xref: /openbmc/linux/drivers/infiniband/core/sysfs.c (revision 45cc842d5b75ba8f9a958f2dd12b95c6dd0452bd)
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 
46 struct ib_port;
47 
48 struct gid_attr_group {
49 	struct ib_port		*port;
50 	struct kobject		kobj;
51 	struct attribute_group	ndev;
52 	struct attribute_group	type;
53 };
54 struct ib_port {
55 	struct kobject         kobj;
56 	struct ib_device      *ibdev;
57 	struct gid_attr_group *gid_attr_group;
58 	struct attribute_group gid_group;
59 	struct attribute_group pkey_group;
60 	struct attribute_group *pma_table;
61 	struct attribute_group *hw_stats_ag;
62 	struct rdma_hw_stats   *hw_stats;
63 	u8                     port_num;
64 };
65 
66 struct port_attribute {
67 	struct attribute attr;
68 	ssize_t (*show)(struct ib_port *, struct port_attribute *, char *buf);
69 	ssize_t (*store)(struct ib_port *, struct port_attribute *,
70 			 const char *buf, size_t count);
71 };
72 
73 #define PORT_ATTR(_name, _mode, _show, _store) \
74 struct port_attribute port_attr_##_name = __ATTR(_name, _mode, _show, _store)
75 
76 #define PORT_ATTR_RO(_name) \
77 struct port_attribute port_attr_##_name = __ATTR_RO(_name)
78 
79 struct port_table_attribute {
80 	struct port_attribute	attr;
81 	char			name[8];
82 	int			index;
83 	__be16			attr_id;
84 };
85 
86 struct hw_stats_attribute {
87 	struct attribute	attr;
88 	ssize_t			(*show)(struct kobject *kobj,
89 					struct attribute *attr, char *buf);
90 	ssize_t			(*store)(struct kobject *kobj,
91 					 struct attribute *attr,
92 					 const char *buf,
93 					 size_t count);
94 	int			index;
95 	u8			port_num;
96 };
97 
98 static ssize_t port_attr_show(struct kobject *kobj,
99 			      struct attribute *attr, char *buf)
100 {
101 	struct port_attribute *port_attr =
102 		container_of(attr, struct port_attribute, attr);
103 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
104 
105 	if (!port_attr->show)
106 		return -EIO;
107 
108 	return port_attr->show(p, port_attr, buf);
109 }
110 
111 static ssize_t port_attr_store(struct kobject *kobj,
112 			       struct attribute *attr,
113 			       const char *buf, size_t count)
114 {
115 	struct port_attribute *port_attr =
116 		container_of(attr, struct port_attribute, attr);
117 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
118 
119 	if (!port_attr->store)
120 		return -EIO;
121 	return port_attr->store(p, port_attr, buf, count);
122 }
123 
124 static const struct sysfs_ops port_sysfs_ops = {
125 	.show	= port_attr_show,
126 	.store	= port_attr_store
127 };
128 
129 static ssize_t gid_attr_show(struct kobject *kobj,
130 			     struct attribute *attr, char *buf)
131 {
132 	struct port_attribute *port_attr =
133 		container_of(attr, struct port_attribute, attr);
134 	struct ib_port *p = container_of(kobj, struct gid_attr_group,
135 					 kobj)->port;
136 
137 	if (!port_attr->show)
138 		return -EIO;
139 
140 	return port_attr->show(p, port_attr, buf);
141 }
142 
143 static const struct sysfs_ops gid_attr_sysfs_ops = {
144 	.show = gid_attr_show
145 };
146 
147 static ssize_t state_show(struct ib_port *p, struct port_attribute *unused,
148 			  char *buf)
149 {
150 	struct ib_port_attr attr;
151 	ssize_t ret;
152 
153 	static const char *state_name[] = {
154 		[IB_PORT_NOP]		= "NOP",
155 		[IB_PORT_DOWN]		= "DOWN",
156 		[IB_PORT_INIT]		= "INIT",
157 		[IB_PORT_ARMED]		= "ARMED",
158 		[IB_PORT_ACTIVE]	= "ACTIVE",
159 		[IB_PORT_ACTIVE_DEFER]	= "ACTIVE_DEFER"
160 	};
161 
162 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
163 	if (ret)
164 		return ret;
165 
166 	return sprintf(buf, "%d: %s\n", attr.state,
167 		       attr.state >= 0 && attr.state < ARRAY_SIZE(state_name) ?
