xref: /openbmc/linux/drivers/infiniband/core/sysfs.c (revision 25b892b5)
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 #include <rdma/rdma_counter.h>
47 #include <rdma/ib_sysfs.h>
48 
49 struct port_table_attribute {
50 	struct ib_port_attribute attr;
51 	char			name[8];
52 	int			index;
53 	__be16			attr_id;
54 };
55 
56 struct gid_attr_group {
57 	struct ib_port *port;
58 	struct kobject kobj;
59 	struct attribute_group groups[2];
60 	const struct attribute_group *groups_list[3];
61 	struct port_table_attribute attrs_list[];
62 };
63 
64 struct ib_port {
65 	struct kobject kobj;
66 	struct ib_device *ibdev;
67 	struct gid_attr_group *gid_attr_group;
68 	struct hw_stats_port_data *hw_stats_data;
69 
70 	struct attribute_group groups[3];
71 	const struct attribute_group *groups_list[5];
72 	u32 port_num;
73 	struct port_table_attribute attrs_list[];
74 };
75 
76 struct hw_stats_device_attribute {
77 	struct device_attribute attr;
78 	ssize_t (*show)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
79 			unsigned int index, unsigned int port_num, char *buf);
80 	ssize_t (*store)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
81 			 unsigned int index, unsigned int port_num,
82 			 const char *buf, size_t count);
83 };
84 
85 struct hw_stats_port_attribute {
86 	struct ib_port_attribute attr;
87 	ssize_t (*show)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
88 			unsigned int index, unsigned int port_num, char *buf);
89 	ssize_t (*store)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
90 			 unsigned int index, unsigned int port_num,
91 			 const char *buf, size_t count);
92 };
93 
94 struct hw_stats_device_data {
95 	struct attribute_group group;
96 	struct rdma_hw_stats *stats;
97 	struct hw_stats_device_attribute attrs[];
98 };
99 
100 struct hw_stats_port_data {
101 	struct rdma_hw_stats *stats;
102 	struct hw_stats_port_attribute attrs[];
103 };
104 
105 static ssize_t port_attr_show(struct kobject *kobj,
106 			      struct attribute *attr, char *buf)
107 {
108 	struct ib_port_attribute *port_attr =
109 		container_of(attr, struct ib_port_attribute, attr);
110 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
111 
112 	if (!port_attr->show)
113 		return -EIO;
114 
115 	return port_attr->show(p->ibdev, p->port_num, port_attr, buf);
116 }
117 
118 static ssize_t port_attr_store(struct kobject *kobj,
119 			       struct attribute *attr,
120 			       const char *buf, size_t count)
121 {
122 	struct ib_port_attribute *port_attr =
123 		container_of(attr, struct ib_port_attribute, attr);
124 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
125 
126 	if (!port_attr->store)
127 		return -EIO;
128 	return port_attr->store(p->ibdev, p->port_num, port_attr, buf, count);
129 }
130 
131 struct ib_device *ib_port_sysfs_get_ibdev_kobj(struct kobject *kobj,
132 					       u32 *port_num)
133 {
134 	struct ib_port *port = container_of(kobj, struct ib_port, kobj);
135 
136 	*port_num = port->port_num;
137 	return port->ibdev;
138 }
139 EXPORT_SYMBOL(ib_port_sysfs_get_ibdev_kobj);
140 
141 static const struct sysfs_ops port_sysfs_ops = {
142 	.show	= port_attr_show,
143 	.store	= port_attr_store
144 };
145 
146 static ssize_t hw_stat_device_show(struct device *dev,
147 				   struct device_attribute *attr, char *buf)
148 {
149 	struct hw_stats_device_attribute *stat_attr =
150 		container_of(attr, struct hw_stats_device_attribute, attr);
151 	struct ib_device *ibdev = container_of(dev, struct ib_device, dev);
152 
153 	return stat_attr->show(ibdev, ibdev->hw_stats_data->stats,
154 			       stat_attr - ibdev->hw_stats_data->attrs, 0, buf);
155 }
156 
157 static ssize_t hw_stat_device_store(struct device *dev,
158 				    struct device_attribute *attr,
159 				    const char *buf, size_t count)
160 {
161 	struct hw_stats_device_attribute *stat_attr =
162 		container_of(attr, struct hw_stats_device_attribute, attr);
163 	struct ib_device *ibdev = container_of(dev, struct ib_device, dev);
164 
165 	return stat_attr->store(ibdev, ibdev->hw_stats_data->stats,
166 				stat_attr - ibdev->hw_stats_data->attrs, 0, buf,
167 				count);
168 }
169 
170 static ssize_t hw_stat_port_show(struct ib_device *ibdev, u32 port_num,
171 				 struct ib_port_attribute *attr, char *buf)
172 {
173 	struct hw_stats_port_attribute *stat_attr =
174 		container_of(attr, struct hw_stats_port_attribute, attr);
175 	struct ib_port *port = ibdev->port_data[port_num].sysfs;
176 
177 	return stat_attr->show(ibdev, port->hw_stats_data->stats,
178 			       stat_attr - port->hw_stats_data->attrs,
179 			       port->port_num, buf);
180 }
181 
182 static ssize_t hw_stat_port_store(struct ib_device *ibdev, u32 port_num,
183 				  struct ib_port_attribute *attr,
184 				  const char *buf, size_t count)
185 {
186 	struct hw_stats_port_attribute *stat_attr =
187 		container_of(attr, struct hw_stats_port_attribute, attr);
188 	struct ib_port *port = ibdev->port_data[port_num].sysfs;
189 
190 	return stat_attr->store(ibdev, port->hw_stats_data->stats,
191 				stat_attr - port->hw_stats_data->attrs,
192 				port->port_num, buf, count);
193 }
194 
195 static ssize_t gid_attr_show(struct kobject *kobj,
196 			     struct attribute *attr, char *buf)
197 {
198 	struct ib_port_attribute *port_attr =
199 		container_of(attr, struct ib_port_attribute, attr);
200 	struct ib_port *p = container_of(kobj, struct gid_attr_group,
201 					 kobj)->port;
202 
203 	if (!port_attr->show)
204 		return -EIO;
205 
206 	return port_attr->show(p->ibdev, p->port_num, port_attr, buf);
207 }
208 
209 static const struct sysfs_ops gid_attr_sysfs_ops = {
210 	.show = gid_attr_show
211 };
212 
213 static ssize_t state_show(struct ib_device *ibdev, u32 port_num,
214 			  struct ib_port_attribute *unused, char *buf)
215 {
216 	struct ib_port_attr attr;
217 	ssize_t ret;
218 
219 	static const char *state_name[] = {
220 		[IB_PORT_NOP]		= "NOP",
221 		[IB_PORT_DOWN]		= "DOWN",
222 		[IB_PORT_INIT]		= "INIT",
223 		[IB_PORT_ARMED]		= "ARMED",
224 		[IB_PORT_ACTIVE]	= "ACTIVE",
225 		[IB_PORT_ACTIVE_DEFER]	= "ACTIVE_DEFER"
226 	};
227 
228 	ret = ib_query_port(ibdev, port_num, &attr);
229 	if (ret)
230 		return ret;
231 
232 	return sysfs_emit(buf, "%d: %s\n", attr.state,
233 			  attr.state >= 0 &&
234 					  attr.state < ARRAY_SIZE(state_name) ?
