struct kobject *))
1314 {
1315 	struct device *class_dev = &device->dev;
1316 	int ret;
1317 	int i;
1318 
1319 	device->groups[0] = &dev_attr_group;
1320 	class_dev->groups = device->groups;
1321 
1322 	ret = device_add(class_dev);
1323 	if (ret)
1324 		goto err;
1325 
1326 	device->ports_kobj = kobject_create_and_add("ports", &class_dev->kobj);
1327 	if (!device->ports_kobj) {
1328 		ret = -ENOMEM;
1329 		goto err_put;
1330 	}
1331 
1332 	if (rdma_cap_ib_switch(device)) {
1333 		ret = add_port(device, 0, port_callback);
1334 		if (ret)
1335 			goto err_put;
1336 	} else {
1337 		for (i = 1; i <= device->phys_port_cnt; ++i) {
1338 			ret = add_port(device, i, port_callback);
1339 			if (ret)
1340 				goto err_put;
1341 		}
1342 	}
1343 
1344 	if (device->ops.alloc_hw_stats)
1345 		setup_hw_stats(device, NULL, 0);
1346 
1347 	return 0;
1348 
1349 err_put:
1350 	free_port_list_attributes(device);
1351 	device_del(class_dev);
1352 err:
1353 	return ret;
1354 }
1355 
1356 void ib_device_unregister_sysfs(struct ib_device *device)
1357 {
1358 	/* Hold device until ib_dealloc_device() */
1359 	get_device(&device->dev);
1360 
1361 	free_port_list_attributes(device);
1362 
1363 	if (device->hw_stats) {
1364 		kfree(device->hw_stats);
1365 		free_hsag(&device->dev.kobj, device->hw_stats_ag);
1366 	}
1367 
1368 	device_unregister(&device->dev);
1369 }
1370