168 		       state_name[attr.state] : "UNKNOWN");
169 }
170 
171 static ssize_t lid_show(struct ib_port *p, struct port_attribute *unused,
172 			char *buf)
173 {
174 	struct ib_port_attr attr;
175 	ssize_t ret;
176 
177 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
178 	if (ret)
179 		return ret;
180 
181 	return sprintf(buf, "0x%x\n", attr.lid);
182 }
183 
184 static ssize_t lid_mask_count_show(struct ib_port *p,
185 				   struct port_attribute *unused,
186 				   char *buf)
187 {
188 	struct ib_port_attr attr;
189 	ssize_t ret;
190 
191 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
192 	if (ret)
193 		return ret;
194 
195 	return sprintf(buf, "%d\n", attr.lmc);
196 }
197 
198 static ssize_t sm_lid_show(struct ib_port *p, struct port_attribute *unused,
199 			   char *buf)
200 {
201 	struct ib_port_attr attr;
202 	ssize_t ret;
203 
204 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
205 	if (ret)
206 		return ret;
207 
208 	return sprintf(buf, "0x%x\n", attr.sm_lid);
209 }
210 
211 static ssize_t sm_sl_show(struct ib_port *p, struct port_attribute *unused,
212 			  char *buf)
213 {
214 	struct ib_port_attr attr;
215 	ssize_t ret;
216 
217 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
218 	if (ret)
219 		return ret;
220 
221 	return sprintf(buf, "%d\n", attr.sm_sl);
222 }
223 
224 static ssize_t cap_mask_show(struct ib_port *p, struct port_attribute *unused,
225 			     char *buf)
226 {
227 	struct ib_port_attr attr;
228 	ssize_t ret;
229 
230 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
231 	if (ret)
232 		return ret;
233 
234 	return sprintf(buf, "0x%08x\n", attr.port_cap_flags);
235 }
236 
237 static ssize_t rate_show(struct ib_port *p, struct port_attribute *unused,
238 			 char *buf)
239 {
240 	struct ib_port_attr attr;
241 	char *speed = "";
242 	int rate;		/* in deci-Gb/sec */
243 	ssize_t ret;
244 
245 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
246 	if (ret)
247 		return ret;
248 
249 	switch (attr.active_speed) {
250 	case IB_SPEED_DDR:
251 		speed = " DDR";
252 		rate = 50;
253 		break;
254 	case IB_SPEED_QDR:
255 		speed = " QDR";
256 		rate = 100;
257 		break;
258 	case IB_SPEED_FDR10:
259 		speed = " FDR10";
260 		rate = 100;
261 		break;
262 	case IB_SPEED_FDR:
263 		speed = " FDR";
264 		rate = 140;
265 		break;
266 	case IB_SPEED_EDR:
267 		speed = " EDR";
268 		rate = 250;
269 		break;
270 	case IB_SPEED_HDR:
271 		speed = " HDR";
272 		rate = 500;
273 		break;
274 	case IB_SPEED_SDR:
275 	default:		/* default to SDR for invalid rates */
276 		rate = 25;
277 		break;
278 	}
279 
280 	rate *= ib_width_enum_to_int(attr.active_width);
281 	if (rate < 0)
282 		return -EINVAL;
283 
284 	return sprintf(buf, "%d%s Gb/sec (%dX%s)\n",
285 		       rate / 10, rate % 10 ? ".5" : "",
286 		       ib_width_enum_to_int(attr.active_width), speed);
287 }
288 
289 static ssize_t phys_state_show(struct ib_port *p, struct port_attribute *unused,
290 			       char *buf)
291 {
292 	struct ib_port_attr attr;
293 
294 	ssize_t ret;
295 
296 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
297 	if (ret)
298 		return ret;
299 
300 	switch (attr.phys_state) {
301 	case 1:  return sprintf(buf, "1: Sleep\n");
302 	case 2:  return sprintf(buf, "2: Polling\n");
303 	case 3:  return sprintf(buf, "3: Disabled\n");
304 	case 4:  return sprintf(buf, "4: PortConfigurationTraining\n");
305 	case 5:  return sprintf(buf, "5: LinkUp\n");
306 	case 6:  return sprintf(buf, "6: LinkErrorRecovery\n");
307 	case 7:  return sprintf(buf, "7: Phy Test\n");
308 	default: return sprintf(buf, "%d: <unknown>\n", attr.phys_state);
309 	}
310 }
311 
312 static ssize_t link_layer_show(struct ib_port *p, struct port_attribute *unused,
313 			       char *buf)
314 {
315 	switch (rdma_port_get_link_layer(p->ibdev, p->port_num)) {
316 	case IB_LINK_LAYER_INFINIBAND:
317 		return sprintf(buf, "%s\n", "InfiniBand");
318 	case IB_LINK_LAYER_ETHERNET:
319 		return sprintf(buf, "%s\n", "Ethernet");
320 	default:
321 		return sprintf(buf, "%s\n", "Unknown");
322 	}
323 }
324 
325 static PORT_ATTR_RO(state);
326 static PORT_ATTR_RO(lid);
327 static PORT_ATTR_RO(lid_mask_count);
328 static PORT_ATTR_RO(sm_lid);
329 static PORT_ATTR_RO(sm_sl);
330 static PORT_ATTR_RO(cap_mask);
331 static PORT_ATTR_RO(rate);
332 static PORT_ATTR_RO(phys_state);
333 static PORT_ATTR_RO(link_layer);
334 
335 static struct attribute *port_default_attrs[] = {
336 	&port_attr_state.attr,
337 	&port_attr_lid.attr,
338 	&port_attr_lid_mask_count.attr,
339 	&port_attr_sm_lid.attr,
340 	&port_attr_sm_sl.attr,
341 	&port_attr_cap_mask.attr,
342 	&port_attr_rate.attr,
343 	&port_attr_phys_state.attr,
344 	&port_attr_link_layer.attr,
345 	NULL
346 };
347 
348 static size_t print_ndev(struct ib_gid_attr *gid_attr, char *buf)
349 {
350 	if (!gid_attr->ndev)
351 		return -EINVAL;
352 
353 	return sprintf(buf, "%s\n", gid_attr->ndev->name);
354 }
355 
356 static size_t print_gid_type(struct ib_gid_attr *gid_attr, char *buf)
357 {