235 				  state_name[attr.state] :
236 				  "UNKNOWN");
237 }
238 
239 static ssize_t lid_show(struct ib_device *ibdev, u32 port_num,
240 			struct ib_port_attribute *unused, char *buf)
241 {
242 	struct ib_port_attr attr;
243 	ssize_t ret;
244 
245 	ret = ib_query_port(ibdev, port_num, &attr);
246 	if (ret)
247 		return ret;
248 
249 	return sysfs_emit(buf, "0x%x\n", attr.lid);
250 }
251 
252 static ssize_t lid_mask_count_show(struct ib_device *ibdev, u32 port_num,
253 				   struct ib_port_attribute *unused, char *buf)
254 {
255 	struct ib_port_attr attr;
256 	ssize_t ret;
257 
258 	ret = ib_query_port(ibdev, port_num, &attr);
259 	if (ret)
260 		return ret;
261 
262 	return sysfs_emit(buf, "%u\n", attr.lmc);
263 }
264 
265 static ssize_t sm_lid_show(struct ib_device *ibdev, u32 port_num,
266 			   struct ib_port_attribute *unused, char *buf)
267 {
268 	struct ib_port_attr attr;
269 	ssize_t ret;
270 
271 	ret = ib_query_port(ibdev, port_num, &attr);
272 	if (ret)
273 		return ret;
274 
275 	return sysfs_emit(buf, "0x%x\n", attr.sm_lid);
276 }
277 
278 static ssize_t sm_sl_show(struct ib_device *ibdev, u32 port_num,
279 			  struct ib_port_attribute *unused, char *buf)
280 {
281 	struct ib_port_attr attr;
282 	ssize_t ret;
283 
284 	ret = ib_query_port(ibdev, port_num, &attr);
285 	if (ret)
286 		return ret;
287 
288 	return sysfs_emit(buf, "%u\n", attr.sm_sl);
289 }
290 
291 static ssize_t cap_mask_show(struct ib_device *ibdev, u32 port_num,
292 			     struct ib_port_attribute *unused, char *buf)
293 {
294 	struct ib_port_attr attr;
295 	ssize_t ret;
296 
297 	ret = ib_query_port(ibdev, port_num, &attr);
298 	if (ret)
299 		return ret;
300 
301 	return sysfs_emit(buf, "0x%08x\n", attr.port_cap_flags);
302 }
303 
304 static ssize_t rate_show(struct ib_device *ibdev, u32 port_num,
305 			 struct ib_port_attribute *unused, char *buf)
306 {
307 	struct ib_port_attr attr;
308 	char *speed = "";
309 	int rate;		/* in deci-Gb/sec */
310 	ssize_t ret;
311 
312 	ret = ib_query_port(ibdev, port_num, &attr);
313 	if (ret)
314 		return ret;
315 
316 	switch (attr.active_speed) {
317 	case IB_SPEED_DDR:
318 		speed = " DDR";
319 		rate = 50;
320 		break;
321 	case IB_SPEED_QDR:
322 		speed = " QDR";
323 		rate = 100;
324 		break;
325 	case IB_SPEED_FDR10:
326 		speed = " FDR10";
327 		rate = 100;
328 		break;
329 	case IB_SPEED_FDR:
330 		speed = " FDR";
331 		rate = 140;
332 		break;
333 	case IB_SPEED_EDR:
334 		speed = " EDR";
335 		rate = 250;
336 		break;
337 	case IB_SPEED_HDR:
338 		speed = " HDR";
339 		rate = 500;
340 		break;
341 	case IB_SPEED_NDR:
342 		speed = " NDR";
343 		rate = 1000;
344 		break;
345 	case IB_SPEED_SDR:
346 	default:		/* default to SDR for invalid rates */
347 		speed = " SDR";
348 		rate = 25;
349 		break;
350 	}
351 
352 	rate *= ib_width_enum_to_int(attr.active_width);
353 	if (rate < 0)
354 		return -EINVAL;
355 
356 	return sysfs_emit(buf, "%d%s Gb/sec (%dX%s)\n", rate / 10,
357 			  rate % 10 ? ".5" : "",
358 			  ib_width_enum_to_int(attr.active_width), speed);
359 }
360 
361 static const char *phys_state_to_str(enum ib_port_phys_state phys_state)
362 {
363 	static const char *phys_state_str[] = {
364 		"<unknown>",
365 		"Sleep",
366 		"Polling",
367 		"Disabled",
368 		"PortConfigurationTraining",
369 		"LinkUp",
370 		"LinkErrorRecovery",
371 		"Phy Test",
372 	};
373 
374 	if (phys_state < ARRAY_SIZE(phys_state_str))
375 		return phys_state_str[phys_state];
376 	return "<unknown>";
377 }
378 
379 static ssize_t phys_state_show(struct ib_device *ibdev, u32 port_num,
380 			       struct ib_port_attribute *unused, char *buf)
381 {
382 	struct ib_port_attr attr;
383 
384 	ssize_t ret;
385 
386 	ret = ib_query_port(ibdev, port_num, &attr);
387 	if (ret)
388 		return ret;
389 
390 	return sysfs_emit(buf, "%u: %s\n", attr.phys_state,
391 			  phys_state_to_str(attr.phys_state));
392 }
393 
394 static ssize_t link_layer_show(struct ib_device *ibdev, u32 port_num,
395 			       struct ib_port_attribute *unused, char *buf)
396 {
397 	const char *output;
398 
399 	switch (rdma_port_get_link_layer(ibdev, port_num)) {
400 	case IB_LINK_LAYER_INFINIBAND:
401 		output = "InfiniBand";
402 		break;
403 	case IB_LINK_LAYER_ETHERNET:
404 		output = "Ethernet";
405 		break;
406 	default:
407 		output = "Unknown";
408 		break;
409 	}
410 
411 	return sysfs_emit(buf, "%s\n", output);
412 }
413 
414 static IB_PORT_ATTR_RO(state);
415 static IB_PORT_ATTR_RO(lid);
416 static IB_PORT_ATTR_RO(lid_mask_count);
417 static IB_PORT_ATTR_RO(sm_lid);
