xref: /linux/drivers/infiniband/core/sysfs.c (revision 69050f8d6d075dc01af7a5f2f550a8067510366f)
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 	ret = ib_query_port(ibdev, port_num, &attr);
220 	if (ret)
221 		return ret;
222 
223 	return sysfs_emit(buf, "%d: %s\n", attr.state,
224 			  ib_port_state_to_str(attr.state));
225 }
226 
227 static ssize_t lid_show(struct ib_device *ibdev, u32 port_num,
228 			struct ib_port_attribute *unused, char *buf)
229 {
230 	struct ib_port_attr attr;
231 	ssize_t ret;
232 
233 	ret = ib_query_port(ibdev, port_num, &attr);
234 	if (ret)
235 		return ret;
236 
237 	return sysfs_emit(buf, "0x%x\n", attr.lid);
238 }
239 
240 static ssize_t lid_mask_count_show(struct ib_device *ibdev, u32 port_num,
241 				   struct ib_port_attribute *unused, char *buf)
242 {
243 	struct ib_port_attr attr;
244 	ssize_t ret;
245 
246 	ret = ib_query_port(ibdev, port_num, &attr);
247 	if (ret)
248 		return ret;
249 
250 	return sysfs_emit(buf, "%u\n", attr.lmc);
251 }
252 
253 static ssize_t sm_lid_show(struct ib_device *ibdev, u32 port_num,
254 			   struct ib_port_attribute *unused, char *buf)
255 {
256 	struct ib_port_attr attr;
257 	ssize_t ret;
258 
259 	ret = ib_query_port(ibdev, port_num, &attr);
260 	if (ret)
261 		return ret;
262 
263 	return sysfs_emit(buf, "0x%x\n", attr.sm_lid);
264 }
265 
266 static ssize_t sm_sl_show(struct ib_device *ibdev, u32 port_num,
267 			  struct ib_port_attribute *unused, char *buf)
268 {
269 	struct ib_port_attr attr;
270 	ssize_t ret;
271 
272 	ret = ib_query_port(ibdev, port_num, &attr);
273 	if (ret)
274 		return ret;
275 
276 	return sysfs_emit(buf, "%u\n", attr.sm_sl);
277 }
278 
279 static ssize_t cap_mask_show(struct ib_device *ibdev, u32 port_num,
280 			     struct ib_port_attribute *unused, char *buf)
281 {
282 	struct ib_port_attr attr;
283 	ssize_t ret;
284 
285 	ret = ib_query_port(ibdev, port_num, &attr);
286 	if (ret)
287 		return ret;
288 
289 	return sysfs_emit(buf, "0x%08x\n", attr.port_cap_flags);
290 }
291 
292 static ssize_t rate_show(struct ib_device *ibdev, u32 port_num,
293 			 struct ib_port_attribute *unused, char *buf)
294 {
295 	struct ib_port_speed_info speed_info;
296 	struct ib_port_attr attr;
297 	ssize_t ret;
298 
299 	ret = ib_query_port(ibdev, port_num, &attr);
300 	if (ret)
301 		return ret;
302 
303 	ret = ib_port_attr_to_speed_info(&attr, &speed_info);
304 	if (ret)
305 		return ret;
306 
307 	return sysfs_emit(buf, "%d%s Gb/sec (%dX%s)\n", speed_info.rate / 10,
308 			  speed_info.rate % 10 ? ".5" : "",
309 			  ib_width_enum_to_int(attr.active_width),
310 			  speed_info.str);
311 }
312 
313 static const char *phys_state_to_str(enum ib_port_phys_state phys_state)
314 {
315 	static const char *phys_state_str[] = {
316 		"<unknown>",
317 		"Sleep",
318 		"Polling",
319 		"Disabled",
320 		"PortConfigurationTraining",
321 		"LinkUp",
322 		"LinkErrorRecovery",
323 		"Phy Test",
324 	};
325 
326 	if (phys_state < ARRAY_SIZE(phys_state_str))
327 		return phys_state_str[phys_state];
328 	return "<unknown>";
329 }
330 
331 static ssize_t phys_state_show(struct ib_device *ibdev, u32 port_num,
332 			       struct ib_port_attribute *unused, char *buf)
333 {
334 	struct ib_port_attr attr;
335 
336 	ssize_t ret;
337 
338 	ret = ib_query_port(ibdev, port_num, &attr);
339 	if (ret)
340 		return ret;
341 
342 	return sysfs_emit(buf, "%u: %s\n", attr.phys_state,
343 			  phys_state_to_str(attr.phys_state));
344 }
345 
346 static ssize_t link_layer_show(struct ib_device *ibdev, u32 port_num,
347 			       struct ib_port_attribute *unused, char *buf)
348 {
349 	const char *output;
350 
351 	switch (rdma_port_get_link_layer(ibdev, port_num)) {
352 	case IB_LINK_LAYER_INFINIBAND:
353 		output = "InfiniBand";
354 		break;
355 	case IB_LINK_LAYER_ETHERNET:
356 		output = "Ethernet";
357 		break;
358 	default:
359 		output = "Unknown";
360 		break;
361 	}
362 
363 	return sysfs_emit(buf, "%s\n", output);
364 }
365 
366 static IB_PORT_ATTR_RO(state);
367 static IB_PORT_ATTR_RO(lid);
368 static IB_PORT_ATTR_RO(lid_mask_count);
369 static IB_PORT_ATTR_RO(sm_lid);
370 static IB_PORT_ATTR_RO(sm_sl);
371 static IB_PORT_ATTR_RO(cap_mask);
372 static IB_PORT_ATTR_RO(rate);
373 static IB_PORT_ATTR_RO(phys_state);
374 static IB_PORT_ATTR_RO(link_layer);
375 