358 	return sprintf(buf, "%s\n", ib_cache_gid_type_str(gid_attr->gid_type));
359 }
360 
361 static ssize_t _show_port_gid_attr(struct ib_port *p,
362 				   struct port_attribute *attr,
363 				   char *buf,
364 				   size_t (*print)(struct ib_gid_attr *gid_attr,
365 						   char *buf))
366 {
367 	struct port_table_attribute *tab_attr =
368 		container_of(attr, struct port_table_attribute, attr);
369 	union ib_gid gid;
370 	struct ib_gid_attr gid_attr = {};
371 	ssize_t ret;
372 
373 	ret = ib_query_gid(p->ibdev, p->port_num, tab_attr->index, &gid,
374 			   &gid_attr);
375 	if (ret)
376 		goto err;
377 
378 	ret = print(&gid_attr, buf);
379 
380 err:
381 	if (gid_attr.ndev)
382 		dev_put(gid_attr.ndev);
383 	return ret;
384 }
385 
386 static ssize_t show_port_gid(struct ib_port *p, struct port_attribute *attr,
387 			     char *buf)
388 {
389 	struct port_table_attribute *tab_attr =
390 		container_of(attr, struct port_table_attribute, attr);
391 	union ib_gid gid;
392 	ssize_t ret;
393 
394 	ret = ib_query_gid(p->ibdev, p->port_num, tab_attr->index, &gid, NULL);
395 	if (ret)
396 		return ret;
397 
398 	return sprintf(buf, "%pI6\n", gid.raw);
399 }
400 
401 static ssize_t show_port_gid_attr_ndev(struct ib_port *p,
402 				       struct port_attribute *attr, char *buf)
403 {
404 	return _show_port_gid_attr(p, attr, buf, print_ndev);
405 }
406 
407 static ssize_t show_port_gid_attr_gid_type(struct ib_port *p,
408 					   struct port_attribute *attr,
409 					   char *buf)
410 {
411 	return _show_port_gid_attr(p, attr, buf, print_gid_type);
412 }
413 
414 static ssize_t show_port_pkey(struct ib_port *p, struct port_attribute *attr,
415 			      char *buf)
416 {
417 	struct port_table_attribute *tab_attr =
418 		container_of(attr, struct port_table_attribute, attr);
419 	u16 pkey;
420 	ssize_t ret;
421 
422 	ret = ib_query_pkey(p->ibdev, p->port_num, tab_attr->index, &pkey);
423 	if (ret)
424 		return ret;
425 
426 	return sprintf(buf, "0x%04x\n", pkey);
427 }
428 
429 #define PORT_PMA_ATTR(_name, _counter, _width, _offset)			\
430 struct port_table_attribute port_pma_attr_##_name = {			\
431 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
432 	.index = (_offset) | ((_width) << 16) | ((_counter) << 24),	\
433 	.attr_id = IB_PMA_PORT_COUNTERS ,				\
434 }
435 
436 #define PORT_PMA_ATTR_EXT(_name, _width, _offset)			\
437 struct port_table_attribute port_pma_attr_ext_##_name = {		\
438 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
439 	.index = (_offset) | ((_width) << 16),				\
440 	.attr_id = IB_PMA_PORT_COUNTERS_EXT ,				\
441 }
442 
443 /*
444  * Get a Perfmgmt MAD block of data.
445  * Returns error code or the number of bytes retrieved.
446  */
447 static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
448 		void *data, int offset, size_t size)
449 {
450 	struct ib_mad *in_mad;
451 	struct ib_mad *out_mad;
452 	size_t mad_size = sizeof(*out_mad);
453 	u16 out_mad_pkey_index = 0;
454 	ssize_t ret;
455 
456 	if (!dev->process_mad)
457 		return -ENOSYS;
458 
459 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
460 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
461 	if (!in_mad || !out_mad) {
462 		ret = -ENOMEM;
463 		goto out;
464 	}
465 
466 	in_mad->mad_hdr.base_version  = 1;
467 	in_mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_PERF_MGMT;
468 	in_mad->mad_hdr.class_version = 1;
469 	in_mad->mad_hdr.method        = IB_MGMT_METHOD_GET;
470 	in_mad->mad_hdr.attr_id       = attr;
471 
472 	if (attr != IB_PMA_CLASS_PORT_INFO)
473 		in_mad->data[41] = port_num;	/* PortSelect field */
474 
475 	if ((dev->process_mad(dev, IB_MAD_IGNORE_MKEY,
476 		 port_num, NULL, NULL,
477 		 (const struct ib_mad_hdr *)in_mad, mad_size,
478 		 (struct ib_mad_hdr *)out_mad, &mad_size,
479 		 &out_mad_pkey_index) &
480 	     (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
481 	    (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
482 		ret = -EINVAL;
483 		goto out;
484 	}
485 	memcpy(data, out_mad->data + offset, size);
486 	ret = size;
487 out:
488 	kfree(in_mad);
489 	kfree(out_mad);
490 	return ret;
491 }
492 
493 static ssize_t show_pma_counter(struct ib_port *p, struct port_attribute *attr,
494 				char *buf)
495 {
496 	struct port_table_attribute *tab_attr =
497 		container_of(attr, struct port_table_attribute, attr);
498 	int offset = tab_attr->index & 0xffff;
499 	int width  = (tab_attr->index >> 16) & 0xff;
500 	ssize_t ret;
501 	u8 data[8];
502 
503 	ret = get_perf_mad(p->ibdev, p->port_num, tab_attr->attr_id, &data,
504 			40 + offset / 8, sizeof(data));
505 	if (ret < 0)
506 		return sprintf(buf, "N/A (no PMA)\n");
507 
508 	switch (width) {
509 	case 4:
510 		ret = sprintf(buf, "%u\n", (*data >>
511 					    (4 - (offset % 8))) & 0xf);
512 		break;
513 	case 8:
514 		ret = sprintf(buf, "%u\n", *data);
515 		break;
516 	case 16:
517 		ret = sprintf(buf, "%u\n",
518 			      be16_to_cpup((__be16 *)data));
519 		break;
520 	case 32:
521 		ret = sprintf(buf, "%u\n",
522 			      be32_to_cpup((__be32 *)data));
523 		break;
524 	case 64:
525 		ret = sprintf(buf, "%llu\n",
526 				be64_to_cpup((__be64 *)data));
527 		break;
528 
529 	default:
530 		ret = 0;
531 	}
532 
533 	return ret;
534 }
535 
536 static PORT_PMA_ATTR(symbol_error		    ,  0, 16,  32);
537 static PORT_PMA_ATTR(link_error_recovery	    ,  1,  8,  48);
538 static PORT_PMA_ATTR(link_downed		    ,  2,  8,  56);
539 static PORT_PMA_ATTR(port_rcv_errors		    ,  3, 16,  64);
540 static PORT_PMA_ATTR(port_rcv_remote_physical_errors,  4, 16,  80);
541 static PORT_PMA_ATTR(port_rcv_switch_relay_errors   ,  5, 16,  96);
542 static PORT_PMA_ATTR(port_xmit_discards		    ,  6, 16, 112);
543 static PORT_PMA_ATTR(port_xmit_constraint_errors    ,  7,  8, 128);
544 static PORT_PMA_ATTR(port_rcv_constraint_errors	    ,  8,  8, 136);
545 static PORT_PMA_ATTR(local_link_integrity_errors    ,  9,  4, 152);
546 static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10,  4, 156);
547 static PORT_PMA_ATTR(VL15_dropped		    , 11, 16, 176);
548 static PORT_PMA_ATTR(port_xmit_data		    , 12, 32, 192);
549 static PORT_PMA_ATTR(port_rcv_data		    , 13, 32, 224);
550 static PORT_PMA_ATTR(port_xmit_packets		    , 14, 32, 256);
551 static PORT_PMA_ATTR(port_rcv_packets		    , 15, 32, 288);
552 static PORT_PMA_ATTR(port_xmit_wait		    ,  0, 32, 320);
553 
554 /*
555  * Counters added by extended set
556  */
557 static PORT_PMA_ATTR_EXT(port_xmit_data		    , 64,  64);
558 static PORT_PMA_ATTR_EXT(port_rcv_data		    , 64, 128);
559 static PORT_PMA_ATTR_EXT(port_xmit_packets	    , 64, 192);
560 static PORT_PMA_ATTR_EXT(port_rcv_packets	    , 64, 256);
561 static PORT_PMA_ATTR_EXT(unicast_xmit_packets	    , 64, 320);
562 static PORT_PMA_ATTR_EXT(unicast_rcv_packets	    , 64, 384);
563 static PORT_PMA_ATTR_EXT(multicast_xmit_packets	    , 64, 448);
564 static PORT_PMA_ATTR_EXT(multicast_rcv_packets	    , 64, 512);
565 
566 static struct attribute *pma_attrs[] = {
567 	&port_pma_attr_symbol_error.attr.attr,
568 	&port_pma_attr_link_error_recovery.attr.attr,
569 	&port_pma_attr_link_downed.attr.attr,
570 	&port_pma_attr_port_rcv_errors.attr.attr,
571 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
572 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
573 	&port_pma_attr_port_xmit_discards.attr.attr,
574 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
575 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
576 	&port_pma_attr_local_link_integrity_errors.attr.attr,
577 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
578 	&port_pma_attr_VL15_dropped.attr.attr,
579 	&port_pma_attr_port_xmit_data.attr.attr,
580 	&port_pma_attr_port_rcv_data.attr.attr,
581 	&port_pma_attr_port_xmit_packets.attr.attr,
582 	&port_pma_attr_port_rcv_packets.attr.attr,
583 	&port_pma_attr_port_xmit_wait.attr.attr,
584 	NULL
585 };
586 
587 static struct attribute *pma_attrs_ext[] = {
588 	&port_pma_attr_symbol_error.attr.attr,
589 	&port_pma_attr_link_error_recovery.attr.attr,
590 	&port_pma_attr_link_downed.attr.attr,
591 	&port_pma_attr_port_rcv_errors.attr.attr,
592 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
593 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
594 	&port_pma_attr_port_xmit_discards.attr.attr,
595 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
596 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
597 	&port_pma_attr_local_link_integrity_errors.attr.attr,
598 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
599 	&port_pma_attr_VL15_dropped.attr.attr,
600 	&port_pma_attr_ext_port_xmit_data.attr.attr,
601 	&port_pma_attr_ext_port_rcv_data.attr.attr,
602 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
603 	&port_pma_attr_port_xmit_wait.attr.attr,
604 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
605 	&port_pma_attr_ext_unicast_rcv_packets.attr.attr,
606 	&port_pma_attr_ext_unicast_xmit_packets.attr.attr,
607 	&port_pma_attr_ext_multicast_rcv_packets.attr.attr,
608 	&port_pma_attr_ext_multicast_xmit_packets.attr.attr,
609 	NULL
610 };
611 
612 static struct attribute *pma_attrs_noietf[] = {
613 	&port_pma_attr_symbol_error.attr.attr,
614 	&port_pma_attr_link_error_recovery.attr.attr,
615 	&port_pma_attr_link_downed.attr.attr,
616 	&port_pma_attr_port_rcv_errors.attr.attr,
617 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
618 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
619 	&port_pma_attr_port_xmit_discards.attr.attr,
620 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
621 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
622 	&port_pma_attr_local_link_integrity_errors.attr.attr,
623 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
624 	&port_pma_attr_VL15_dropped.attr.attr,
625 	&port_pma_attr_ext_port_xmit_data.attr.attr,
626 	&port_pma_attr_ext_port_rcv_data.attr.attr,
627 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
628 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
629 	&port_pma_attr_port_xmit_wait.attr.attr,
630 	NULL
631 };
632 
633 static struct attribute_group pma_group = {
634 	.name  = "counters",
635 	.attrs  = pma_attrs
