xref: /freebsd/sys/ofed/drivers/infiniband/core/ib_sysfs.c (revision 4726b80d9379cdefd00ae77d0b91973bde573790)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
3  *
4  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
5  * Copyright (c) 2005 Mellanox Technologies Ltd.  All rights reserved.
6  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
7  *
8  * This software is available to you under a choice of one of two
9  * licenses.  You may choose to be licensed under the terms of the GNU
10  * General Public License (GPL) Version 2, available from the file
11  * COPYING in the main directory of this source tree, or the
12  * OpenIB.org BSD license below:
13  *
14  *     Redistribution and use in source and binary forms, with or
15  *     without modification, are permitted provided that the following
16  *     conditions are met:
17  *
18  *      - Redistributions of source code must retain the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer.
21  *
22  *      - Redistributions in binary form must reproduce the above
23  *        copyright notice, this list of conditions and the following
24  *        disclaimer in the documentation and/or other materials
25  *        provided with the distribution.
26  *
27  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34  * SOFTWARE.
35  */
36 
37 #include <sys/cdefs.h>
38 #include "core_priv.h"
39 
40 #include <linux/slab.h>
41 #include <linux/string.h>
42 #include <linux/netdevice.h>
43 #include <linux/fs.h>
44 #include <linux/printk.h>
45 
46 #include <rdma/ib_addr.h>
47 #include <rdma/ib_mad.h>
48 #include <rdma/ib_pma.h>
49 #include <rdma/ib_cache.h>
50 
51 struct ib_port;
52 
53 struct gid_attr_group {
54 	struct ib_port		*port;
55 	struct kobject		kobj;
56 	struct attribute_group	ndev;
57 	struct attribute_group	type;
58 };
59 struct ib_port {
60 	struct kobject         kobj;
61 	struct ib_device      *ibdev;
62 	struct gid_attr_group *gid_attr_group;
63 	struct attribute_group gid_group;
64 	struct attribute_group pkey_group;
65 	struct attribute_group *pma_table;
66 	struct attribute_group *hw_stats_ag;
67 	struct rdma_hw_stats   *hw_stats;
68 	u8                     port_num;
69 };
70 
71 struct port_attribute {
72 	struct attribute attr;
73 	ssize_t (*show)(struct ib_port *, struct port_attribute *, char *buf);
74 	ssize_t (*store)(struct ib_port *, struct port_attribute *,
75 			 const char *buf, size_t count);
76 };
77 
78 #define PORT_ATTR(_name, _mode, _show, _store) \
79 struct port_attribute port_attr_##_name = __ATTR(_name, _mode, _show, _store)
80 
81 #define PORT_ATTR_RO(_name) \
82 struct port_attribute port_attr_##_name = __ATTR_RO(_name)
83 
84 struct port_table_attribute {
85 	struct port_attribute	attr;
86 	char			name[8];
87 	int			index;
88 	__be16			attr_id;
89 };
90 
91 struct hw_stats_attribute {
92 	struct attribute	attr;
93 	ssize_t			(*show)(struct kobject *kobj,
94 					struct attribute *attr, char *buf);
95 	ssize_t			(*store)(struct kobject *kobj,
96 					 struct attribute *attr,
97 					 const char *buf,
98 					 size_t count);
99 	int			index;
100 	u8			port_num;
101 };
102 
103 static ssize_t port_attr_show(struct kobject *kobj,
104 			      struct attribute *attr, char *buf)
105 {
106 	struct port_attribute *port_attr =
107 		container_of(attr, struct port_attribute, attr);
108 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
109 
110 	if (!port_attr->show)
111 		return -EIO;
112 
113 	return port_attr->show(p, port_attr, buf);
114 }
115 
116 static ssize_t port_attr_store(struct kobject *kobj,
117 			       struct attribute *attr,
118 			       const char *buf, size_t count)
119 {
120 	struct port_attribute *port_attr =
121 		container_of(attr, struct port_attribute, attr);
122 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
123 
124 	if (!port_attr->store)
125 		return -EIO;
126 	return port_attr->store(p, port_attr, buf, count);
127 }
128 
129 static const struct sysfs_ops port_sysfs_ops = {
130 	.show	= port_attr_show,
131 	.store	= port_attr_store
132 };
133 
134 static ssize_t gid_attr_show(struct kobject *kobj,
135 			     struct attribute *attr, char *buf)
136 {
137 	struct port_attribute *port_attr =
138 		container_of(attr, struct port_attribute, attr);
139 	struct ib_port *p = container_of(kobj, struct gid_attr_group,
140 					 kobj)->port;
141 
142 	if (!port_attr->show)
143 		return -EIO;
144 
145 	return port_attr->show(p, port_attr, buf);
146 }
147 
148 static const struct sysfs_ops gid_attr_sysfs_ops = {
149 	.show = gid_attr_show
150 };
151 
152 static ssize_t state_show(struct ib_port *p, struct port_attribute *unused,
153 			  char *buf)
154 {
155 	struct ib_port_attr attr;
156 	ssize_t ret;
157 
158 	static const char *state_name[] = {
159 		[IB_PORT_NOP]		= "NOP",
160 		[IB_PORT_DOWN]		= "DOWN",
161 		[IB_PORT_INIT]		= "INIT",
162 		[IB_PORT_ARMED]		= "ARMED",
163 		[IB_PORT_ACTIVE]	= "ACTIVE",
164 		[IB_PORT_ACTIVE_DEFER]	= "ACTIVE_DEFER"
165 	};
166 
167 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
168 	if (ret)
169 		return ret;
170 
171 	return sprintf(buf, "%d: %s\n", attr.state,
172 		       attr.state >= 0 && attr.state < ARRAY_SIZE(state_name) ?