376 static struct attribute *port_default_attrs[] = {
377 	&ib_port_attr_state.attr,
378 	&ib_port_attr_lid.attr,
379 	&ib_port_attr_lid_mask_count.attr,
380 	&ib_port_attr_sm_lid.attr,
381 	&ib_port_attr_sm_sl.attr,
382 	&ib_port_attr_cap_mask.attr,
383 	&ib_port_attr_rate.attr,
384 	&ib_port_attr_phys_state.attr,
385 	&ib_port_attr_link_layer.attr,
386 	NULL
387 };
388 ATTRIBUTE_GROUPS(port_default);
389 
390 static ssize_t print_ndev(const struct ib_gid_attr *gid_attr, char *buf)
391 {
392 	struct net_device *ndev;
393 	int ret = -EINVAL;
394 
395 	rcu_read_lock();
396 	ndev = rcu_dereference(gid_attr->ndev);
397 	if (ndev)
398 		ret = sysfs_emit(buf, "%s\n", ndev->name);
399 	rcu_read_unlock();
400 	return ret;
401 }
402 
403 static ssize_t print_gid_type(const struct ib_gid_attr *gid_attr, char *buf)
404 {
405 	return sysfs_emit(buf, "%s\n",
406 			  ib_cache_gid_type_str(gid_attr->gid_type));
407 }
408 
409 static ssize_t _show_port_gid_attr(
410 	struct ib_device *ibdev, u32 port_num, struct ib_port_attribute *attr,
411 	char *buf,
412 	ssize_t (*print)(const struct ib_gid_attr *gid_attr, char *buf))
413 {
414 	struct port_table_attribute *tab_attr =
415 		container_of(attr, struct port_table_attribute, attr);
416 	const struct ib_gid_attr *gid_attr;
417 	ssize_t ret;
418 
419 	gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
420 	if (IS_ERR(gid_attr))
421 		/* -EINVAL is returned for user space compatibility reasons. */
422 		return -EINVAL;
423 
424 	ret = print(gid_attr, buf);
425 	rdma_put_gid_attr(gid_attr);
426 	return ret;
427 }
428 
429 static ssize_t show_port_gid(struct ib_device *ibdev, u32 port_num,
430 			     struct ib_port_attribute *attr, char *buf)
431 {
432 	struct port_table_attribute *tab_attr =
433 		container_of(attr, struct port_table_attribute, attr);
434 	const struct ib_gid_attr *gid_attr;
435 	int len;
436 
437 	gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
438 	if (IS_ERR(gid_attr)) {
439 		const union ib_gid zgid = {};
440 
441 		/* If reading GID fails, it is likely due to GID entry being
442 		 * empty (invalid) or reserved GID in the table.  User space
443 		 * expects to read GID table entries as long as it given index
444 		 * is within GID table size.  Administrative/debugging tool
445 		 * fails to query rest of the GID entries if it hits error
446 		 * while querying a GID of the given index.  To avoid user
447 		 * space throwing such error on fail to read gid, return zero
448 		 * GID as before. This maintains backward compatibility.
449 		 */
450 		return sysfs_emit(buf, "%pI6\n", zgid.raw);
451 	}
452 
453 	len = sysfs_emit(buf, "%pI6\n", gid_attr->gid.raw);
454 	rdma_put_gid_attr(gid_attr);
455 	return len;
456 }
457 
458 static ssize_t show_port_gid_attr_ndev(struct ib_device *ibdev, u32 port_num,
459 				       struct ib_port_attribute *attr,
460 				       char *buf)
461 {
462 	return _show_port_gid_attr(ibdev, port_num, attr, buf, print_ndev);
463 }
464 
465 static ssize_t show_port_gid_attr_gid_type(struct ib_device *ibdev,
466 					   u32 port_num,
467 					   struct ib_port_attribute *attr,
468 					   char *buf)
469 {
470 	return _show_port_gid_attr(ibdev, port_num, attr, buf, print_gid_type);
471 }
472 
473 static ssize_t show_port_pkey(struct ib_device *ibdev, u32 port_num,
474 			      struct ib_port_attribute *attr, char *buf)
475 {
476 	struct port_table_attribute *tab_attr =
477 		container_of(attr, struct port_table_attribute, attr);
478 	u16 pkey;
479 	int ret;
480 
481 	ret = ib_query_pkey(ibdev, port_num, tab_attr->index, &pkey);
482 	if (ret)
483 		return ret;
484 
485 	return sysfs_emit(buf, "0x%04x\n", pkey);
486 }
487 
488 #define PORT_PMA_ATTR(_name, _counter, _width, _offset)			\
489 struct port_table_attribute port_pma_attr_##_name = {			\
490 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
491 	.index = (_offset) | ((_width) << 16) | ((_counter) << 24),	\
492 	.attr_id = IB_PMA_PORT_COUNTERS,				\
493 }
494 
495 #define PORT_PMA_ATTR_EXT(_name, _width, _offset)			\
496 struct port_table_attribute port_pma_attr_ext_##_name = {		\
497 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
498 	.index = (_offset) | ((_width) << 16),				\
499 	.attr_id = IB_PMA_PORT_COUNTERS_EXT,				\
500 }
501 
502 /*
503  * Get a Perfmgmt MAD block of data.