636 };
637 
638 static struct attribute_group pma_group_ext = {
639 	.name  = "counters",
640 	.attrs  = pma_attrs_ext
641 };
642 
643 static struct attribute_group pma_group_noietf = {
644 	.name  = "counters",
645 	.attrs  = pma_attrs_noietf
646 };
647 
648 static void ib_port_release(struct kobject *kobj)
649 {
650 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
651 	struct attribute *a;
652 	int i;
653 
654 	if (p->gid_group.attrs) {
655 		for (i = 0; (a = p->gid_group.attrs[i]); ++i)
656 			kfree(a);
657 
658 		kfree(p->gid_group.attrs);
659 	}
660 
661 	if (p->pkey_group.attrs) {
662 		for (i = 0; (a = p->pkey_group.attrs[i]); ++i)
663 			kfree(a);
664 
665 		kfree(p->pkey_group.attrs);
666 	}
667 
668 	kfree(p);
669 }
670 
671 static void ib_port_gid_attr_release(struct kobject *kobj)
672 {
673 	struct gid_attr_group *g = container_of(kobj, struct gid_attr_group,
674 						kobj);
675 	struct attribute *a;
676 	int i;
677 
678 	if (g->ndev.attrs) {
679 		for (i = 0; (a = g->ndev.attrs[i]); ++i)
680 			kfree(a);
681 
682 		kfree(g->ndev.attrs);
683 	}
684 
685 	if (g->type.attrs) {
686 		for (i = 0; (a = g->type.attrs[i]); ++i)
687 			kfree(a);
688 
689 		kfree(g->type.attrs);
690 	}
691 
692 	kfree(g);
693 }
694 
695 static struct kobj_type port_type = {
696 	.release       = ib_port_release,
697 	.sysfs_ops     = &port_sysfs_ops,
698 	.default_attrs = port_default_attrs
699 };
700 
701 static struct kobj_type gid_attr_type = {
702 	.sysfs_ops      = &gid_attr_sysfs_ops,
703 	.release        = ib_port_gid_attr_release
704 };
705 
706 static struct attribute **
707 alloc_group_attrs(ssize_t (*show)(struct ib_port *,
708 				  struct port_attribute *, char *buf),
709 		  int len)
710 {
711 	struct attribute **tab_attr;
712 	struct port_table_attribute *element;
713 	int i;
714 
715 	tab_attr = kcalloc(1 + len, sizeof(struct attribute *), GFP_KERNEL);
716 	if (!tab_attr)
717 		return NULL;
718 
719 	for (i = 0; i < len; i++) {
720 		element = kzalloc(sizeof(struct port_table_attribute),
721 				  GFP_KERNEL);
722 		if (!element)
723 			goto err;
724 
725 		if (snprintf(element->name, sizeof(element->name),
726 			     "%d", i) >= sizeof(element->name)) {
727 			kfree(element);
728 			goto err;
729 		}
730 
731 		element->attr.attr.name  = element->name;
732 		element->attr.attr.mode  = S_IRUGO;
733 		element->attr.show       = show;
734 		element->index		 = i;
735 		sysfs_attr_init(&element->attr.attr);
736 
737 		tab_attr[i] = &element->attr.attr;
738 	}
739 
740 	return tab_attr;
741 
742 err:
743 	while (--i >= 0)
744 		kfree(tab_attr[i]);
745 	kfree(tab_attr);
746 	return NULL;
747 }
748 
749 /*
750  * Figure out which counter table to use depending on
751  * the device capabilities.
752  */
753 static struct attribute_group *get_counter_table(struct ib_device *dev,
754 						 int port_num)
755 {
756 	struct ib_class_port_info cpi;
757 
758 	if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
759 				&cpi, 40, sizeof(cpi)) >= 0) {
760 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
761 			/* We have extended counters */
762 			return &pma_group_ext;
763 
764 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
765 			/* But not the IETF ones */
766 			return &pma_group_noietf;
767 	}
768 
769 	/* Fall back to normal counters */
770 	return &pma_group;
771 }
772 
773 static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
774 			   u8 port_num, int index)
775 {
776 	int ret;
777 
778 	if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
779 		return 0;
780 	ret = dev->get_hw_stats(dev, stats, port_num, index);
781 	if (ret < 0)
782 		return ret;
783 	if (ret == stats->num_counters)
784 		stats->timestamp = jiffies;
785 
786 	return 0;
787 }
788 
789 static ssize_t print_hw_stat(struct rdma_hw_stats *stats, int index, char *buf)
790 {
791 	return sprintf(buf, "%llu\n", stats->value[index]);
792 }
793 
794 static ssize_t show_hw_stats(struct kobject *kobj, struct attribute *attr,
795 			     char *buf)
796 {
797 	struct ib_device *dev;
798 	struct ib_port *port;
799 	struct hw_stats_attribute *hsa;
800 	struct rdma_hw_stats *stats;
801 	int ret;
802 
803 	hsa = container_of(attr, struct hw_stats_attribute, attr);
804 	if (!hsa->port_num) {
805 		dev = container_of((struct device *)kobj,
806 				   struct ib_device, dev);
807 		stats = dev->hw_stats;
808 	} else {
809 		port = container_of(kobj, struct ib_port, kobj);
810 		dev = port->ibdev;
811 		stats = port->hw_stats;
812 	}
813 	ret = update_hw_stats(dev, stats, hsa->port_num, hsa->index);
814 	if (ret)
815 		return ret;
816 	return print_hw_stat(stats, hsa->index, buf);
817 }
818 
819 static ssize_t show_stats_lifespan(struct kobject *kobj,
820 				   struct attribute *attr,
821 				   char *buf)
822 {
823 	struct hw_stats_attribute *hsa;
824 	int msecs;
825 
826 	hsa = container_of(attr, struct hw_stats_attribute, attr);
827 	if (!hsa->port_num) {
828 		struct ib_device *dev = container_of((struct device *)kobj,
829 						     struct ib_device, dev);