173 		       state_name[attr.state] : "UNKNOWN");
174 }
175 
176 static ssize_t lid_show(struct ib_port *p, struct port_attribute *unused,
177 			char *buf)
178 {
179 	struct ib_port_attr attr;
180 	ssize_t ret;
181 
182 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
183 	if (ret)
184 		return ret;
185 
186 	return sprintf(buf, "0x%x\n", attr.lid);
187 }
188 
189 static ssize_t lid_mask_count_show(struct ib_port *p,
190 				   struct port_attribute *unused,
191 				   char *buf)
192 {
193 	struct ib_port_attr attr;
194 	ssize_t ret;
195 
196 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
197 	if (ret)
198 		return ret;
199 
200 	return sprintf(buf, "%d\n", attr.lmc);
201 }
202 
203 static ssize_t sm_lid_show(struct ib_port *p, struct port_attribute *unused,
204 			   char *buf)
205 {
206 	struct ib_port_attr attr;
207 	ssize_t ret;
208 
209 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
210 	if (ret)
211 		return ret;
212 
213 	return sprintf(buf, "0x%x\n", attr.sm_lid);
214 }
215 
216 static ssize_t sm_sl_show(struct ib_port *p, struct port_attribute *unused,
217 			  char *buf)
218 {
219 	struct ib_port_attr attr;
220 	ssize_t ret;
221 
222 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
223 	if (ret)
224 		return ret;
225 
226 	return sprintf(buf, "%d\n", attr.sm_sl);
227 }
228 
229 static ssize_t cap_mask_show(struct ib_port *p, struct port_attribute *unused,
230 			     char *buf)
231 {
232 	struct ib_port_attr attr;
233 	ssize_t ret;
234 
235 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
236 	if (ret)
237 		return ret;
238 
239 	return sprintf(buf, "0x%08x\n", attr.port_cap_flags);
240 }
241 
242 static ssize_t rate_show(struct ib_port *p, struct port_attribute *unused,
243 			 char *buf)
244 {
245 	struct ib_port_attr attr;
246 	char *speed = "";
247 	int rate;		/* in deci-Gb/sec */
248 	ssize_t ret;
249 
250 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
251 	if (ret)
252 		return ret;
253 
254 	switch (attr.active_speed) {
255 	case IB_SPEED_DDR:
256 		speed = " DDR";
257 		rate = 50;
258 		break;
259 	case IB_SPEED_QDR:
260 		speed = " QDR";
261 		rate = 100;
262 		break;
263 	case IB_SPEED_FDR10:
264 		speed = " FDR10";
265 		rate = 100;
266 		break;
267 	case IB_SPEED_FDR:
268 		speed = " FDR";
269 		rate = 140;
270 		break;
271 	case IB_SPEED_EDR:
272 		speed = " EDR";
273 		rate = 250;
274 		break;
275 	case IB_SPEED_HDR:
276 		speed = " HDR";
277 		rate = 500;
278 		break;
279 	case IB_SPEED_NDR:
280 		speed = " NDR";
281 		rate = 1000;
282 		break;
283 	case IB_SPEED_XDR:
284 		speed = " XDR";
285 		rate = 2000;
286 		break;
287 	case IB_SPEED_SDR:
288 	default:		/* default to SDR for invalid rates */
289 		speed = " SDR";
290 		rate = 25;
291 		break;
292 	}
293 
294 	rate *= ib_width_enum_to_int(attr.active_width);
295 	if (rate < 0)
296 		return -EINVAL;
297 
298 	return sprintf(buf, "%d%s Gb/sec (%dX%s)\n",
299 		       rate / 10, rate % 10 ? ".5" : "",
300 		       ib_width_enum_to_int(attr.active_width), speed);
301 }
302 
303 static const char *phys_state_to_str(enum ib_port_phys_state phys_state)
304 {
305 	static const char * phys_state_str[] = {
306 		"<unknown>",
307 		"Sleep",
308 		"Polling",
309 		"Disabled",
310 		"PortConfigurationTraining",
311 		"LinkUp",
312 		"LinkErrorRecovery",
313 		"Phy Test",
314 	};
315 
316 	if (phys_state < ARRAY_SIZE(phys_state_str))
317 		return phys_state_str[phys_state];
318 	return "<unknown>";
319 }
320 
321 static ssize_t phys_state_show(struct ib_port *p, struct port_attribute *unused,
322 			       char *buf)
323 {
324 	struct ib_port_attr attr;
325 
326 	ssize_t ret;
327 
328 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
329 	if (ret)
330 		return ret;
331 
332 	return sprintf(buf, "%d: %s\n", attr.phys_state,
333 		       phys_state_to_str(attr.phys_state));
334 }
335 
336 static ssize_t link_layer_show(struct ib_port *p, struct port_attribute *unused,
337 			       char *buf)
338 {
339 	switch (rdma_port_get_link_layer(p->ibdev, p->port_num)) {
340 	case IB_LINK_LAYER_INFINIBAND:
341 		return sprintf(buf, "%s\n", "InfiniBand");
342 	case IB_LINK_LAYER_ETHERNET:
343 		return sprintf(buf, "%s\n", "Ethernet");
344 	default:
345 		return sprintf(buf, "%s\n", "Unknown");
346 	}
347 }
348 
349 static PORT_ATTR_RO(state);
350 static PORT_ATTR_RO(lid);
351 static PORT_ATTR_RO(lid_mask_count);
352 static PORT_ATTR_RO(sm_lid);
353 static PORT_ATTR_RO(sm_sl);
354 static PORT_ATTR_RO(cap_mask);
355 static PORT_ATTR_RO(rate);
356 static PORT_ATTR_RO(phys_state);
357 static PORT_ATTR_RO(link_layer);
358 
359 static struct attribute *port_default_attrs[] = {
360 	&port_attr_state.attr,
361 	&port_attr_lid.attr,
362 	&port_attr_lid_mask_count.attr,
363 	&port_attr_sm_lid.attr,
364 	&port_attr_sm_sl.attr,
365 	&port_attr_cap_mask.attr,
366 	&port_attr_rate.attr,
367 	&port_attr_phys_state.attr,
368 	&port_attr_link_layer.attr,
369 	NULL
370 };
371 
372 static size_t print_ndev(const struct ib_gid_attr *gid_attr, char *buf)
373 {
374 	if (!gid_attr->ndev)
375 		return -EINVAL;
376 
377 	return sprintf(buf, "%s\n", if_name(gid_attr->ndev));
378 }
379 
380 static size_t print_gid_type(const struct ib_gid_attr *gid_attr, char *buf)
381 {
382 	return sprintf(buf, "%s\n", ib_cache_gid_type_str(gid_attr->gid_type));
383 }
384 
385 static ssize_t _show_port_gid_attr(
386 	struct ib_port *p, struct port_attribute *attr, char *buf,
387 	size_t (*print)(const struct ib_gid_attr *gid_attr, char *buf))
388 {
389 	struct port_table_attribute *tab_attr =
390 		container_of(attr, struct port_table_attribute, attr);
391 	const struct ib_gid_attr *gid_attr;
392 	ssize_t ret;
393 
394 	gid_attr = rdma_get_gid_attr(p->ibdev, p->port_num, tab_attr->index);
395 	if (IS_ERR(gid_attr))
396 		return PTR_ERR(gid_attr);
397 
398 	ret = print(gid_attr, buf);
399 	rdma_put_gid_attr(gid_attr);
400 	return ret;
401 }
402 
403 static ssize_t show_port_gid(struct ib_port *p, struct port_attribute *attr,
404 			     char *buf)
405 {
406 	struct port_table_attribute *tab_attr =
407 		container_of(attr, struct port_table_attribute, attr);
408 	const struct ib_gid_attr *gid_attr;
409 	ssize_t ret;
410 
411 	gid_attr = rdma_get_gid_attr(p->ibdev, p->port_num, tab_attr->index);
412 	if (IS_ERR(gid_attr)) {
413 		const union ib_gid zgid = {};
414 
415 		/* If reading GID fails, it is likely due to GID entry being
416 		 * empty (invalid) or reserved GID in the table.  User space
417 		 * expects to read GID table entries as long as it given index
418 		 * is within GID table size.  Administrative/debugging tool
419 		 * fails to query rest of the GID entries if it hits error
420 		 * while querying a GID of the given index.  To avoid user
421 		 * space throwing such error on fail to read gid, return zero
422 		 * GID as before. This maintains backward compatibility.