418 static IB_PORT_ATTR_RO(sm_sl);
419 static IB_PORT_ATTR_RO(cap_mask);
420 static IB_PORT_ATTR_RO(rate);
421 static IB_PORT_ATTR_RO(phys_state);
422 static IB_PORT_ATTR_RO(link_layer);
423 
424 static struct attribute *port_default_attrs[] = {
425 	&ib_port_attr_state.attr,
426 	&ib_port_attr_lid.attr,
427 	&ib_port_attr_lid_mask_count.attr,
428 	&ib_port_attr_sm_lid.attr,
429 	&ib_port_attr_sm_sl.attr,
430 	&ib_port_attr_cap_mask.attr,
431 	&ib_port_attr_rate.attr,
432 	&ib_port_attr_phys_state.attr,
433 	&ib_port_attr_link_layer.attr,
434 	NULL
435 };
436 
437 static ssize_t print_ndev(const struct ib_gid_attr *gid_attr, char *buf)
438 {
439 	struct net_device *ndev;
440 	int ret = -EINVAL;
441 
442 	rcu_read_lock();
443 	ndev = rcu_dereference(gid_attr->ndev);
444 	if (ndev)
445 		ret = sysfs_emit(buf, "%s\n", ndev->name);
446 	rcu_read_unlock();
447 	return ret;
448 }
449 
450 static ssize_t print_gid_type(const struct ib_gid_attr *gid_attr, char *buf)
451 {
452 	return sysfs_emit(buf, "%s\n",
453 			  ib_cache_gid_type_str(gid_attr->gid_type));
454 }
455 
456 static ssize_t _show_port_gid_attr(
457 	struct ib_device *ibdev, u32 port_num, struct ib_port_attribute *attr,
458 	char *buf,
459 	ssize_t (*print)(const struct ib_gid_attr *gid_attr, char *buf))
460 {
461 	struct port_table_attribute *tab_attr =
462 		container_of(attr, struct port_table_attribute, attr);
463 	const struct ib_gid_attr *gid_attr;
464 	ssize_t ret;
465 
466 	gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
467 	if (IS_ERR(gid_attr))
468 		/* -EINVAL is returned for user space compatibility reasons. */
469 		return -EINVAL;
470 
471 	ret = print(gid_attr, buf);
472 	rdma_put_gid_attr(gid_attr);
473 	return ret;
474 }
475 
476 static ssize_t show_port_gid(struct ib_device *ibdev, u32 port_num,
477 			     struct ib_port_attribute *attr, char *buf)
478 {
479 	struct port_table_attribute *tab_attr =
480 		container_of(attr, struct port_table_attribute, attr);
481 	const struct ib_gid_attr *gid_attr;
482 	int len;
483 
484 	gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
485 	if (IS_ERR(gid_attr)) {
486 		const union ib_gid zgid = {};
487 
488 		/* If reading GID fails, it is likely due to GID entry being
489 		 * empty (invalid) or reserved GID in the table.  User space
490 		 * expects to read GID table entries as long as it given index
491 		 * is within GID table size.  Administrative/debugging tool
492 		 * fails to query rest of the GID entries if it hits error
493 		 * while querying a GID of the given index.  To avoid user
494 		 * space throwing such error on fail to read gid, return zero
495 		 * GID as before. This maintains backward compatibility.
496 		 */
497 		return sysfs_emit(buf, "%pI6\n", zgid.raw);
498 	}
499 
500 	len = sysfs_emit(buf, "%pI6\n", gid_attr->gid.raw);
501 	rdma_put_gid_attr(gid_attr);
502 	return len;
503 }
504 
505 static ssize_t show_port_gid_attr_ndev(struct ib_device *ibdev, u32 port_num,
506 				       struct ib_port_attribute *attr,
507 				       char *buf)
508 {
509 	return _show_port_gid_attr(ibdev, port_num, attr, buf, print_ndev);
510 }
511 
512 static ssize_t show_port_gid_attr_gid_type(struct ib_device *ibdev,
513 					   u32 port_num,
514 					   struct ib_port_attribute *attr,
515 					   char *buf)
516 {
517 	return _show_port_gid_attr(ibdev, port_num, attr, buf, print_gid_type);
518 }
519 
520 static ssize_t show_port_pkey(struct ib_device *ibdev, u32 port_num,
521 			      struct ib_port_attribute *attr, char *buf)
522 {
523 	struct port_table_attribute *tab_attr =
524 		container_of(attr, struct port_table_attribute, attr);
525 	u16 pkey;
526 	int ret;
527 
528 	ret = ib_query_pkey(ibdev, port_num, tab_attr->index, &pkey);
529 	if (ret)
530 		return ret;
531 
532 	return sysfs_emit(buf, "0x%04x\n", pkey);
533 }
534 
535 #define PORT_PMA_ATTR(_name, _counter, _width, _offset)			\
536 struct port_table_attribute port_pma_attr_##_name = {			\
537 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
538 	.index = (_offset) | ((_width) << 16) | ((_counter) << 24),	\
539 	.attr_id = IB_PMA_PORT_COUNTERS,				\
540 }
541 
542 #define PORT_PMA_ATTR_EXT(_name, _width, _offset)			\
543 struct port_table_attribute port_pma_attr_ext_##_name = {		\
544 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
545 	.index = (_offset) | ((_width) << 16),				\
546 	.attr_id = IB_PMA_PORT_COUNTERS_EXT,				\
547 }
548 
549 /*
550  * Get a Perfmgmt MAD block of data.
551  * Returns error code or the number of bytes retrieved.