830 		msecs = jiffies_to_msecs(dev->hw_stats->lifespan);
831 	} else {
832 		struct ib_port *p = container_of(kobj, struct ib_port, kobj);
833 		msecs = jiffies_to_msecs(p->hw_stats->lifespan);
834 	}
835 	return sprintf(buf, "%d\n", msecs);
836 }
837 
838 static ssize_t set_stats_lifespan(struct kobject *kobj,
839 				  struct attribute *attr,
840 				  const char *buf, size_t count)
841 {
842 	struct hw_stats_attribute *hsa;
843 	int msecs;
844 	int jiffies;
845 	int ret;
846 
847 	ret = kstrtoint(buf, 10, &msecs);
848 	if (ret)
849 		return ret;
850 	if (msecs < 0 || msecs > 10000)
851 		return -EINVAL;
852 	jiffies = msecs_to_jiffies(msecs);
853 	hsa = container_of(attr, struct hw_stats_attribute, attr);
854 	if (!hsa->port_num) {
855 		struct ib_device *dev = container_of((struct device *)kobj,
856 						     struct ib_device, dev);
857 		dev->hw_stats->lifespan = jiffies;
858 	} else {
859 		struct ib_port *p = container_of(kobj, struct ib_port, kobj);
860 		p->hw_stats->lifespan = jiffies;
861 	}
862 	return count;
863 }
864 
865 static void free_hsag(struct kobject *kobj, struct attribute_group *attr_group)
866 {
867 	struct attribute **attr;
868 
869 	sysfs_remove_group(kobj, attr_group);
870 
871 	for (attr = attr_group->attrs; *attr; attr++)
872 		kfree(*attr);
873 	kfree(attr_group);
874 }
875 
876 static struct attribute *alloc_hsa(int index, u8 port_num, const char *name)
877 {
878 	struct hw_stats_attribute *hsa;
879 
880 	hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
881 	if (!hsa)
882 		return NULL;
883 
884 	hsa->attr.name = (char *)name;
885 	hsa->attr.mode = S_IRUGO;
886 	hsa->show = show_hw_stats;
887 	hsa->store = NULL;
888 	hsa->index = index;
889 	hsa->port_num = port_num;
890 
891 	return &hsa->attr;
892 }
893 
894 static struct attribute *alloc_hsa_lifespan(char *name, u8 port_num)
895 {
896 	struct hw_stats_attribute *hsa;
897 
898 	hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
899 	if (!hsa)
900 		return NULL;
901 
902 	hsa->attr.name = name;
903 	hsa->attr.mode = S_IWUSR | S_IRUGO;
904 	hsa->show = show_stats_lifespan;
905 	hsa->store = set_stats_lifespan;
906 	hsa->index = 0;
907 	hsa->port_num = port_num;
908 
909 	return &hsa->attr;
910 }
911 
912 static void setup_hw_stats(struct ib_device *device, struct ib_port *port,
913 			   u8 port_num)
914 {
915 	struct attribute_group *hsag;
916 	struct rdma_hw_stats *stats;
917 	int i, ret;
918 
919 	stats = device->alloc_hw_stats(device, port_num);
920 
921 	if (!stats)
922 		return;
923 
924 	if (!stats->names || stats->num_counters <= 0)
925 		goto err_free_stats;
926 
927 	/*
928 	 * Two extra attribue elements here, one for the lifespan entry and
929 	 * one to NULL terminate the list for the sysfs core code
930 	 */
931 	hsag = kzalloc(sizeof(*hsag) +
932 		       sizeof(void *) * (stats->num_counters + 2),
933 		       GFP_KERNEL);
934 	if (!hsag)
935 		goto err_free_stats;
936 
937 	ret = device->get_hw_stats(device, stats, port_num,
938 				   stats->num_counters);
939 	if (ret != stats->num_counters)
940 		goto err_free_hsag;
941 
942 	stats->timestamp = jiffies;
943 
944 	hsag->name = "hw_counters";
945 	hsag->attrs = (void *)hsag + sizeof(*hsag);
946 
947 	for (i = 0; i < stats->num_counters; i++) {
948 		hsag->attrs[i] = alloc_hsa(i, port_num, stats->names[i]);
949 		if (!hsag->attrs[i])
950 			goto err;
951 		sysfs_attr_init(hsag->attrs[i]);
952 	}
953 
954 	/* treat an error here as non-fatal */
955 	hsag->attrs[i] = alloc_hsa_lifespan("lifespan", port_num);
956 	if (hsag->attrs[i])
957 		sysfs_attr_init(hsag->attrs[i]);
958 
959 	if (port) {
960 		struct kobject *kobj = &port->kobj;
961 		ret = sysfs_create_group(kobj, hsag);
962 		if (ret)
963 			goto err;
964 		port->hw_stats_ag = hsag;
965 		port->hw_stats = stats;
966 	} else {
967 		struct kobject *kobj = &device->dev.kobj;
968 		ret = sysfs_create_group(kobj, hsag);
969 		if (ret)
970 			goto err;
971 		device->hw_stats_ag = hsag;
972 		device->hw_stats = stats;
973 	}
974 
975 	return;
976 
977 err:
978 	for (; i >= 0; i--)
979 		kfree(hsag->attrs[i]);
980 err_free_hsag:
981 	kfree(hsag);
982 err_free_stats:
983 	kfree(stats);
984 	return;
985 }
986 
987 static int add_port(struct ib_device *device, int port_num,
988 		    int (*port_callback)(struct ib_device *,
989 					 u8, struct kobject *))
990 {
991 	struct ib_port *p;
992 	struct ib_port_attr attr;
993 	int i;
994 	int ret;
995 
996 	ret = ib_query_port(device, port_num, &attr);
997 	if (ret)
998 		return ret;
999 
1000 	p = kzalloc(sizeof *p, GFP_KERNEL);
1001 	if (!p)
1002 		return -ENOMEM;
1003 
1004 	p->ibdev      = device;
1005 	p->port_num   = port_num;
1006 
1007 	ret = kobject_init_and_add(&p->kobj, &port_type,
1008 				   device->ports_parent,
1009 				   "%d", port_num);
1010 	if (ret) {
1011 		kfree(p);
1012 		return ret;
1013 	}
1014 
1015 	p->gid_attr_group = kzalloc(sizeof(*p->gid_attr_group), GFP_KERNEL);
1016 	if (!p->gid_attr_group) {
1017 		ret = -ENOMEM;
1018 		goto err_put;
1019 	}
1020 