423 		 */
424 		return sprintf(buf, "%pI6\n", zgid.raw);
425 	}
426 
427 	ret = sprintf(buf, GID_PRINT_FMT"\n", GID_PRINT_ARGS(gid_attr->gid.raw));
428 	rdma_put_gid_attr(gid_attr);
429 	return ret;
430 }
431 
432 static ssize_t show_port_gid_attr_ndev(struct ib_port *p,
433 				       struct port_attribute *attr, char *buf)
434 {
435 	return _show_port_gid_attr(p, attr, buf, print_ndev);
436 }
437 
438 static ssize_t show_port_gid_attr_gid_type(struct ib_port *p,
439 					   struct port_attribute *attr,
440 					   char *buf)
441 {
442 	return _show_port_gid_attr(p, attr, buf, print_gid_type);
443 }
444 
445 static ssize_t show_port_pkey(struct ib_port *p, struct port_attribute *attr,
446 			      char *buf)
447 {
448 	struct port_table_attribute *tab_attr =
449 		container_of(attr, struct port_table_attribute, attr);
450 	u16 pkey;
451 	ssize_t ret;
452 
453 	ret = ib_query_pkey(p->ibdev, p->port_num, tab_attr->index, &pkey);
454 	if (ret)
455 		return ret;
456 
457 	return sprintf(buf, "0x%04x\n", pkey);
458 }
459 
460 #define PORT_PMA_ATTR(_name, _counter, _width, _offset)			\
461 struct port_table_attribute port_pma_attr_##_name = {			\
462 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
463 	.index = (_offset) | ((_width) << 16) | ((_counter) << 24),	\
464 	.attr_id = IB_PMA_PORT_COUNTERS ,				\
465 }
466 
467 #define PORT_PMA_ATTR_EXT(_name, _width, _offset)			\
468 struct port_table_attribute port_pma_attr_ext_##_name = {		\
469 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
470 	.index = (_offset) | ((_width) << 16),				\
471 	.attr_id = IB_PMA_PORT_COUNTERS_EXT ,				\
472 }
473 
474 /*
475  * Get a Perfmgmt MAD block of data.
476  * Returns error code or the number of bytes retrieved.
477  */
478 static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
479 		void *data, int offset, size_t size)
480 {
481 	struct ib_mad *in_mad;
482 	struct ib_mad *out_mad;
483 	size_t mad_size = sizeof(*out_mad);
484 	u16 out_mad_pkey_index = 0;
485 	ssize_t ret;
486 
487 	if (!dev->process_mad)
488 		return -ENOSYS;
489 
490 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
491 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
492 	if (!in_mad || !out_mad) {
493 		ret = -ENOMEM;
494 		goto out;
495 	}
496 
497 	in_mad->mad_hdr.base_version  = 1;
498 	in_mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_PERF_MGMT;
499 	in_mad->mad_hdr.class_version = 1;
500 	in_mad->mad_hdr.method        = IB_MGMT_METHOD_GET;
501 	in_mad->mad_hdr.attr_id       = attr;
502 
503 	if (attr != IB_PMA_CLASS_PORT_INFO)
504 		in_mad->data[41] = port_num;	/* PortSelect field */
505 
506 	if ((dev->process_mad(dev, IB_MAD_IGNORE_MKEY,
507 		 port_num, NULL, NULL,
508 		 (const struct ib_mad_hdr *)in_mad, mad_size,
509 		 (struct ib_mad_hdr *)out_mad, &mad_size,
510 		 &out_mad_pkey_index) &
511 	     (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
512 	    (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
513 		ret = -EINVAL;
514 		goto out;
515 	}
516 	memcpy(data, out_mad->data + offset, size);
517 	ret = size;
518 out:
519 	kfree(in_mad);
520 	kfree(out_mad);
521 	return ret;
522 }
523 
524 static ssize_t show_pma_counter(struct ib_port *p, struct port_attribute *attr,
525 				char *buf)
526 {
527 	struct port_table_attribute *tab_attr =
528 		container_of(attr, struct port_table_attribute, attr);
529 	int offset = tab_attr->index & 0xffff;
530 	int width  = (tab_attr->index >> 16) & 0xff;
531 	ssize_t ret;
532 	u8 data[8];
533 
534 	ret = get_perf_mad(p->ibdev, p->port_num, tab_attr->attr_id, &data,
535 			40 + offset / 8, sizeof(data));
536 	if (ret < 0)
537 		return ret;
538 
539 	switch (width) {
540 	case 4:
541 		ret = sprintf(buf, "%u\n", (*data >>
542 					    (4 - (offset % 8))) & 0xf);
543 		break;
544 	case 8:
545 		ret = sprintf(buf, "%u\n", *data);
546 		break;
547 	case 16:
548 		ret = sprintf(buf, "%u\n",
549 			      be16_to_cpup((__be16 *)data));
550 		break;
551 	case 32:
552 		ret = sprintf(buf, "%u\n",
553 			      be32_to_cpup((__be32 *)data));
554 		break;
555 	case 64:
556 		ret = sprintf(buf, "%llu\n",
557 			      (unsigned long long)be64_to_cpup((__be64 *)data));
558 		break;
559 
560 	default:
561 		ret = 0;
562 	}
563 
564 	return ret;
565 }
566 
567 static PORT_PMA_ATTR(symbol_error		    ,  0, 16,  32);
568 static PORT_PMA_ATTR(link_error_recovery	    ,  1,  8,  48);
569 static PORT_PMA_ATTR(link_downed		    ,  2,  8,  56);
570 static PORT_PMA_ATTR(port_rcv_errors		    ,  3, 16,  64);
571 static PORT_PMA_ATTR(port_rcv_remote_physical_errors,  4, 16,  80);
572 static PORT_PMA_ATTR(port_rcv_switch_relay_errors   ,  5, 16,  96);
573 static PORT_PMA_ATTR(port_xmit_discards		    ,  6, 16, 112);
574 static PORT_PMA_ATTR(port_xmit_constraint_errors    ,  7,  8, 128);
575 static PORT_PMA_ATTR(port_rcv_constraint_errors	    ,  8,  8, 136);
576 static PORT_PMA_ATTR(local_link_integrity_errors    ,  9,  4, 152);
577 static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10,  4, 156);
578 static PORT_PMA_ATTR(VL15_dropped		    , 11, 16, 176);
579 static PORT_PMA_ATTR(port_xmit_data		    , 12, 32, 192);