552  */
553 static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
554 		void *data, int offset, size_t size)
555 {
556 	struct ib_mad *in_mad;
557 	struct ib_mad *out_mad;
558 	size_t mad_size = sizeof(*out_mad);
559 	u16 out_mad_pkey_index = 0;
560 	ssize_t ret;
561 
562 	if (!dev->ops.process_mad)
563 		return -ENOSYS;
564 
565 	in_mad = kzalloc(sizeof(*in_mad), GFP_KERNEL);
566 	out_mad = kzalloc(sizeof(*out_mad), GFP_KERNEL);
567 	if (!in_mad || !out_mad) {
568 		ret = -ENOMEM;
569 		goto out;
570 	}
571 
572 	in_mad->mad_hdr.base_version  = 1;
573 	in_mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_PERF_MGMT;
574 	in_mad->mad_hdr.class_version = 1;
575 	in_mad->mad_hdr.method        = IB_MGMT_METHOD_GET;
576 	in_mad->mad_hdr.attr_id       = attr;
577 
578 	if (attr != IB_PMA_CLASS_PORT_INFO)
579 		in_mad->data[41] = port_num;	/* PortSelect field */
580 
581 	if ((dev->ops.process_mad(dev, IB_MAD_IGNORE_MKEY, port_num, NULL, NULL,
582 				  in_mad, out_mad, &mad_size,
583 				  &out_mad_pkey_index) &
584 	     (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
585 	    (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
586 		ret = -EINVAL;
587 		goto out;
588 	}
589 	memcpy(data, out_mad->data + offset, size);
590 	ret = size;
591 out:
592 	kfree(in_mad);
593 	kfree(out_mad);
594 	return ret;
595 }
596 
597 static ssize_t show_pma_counter(struct ib_device *ibdev, u32 port_num,
598 				struct ib_port_attribute *attr, char *buf)
599 {
600 	struct port_table_attribute *tab_attr =
601 		container_of(attr, struct port_table_attribute, attr);
602 	int offset = tab_attr->index & 0xffff;
603 	int width  = (tab_attr->index >> 16) & 0xff;
604 	int ret;
605 	u8 data[8];
606 	int len;
607 
608 	ret = get_perf_mad(ibdev, port_num, tab_attr->attr_id, &data,
609 			40 + offset / 8, sizeof(data));
610 	if (ret < 0)
611 		return ret;
612 
613 	switch (width) {
614 	case 4:
615 		len = sysfs_emit(buf, "%d\n",
616 				 (*data >> (4 - (offset % 8))) & 0xf);
617 		break;
618 	case 8:
619 		len = sysfs_emit(buf, "%u\n", *data);
620 		break;
621 	case 16:
622 		len = sysfs_emit(buf, "%u\n", be16_to_cpup((__be16 *)data));
623 		break;
624 	case 32:
625 		len = sysfs_emit(buf, "%u\n", be32_to_cpup((__be32 *)data));
626 		break;
627 	case 64:
628 		len = sysfs_emit(buf, "%llu\n", be64_to_cpup((__be64 *)data));
629 		break;
630 	default:
631 		len = 0;
632 		break;
633 	}
634 
635 	return len;
636 }
637 
638 static PORT_PMA_ATTR(symbol_error		    ,  0, 16,  32);
639 static PORT_PMA_ATTR(link_error_recovery	    ,  1,  8,  48);
640 static PORT_PMA_ATTR(link_downed		    ,  2,  8,  56);
641 static PORT_PMA_ATTR(port_rcv_errors		    ,  3, 16,  64);
642 static PORT_PMA_ATTR(port_rcv_remote_physical_errors,  4, 16,  80);
643 static PORT_PMA_ATTR(port_rcv_switch_relay_errors   ,  5, 16,  96);
644 static PORT_PMA_ATTR(port_xmit_discards		    ,  6, 16, 112);
645 static PORT_PMA_ATTR(port_xmit_constraint_errors    ,  7,  8, 128);
646 static PORT_PMA_ATTR(port_rcv_constraint_errors	    ,  8,  8, 136);
647 static PORT_PMA_ATTR(local_link_integrity_errors    ,  9,  4, 152);
648 static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10,  4, 156);
649 static PORT_PMA_ATTR(VL15_dropped		    , 11, 16, 176);
650 static PORT_PMA_ATTR(port_xmit_data		    , 12, 32, 192);
651 static PORT_PMA_ATTR(port_rcv_data		    , 13, 32, 224);
652 static PORT_PMA_ATTR(port_xmit_packets		    , 14, 32, 256);
653 static PORT_PMA_ATTR(port_rcv_packets		    , 15, 32, 288);
654 static PORT_PMA_ATTR(port_xmit_wait		    ,  0, 32, 320);
655 
656 /*
657  * Counters added by extended set
658  */
659 static PORT_PMA_ATTR_EXT(port_xmit_data		    , 64,  64);
660 static PORT_PMA_ATTR_EXT(port_rcv_data		    , 64, 128);
661 static PORT_PMA_ATTR_EXT(port_xmit_packets	    , 64, 192);
662 static PORT_PMA_ATTR_EXT(port_rcv_packets	    , 64, 256);
663 static PORT_PMA_ATTR_EXT(unicast_xmit_packets	    , 64, 320);
664 static PORT_PMA_ATTR_EXT(unicast_rcv_packets	    , 64, 384);
665 static PORT_PMA_ATTR_EXT(multicast_xmit_packets	    , 64, 448);
666 static PORT_PMA_ATTR_EXT(multicast_rcv_packets	    , 64, 512);
667 
668 static struct attribute *pma_attrs[] = {
669 	&port_pma_attr_symbol_error.attr.attr,
670 	&port_pma_attr_link_error_recovery.attr.attr,
671 	&port_pma_attr_link_downed.attr.attr,
672 	&port_pma_attr_port_rcv_errors.attr.attr,
673 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
674 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
675 	&port_pma_attr_port_xmit_discards.attr.attr,
676 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
677 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
678 	&port_pma_attr_local_link_integrity_errors.attr.attr,
679 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
680 	&port_pma_attr_VL15_dropped.attr.attr,
681 	&port_pma_attr_port_xmit_data.attr.attr,
682 	&port_pma_attr_port_rcv_data.attr.attr,
683 	&port_pma_attr_port_xmit_packets.attr.attr,
684 	&port_pma_attr_port_rcv_packets.attr.attr,
685 	&port_pma_attr_port_xmit_wait.attr.attr,
686 	NULL
687 };
688 
689 static struct attribute *pma_attrs_ext[] = {
690 	&port_pma_attr_symbol_error.attr.attr,
691 	&port_pma_attr_link_error_recovery.attr.attr,
692 	&port_pma_attr_link_downed.attr.attr,
693 	&port_pma_attr_port_rcv_errors.attr.attr,
694 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
695 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
696 	&port_pma_attr_port_xmit_discards.attr.attr,
697 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
698 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
699 	&port_pma_attr_local_link_integrity_errors.attr.attr,
700 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
701 	&port_pma_attr_VL15_dropped.attr.attr,
702 	&port_pma_attr_ext_port_xmit_data.attr.attr,
703 	&port_pma_attr_ext_port_rcv_data.attr.attr,
704 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
705 	&port_pma_attr_port_xmit_wait.attr.attr,
706 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
707 	&port_pma_attr_ext_unicast_rcv_packets.attr.attr,
708 	&port_pma_attr_ext_unicast_xmit_packets.attr.attr,
709 	&port_pma_attr_ext_multicast_rcv_packets.attr.attr,
710 	&port_pma_attr_ext_multicast_xmit_packets.attr.attr,
711 	NULL
712 };
713 
714 static struct attribute *pma_attrs_noietf[] = {
715 	&port_pma_attr_symbol_error.attr.attr,
716 	&port_pma_attr_link_error_recovery.attr.attr,
717 	&port_pma_attr_link_downed.attr.attr,
718 	&port_pma_attr_port_rcv_errors.attr.attr,
719 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
720 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
721 	&port_pma_attr_port_xmit_discards.attr.attr,
722 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
723 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
724 	&port_pma_attr_local_link_integrity_errors.attr.attr,
725 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
726 	&port_pma_attr_VL15_dropped.attr.attr,
727 	&port_pma_attr_ext_port_xmit_data.attr.attr,
728 	&port_pma_attr_ext_port_rcv_data.attr.attr,
729 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
730 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
731 	&port_pma_attr_port_xmit_wait.attr.attr,
732 	NULL
733 };
734 
735 static const struct attribute_group pma_group = {
736 	.name  = "counters",
737 	.attrs  = pma_attrs
738 };
739 
740 static const struct attribute_group pma_group_ext = {
741 	.name  = "counters",
742 	.attrs  = pma_attrs_ext
743 };
744 
745 static const struct attribute_group pma_group_noietf = {
746 	.name  = "counters",
747 	.attrs  = pma_attrs_noietf
748 };
749 
750 static void ib_port_release(struct kobject *kobj)
751 {
752 	struct ib_port *port = container_of(kobj, struct ib_port, kobj);
753 	int i;
754 
755 	for (i = 0; i != ARRAY_SIZE(port->groups); i++)
756 		kfree(port->groups[i].attrs);
757 	if (port->hw_stats_data)
758 		kfree(port->hw_stats_data->stats);
759 	kfree(port->hw_stats_data);
760 	kfree(port);
761 }
762 
763 static void ib_port_gid_attr_release(struct kobject *kobj)
764 {
765 	struct gid_attr_group *gid_attr_group =
766 		container_of(kobj, struct gid_attr_group, kobj);
767 	int i;
768 
769 	for (i = 0; i != ARRAY_SIZE(gid_attr_group->groups); i++)
770 		kfree(gid_attr_group->groups[i].attrs);
771 	kfree(gid_attr_group);
772 }
773 
774 static struct kobj_type port_type = {
775 	.release       = ib_port_release,
776 	.sysfs_ops     = &port_sysfs_ops,
777 	.default_attrs = port_default_attrs
778 };
779 
780 static struct kobj_type gid_attr_type = {
781 	.sysfs_ops      = &gid_attr_sysfs_ops,
782 	.release        = ib_port_gid_attr_release
783 };
784 
785 /*
786  * Figure out which counter table to use depending on
787  * the device capabilities.