1021 	p->gid_attr_group->port = p;
1022 	ret = kobject_init_and_add(&p->gid_attr_group->kobj, &gid_attr_type,
1023 				   &p->kobj, "gid_attrs");
1024 	if (ret) {
1025 		kfree(p->gid_attr_group);
1026 		goto err_put;
1027 	}
1028 
1029 	p->pma_table = get_counter_table(device, port_num);
1030 	ret = sysfs_create_group(&p->kobj, p->pma_table);
1031 	if (ret)
1032 		goto err_put_gid_attrs;
1033 
1034 	p->gid_group.name  = "gids";
1035 	p->gid_group.attrs = alloc_group_attrs(show_port_gid, attr.gid_tbl_len);
1036 	if (!p->gid_group.attrs) {
1037 		ret = -ENOMEM;
1038 		goto err_remove_pma;
1039 	}
1040 
1041 	ret = sysfs_create_group(&p->kobj, &p->gid_group);
1042 	if (ret)
1043 		goto err_free_gid;
1044 
1045 	p->gid_attr_group->ndev.name = "ndevs";
1046 	p->gid_attr_group->ndev.attrs = alloc_group_attrs(show_port_gid_attr_ndev,
1047 							  attr.gid_tbl_len);
1048 	if (!p->gid_attr_group->ndev.attrs) {
1049 		ret = -ENOMEM;
1050 		goto err_remove_gid;
1051 	}
1052 
1053 	ret = sysfs_create_group(&p->gid_attr_group->kobj,
1054 				 &p->gid_attr_group->ndev);
1055 	if (ret)
1056 		goto err_free_gid_ndev;
1057 
1058 	p->gid_attr_group->type.name = "types";
1059 	p->gid_attr_group->type.attrs = alloc_group_attrs(show_port_gid_attr_gid_type,
1060 							  attr.gid_tbl_len);
1061 	if (!p->gid_attr_group->type.attrs) {
1062 		ret = -ENOMEM;
1063 		goto err_remove_gid_ndev;
1064 	}
1065 
1066 	ret = sysfs_create_group(&p->gid_attr_group->kobj,
1067 				 &p->gid_attr_group->type);
1068 	if (ret)
1069 		goto err_free_gid_type;
1070 
1071 	p->pkey_group.name  = "pkeys";
1072 	p->pkey_group.attrs = alloc_group_attrs(show_port_pkey,
1073 						attr.pkey_tbl_len);
1074 	if (!p->pkey_group.attrs) {
1075 		ret = -ENOMEM;
1076 		goto err_remove_gid_type;
1077 	}
1078 
1079 	ret = sysfs_create_group(&p->kobj, &p->pkey_group);
1080 	if (ret)
1081 		goto err_free_pkey;
1082 
1083 	if (port_callback) {
1084 		ret = port_callback(device, port_num, &p->kobj);
1085 		if (ret)
1086 			goto err_remove_pkey;
1087 	}
1088 
1089 	/*
1090 	 * If port == 0, it means we have only one port and the parent
1091 	 * device, not this port device, should be the holder of the
1092 	 * hw_counters
1093 	 */
1094 	if (device->alloc_hw_stats && port_num)
1095 		setup_hw_stats(device, p, port_num);
1096 
1097 	list_add_tail(&p->kobj.entry, &device->port_list);
1098 
1099 	kobject_uevent(&p->kobj, KOBJ_ADD);
1100 	return 0;
1101 
1102 err_remove_pkey:
1103 	sysfs_remove_group(&p->kobj, &p->pkey_group);
1104 
1105 err_free_pkey:
1106 	for (i = 0; i < attr.pkey_tbl_len; ++i)
1107 		kfree(p->pkey_group.attrs[i]);
1108 
1109 	kfree(p->pkey_group.attrs);
1110 	p->pkey_group.attrs = NULL;
1111 
1112 err_remove_gid_type:
1113 	sysfs_remove_group(&p->gid_attr_group->kobj,
1114 			   &p->gid_attr_group->type);
1115 
1116 err_free_gid_type:
1117 	for (i = 0; i < attr.gid_tbl_len; ++i)
1118 		kfree(p->gid_attr_group->type.attrs[i]);
1119 
1120 	kfree(p->gid_attr_group->type.attrs);
1121 	p->gid_attr_group->type.attrs = NULL;
1122 
1123 err_remove_gid_ndev:
1124 	sysfs_remove_group(&p->gid_attr_group->kobj,
1125 			   &p->gid_attr_group->ndev);
1126 
1127 err_free_gid_ndev:
1128 	for (i = 0; i < attr.gid_tbl_len; ++i)
1129 		kfree(p->gid_attr_group->ndev.attrs[i]);
1130 
1131 	kfree(p->gid_attr_group->ndev.attrs);
1132 	p->gid_attr_group->ndev.attrs = NULL;
1133 
1134 err_remove_gid:
1135 	sysfs_remove_group(&p->kobj, &p->gid_group);
1136 
1137 err_free_gid:
1138 	for (i = 0; i < attr.gid_tbl_len; ++i)
1139 		kfree(p->gid_group.attrs[i]);
1140 
1141 	kfree(p->gid_group.attrs);
1142 	p->gid_group.attrs = NULL;
1143 
1144 err_remove_pma:
1145 	sysfs_remove_group(&p->kobj, p->pma_table);
1146 
1147 err_put_gid_attrs:
1148 	kobject_put(&p->gid_attr_group->kobj);
1149 
1150 err_put:
1151 	kobject_put(&p->kobj);
1152 	return ret;
1153 }
1154 
1155 static ssize_t show_node_type(struct device *device,
1156 			      struct device_attribute *attr, char *buf)
1157 {
1158 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1159 
1160 	switch (dev->node_type) {
1161 	case RDMA_NODE_IB_CA:	  return sprintf(buf, "%d: CA\n", dev->node_type);
1162 	case RDMA_NODE_RNIC:	  return sprintf(buf, "%d: RNIC\n", dev->node_type);
1163 	case RDMA_NODE_USNIC:	  return sprintf(buf, "%d: usNIC\n", dev->node_type);
1164 	case RDMA_NODE_USNIC_UDP: return sprintf(buf, "%d: usNIC UDP\n", dev->node_type);
1165 	case RDMA_NODE_IB_SWITCH: return sprintf(buf, "%d: switch\n", dev->node_type);
1166 	case RDMA_NODE_IB_ROUTER: return sprintf(buf, "%d: router\n", dev->node_type);
1167 	default:		  return sprintf(buf, "%d: <unknown>\n", dev->node_type);
1168 	}
1169 }
1170 
1171 static ssize_t show_sys_image_guid(struct device *device,
1172 				   struct device_attribute *dev_attr, char *buf)
1173 {
1174 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1175 
1176 	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1177 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[0]),
1178 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[1]),