580 static PORT_PMA_ATTR(port_rcv_data		    , 13, 32, 224);
581 static PORT_PMA_ATTR(port_xmit_packets		    , 14, 32, 256);
582 static PORT_PMA_ATTR(port_rcv_packets		    , 15, 32, 288);
583 static PORT_PMA_ATTR(port_xmit_wait		    ,  0, 32, 320);
584 
585 /*
586  * Counters added by extended set
587  */
588 static PORT_PMA_ATTR_EXT(port_xmit_data		    , 64,  64);
589 static PORT_PMA_ATTR_EXT(port_rcv_data		    , 64, 128);
590 static PORT_PMA_ATTR_EXT(port_xmit_packets	    , 64, 192);
591 static PORT_PMA_ATTR_EXT(port_rcv_packets	    , 64, 256);
592 static PORT_PMA_ATTR_EXT(unicast_xmit_packets	    , 64, 320);
593 static PORT_PMA_ATTR_EXT(unicast_rcv_packets	    , 64, 384);
594 static PORT_PMA_ATTR_EXT(multicast_xmit_packets	    , 64, 448);
595 static PORT_PMA_ATTR_EXT(multicast_rcv_packets	    , 64, 512);
596 
597 static struct attribute *pma_attrs[] = {
598 	&port_pma_attr_symbol_error.attr.attr,
599 	&port_pma_attr_link_error_recovery.attr.attr,
600 	&port_pma_attr_link_downed.attr.attr,
601 	&port_pma_attr_port_rcv_errors.attr.attr,
602 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
603 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
604 	&port_pma_attr_port_xmit_discards.attr.attr,
605 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
606 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
607 	&port_pma_attr_local_link_integrity_errors.attr.attr,
608 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
609 	&port_pma_attr_VL15_dropped.attr.attr,
610 	&port_pma_attr_port_xmit_data.attr.attr,
611 	&port_pma_attr_port_rcv_data.attr.attr,
612 	&port_pma_attr_port_xmit_packets.attr.attr,
613 	&port_pma_attr_port_rcv_packets.attr.attr,
614 	&port_pma_attr_port_xmit_wait.attr.attr,
615 	NULL
616 };
617 
618 static struct attribute *pma_attrs_ext[] = {
619 	&port_pma_attr_symbol_error.attr.attr,
620 	&port_pma_attr_link_error_recovery.attr.attr,
621 	&port_pma_attr_link_downed.attr.attr,
622 	&port_pma_attr_port_rcv_errors.attr.attr,
623 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
624 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
625 	&port_pma_attr_port_xmit_discards.attr.attr,
626 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
627 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
628 	&port_pma_attr_local_link_integrity_errors.attr.attr,
629 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
630 	&port_pma_attr_VL15_dropped.attr.attr,
631 	&port_pma_attr_ext_port_xmit_data.attr.attr,
632 	&port_pma_attr_ext_port_rcv_data.attr.attr,
633 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
634 	&port_pma_attr_port_xmit_wait.attr.attr,
635 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
636 	&port_pma_attr_ext_unicast_rcv_packets.attr.attr,
637 	&port_pma_attr_ext_unicast_xmit_packets.attr.attr,
638 	&port_pma_attr_ext_multicast_rcv_packets.attr.attr,
639 	&port_pma_attr_ext_multicast_xmit_packets.attr.attr,
640 	NULL
641 };
642 
643 static struct attribute *pma_attrs_noietf[] = {
644 	&port_pma_attr_symbol_error.attr.attr,
645 	&port_pma_attr_link_error_recovery.attr.attr,
646 	&port_pma_attr_link_downed.attr.attr,
647 	&port_pma_attr_port_rcv_errors.attr.attr,
648 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
649 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
650 	&port_pma_attr_port_xmit_discards.attr.attr,
651 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
652 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
653 	&port_pma_attr_local_link_integrity_errors.attr.attr,
654 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
655 	&port_pma_attr_VL15_dropped.attr.attr,
656 	&port_pma_attr_ext_port_xmit_data.attr.attr,
657 	&port_pma_attr_ext_port_rcv_data.attr.attr,
658 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
659 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
660 	&port_pma_attr_port_xmit_wait.attr.attr,
661 	NULL
662 };
663 
664 static struct attribute_group pma_group = {
665 	.name  = "counters",
666 	.attrs  = pma_attrs
667 };
668 
669 static struct attribute_group pma_group_ext = {
670 	.name  = "counters",
671 	.attrs  = pma_attrs_ext
672 };
673 
674 static struct attribute_group pma_group_noietf = {
675 	.name  = "counters",
676 	.attrs  = pma_attrs_noietf
677 };
678 
679 static void ib_port_release(struct kobject *kobj)
680 {
681 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
682 	struct attribute *a;
683 	int i;
684 
685 	if (p->gid_group.attrs) {
686 		for (i = 0; (a = p->gid_group.attrs[i]); ++i)
687 			kfree(a);
688 
689 		kfree(p->gid_group.attrs);
690 	}
691 
692 	if (p->pkey_group.attrs) {
693 		for (i = 0; (a = p->pkey_group.attrs[i]); ++i)
694 			kfree(a);
695 
696 		kfree(p->pkey_group.attrs);
697 	}
698 
699 	kfree(p);
700 }
701 
702 static void ib_port_gid_attr_release(struct kobject *kobj)
703 {
704 	struct gid_attr_group *g = container_of(kobj, struct gid_attr_group,
705 						kobj);
706 	struct attribute *a;
707 	int i;
708 
709 	if (g->ndev.attrs) {
710 		for (i = 0; (a = g->ndev.attrs[i]); ++i)
711 			kfree(a);
712 
713 		kfree(g->ndev.attrs);
714 	}
715 
716 	if (g->type.attrs) {
717 		for (i = 0; (a = g->type.attrs[i]); ++i)