788  */
789 static const struct attribute_group *get_counter_table(struct ib_device *dev,
790 						       int port_num)
791 {
792 	struct ib_class_port_info cpi;
793 
794 	if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
795 				&cpi, 40, sizeof(cpi)) >= 0) {
796 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
797 			/* We have extended counters */
798 			return &pma_group_ext;
799 
800 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
801 			/* But not the IETF ones */
802 			return &pma_group_noietf;
803 	}
804 
805 	/* Fall back to normal counters */
806 	return &pma_group;
807 }
808 
809 static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
810 			   u32 port_num, int index)
811 {
812 	int ret;
813 
814 	if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
815 		return 0;
816 	ret = dev->ops.get_hw_stats(dev, stats, port_num, index);
817 	if (ret < 0)
818 		return ret;
819 	if (ret == stats->num_counters)
820 		stats->timestamp = jiffies;
821 
822 	return 0;
823 }
824 
825 static int print_hw_stat(struct ib_device *dev, int port_num,
826 			 struct rdma_hw_stats *stats, int index, char *buf)
827 {
828 	u64 v = rdma_counter_get_hwstat_value(dev, port_num, index);
829 
830 	return sysfs_emit(buf, "%llu\n", stats->value[index] + v);
831 }
832 
833 static ssize_t show_hw_stats(struct ib_device *ibdev,
834 			     struct rdma_hw_stats *stats, unsigned int index,
835 			     unsigned int port_num, char *buf)
836 {
837 	int ret;
838 
839 	mutex_lock(&stats->lock);
840 	ret = update_hw_stats(ibdev, stats, port_num, index);
841 	if (ret)
842 		goto unlock;
843 	ret = print_hw_stat(ibdev, port_num, stats, index, buf);
844 unlock:
845 	mutex_unlock(&stats->lock);
846 
847 	return ret;
848 }
849 
850 static ssize_t show_stats_lifespan(struct ib_device *ibdev,
851 				   struct rdma_hw_stats *stats,
852 				   unsigned int index, unsigned int port_num,
853 				   char *buf)
854 {
855 	int msecs;
856 
857 	mutex_lock(&stats->lock);
858 	msecs = jiffies_to_msecs(stats->lifespan);
859 	mutex_unlock(&stats->lock);
860 
861 	return sysfs_emit(buf, "%d\n", msecs);
862 }
863 
864 static ssize_t set_stats_lifespan(struct ib_device *ibdev,
865 				   struct rdma_hw_stats *stats,
866 				   unsigned int index, unsigned int port_num,
867 				   const char *buf, size_t count)
868 {
869 	int msecs;
870 	int jiffies;
871 	int ret;
872 
873 	ret = kstrtoint(buf, 10, &msecs);
874 	if (ret)
875 		return ret;
876 	if (msecs < 0 || msecs > 10000)
877 		return -EINVAL;
878 	jiffies = msecs_to_jiffies(msecs);
879 
880 	mutex_lock(&stats->lock);
881 	stats->lifespan = jiffies;
882 	mutex_unlock(&stats->lock);
883 
884 	return count;
885 }
886 
887 static struct hw_stats_device_data *
888 alloc_hw_stats_device(struct ib_device *ibdev)
889 {
890 	struct hw_stats_device_data *data;
891 	struct rdma_hw_stats *stats;
892 
893 	if (!ibdev->ops.alloc_hw_device_stats)
894 		return ERR_PTR(-EOPNOTSUPP);
895 	stats = ibdev->ops.alloc_hw_device_stats(ibdev);
896 	if (!stats)
897 		return ERR_PTR(-ENOMEM);
898 	if (!stats->names || stats->num_counters <= 0)
899 		goto err_free_stats;
900 
901 	/*
902 	 * Two extra attribue elements here, one for the lifespan entry and
903 	 * one to NULL terminate the list for the sysfs core code
904 	 */
905 	data = kzalloc(struct_size(data, attrs, stats->num_counters + 1),
906 		       GFP_KERNEL);
907 	if (!data)
908 		goto err_free_stats;
909 	data->group.attrs = kcalloc(stats->num_counters + 2,
910 				    sizeof(*data->group.attrs), GFP_KERNEL);
911 	if (!data->group.attrs)
912 		goto err_free_data;
913 
914 	mutex_init(&stats->lock);
915 	data->group.name = "hw_counters";
916 	data->stats = stats;
917 	return data;
918 
919 err_free_data:
920 	kfree(data);
921 err_free_stats:
922 	kfree(stats);
923 	return ERR_PTR(-ENOMEM);
924 }
925 
926 void ib_device_release_hw_stats(struct hw_stats_device_data *data)
927 {
928 	kfree(data->group.attrs);
929 	kfree(data->stats);
930 	kfree(data);
931 }
932 
933 int ib_setup_device_attrs(struct ib_device *ibdev)
934 {
935 	struct hw_stats_device_attribute *attr;
936 	struct hw_stats_device_data *data;
937 	int i, ret;
938 
939 	data = alloc_hw_stats_device(ibdev);
940 	if (IS_ERR(data)) {
941 		if (PTR_ERR(data) == -EOPNOTSUPP)
942 			return 0;
943 		return PTR_ERR(data);
944 	}
945 	ibdev->hw_stats_data = data;
946 
947 	ret = ibdev->ops.get_hw_stats(ibdev, data->stats, 0,
948 				      data->stats->num_counters);
949 	if (ret != data->stats->num_counters) {
950 		if (WARN_ON(ret >= 0))
951 			return -EINVAL;
952 		return ret;
953 	}
954 
955 	data->stats->timestamp = jiffies;
956 
957 	for (i = 0; i < data->stats->num_counters; i++) {
958 		attr = &data->attrs[i];
959 		sysfs_attr_init(&attr->attr.attr);
960 		attr->attr.attr.name = data->stats->names[i];
961 		attr->attr.attr.mode = 0444;
962 		attr->attr.show = hw_stat_device_show;
963 		attr->show = show_hw_stats;
964 		data->group.attrs[i] = &attr->attr.attr;
965 	}
966 
967 	attr = &data->attrs[i];
968 	sysfs_attr_init(&attr->attr.attr);