504  * Returns error code or the number of bytes retrieved.
505  */
506 static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
507 		void *data, int offset, size_t size)
508 {
509 	struct ib_mad *in_mad;
510 	struct ib_mad *out_mad;
511 	size_t mad_size = sizeof(*out_mad);
512 	u16 out_mad_pkey_index = 0;
513 	ssize_t ret;
514 
515 	if (!dev->ops.process_mad)
516 		return -ENOSYS;
517 
518 	in_mad = kzalloc_obj(*in_mad, GFP_KERNEL);
519 	out_mad = kzalloc_obj(*out_mad, GFP_KERNEL);
520 	if (!in_mad || !out_mad) {
521 		ret = -ENOMEM;
522 		goto out;
523 	}
524 
525 	in_mad->mad_hdr.base_version  = 1;
526 	in_mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_PERF_MGMT;
527 	in_mad->mad_hdr.class_version = 1;
528 	in_mad->mad_hdr.method        = IB_MGMT_METHOD_GET;
529 	in_mad->mad_hdr.attr_id       = attr;
530 
531 	if (attr != IB_PMA_CLASS_PORT_INFO)
532 		in_mad->data[41] = port_num;	/* PortSelect field */
533 
534 	if ((dev->ops.process_mad(dev, IB_MAD_IGNORE_MKEY, port_num, NULL, NULL,
535 				  in_mad, out_mad, &mad_size,
536 				  &out_mad_pkey_index) &
537 	     (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
538 	    (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
539 		ret = -EINVAL;
540 		goto out;
541 	}
542 	memcpy(data, out_mad->data + offset, size);
543 	ret = size;
544 out:
545 	kfree(in_mad);
546 	kfree(out_mad);
547 	return ret;
548 }
549 
550 static ssize_t show_pma_counter(struct ib_device *ibdev, u32 port_num,
551 				struct ib_port_attribute *attr, char *buf)
552 {
553 	struct port_table_attribute *tab_attr =
554 		container_of(attr, struct port_table_attribute, attr);
555 	int offset = tab_attr->index & 0xffff;
556 	int width  = (tab_attr->index >> 16) & 0xff;
557 	int ret;
558 	u8 data[8];
559 	int len;
560 
561 	ret = get_perf_mad(ibdev, port_num, tab_attr->attr_id, &data,
562 			40 + offset / 8, sizeof(data));
563 	if (ret < 0)
564 		return ret;
565 
566 	switch (width) {
567 	case 4:
568 		len = sysfs_emit(buf, "%d\n",
569 				 (*data >> (4 - (offset % 8))) & 0xf);
570 		break;
571 	case 8:
572 		len = sysfs_emit(buf, "%u\n", *data);
573 		break;
574 	case 16:
575 		len = sysfs_emit(buf, "%u\n", be16_to_cpup((__be16 *)data));
576 		break;
577 	case 32:
578 		len = sysfs_emit(buf, "%u\n", be32_to_cpup((__be32 *)data));
579 		break;
580 	case 64:
581 		len = sysfs_emit(buf, "%llu\n", be64_to_cpup((__be64 *)data));
582 		break;
583 	default:
584 		len = 0;
585 		break;
586 	}
587 
588 	return len;
589 }
590 
591 static PORT_PMA_ATTR(symbol_error		    ,  0, 16,  32);
592 static PORT_PMA_ATTR(link_error_recovery	    ,  1,  8,  48);
593 static PORT_PMA_ATTR(link_downed		    ,  2,  8,  56);
594 static PORT_PMA_ATTR(port_rcv_errors		    ,  3, 16,  64);
595 static PORT_PMA_ATTR(port_rcv_remote_physical_errors,  4, 16,  80);
596 static PORT_PMA_ATTR(port_rcv_switch_relay_errors   ,  5, 16,  96);
597 static PORT_PMA_ATTR(port_xmit_discards		    ,  6, 16, 112);
598 static PORT_PMA_ATTR(port_xmit_constraint_errors    ,  7,  8, 128);
599 static PORT_PMA_ATTR(port_rcv_constraint_errors	    ,  8,  8, 136);
600 static PORT_PMA_ATTR(local_link_integrity_errors    ,  9,  4, 152);
601 static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10,  4, 156);
602 static PORT_PMA_ATTR(VL15_dropped		    , 11, 16, 176);
603 static PORT_PMA_ATTR(port_xmit_data		    , 12, 32, 192);
604 static PORT_PMA_ATTR(port_rcv_data		    , 13, 32, 224);
605 static PORT_PMA_ATTR(port_xmit_packets		    , 14, 32, 256);
606 static PORT_PMA_ATTR(port_rcv_packets		    , 15, 32, 288);
607 static PORT_PMA_ATTR(port_xmit_wait		    ,  0, 32, 320);
608 
609 /*
610  * Counters added by extended set
611  */
612 static PORT_PMA_ATTR_EXT(port_xmit_data		    , 64,  64);
613 static PORT_PMA_ATTR_EXT(port_rcv_data		    , 64, 128);
614 static PORT_PMA_ATTR_EXT(port_xmit_packets	    , 64, 192);
615 static PORT_PMA_ATTR_EXT(port_rcv_packets	    , 64, 256);
616 static PORT_PMA_ATTR_EXT(unicast_xmit_packets	    , 64, 320);
617 static PORT_PMA_ATTR_EXT(unicast_rcv_packets	    , 64, 384);
618 static PORT_PMA_ATTR_EXT(multicast_xmit_packets	    , 64, 448);
619 static PORT_PMA_ATTR_EXT(multicast_rcv_packets	    , 64, 512);
620 
621 static struct attribute *pma_attrs[] = {
622 	&port_pma_attr_symbol_error.attr.attr,
623 	&port_pma_attr_link_error_recovery.attr.attr,
624 	&port_pma_attr_link_downed.attr.attr,