718 			kfree(a);
719 
720 		kfree(g->type.attrs);
721 	}
722 
723 	kfree(g);
724 }
725 
726 static struct kobj_type port_type = {
727 	.release       = ib_port_release,
728 	.sysfs_ops     = &port_sysfs_ops,
729 	.default_attrs = port_default_attrs
730 };
731 
732 static struct kobj_type gid_attr_type = {
733 	.sysfs_ops      = &gid_attr_sysfs_ops,
734 	.release        = ib_port_gid_attr_release
735 };
736 
737 static struct attribute **
738 alloc_group_attrs(ssize_t (*show)(struct ib_port *,
739 				  struct port_attribute *, char *buf),
740 		  int len)
741 {
742 	struct attribute **tab_attr;
743 	struct port_table_attribute *element;
744 	int i;
745 
746 	tab_attr = kcalloc(1 + len, sizeof(struct attribute *), GFP_KERNEL);
747 	if (!tab_attr)
748 		return NULL;
749 
750 	for (i = 0; i < len; i++) {
751 		element = kzalloc(sizeof(struct port_table_attribute),
752 				  GFP_KERNEL);
753 		if (!element)
754 			goto err;
755 
756 		if (snprintf(element->name, sizeof(element->name),
757 			     "%d", i) >= sizeof(element->name)) {
758 			kfree(element);
759 			goto err;
760 		}
761 
762 		element->attr.attr.name  = element->name;
763 		element->attr.attr.mode  = S_IRUGO;
764 		element->attr.show       = show;
765 		element->index		 = i;
766 		sysfs_attr_init(&element->attr.attr);
767 
768 		tab_attr[i] = &element->attr.attr;
769 	}
770 
771 	return tab_attr;
772 
773 err:
774 	while (--i >= 0)
775 		kfree(tab_attr[i]);
776 	kfree(tab_attr);
777 	return NULL;
778 }
779 
780 /*
781  * Figure out which counter table to use depending on
782  * the device capabilities.
783  */
784 static struct attribute_group *get_counter_table(struct ib_device *dev,
785 						 int port_num)
786 {
787 	struct ib_class_port_info cpi;
788 
789 	if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
790 				&cpi, 40, sizeof(cpi)) >= 0) {
791 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
792 			/* We have extended counters */
793 			return &pma_group_ext;
794 
795 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
796 			/* But not the IETF ones */
797 			return &pma_group_noietf;
798 	}
799 
800 	/* Fall back to normal counters */
801 	return &pma_group;
802 }
803 
804 static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
805 			   u8 port_num, int index)
806 {
807 	int ret;
808 
809 	if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
810 		return 0;
811 	ret = dev->get_hw_stats(dev, stats, port_num, index);
812 	if (ret < 0)
813 		return ret;
814 	if (ret == stats->num_counters)
815 		stats->timestamp = jiffies;
816 
817 	return 0;
818 }
819 
820 static ssize_t print_hw_stat(struct rdma_hw_stats *stats, int index, char *buf)
821 {
822 	return sprintf(buf, "%llu\n", (unsigned long long)stats->value[index]);
823 }
824 
825 static ssize_t show_hw_stats(struct kobject *kobj, struct attribute *attr,
826 			     char *buf)
827 {
828 	struct ib_device *dev;
829 	struct ib_port *port;
830 	struct hw_stats_attribute *hsa;
831 	struct rdma_hw_stats *stats;
832 	int ret;
833 
834 	hsa = container_of(attr, struct hw_stats_attribute, attr);
835 	if (!hsa->port_num) {
836 		dev = container_of((struct device *)kobj,
837 				   struct ib_device, dev);
838 		stats = dev->hw_stats;
839 	} else {
840 		port = container_of(kobj, struct ib_port, kobj);
841 		dev = port->ibdev;
842 		stats = port->hw_stats;
843 	}
844 	mutex_lock(&stats->lock);
845 	ret = update_hw_stats(dev, stats, hsa->port_num, hsa->index);
846 	if (ret)
847 		goto unlock;
848 	ret = print_hw_stat(stats, hsa->index, buf);
849 unlock:
850 	mutex_unlock(&stats->lock);
851 
852 	return ret;
853 }
854 
855 static ssize_t show_stats_lifespan(struct kobject *kobj,
856 				   struct attribute *attr,
857 				   char *buf)
858 {
859 	struct hw_stats_attribute *hsa;
860 	struct rdma_hw_stats *stats;
861 	int msecs;
862 
863 	hsa = container_of(attr, struct hw_stats_attribute, attr);
864 	if (!hsa->port_num) {
865 		struct ib_device *dev = container_of((struct device *)kobj,
866 						     struct ib_device, dev);
867 
868 		stats = dev->hw_stats;
869 	} else {
870 		struct ib_port *p = container_of(kobj, struct ib_port, kobj);
871 
872 		stats = p->hw_stats;
873 	}
874 
875 	mutex_lock(&stats->lock);
876 	msecs = jiffies_to_msecs(stats->lifespan);
877 	mutex_unlock(&stats->lock);
878 
879 	return sprintf(buf, "%d\n", msecs);
880 }
881 
882 static ssize_t set_stats_lifespan(struct kobject *kobj,
883 				  struct attribute *attr,
884 				  const char *buf, size_t count)
885 {
886 	struct hw_stats_attribute *hsa;
887 	struct rdma_hw_stats *stats;
888 	int msecs;
889 	unsigned long jiffies;
890 	int ret;
891 
892 	ret = kstrtoint(buf, 10, &msecs);
893 	if (ret)
894 		return ret;
895 	if (msecs < 0 || msecs > 10000)
896 		return -EINVAL;
897 	jiffies = msecs_to_jiffies(msecs);
898 	hsa = container_of(attr, struct hw_stats_attribute, attr);
899 	if (!hsa->port_num) {
900 		struct ib_device *dev = container_of((struct device *)kobj,
901 						     struct ib_device, dev);
902 
903 		stats = dev->hw_stats;
904 	} else {
905 		struct ib_port *p = container_of(kobj, struct ib_port, kobj);
906 
907 		stats = p->hw_stats;
908 	}
909 