969 	attr->attr.attr.name = "lifespan";
970 	attr->attr.attr.mode = 0644;
971 	attr->attr.show = hw_stat_device_show;
972 	attr->show = show_stats_lifespan;
973 	attr->attr.store = hw_stat_device_store;
974 	attr->store = set_stats_lifespan;
975 	data->group.attrs[i] = &attr->attr.attr;
976 	for (i = 0; i != ARRAY_SIZE(ibdev->groups); i++)
977 		if (!ibdev->groups[i]) {
978 			ibdev->groups[i] = &data->group;
979 			return 0;
980 		}
981 	WARN(true, "struct ib_device->groups is too small");
982 	return -EINVAL;
983 }
984 
985 static struct hw_stats_port_data *
986 alloc_hw_stats_port(struct ib_port *port, struct attribute_group *group)
987 {
988 	struct ib_device *ibdev = port->ibdev;
989 	struct hw_stats_port_data *data;
990 	struct rdma_hw_stats *stats;
991 
992 	if (!ibdev->ops.alloc_hw_port_stats)
993 		return ERR_PTR(-EOPNOTSUPP);
994 	stats = ibdev->ops.alloc_hw_port_stats(port->ibdev, port->port_num);
995 	if (!stats)
996 		return ERR_PTR(-ENOMEM);
997 	if (!stats->names || stats->num_counters <= 0)
998 		goto err_free_stats;
999 
1000 	/*
1001 	 * Two extra attribue elements here, one for the lifespan entry and
1002 	 * one to NULL terminate the list for the sysfs core code
1003 	 */
1004 	data = kzalloc(struct_size(data, attrs, stats->num_counters + 1),
1005 		       GFP_KERNEL);
1006 	if (!data)
1007 		goto err_free_stats;
1008 	group->attrs = kcalloc(stats->num_counters + 2,
1009 				    sizeof(*group->attrs), GFP_KERNEL);
1010 	if (!group->attrs)
1011 		goto err_free_data;
1012 
1013 	mutex_init(&stats->lock);
1014 	group->name = "hw_counters";
1015 	data->stats = stats;
1016 	return data;
1017 
1018 err_free_data:
1019 	kfree(data);
1020 err_free_stats:
1021 	kfree(stats);
1022 	return ERR_PTR(-ENOMEM);
1023 }
1024 
1025 static int setup_hw_port_stats(struct ib_port *port,
1026 			       struct attribute_group *group)
1027 {
1028 	struct hw_stats_port_attribute *attr;
1029 	struct hw_stats_port_data *data;
1030 	int i, ret;
1031 
1032 	data = alloc_hw_stats_port(port, group);
1033 	if (IS_ERR(data))
1034 		return PTR_ERR(data);
1035 
1036 	ret = port->ibdev->ops.get_hw_stats(port->ibdev, data->stats,
1037 					    port->port_num,
1038 					    data->stats->num_counters);
1039 	if (ret != data->stats->num_counters) {
1040 		if (WARN_ON(ret >= 0))
1041 			return -EINVAL;
1042 		return ret;
1043 	}
1044 
1045 	data->stats->timestamp = jiffies;
1046 
1047 	for (i = 0; i < data->stats->num_counters; i++) {
1048 		attr = &data->attrs[i];
1049 		sysfs_attr_init(&attr->attr.attr);
1050 		attr->attr.attr.name = data->stats->names[i];
1051 		attr->attr.attr.mode = 0444;
1052 		attr->attr.show = hw_stat_port_show;
1053 		attr->show = show_hw_stats;
1054 		group->attrs[i] = &attr->attr.attr;
1055 	}
1056 
1057 	attr = &data->attrs[i];
1058 	sysfs_attr_init(&attr->attr.attr);
1059 	attr->attr.attr.name = "lifespan";
1060 	attr->attr.attr.mode = 0644;
1061 	attr->attr.show = hw_stat_port_show;
1062 	attr->show = show_stats_lifespan;
1063 	attr->attr.store = hw_stat_port_store;
1064 	attr->store = set_stats_lifespan;
1065 	group->attrs[i] = &attr->attr.attr;
1066 
1067 	port->hw_stats_data = data;
1068 	return 0;
1069 }
1070 
1071 struct rdma_hw_stats *ib_get_hw_stats_port(struct ib_device *ibdev,
1072 					   u32 port_num)
1073 {
1074 	if (!ibdev->port_data || !rdma_is_port_valid(ibdev, port_num) ||
1075 	    !ibdev->port_data[port_num].sysfs->hw_stats_data)
1076 		return NULL;
1077 	return ibdev->port_data[port_num].sysfs->hw_stats_data->stats;
1078 }
1079 
1080 static int
1081 alloc_port_table_group(const char *name, struct attribute_group *group,
1082 		       struct port_table_attribute *attrs, size_t num,
1083 		       ssize_t (*show)(struct ib_device *ibdev, u32 port_num,
1084 				       struct ib_port_attribute *, char *buf))
1085 {
1086 	struct attribute **attr_list;
1087 	int i;
1088 
1089 	attr_list = kcalloc(num + 1, sizeof(*attr_list), GFP_KERNEL);
1090 	if (!attr_list)
1091 		return -ENOMEM;
1092 
1093 	for (i = 0; i < num; i++) {
1094 		struct port_table_attribute *element = &attrs[i];
1095 
1096 		if (snprintf(element->name, sizeof(element->name), "%d", i) >=
1097 		    sizeof(element->name))
1098 			goto err;
1099 
1100 		sysfs_attr_init(&element->attr.attr);
1101 		element->attr.attr.name = element->name;
1102 		element->attr.attr.mode = 0444;
1103 		element->attr.show = show;
1104 		element->index = i;
1105 
1106 		attr_list[i] = &element->attr.attr;
1107 	}
1108 	group->name = name;
1109 	group->attrs = attr_list;
1110 	return 0;
1111 err:
1112 	kfree(attr_list);
1113 	return -EINVAL;
1114 }
1115 
1116 /*
1117  * Create the sysfs:
1118  *  ibp0s9/ports/XX/gid_attrs/{ndevs,types}/YYY
1119  * YYY is the gid table index in decimal
1120  */
1121 static int setup_gid_attrs(struct ib_port *port,
1122 			   const struct ib_port_attr *attr)
1123 {
1124 	struct gid_attr_group *gid_attr_group;
1125 	int ret;
1126 
1127 	gid_attr_group = kzalloc(struct_size(gid_attr_group, attrs_list,
1128 					     attr->gid_tbl_len * 2),
1129 				 GFP_KERNEL);
1130 	if (!gid_attr_group)
1131 		return -ENOMEM;