1179 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[2]),
1180 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[3]));
1181 }
1182 
1183 static ssize_t show_node_guid(struct device *device,
1184 			      struct device_attribute *attr, char *buf)
1185 {
1186 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1187 
1188 	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1189 		       be16_to_cpu(((__be16 *) &dev->node_guid)[0]),
1190 		       be16_to_cpu(((__be16 *) &dev->node_guid)[1]),
1191 		       be16_to_cpu(((__be16 *) &dev->node_guid)[2]),
1192 		       be16_to_cpu(((__be16 *) &dev->node_guid)[3]));
1193 }
1194 
1195 static ssize_t show_node_desc(struct device *device,
1196 			      struct device_attribute *attr, char *buf)
1197 {
1198 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1199 
1200 	return sprintf(buf, "%.64s\n", dev->node_desc);
1201 }
1202 
1203 static ssize_t set_node_desc(struct device *device,
1204 			     struct device_attribute *attr,
1205 			     const char *buf, size_t count)
1206 {
1207 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1208 	struct ib_device_modify desc = {};
1209 	int ret;
1210 
1211 	if (!dev->modify_device)
1212 		return -EIO;
1213 
1214 	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1215 	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1216 	if (ret)
1217 		return ret;
1218 
1219 	return count;
1220 }
1221 
1222 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1223 			   char *buf)
1224 {
1225 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1226 
1227 	ib_get_device_fw_str(dev, buf);
1228 	strlcat(buf, "\n", IB_FW_VERSION_NAME_MAX);
1229 	return strlen(buf);
1230 }
1231 
1232 static DEVICE_ATTR(node_type, S_IRUGO, show_node_type, NULL);
1233 static DEVICE_ATTR(sys_image_guid, S_IRUGO, show_sys_image_guid, NULL);
1234 static DEVICE_ATTR(node_guid, S_IRUGO, show_node_guid, NULL);
1235 static DEVICE_ATTR(node_desc, S_IRUGO | S_IWUSR, show_node_desc, set_node_desc);
1236 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1237 
1238 static struct device_attribute *ib_class_attributes[] = {
1239 	&dev_attr_node_type,
1240 	&dev_attr_sys_image_guid,
1241 	&dev_attr_node_guid,
1242 	&dev_attr_node_desc,
1243 	&dev_attr_fw_ver,
1244 };
1245 
1246 static void free_port_list_attributes(struct ib_device *device)
1247 {
1248 	struct kobject *p, *t;
1249 
1250 	list_for_each_entry_safe(p, t, &device->port_list, entry) {
1251 		struct ib_port *port = container_of(p, struct ib_port, kobj);
1252 		list_del(&p->entry);
1253 		if (port->hw_stats) {
1254 			kfree(port->hw_stats);
1255 			free_hsag(&port->kobj, port->hw_stats_ag);
1256 		}
1257 		sysfs_remove_group(p, port->pma_table);
1258 		sysfs_remove_group(p, &port->pkey_group);
1259 		sysfs_remove_group(p, &port->gid_group);
1260 		sysfs_remove_group(&port->gid_attr_group->kobj,
1261 				   &port->gid_attr_group->ndev);
1262 		sysfs_remove_group(&port->gid_attr_group->kobj,
1263 				   &port->gid_attr_group->type);
1264 		kobject_put(&port->gid_attr_group->kobj);
1265 		kobject_put(p);
1266 	}
1267 
1268 	kobject_put(device->ports_parent);
1269 }
1270 
1271 int ib_device_register_sysfs(struct ib_device *device,
1272 			     int (*port_callback)(struct ib_device *,
1273 						  u8, struct kobject *))
1274 {
1275 	struct device *class_dev = &device->dev;
1276 	int ret;
1277 	int i;
1278 
1279 	ret = dev_set_name(class_dev, "%s", device->name);
1280 	if (ret)
1281 		return ret;
1282 
1283 	ret = device_add(class_dev);
1284 	if (ret)
1285 		goto err;
1286 
1287 	for (i = 0; i < ARRAY_SIZE(ib_class_attributes); ++i) {
1288 		ret = device_create_file(class_dev, ib_class_attributes[i]);
1289 		if (ret)
1290 			goto err_unregister;
1291 	}
1292 
1293 	device->ports_parent = kobject_create_and_add("ports",
1294 						      &class_dev->kobj);
1295 	if (!device->ports_parent) {
1296 		ret = -ENOMEM;
1297 		goto err_put;
1298 	}
1299 
1300 	if (rdma_cap_ib_switch(device)) {
1301 		ret = add_port(device, 0, port_callback);
1302 		if (ret)
1303 			goto err_put;
1304 	} else {
1305 		for (i = 1; i <= device->phys_port_cnt; ++i) {
1306 			ret = add_port(device, i, port_callback);
1307 			if (ret)
1308 				goto err_put;
1309 		}
1310 	}
1311 
1312 	if (device->alloc_hw_stats)
1313 		setup_hw_stats(device, NULL, 0);
1314 
1315 	return 0;
1316 
1317 err_put:
1318 	free_port_list_attributes(device);
1319 
1320 err_unregister:
1321 	device_del(class_dev);
1322 
1323 err:
1324 	return ret;
1325 }
1326 
1327 void ib_device_unregister_sysfs(struct ib_device *device)
1328 {
1329 	int i;
1330 
1331 	/* Hold kobject until ib_dealloc_device() */
1332 	kobject_get(&device->dev.kobj);
1333 
1334 	free_port_list_attributes(device);
1335 
1336 	if (device->hw_stats) {
1337 		kfree(device->hw_stats);
1338 		free_hsag(&device->dev.kobj, device->hw_stats_ag);
1339 	}
1340 
1341 	for (i = 0; i < ARRAY_SIZE(ib_class_attributes); ++i)
1342 		device_remove_file(&device->dev, ib_class_attributes[i]);
1343 
1344 	device_unregister(&device->dev);
1345 }
1346