625 	&port_pma_attr_port_rcv_errors.attr.attr,
626 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
627 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
628 	&port_pma_attr_port_xmit_discards.attr.attr,
629 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
630 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
631 	&port_pma_attr_local_link_integrity_errors.attr.attr,
632 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
633 	&port_pma_attr_VL15_dropped.attr.attr,
634 	&port_pma_attr_port_xmit_data.attr.attr,
635 	&port_pma_attr_port_rcv_data.attr.attr,
636 	&port_pma_attr_port_xmit_packets.attr.attr,
637 	&port_pma_attr_port_rcv_packets.attr.attr,
638 	&port_pma_attr_port_xmit_wait.attr.attr,
639 	NULL
640 };
641 
642 static struct attribute *pma_attrs_ext[] = {
643 	&port_pma_attr_symbol_error.attr.attr,
644 	&port_pma_attr_link_error_recovery.attr.attr,
645 	&port_pma_attr_link_downed.attr.attr,
646 	&port_pma_attr_port_rcv_errors.attr.attr,
647 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
648 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
649 	&port_pma_attr_port_xmit_discards.attr.attr,
650 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
651 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
652 	&port_pma_attr_local_link_integrity_errors.attr.attr,
653 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
654 	&port_pma_attr_VL15_dropped.attr.attr,
655 	&port_pma_attr_ext_port_xmit_data.attr.attr,
656 	&port_pma_attr_ext_port_rcv_data.attr.attr,
657 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
658 	&port_pma_attr_port_xmit_wait.attr.attr,
659 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
660 	&port_pma_attr_ext_unicast_rcv_packets.attr.attr,
661 	&port_pma_attr_ext_unicast_xmit_packets.attr.attr,
662 	&port_pma_attr_ext_multicast_rcv_packets.attr.attr,
663 	&port_pma_attr_ext_multicast_xmit_packets.attr.attr,
664 	NULL
665 };
666 
667 static struct attribute *pma_attrs_noietf[] = {
668 	&port_pma_attr_symbol_error.attr.attr,
669 	&port_pma_attr_link_error_recovery.attr.attr,
670 	&port_pma_attr_link_downed.attr.attr,
671 	&port_pma_attr_port_rcv_errors.attr.attr,
672 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
673 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
674 	&port_pma_attr_port_xmit_discards.attr.attr,
675 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
676 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
677 	&port_pma_attr_local_link_integrity_errors.attr.attr,
678 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
679 	&port_pma_attr_VL15_dropped.attr.attr,
680 	&port_pma_attr_ext_port_xmit_data.attr.attr,
681 	&port_pma_attr_ext_port_rcv_data.attr.attr,
682 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
683 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
684 	&port_pma_attr_port_xmit_wait.attr.attr,
685 	NULL
686 };
687 
688 static const struct attribute_group pma_group = {
689 	.name  = "counters",
690 	.attrs  = pma_attrs
691 };
692 
693 static const struct attribute_group pma_group_ext = {
694 	.name  = "counters",
695 	.attrs  = pma_attrs_ext
696 };
697 
698 static const struct attribute_group pma_group_noietf = {
699 	.name  = "counters",
700 	.attrs  = pma_attrs_noietf
701 };
702 
703 static void ib_port_release(struct kobject *kobj)
704 {
705 	struct ib_port *port = container_of(kobj, struct ib_port, kobj);
706 	int i;
707 
708 	for (i = 0; i != ARRAY_SIZE(port->groups); i++)
709 		kfree(port->groups[i].attrs);
710 	if (port->hw_stats_data)
711 		rdma_free_hw_stats_struct(port->hw_stats_data->stats);
712 	kfree(port->hw_stats_data);
713 	kvfree(port);
714 }
715 
716 static void ib_port_gid_attr_release(struct kobject *kobj)
717 {
718 	struct gid_attr_group *gid_attr_group =
719 		container_of(kobj, struct gid_attr_group, kobj);
720 	int i;
721 
722 	for (i = 0; i != ARRAY_SIZE(gid_attr_group->groups); i++)
723 		kfree(gid_attr_group->groups[i].attrs);
724 	kfree(gid_attr_group);
725 }
726 
727 static struct kobj_type port_type = {
728 	.release       = ib_port_release,
729 	.sysfs_ops     = &port_sysfs_ops,
730 	.default_groups = port_default_groups,
731 };
732 
733 static struct kobj_type gid_attr_type = {
734 	.sysfs_ops      = &gid_attr_sysfs_ops,
735 	.release        = ib_port_gid_attr_release
736 };
737 
738 /*
739  * Figure out which counter table to use depending on
740  * the device capabilities.