910 	mutex_lock(&stats->lock);
911 	stats->lifespan = jiffies;
912 	mutex_unlock(&stats->lock);
913 
914 	return count;
915 }
916 
917 static void free_hsag(struct kobject *kobj, struct attribute_group *attr_group)
918 {
919 	struct attribute **attr;
920 
921 	sysfs_remove_group(kobj, attr_group);
922 
923 	for (attr = attr_group->attrs; *attr; attr++)
924 		kfree(*attr);
925 	kfree(attr_group);
926 }
927 
928 static struct attribute *alloc_hsa(int index, u8 port_num, const char *name)
929 {
930 	struct hw_stats_attribute *hsa;
931 
932 	hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
933 	if (!hsa)
934 		return NULL;
935 
936 	hsa->attr.name = __DECONST(char *, name);
937 	hsa->attr.mode = S_IRUGO;
938 	hsa->show = show_hw_stats;
939 	hsa->store = NULL;
940 	hsa->index = index;
941 	hsa->port_num = port_num;
942 
943 	return &hsa->attr;
944 }
945 
946 static struct attribute *alloc_hsa_lifespan(char *name, u8 port_num)
947 {
948 	struct hw_stats_attribute *hsa;
949 
950 	hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
951 	if (!hsa)
952 		return NULL;
953 
954 	hsa->attr.name = name;
955 	hsa->attr.mode = S_IWUSR | S_IRUGO;
956 	hsa->show = show_stats_lifespan;
957 	hsa->store = set_stats_lifespan;
958 	hsa->index = 0;
959 	hsa->port_num = port_num;
960 
961 	return &hsa->attr;
962 }
963 
964 static void setup_hw_stats(struct ib_device *device, struct ib_port *port,
965 			   u8 port_num)
966 {
967 	struct attribute_group *hsag;
968 	struct rdma_hw_stats *stats;
969 	int i, ret;
970 
971 	stats = device->alloc_hw_stats(device, port_num);
972 
973 	if (!stats)
974 		return;
975 
976 	if (!stats->names || stats->num_counters <= 0)
977 		goto err_free_stats;
978 
979 	/*
980 	 * Two extra attribue elements here, one for the lifespan entry and
981 	 * one to NULL terminate the list for the sysfs core code
982 	 */
983 	hsag = kzalloc(sizeof(*hsag) +
984 		       sizeof(void *) * (stats->num_counters + 2),
985 		       GFP_KERNEL);
986 	if (!hsag)
987 		goto err_free_stats;
988 
989 	ret = device->get_hw_stats(device, stats, port_num,
990 				   stats->num_counters);
991 	if (ret != stats->num_counters)
992 		goto err_free_hsag;
993 
994 	stats->timestamp = jiffies;
995 
996 	hsag->name = "hw_counters";
997 	hsag->attrs = (void *)((char *)hsag + sizeof(*hsag));
998 
999 	for (i = 0; i < stats->num_counters; i++) {
1000 		hsag->attrs[i] = alloc_hsa(i, port_num, stats->names[i]);
1001 		if (!hsag->attrs[i])
1002 			goto err;
1003 		sysfs_attr_init(hsag->attrs[i]);
1004 	}
1005 
1006 	mutex_init(&stats->lock);
1007 	/* treat an error here as non-fatal */
1008 	hsag->attrs[i] = alloc_hsa_lifespan("lifespan", port_num);
1009 	if (hsag->attrs[i])
1010 		sysfs_attr_init(hsag->attrs[i]);
1011 
1012 	if (port) {
1013 		struct kobject *kobj = &port->kobj;
1014 		ret = sysfs_create_group(kobj, hsag);
1015 		if (ret)
1016 			goto err;
1017 		port->hw_stats_ag = hsag;
1018 		port->hw_stats = stats;
1019 	} else {
1020 		struct kobject *kobj = &device->dev.kobj;
1021 		ret = sysfs_create_group(kobj, hsag);
1022 		if (ret)
1023 			goto err;
1024 		device->hw_stats_ag = hsag;
1025 		device->hw_stats = stats;
1026 	}
1027 
1028 	return;
1029 
1030 err:
1031 	for (; i >= 0; i--)
1032 		kfree(hsag->attrs[i]);
1033 err_free_hsag:
1034 	kfree(hsag);
1035 err_free_stats:
1036 	kfree(stats);
1037 	return;
1038 }
1039 
1040 static int add_port(struct ib_device *device, int port_num,
1041 		    int (*port_callback)(struct ib_device *,
1042 					 u8, struct kobject *))
1043 {
1044 	struct ib_port *p;
1045 	struct ib_port_attr attr;
1046 	int i;
1047 	int ret;
1048 
1049 	ret = ib_query_port(device, port_num, &attr);
1050 	if (ret)
1051 		return ret;
1052 
1053 	p = kzalloc(sizeof *p, GFP_KERNEL);
1054 	if (!p)
1055 		return -ENOMEM;
1056 
1057 	p->ibdev      = device;
1058 	p->port_num   = port_num;
1059 
1060 	ret = kobject_init_and_add(&p->kobj, &port_type,
1061 				   device->ports_parent,
1062 				   "%d", port_num);
1063 	if (ret) {
1064 		kfree(p);
1065 		return ret;
1066 	}
1067 
1068 	p->gid_attr_group = kzalloc(sizeof(*p->gid_attr_group), GFP_KERNEL);
1069 	if (!p->gid_attr_group) {
1070 		ret = -ENOMEM;
1071 		goto err_put;
1072 	}
1073 
1074 	p->gid_attr_group->port = p;
1075 	ret = kobject_init_and_add(&p->gid_attr_group->kobj, &gid_attr_type,
1076 				   &p->kobj, "gid_attrs");
1077 	if (ret) {
1078 		kfree(p->gid_attr_group);
1079 		goto err_put;
1080 	}
1081 
1082 	if (device->process_mad) {
1083 		p->pma_table = get_counter_table(device, port_num);
1084 		ret = sysfs_create_group(&p->kobj, p->pma_table);
1085 		if (ret)
1086 			goto err_put_gid_attrs;
1087 	}
1088 
1089 	p->gid_group.name  = "gids";
1090 	p->gid_group.attrs = alloc_group_attrs(show_port_gid, attr.gid_tbl_len);
1091 	if (!p->gid_group.attrs) {
1092 		ret = -ENOMEM;
1093 		goto err_remove_pma;
1094 	}
1095 
1096 	ret = sysfs_create_group(&p->kobj, &p->gid_group);
1097 	if (ret)
1098 		goto err_free_gid;
1099 
1100 	p->gid_attr_group->ndev.name = "ndevs";
1101 	p->gid_attr_group->ndev.attrs = alloc_group_attrs(show_port_gid_attr_ndev,
1102 							  attr.gid_tbl_len);
1103 	if (!p->gid_attr_group->ndev.attrs) {