1132 	gid_attr_group->port = port;
1133 	kobject_init(&gid_attr_group->kobj, &gid_attr_type);
1134 
1135 	ret = alloc_port_table_group("ndevs", &gid_attr_group->groups[0],
1136 				     gid_attr_group->attrs_list,
1137 				     attr->gid_tbl_len,
1138 				     show_port_gid_attr_ndev);
1139 	if (ret)
1140 		goto err_put;
1141 	gid_attr_group->groups_list[0] = &gid_attr_group->groups[0];
1142 
1143 	ret = alloc_port_table_group(
1144 		"types", &gid_attr_group->groups[1],
1145 		gid_attr_group->attrs_list + attr->gid_tbl_len,
1146 		attr->gid_tbl_len, show_port_gid_attr_gid_type);
1147 	if (ret)
1148 		goto err_put;
1149 	gid_attr_group->groups_list[1] = &gid_attr_group->groups[1];
1150 
1151 	ret = kobject_add(&gid_attr_group->kobj, &port->kobj, "gid_attrs");
1152 	if (ret)
1153 		goto err_put;
1154 	ret = sysfs_create_groups(&gid_attr_group->kobj,
1155 				  gid_attr_group->groups_list);
1156 	if (ret)
1157 		goto err_del;
1158 	port->gid_attr_group = gid_attr_group;
1159 	return 0;
1160 
1161 err_del:
1162 	kobject_del(&gid_attr_group->kobj);
1163 err_put:
1164 	kobject_put(&gid_attr_group->kobj);
1165 	return ret;
1166 }
1167 
1168 static void destroy_gid_attrs(struct ib_port *port)
1169 {
1170 	struct gid_attr_group *gid_attr_group = port->gid_attr_group;
1171 
1172 	if (!gid_attr_group)
1173 		return;
1174 	sysfs_remove_groups(&gid_attr_group->kobj, gid_attr_group->groups_list);
1175 	kobject_del(&gid_attr_group->kobj);
1176 	kobject_put(&gid_attr_group->kobj);
1177 }
1178 
1179 /*
1180  * Create the sysfs:
1181  *  ibp0s9/ports/XX/{gids,pkeys,counters}/YYY
1182  */
1183 static struct ib_port *setup_port(struct ib_core_device *coredev, int port_num,
1184 				  const struct ib_port_attr *attr)
1185 {
1186 	struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1187 	bool is_full_dev = &device->coredev == coredev;
1188 	const struct attribute_group **cur_group;
1189 	struct ib_port *p;
1190 	int ret;
1191 
1192 	p = kzalloc(struct_size(p, attrs_list,
1193 				attr->gid_tbl_len + attr->pkey_tbl_len),
1194 		    GFP_KERNEL);
1195 	if (!p)
1196 		return ERR_PTR(-ENOMEM);
1197 	p->ibdev = device;
1198 	p->port_num = port_num;
1199 	kobject_init(&p->kobj, &port_type);
1200 
1201 	cur_group = p->groups_list;
1202 	ret = alloc_port_table_group("gids", &p->groups[0], p->attrs_list,
1203 				     attr->gid_tbl_len, show_port_gid);
1204 	if (ret)
1205 		goto err_put;
1206 	*cur_group++ = &p->groups[0];
1207 
1208 	if (attr->pkey_tbl_len) {
1209 		ret = alloc_port_table_group("pkeys", &p->groups[1],
1210 					     p->attrs_list + attr->gid_tbl_len,
1211 					     attr->pkey_tbl_len, show_port_pkey);
1212 		if (ret)
1213 			goto err_put;
1214 		*cur_group++ = &p->groups[1];
1215 	}
1216 
1217 	/*
1218 	 * If port == 0, it means hw_counters are per device and not per
1219 	 * port, so holder should be device. Therefore skip per port
1220 	 * counter initialization.
1221 	 */
1222 	if (port_num && is_full_dev) {
1223 		ret = setup_hw_port_stats(p, &p->groups[2]);
1224 		if (ret && ret != -EOPNOTSUPP)
1225 			goto err_put;
1226 		if (!ret)
1227 			*cur_group++ = &p->groups[2];
1228 	}
1229 
1230 	if (device->ops.process_mad && is_full_dev)
1231 		*cur_group++ = get_counter_table(device, port_num);
1232 
1233 	ret = kobject_add(&p->kobj, coredev->ports_kobj, "%d", port_num);
1234 	if (ret)
1235 		goto err_put;
1236 	ret = sysfs_create_groups(&p->kobj, p->groups_list);
1237 	if (ret)
1238 		goto err_del;
1239 	if (is_full_dev) {
1240 		ret = sysfs_create_groups(&p->kobj, device->ops.port_groups);
1241 		if (ret)
1242 			goto err_groups;
1243 	}
1244 
1245 	list_add_tail(&p->kobj.entry, &coredev->port_list);
1246 	if (device->port_data && is_full_dev)
1247 		device->port_data[port_num].sysfs = p;
1248 
1249 	return p;
1250 
1251 err_groups:
1252 	sysfs_remove_groups(&p->kobj, p->groups_list);
1253 err_del:
1254 	kobject_del(&p->kobj);
1255 err_put:
1256 	kobject_put(&p->kobj);
1257 	return ERR_PTR(ret);
1258 }
1259 
1260 static void destroy_port(struct ib_core_device *coredev, struct ib_port *port)
1261 {
1262 	bool is_full_dev = &port->ibdev->coredev == coredev;
1263 
1264 	if (port->ibdev->port_data &&
1265 	    port->ibdev->port_data[port->port_num].sysfs == port)
1266 		port->ibdev->port_data[port->port_num].sysfs = NULL;
1267 	list_del(&port->kobj.entry);
1268 	if (is_full_dev)
1269 		sysfs_remove_groups(&port->kobj, port->ibdev->ops.port_groups);
1270 	sysfs_remove_groups(&port->kobj, port->groups_list);
1271 	kobject_del(&port->kobj);
1272 	kobject_put(&port->kobj);
1273 }
1274 
1275 static const char *node_type_string(int node_type)
1276 {
1277 	switch (node_type) {
1278 	case RDMA_NODE_IB_CA:
1279 		return "CA";
1280 	case RDMA_NODE_IB_SWITCH:
1281 		return "switch";
1282 	case RDMA_NODE_IB_ROUTER:
1283 		return "router";
1284 	case RDMA_NODE_RNIC:
1285 		return "RNIC";
1286 	case RDMA_NODE_USNIC:
1287 		return "usNIC";
1288 	case RDMA_NODE_USNIC_UDP:
1289 		return "usNIC UDP";
1290 	case RDMA_NODE_UNSPECIFIED:
1291 		return "unspecified";
1292 	}
1293 	return "<unknown>";