741  */
742 static const struct attribute_group *get_counter_table(struct ib_device *dev,
743 						       int port_num)
744 {
745 	struct ib_class_port_info cpi;
746 
747 	if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
748 				&cpi, 40, sizeof(cpi)) >= 0) {
749 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
750 			/* We have extended counters */
751 			return &pma_group_ext;
752 
753 		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
754 			/* But not the IETF ones */
755 			return &pma_group_noietf;
756 	}
757 
758 	/* Fall back to normal counters */
759 	return &pma_group;
760 }
761 
762 static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
763 			   u32 port_num, int index)
764 {
765 	int ret;
766 
767 	if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
768 		return 0;
769 	ret = dev->ops.get_hw_stats(dev, stats, port_num, index);
770 	if (ret < 0)
771 		return ret;
772 	if (ret == stats->num_counters)
773 		stats->timestamp = jiffies;
774 
775 	return 0;
776 }
777 
778 static int print_hw_stat(struct ib_device *dev, int port_num,
779 			 struct rdma_hw_stats *stats, int index, char *buf)
780 {
781 	u64 v = rdma_counter_get_hwstat_value(dev, port_num, index);
782 
783 	return sysfs_emit(buf, "%llu\n", stats->value[index] + v);
784 }
785 
786 static ssize_t show_hw_stats(struct ib_device *ibdev,
787 			     struct rdma_hw_stats *stats, unsigned int index,
788 			     unsigned int port_num, char *buf)
789 {
790 	int ret;
791 
792 	mutex_lock(&stats->lock);
793 	ret = update_hw_stats(ibdev, stats, port_num, index);
794 	if (ret)
795 		goto unlock;
796 	ret = print_hw_stat(ibdev, port_num, stats, index, buf);
797 unlock:
798 	mutex_unlock(&stats->lock);
799 
800 	return ret;
801 }
802 
803 static ssize_t show_stats_lifespan(struct ib_device *ibdev,
804 				   struct rdma_hw_stats *stats,
805 				   unsigned int index, unsigned int port_num,
806 				   char *buf)
807 {
808 	int msecs;
809 
810 	mutex_lock(&stats->lock);
811 	msecs = jiffies_to_msecs(stats->lifespan);
812 	mutex_unlock(&stats->lock);
813 
814 	return sysfs_emit(buf, "%d\n", msecs);
815 }
816 
817 static ssize_t set_stats_lifespan(struct ib_device *ibdev,
818 				   struct rdma_hw_stats *stats,
819 				   unsigned int index, unsigned int port_num,
820 				   const char *buf, size_t count)
821 {
822 	int msecs;
823 	int jiffies;
824 	int ret;
825 
826 	ret = kstrtoint(buf, 10, &msecs);
827 	if (ret)
828 		return ret;
829 	if (msecs < 0 || msecs > 10000)
830 		return -EINVAL;
831 	jiffies = msecs_to_jiffies(msecs);
832 
833 	mutex_lock(&stats->lock);
834 	stats->lifespan = jiffies;
835 	mutex_unlock(&stats->lock);
836 
837 	return count;
838 }
839 
840 static struct hw_stats_device_data *
841 alloc_hw_stats_device(struct ib_device *ibdev)
842 {
843 	struct hw_stats_device_data *data;
844 	struct rdma_hw_stats *stats;
845 
846 	if (!ibdev->ops.alloc_hw_device_stats)
847 		return ERR_PTR(-EOPNOTSUPP);
848 	stats = ibdev->ops.alloc_hw_device_stats(ibdev);
849 	if (!stats)
850 		return ERR_PTR(-ENOMEM);
851 	if (!stats->descs || stats->num_counters <= 0)
852 		goto err_free_stats;
853 
854 	/*
855 	 * Two extra attribue elements here, one for the lifespan entry and
856 	 * one to NULL terminate the list for the sysfs core code
857 	 */
858 	data = kzalloc_flex(*data, attrs, size_add(stats->num_counters, 1),
859 			    GFP_KERNEL);
860 	if (!data)
861 		goto err_free_stats;
862 	data->group.attrs = kzalloc_objs(*data->group.attrs,
863 					 stats->num_counters + 2, GFP_KERNEL);
864 	if (!data->group.attrs)
865 		goto err_free_data;
866 
867 	data->group.name = "hw_counters";
868 	data->stats = stats;
869 	return data;
870 
871 err_free_data:
872 	kfree(data);
873 err_free_stats:
874 	rdma_free_hw_stats_struct(stats);
875 	return ERR_PTR(-ENOMEM);
876 }
877 
878 void ib_device_release_hw_stats(struct hw_stats_device_data *data)
879 {
880 	kfree(data->group.attrs);
881 	rdma_free_hw_stats_struct(data->stats);
882 	kfree(data);
883 }
884 
885 int ib_setup_device_attrs(struct ib_device *ibdev)
886 {
887 	struct hw_stats_device_attribute *attr;
888 	struct hw_stats_device_data *data;
889 	bool opstat_skipped = false;
890 	int i, ret, pos = 0;
891 
892 	data = alloc_hw_stats_device(ibdev);
893 	if (IS_ERR(data)) {
894 		if (PTR_ERR(data) == -EOPNOTSUPP)
895 			return 0;
896 		return PTR_ERR(data);
897 	}
898 	ibdev->hw_stats_data = data;
899 
900 	ret = ibdev->ops.get_hw_stats(ibdev, data->stats, 0,
901 				      data->stats->num_counters);
902 	if (ret != data->stats->num_counters) {