1104 		ret = -ENOMEM;
1105 		goto err_remove_gid;
1106 	}
1107 
1108 	ret = sysfs_create_group(&p->gid_attr_group->kobj,
1109 				 &p->gid_attr_group->ndev);
1110 	if (ret)
1111 		goto err_free_gid_ndev;
1112 
1113 	p->gid_attr_group->type.name = "types";
1114 	p->gid_attr_group->type.attrs = alloc_group_attrs(show_port_gid_attr_gid_type,
1115 							  attr.gid_tbl_len);
1116 	if (!p->gid_attr_group->type.attrs) {
1117 		ret = -ENOMEM;
1118 		goto err_remove_gid_ndev;
1119 	}
1120 
1121 	ret = sysfs_create_group(&p->gid_attr_group->kobj,
1122 				 &p->gid_attr_group->type);
1123 	if (ret)
1124 		goto err_free_gid_type;
1125 
1126 	p->pkey_group.name  = "pkeys";
1127 	p->pkey_group.attrs = alloc_group_attrs(show_port_pkey,
1128 						attr.pkey_tbl_len);
1129 	if (!p->pkey_group.attrs) {
1130 		ret = -ENOMEM;
1131 		goto err_remove_gid_type;
1132 	}
1133 
1134 	ret = sysfs_create_group(&p->kobj, &p->pkey_group);
1135 	if (ret)
1136 		goto err_free_pkey;
1137 
1138 	if (port_callback) {
1139 		ret = port_callback(device, port_num, &p->kobj);
1140 		if (ret)
1141 			goto err_remove_pkey;
1142 	}
1143 
1144 	/*
1145 	 * If port == 0, it means we have only one port and the parent
1146 	 * device, not this port device, should be the holder of the
1147 	 * hw_counters
1148 	 */
1149 	if (device->alloc_hw_stats && port_num)
1150 		setup_hw_stats(device, p, port_num);
1151 
1152 	list_add_tail(&p->kobj.entry, &device->port_list);
1153 
1154 	return 0;
1155 
1156 err_remove_pkey:
1157 	sysfs_remove_group(&p->kobj, &p->pkey_group);
1158 
1159 err_free_pkey:
1160 	for (i = 0; i < attr.pkey_tbl_len; ++i)
1161 		kfree(p->pkey_group.attrs[i]);
1162 
1163 	kfree(p->pkey_group.attrs);
1164 	p->pkey_group.attrs = NULL;
1165 
1166 err_remove_gid_type:
1167 	sysfs_remove_group(&p->gid_attr_group->kobj,
1168 			   &p->gid_attr_group->type);
1169 
1170 err_free_gid_type:
1171 	for (i = 0; i < attr.gid_tbl_len; ++i)
1172 		kfree(p->gid_attr_group->type.attrs[i]);
1173 
1174 	kfree(p->gid_attr_group->type.attrs);
1175 	p->gid_attr_group->type.attrs = NULL;
1176 
1177 err_remove_gid_ndev:
1178 	sysfs_remove_group(&p->gid_attr_group->kobj,
1179 			   &p->gid_attr_group->ndev);
1180 
1181 err_free_gid_ndev:
1182 	for (i = 0; i < attr.gid_tbl_len; ++i)
1183 		kfree(p->gid_attr_group->ndev.attrs[i]);
1184 
1185 	kfree(p->gid_attr_group->ndev.attrs);
1186 	p->gid_attr_group->ndev.attrs = NULL;
1187 
1188 err_remove_gid:
1189 	sysfs_remove_group(&p->kobj, &p->gid_group);
1190 
1191 err_free_gid:
1192 	for (i = 0; i < attr.gid_tbl_len; ++i)
1193 		kfree(p->gid_group.attrs[i]);
1194 
1195 	kfree(p->gid_group.attrs);
1196 	p->gid_group.attrs = NULL;
1197 
1198 err_remove_pma:
1199 	if (p->pma_table)
1200 		sysfs_remove_group(&p->kobj, p->pma_table);
1201 
1202 err_put_gid_attrs:
1203 	kobject_put(&p->gid_attr_group->kobj);
1204 
1205 err_put:
1206 	kobject_put(&p->kobj);
1207 	return ret;
1208 }
1209 
1210 static ssize_t node_type_show(struct device *device,
1211 			      struct device_attribute *attr, char *buf)
1212 {
1213 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1214 
1215 	switch (dev->node_type) {
1216 	case RDMA_NODE_IB_CA:	  return sprintf(buf, "%d: CA\n", dev->node_type);
1217 	case RDMA_NODE_RNIC:	  return sprintf(buf, "%d: RNIC\n", dev->node_type);
1218 	case RDMA_NODE_USNIC:	  return sprintf(buf, "%d: usNIC\n", dev->node_type);
1219 	case RDMA_NODE_USNIC_UDP: return sprintf(buf, "%d: usNIC UDP\n", dev->node_type);
1220 	case RDMA_NODE_IB_SWITCH: return sprintf(buf, "%d: switch\n", dev->node_type);
1221 	case RDMA_NODE_IB_ROUTER: return sprintf(buf, "%d: router\n", dev->node_type);
1222 	default:		  return sprintf(buf, "%d: <unknown>\n", dev->node_type);
1223 	}
1224 }
1225 static DEVICE_ATTR_RO(node_type);
1226 
1227 static ssize_t sys_image_guid_show(struct device *device,
1228 				   struct device_attribute *dev_attr, char *buf)
1229 {
1230 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1231 
1232 	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1233 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[0]),
1234 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[1]),
1235 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[2]),
1236 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[3]));
1237 }
1238 static DEVICE_ATTR_RO(sys_image_guid);
1239 
1240 static ssize_t node_guid_show(struct device *device,
1241 			      struct device_attribute *attr, char *buf)
1242 {
1243 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1244 
1245 	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1246 		       be16_to_cpu(((__be16 *) &dev->node_guid)[0]),
1247 		       be16_to_cpu(((__be16 *) &dev->node_guid)[1]),
1248 		       be16_to_cpu(((__be16 *) &dev->node_guid)[2]),
1249 		       be16_to_cpu(((__be16 *) &dev->node_guid)[3]));
1250 }
1251 static DEVICE_ATTR_RO(node_guid);
1252 
1253 static ssize_t node_desc_show(struct device *device,
1254 			      struct device_attribute *attr, char *buf)
1255 {
1256 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1257 
1258 	return sprintf(buf, "%.64s\n", dev->node_desc);
1259 }
1260 