1294 }
1295 
1296 static ssize_t node_type_show(struct device *device,
1297 			      struct device_attribute *attr, char *buf)
1298 {
1299 	struct ib_device *dev = rdma_device_to_ibdev(device);
1300 
1301 	return sysfs_emit(buf, "%u: %s\n", dev->node_type,
1302 			  node_type_string(dev->node_type));
1303 }
1304 static DEVICE_ATTR_RO(node_type);
1305 
1306 static ssize_t sys_image_guid_show(struct device *device,
1307 				   struct device_attribute *dev_attr, char *buf)
1308 {
1309 	struct ib_device *dev = rdma_device_to_ibdev(device);
1310 	__be16 *guid = (__be16 *)&dev->attrs.sys_image_guid;
1311 
1312 	return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1313 			  be16_to_cpu(guid[0]),
1314 			  be16_to_cpu(guid[1]),
1315 			  be16_to_cpu(guid[2]),
1316 			  be16_to_cpu(guid[3]));
1317 }
1318 static DEVICE_ATTR_RO(sys_image_guid);
1319 
1320 static ssize_t node_guid_show(struct device *device,
1321 			      struct device_attribute *attr, char *buf)
1322 {
1323 	struct ib_device *dev = rdma_device_to_ibdev(device);
1324 	__be16 *node_guid = (__be16 *)&dev->node_guid;
1325 
1326 	return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1327 			  be16_to_cpu(node_guid[0]),
1328 			  be16_to_cpu(node_guid[1]),
1329 			  be16_to_cpu(node_guid[2]),
1330 			  be16_to_cpu(node_guid[3]));
1331 }
1332 static DEVICE_ATTR_RO(node_guid);
1333 
1334 static ssize_t node_desc_show(struct device *device,
1335 			      struct device_attribute *attr, char *buf)
1336 {
1337 	struct ib_device *dev = rdma_device_to_ibdev(device);
1338 
1339 	return sysfs_emit(buf, "%.64s\n", dev->node_desc);
1340 }
1341 
1342 static ssize_t node_desc_store(struct device *device,
1343 			       struct device_attribute *attr,
1344 			       const char *buf, size_t count)
1345 {
1346 	struct ib_device *dev = rdma_device_to_ibdev(device);
1347 	struct ib_device_modify desc = {};
1348 	int ret;
1349 
1350 	if (!dev->ops.modify_device)
1351 		return -EOPNOTSUPP;
1352 
1353 	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1354 	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1355 	if (ret)
1356 		return ret;
1357 
1358 	return count;
1359 }
1360 static DEVICE_ATTR_RW(node_desc);
1361 
1362 static ssize_t fw_ver_show(struct device *device, struct device_attribute *attr,
1363 			   char *buf)
1364 {
1365 	struct ib_device *dev = rdma_device_to_ibdev(device);
1366 	char version[IB_FW_VERSION_NAME_MAX] = {};
1367 
1368 	ib_get_device_fw_str(dev, version);
1369 
1370 	return sysfs_emit(buf, "%s\n", version);
1371 }
1372 static DEVICE_ATTR_RO(fw_ver);
1373 
1374 static struct attribute *ib_dev_attrs[] = {
1375 	&dev_attr_node_type.attr,
1376 	&dev_attr_node_guid.attr,
1377 	&dev_attr_sys_image_guid.attr,
1378 	&dev_attr_fw_ver.attr,
1379 	&dev_attr_node_desc.attr,
1380 	NULL,
1381 };
1382 
1383 const struct attribute_group ib_dev_attr_group = {
1384 	.attrs = ib_dev_attrs,
1385 };
1386 
1387 void ib_free_port_attrs(struct ib_core_device *coredev)
1388 {
1389 	struct kobject *p, *t;
1390 
1391 	list_for_each_entry_safe(p, t, &coredev->port_list, entry) {
1392 		struct ib_port *port = container_of(p, struct ib_port, kobj);
1393 
1394 		destroy_gid_attrs(port);
1395 		destroy_port(coredev, port);
1396 	}
1397 
1398 	kobject_put(coredev->ports_kobj);
1399 }
1400 
1401 int ib_setup_port_attrs(struct ib_core_device *coredev)
1402 {
1403 	struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1404 	u32 port_num;
1405 	int ret;
1406 
1407 	coredev->ports_kobj = kobject_create_and_add("ports",
1408 						     &coredev->dev.kobj);
1409 	if (!coredev->ports_kobj)
1410 		return -ENOMEM;
1411 
1412 	rdma_for_each_port (device, port_num) {
1413 		struct ib_port_attr attr;
1414 		struct ib_port *port;
1415 
1416 		ret = ib_query_port(device, port_num, &attr);
1417 		if (ret)
1418 			goto err_put;
1419 
1420 		port = setup_port(coredev, port_num, &attr);
1421 		if (IS_ERR(port)) {
1422 			ret = PTR_ERR(port);
1423 			goto err_put;
1424 		}
1425 
1426 		ret = setup_gid_attrs(port, &attr);
1427 		if (ret)
1428 			goto err_put;
1429 	}
1430 	return 0;
1431 
1432 err_put:
1433 	ib_free_port_attrs(coredev);
1434 	return ret;
1435 }
1436 
1437 /**
1438  * ib_port_register_client_groups - Add an ib_client's attributes to the port
1439  *
1440  * @ibdev: IB device to add counters
1441  * @port_num: valid port number
1442  * @groups: Group list of attributes
1443  *
1444  * Do not use. Only for legacy sysfs compatibility.
1445  */
1446 int ib_port_register_client_groups(struct ib_device *ibdev, u32 port_num,
1447 				   const struct attribute_group **groups)
1448 {
1449 	return sysfs_create_groups(&ibdev->port_data[port_num].sysfs->kobj,
1450 				   groups);
1451 }
1452 EXPORT_SYMBOL(ib_port_register_client_groups);
1453 
1454 void ib_port_unregister_client_groups(struct ib_device *ibdev, u32 port_num,
1455 				      const struct attribute_group **groups)
1456 {
1457 	return sysfs_remove_groups(&ibdev->port_data[port_num].sysfs->kobj,
1458 				   groups);
1459 }
1460 EXPORT_SYMBOL(ib_port_unregister_client_groups);
1461