903 		if (WARN_ON(ret >= 0))
904 			return -EINVAL;
905 		return ret;
906 	}
907 
908 	data->stats->timestamp = jiffies;
909 
910 	for (i = 0; i < data->stats->num_counters; i++) {
911 		if (data->stats->descs[i].flags & IB_STAT_FLAG_OPTIONAL) {
912 			opstat_skipped = true;
913 			continue;
914 		}
915 
916 		WARN_ON(opstat_skipped);
917 		attr = &data->attrs[pos];
918 		sysfs_attr_init(&attr->attr.attr);
919 		attr->attr.attr.name = data->stats->descs[i].name;
920 		attr->attr.attr.mode = 0444;
921 		attr->attr.show = hw_stat_device_show;
922 		attr->show = show_hw_stats;
923 		data->group.attrs[pos] = &attr->attr.attr;
924 		pos++;
925 	}
926 
927 	attr = &data->attrs[pos];
928 	sysfs_attr_init(&attr->attr.attr);
929 	attr->attr.attr.name = "lifespan";
930 	attr->attr.attr.mode = 0644;
931 	attr->attr.show = hw_stat_device_show;
932 	attr->show = show_stats_lifespan;
933 	attr->attr.store = hw_stat_device_store;
934 	attr->store = set_stats_lifespan;
935 	data->group.attrs[pos] = &attr->attr.attr;
936 	for (i = 0; i != ARRAY_SIZE(ibdev->groups); i++)
937 		if (!ibdev->groups[i]) {
938 			ibdev->groups[i] = &data->group;
939 			ibdev->hw_stats_attr_index = i;
940 			return 0;
941 		}
942 	WARN(true, "struct ib_device->groups is too small");
943 	return -EINVAL;
944 }
945 
946 static struct hw_stats_port_data *
947 alloc_hw_stats_port(struct ib_port *port, struct attribute_group *group)
948 {
949 	struct ib_device *ibdev = port->ibdev;
950 	struct hw_stats_port_data *data;
951 	struct rdma_hw_stats *stats;
952 
953 	if (!ibdev->ops.alloc_hw_port_stats)
954 		return ERR_PTR(-EOPNOTSUPP);
955 	stats = ibdev->ops.alloc_hw_port_stats(port->ibdev, port->port_num);
956 	if (!stats)
957 		return ERR_PTR(-ENOMEM);
958 	if (!stats->descs || stats->num_counters <= 0)
959 		goto err_free_stats;
960 
961 	/*
962 	 * Two extra attribue elements here, one for the lifespan entry and
963 	 * one to NULL terminate the list for the sysfs core code
964 	 */
965 	data = kzalloc_flex(*data, attrs, size_add(stats->num_counters, 1),
966 			    GFP_KERNEL);
967 	if (!data)
968 		goto err_free_stats;
969 	group->attrs = kzalloc_objs(*group->attrs, stats->num_counters + 2,
970 				    GFP_KERNEL);
971 	if (!group->attrs)
972 		goto err_free_data;
973 
974 	group->name = "hw_counters";
975 	data->stats = stats;
976 	return data;
977 
978 err_free_data:
979 	kfree(data);
980 err_free_stats:
981 	rdma_free_hw_stats_struct(stats);
982 	return ERR_PTR(-ENOMEM);
983 }
984 
985 static int setup_hw_port_stats(struct ib_port *port,
986 			       struct attribute_group *group)
987 {
988 	struct hw_stats_port_attribute *attr;
989 	struct hw_stats_port_data *data;
990 	bool opstat_skipped = false;
991 	int i, ret, pos = 0;
992 
993 	data = alloc_hw_stats_port(port, group);
994 	if (IS_ERR(data))
995 		return PTR_ERR(data);
996 
997 	ret = port->ibdev->ops.get_hw_stats(port->ibdev, data->stats,
998 					    port->port_num,
999 					    data->stats->num_counters);
1000 	if (ret != data->stats->num_counters) {
1001 		if (WARN_ON(ret >= 0))
1002 			return -EINVAL;
1003 		return ret;
1004 	}
1005 
1006 	data->stats->timestamp = jiffies;
1007 
1008 	for (i = 0; i < data->stats->num_counters; i++) {
1009 		if (data->stats->descs[i].flags & IB_STAT_FLAG_OPTIONAL) {
1010 			opstat_skipped = true;
1011 			continue;
1012 		}
1013 
1014 		WARN_ON(opstat_skipped);
1015 		attr = &data->attrs[pos];
1016 		sysfs_attr_init(&attr->attr.attr);
1017 		attr->attr.attr.name = data->stats->descs[i].name;
1018 		attr->attr.attr.mode = 0444;
1019 		attr->attr.show = hw_stat_port_show;
1020 		attr->show = show_hw_stats;
1021 		group->attrs[pos] = &attr->attr.attr;
1022 		pos++;
1023 	}
1024 
1025 	attr = &data->attrs[pos];
1026 	sysfs_attr_init(&attr->attr.attr);
1027 	attr->attr.attr.name = "lifespan";
1028 	attr->attr.attr.mode = 0644;
1029 	attr->attr.show = hw_stat_port_show;
1030 	attr->show = show_stats_lifespan;
1031 	attr->attr.store = hw_stat_port_store;
1032 	attr->store = set_stats_lifespan;
1033 	group->attrs[pos] = &attr->attr.attr;
1034 
1035 	port->hw_stats_data = data;
1036 	return 0;
1037 }
1038 
1039 struct rdma_hw_stats *ib_get_hw_stats_port(struct ib_device *ibdev,
1040 					   u32 port_num)
1041 {
1042 	if (!ibdev->port_data || !rdma_is_port_valid(ibdev, port_num) ||
1043 	    !ibdev->port_data[port_num].sysfs->hw_stats_data)
1044 		return NULL;
1045 	return ibdev->port_data[port_num].sysfs->hw_stats_data->stats;
1046 }
1047 
1048 static int