1261 static ssize_t node_desc_store(struct device *device,
1262 			       struct device_attribute *attr,
1263 			       const char *buf, size_t count)
1264 {
1265 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1266 	struct ib_device_modify desc = {};
1267 	int ret;
1268 
1269 	if (!dev->modify_device)
1270 		return -EIO;
1271 
1272 	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1273 	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1274 	if (ret)
1275 		return ret;
1276 
1277 	return count;
1278 }
1279 static DEVICE_ATTR_RW(node_desc);
1280 
1281 static ssize_t fw_ver_show(struct device *device, struct device_attribute *attr,
1282 			   char *buf)
1283 {
1284 	struct ib_device *dev = container_of(device, struct ib_device, dev);
1285 
1286 	ib_get_device_fw_str(dev, buf, PAGE_SIZE);
1287 	strlcat(buf, "\n", PAGE_SIZE);
1288 	return strlen(buf);
1289 }
1290 static DEVICE_ATTR_RO(fw_ver);
1291 
1292 static struct attribute *ib_dev_attrs[] = {
1293 	&dev_attr_node_type.attr,
1294 	&dev_attr_node_guid.attr,
1295 	&dev_attr_sys_image_guid.attr,
1296 	&dev_attr_fw_ver.attr,
1297 	&dev_attr_node_desc.attr,
1298 	NULL,
1299 };
1300 
1301 static const struct attribute_group dev_attr_group = {
1302 	.attrs = ib_dev_attrs,
1303 };
1304 
1305 static void free_port_list_attributes(struct ib_device *device)
1306 {
1307 	struct kobject *p, *t;
1308 
1309 	list_for_each_entry_safe(p, t, &device->port_list, entry) {
1310 		struct ib_port *port = container_of(p, struct ib_port, kobj);
1311 		list_del(&p->entry);
1312 		if (port->hw_stats) {
1313 			kfree(port->hw_stats);
1314 			free_hsag(&port->kobj, port->hw_stats_ag);
1315 		}
1316 
1317 		if (port->pma_table)
1318 			sysfs_remove_group(p, port->pma_table);
1319 		sysfs_remove_group(p, &port->pkey_group);
1320 		sysfs_remove_group(p, &port->gid_group);
1321 		sysfs_remove_group(&port->gid_attr_group->kobj,
1322 				   &port->gid_attr_group->ndev);
1323 		sysfs_remove_group(&port->gid_attr_group->kobj,
1324 				   &port->gid_attr_group->type);
1325 		kobject_put(&port->gid_attr_group->kobj);
1326 		kobject_put(p);
1327 	}
1328 
1329 	kobject_put(device->ports_parent);
1330 }
1331 
1332 int ib_device_register_sysfs(struct ib_device *device,
1333 			     int (*port_callback)(struct ib_device *,
1334 						  u8, struct kobject *))
1335 {
1336 	struct device *class_dev = &device->dev;
1337 	int ret;
1338 	int i;
1339 
1340 	device->dev.parent = device->dma_device;
1341 	ret = dev_set_name(class_dev, "%s", device->name);
1342 	if (ret)
1343 		return ret;
1344 
1345 	device->groups[0] = &dev_attr_group;
1346 	class_dev->groups = device->groups;
1347 
1348 	ret = device_add(class_dev);
1349 	if (ret)
1350 		goto err;
1351 
1352 	device->ports_parent = kobject_create_and_add("ports",
1353 						      &class_dev->kobj);
1354 	if (!device->ports_parent) {
1355 		ret = -ENOMEM;
1356 		goto err_put;
1357 	}
1358 
1359 	if (rdma_cap_ib_switch(device)) {
1360 		ret = add_port(device, 0, port_callback);
1361 		if (ret)
1362 			goto err_put;
1363 	} else {
1364 		for (i = 1; i <= device->phys_port_cnt; ++i) {
1365 			ret = add_port(device, i, port_callback);
1366 			if (ret)
1367 				goto err_put;
1368 		}
1369 	}
1370 
1371 	if (device->alloc_hw_stats)
1372 		setup_hw_stats(device, NULL, 0);
1373 
1374 	return 0;
1375 
1376 err_put:
1377 	free_port_list_attributes(device);
1378 	device_del(class_dev);
1379 err:
1380 	return ret;
1381 }
1382 
1383 void ib_device_unregister_sysfs(struct ib_device *device)
1384 {
1385 	/* Hold device until ib_dealloc_device() */
1386 	get_device(&device->dev);
1387 
1388 	free_port_list_attributes(device);
1389 
1390 	if (device->hw_stats) {
1391 		kfree(device->hw_stats);
1392 		free_hsag(&device->dev.kobj, device->hw_stats_ag);
1393 	}
1394 
1395 	device_unregister(&device->dev);
1396 }
1397 
1398 /**
1399  * ib_port_register_module_stat - add module counters under relevant port
1400  *  of IB device.
1401  *
1402  * @device: IB device to add counters
1403  * @port_num: valid port number
1404  * @kobj: pointer to the kobject to initialize
1405  * @ktype: pointer to the ktype for this kobject.
1406  * @name: the name of the kobject
1407  */
1408 int ib_port_register_module_stat(struct ib_device *device, u8 port_num,
1409 				 struct kobject *kobj, struct kobj_type *ktype,
1410 				 const char *name)
1411 {
1412 	struct kobject *p, *t;
1413 	int ret;
1414 
1415 	list_for_each_entry_safe(p, t, &device->port_list, entry) {
1416 		struct ib_port *port = container_of(p, struct ib_port, kobj);
1417 
1418 		if (port->port_num != port_num)
1419 			continue;
1420 
1421 		ret = kobject_init_and_add(kobj, ktype, &port->kobj, "%s",
1422 					   name);
1423 		if (ret) {
1424 			kobject_put(kobj);
1425 			return ret;
1426 		}
1427 	}
1428 
1429 	return 0;
1430 }
1431 EXPORT_SYMBOL(ib_port_register_module_stat);
1432 
1433 /**
1434  * ib_port_unregister_module_stat - release module counters
1435  * @kobj: pointer to the kobject to release
1436  */
1437 void ib_port_unregister_module_stat(struct kobject *kobj)
1438 {
1439 	kobject_put(kobj);
1440 }
1441 EXPORT_SYMBOL(ib_port_unregister_module_stat);
1442