1049 alloc_port_table_group(const char *name, struct attribute_group *group,
1050 		       struct port_table_attribute *attrs, size_t num,
1051 		       ssize_t (*show)(struct ib_device *ibdev, u32 port_num,
1052 				       struct ib_port_attribute *, char *buf))
1053 {
1054 	struct attribute **attr_list;
1055 	int i;
1056 
1057 	attr_list = kzalloc_objs(*attr_list, num + 1, GFP_KERNEL);
1058 	if (!attr_list)
1059 		return -ENOMEM;
1060 
1061 	for (i = 0; i < num; i++) {
1062 		struct port_table_attribute *element = &attrs[i];
1063 
1064 		if (snprintf(element->name, sizeof(element->name), "%d", i) >=
1065 		    sizeof(element->name))
1066 			goto err;
1067 
1068 		sysfs_attr_init(&element->attr.attr);
1069 		element->attr.attr.name = element->name;
1070 		element->attr.attr.mode = 0444;
1071 		element->attr.show = show;
1072 		element->index = i;
1073 
1074 		attr_list[i] = &element->attr.attr;
1075 	}
1076 	group->name = name;
1077 	group->attrs = attr_list;
1078 	return 0;
1079 err:
1080 	kfree(attr_list);
1081 	return -EINVAL;
1082 }
1083 
1084 /*
1085  * Create the sysfs:
1086  *  ibp0s9/ports/XX/gid_attrs/{ndevs,types}/YYY
1087  * YYY is the gid table index in decimal
1088  */
1089 static int setup_gid_attrs(struct ib_port *port,
1090 			   const struct ib_port_attr *attr)
1091 {
1092 	struct gid_attr_group *gid_attr_group;
1093 	int ret;
1094 
1095 	gid_attr_group = kzalloc_flex(*gid_attr_group, attrs_list,
1096 				      size_mul(attr->gid_tbl_len, 2),
1097 				      GFP_KERNEL);
1098 	if (!gid_attr_group)
1099 		return -ENOMEM;
1100 	gid_attr_group->port = port;
1101 	kobject_init(&gid_attr_group->kobj, &gid_attr_type);
1102 
1103 	ret = alloc_port_table_group("ndevs", &gid_attr_group->groups[0],
1104 				     gid_attr_group->attrs_list,
1105 				     attr->gid_tbl_len,
1106 				     show_port_gid_attr_ndev);
1107 	if (ret)
1108 		goto err_put;
1109 	gid_attr_group->groups_list[0] = &gid_attr_group->groups[0];
1110 
1111 	ret = alloc_port_table_group(
1112 		"types", &gid_attr_group->groups[1],
1113 		gid_attr_group->attrs_list + attr->gid_tbl_len,
1114 		attr->gid_tbl_len, show_port_gid_attr_gid_type);
1115 	if (ret)
1116 		goto err_put;
1117 	gid_attr_group->groups_list[1] = &gid_attr_group->groups[1];
1118 
1119 	ret = kobject_add(&gid_attr_group->kobj, &port->kobj, "gid_attrs");
1120 	if (ret)
1121 		goto err_put;
1122 	ret = sysfs_create_groups(&gid_attr_group->kobj,
1123 				  gid_attr_group->groups_list);
1124 	if (ret)
1125 		goto err_del;
1126 	port->gid_attr_group = gid_attr_group;
1127 	return 0;
1128 
1129 err_del:
1130 	kobject_del(&gid_attr_group->kobj);
1131 err_put:
1132 	kobject_put(&gid_attr_group->kobj);
1133 	return ret;
1134 }
1135 
1136 static void destroy_gid_attrs(struct ib_port *port)
1137 {
1138 	struct gid_attr_group *gid_attr_group = port->gid_attr_group;
1139 
1140 	if (!gid_attr_group)
1141 		return;
1142 	sysfs_remove_groups(&gid_attr_group->kobj, gid_attr_group->groups_list);
1143 	kobject_del(&gid_attr_group->kobj);
1144 	kobject_put(&gid_attr_group->kobj);
1145 }
1146 
1147 /*
1148  * Create the sysfs:
1149  *  ibp0s9/ports/XX/{gids,pkeys,counters}/YYY
1150  */
1151 static struct ib_port *setup_port(struct ib_core_device *coredev, int port_num,
1152 				  const struct ib_port_attr *attr)
1153 {
1154 	struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1155 	bool is_full_dev = &device->coredev == coredev;
1156 	const struct attribute_group **cur_group;
1157 	struct ib_port *p;
1158 	int ret;
1159 
1160 	p = kvzalloc_flex(*p, attrs_list,
1161 			  size_add(attr->gid_tbl_len, attr->pkey_tbl_len),
1162 			  GFP_KERNEL);
1163 	if (!p)
1164 		return ERR_PTR(-ENOMEM);
1165 	p->ibdev = device;
1166 	p->port_num = port_num;
1167 	kobject_init(&p->kobj, &port_type);
1168 
1169 	if (device->port_data && is_full_dev)
1170 		device->port_data[port_num].sysfs = p;
1171 
1172 	cur_group = p->groups_list;
1173 	ret = alloc_port_table_group("gids", &p->groups[0], p->attrs_list,
1174 				     attr->gid_tbl_len, show_port_gid);
1175 	if (ret)
1176 		goto err_put;
1177 	*cur_group++ = &p->groups[0];
1178 
1179 	if (attr->pkey_tbl_len) {
1180 		ret = alloc_port_table_group("pkeys", &p->groups[1],
1181 					     p->attrs_list + attr->gid_tbl_len,
1182 					     attr->pkey_tbl_len, show_port_pkey);
1183 		if (ret)
1184 			goto err_put;
1185 		*cur_group++ = &p->groups[1];
1186 	}
1187 
1188 	/*
1189 	 * If port == 0, it means hw_counters are per device and not per
1190 	 * port, so holder should be device. Therefore skip per port
1191 	 * counter initialization.
1192 	 */
1193 	if (port_num && is_full_dev) {
1194 		ret = setup_hw_port_stats(p, &p->groups[2]);
1195 		if (ret && ret != -EOPNOTSUPP)
1196 			goto err_put;
1197 		if (!ret)
1198 			*cur_group++ = &p->groups[2];
1199 	}
1200 
1201 	if (device->ops.process_mad && is_full_dev)
1202 		*cur_group++ = get_counter_table(device, port_num);
1203 
1204 	ret = kobject_add(&p->kobj, coredev->ports_kobj, "%d", port_num);
1205 	if (ret)
1206 		goto err_put;
1207 	ret = sysfs_create_groups(&p->kobj, p->groups_list);
1208 	if (ret)
1209 		goto err_del;
1210 	if (is_full_dev) {
1211 		ret = sysfs_create_groups(&p->kobj, device->ops.port_groups);
1212 		if (ret)
1213 			goto err_groups;
1214 	}
1215 
1216 	list_add_tail(&p->kobj.entry, &coredev->port_list);
1217 	return p;
1218 
1219 err_groups:
1220 	sysfs_remove_groups(&p->kobj, p->groups_list);
1221 err_del:
1222 	kobject_del(&p->kobj);
1223 err_put:
1224 	if (device->port_data && is_full_dev)
1225 		device->port_data[port_num].sysfs = NULL;
1226 	kobject_put(&p->kobj);
1227 	return ERR_PTR(ret);
1228 }
1229 
1230 static void destroy_port(struct ib_core_device *coredev, struct ib_port *port)
1231 {
1232 	bool is_full_dev = &port->ibdev->coredev == coredev;
1233 
1234 	list_del(&port->kobj.entry);
1235 	if (is_full_dev)
1236 		sysfs_remove_groups(&port->kobj, port->ibdev->ops.port_groups);
1237 
1238 	sysfs_remove_groups(&port->kobj, port->groups_list);
1239 	kobject_del(&port->kobj);
1240 
1241 	if (port->ibdev->port_data &&
1242 	    port->ibdev->port_data[port->port_num].sysfs == port)
1243 		port->ibdev->port_data[port->port_num].sysfs = NULL;
1244 
1245 	kobject_put(&port->kobj);
1246 }
1247 
1248 static const char *node_type_string(int node_type)
1249 {
1250 	switch (node_type) {
1251 	case RDMA_NODE_IB_CA:
1252 		return "CA";
1253 	case RDMA_NODE_IB_SWITCH:
1254 		return "switch";
1255 	case RDMA_NODE_IB_ROUTER:
1256 		return "router";
1257 	case RDMA_NODE_RNIC:
1258 		return "RNIC";
1259 	case RDMA_NODE_USNIC:
1260 		return "usNIC";
1261 	case RDMA_NODE_USNIC_UDP:
1262 		return "usNIC UDP";
1263 	case RDMA_NODE_UNSPECIFIED:
1264 		return "unspecified";
1265 	}
1266 	return "<unknown>";
1267 }
1268 
1269 static ssize_t node_type_show(struct device *device,
1270 			      struct device_attribute *attr, char *buf)
1271 {
1272 	struct ib_device *dev = rdma_device_to_ibdev(device);
1273 
1274 	return sysfs_emit(buf, "%u: %s\n", dev->node_type,
1275 			  node_type_string(dev->node_type));
1276 }
1277 static DEVICE_ATTR_RO(node_type);
1278 
1279 static ssize_t sys_image_guid_show(struct device *device,
1280 				   struct device_attribute *dev_attr, char *buf)
1281 {
1282 	struct ib_device *dev = rdma_device_to_ibdev(device);
1283 	__be16 *guid = (__be16 *)&dev->attrs.sys_image_guid;
1284 
1285 	return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1286 			  be16_to_cpu(guid[0]),
1287 			  be16_to_cpu(guid[1]),
1288 			  be16_to_cpu(guid[2]),
1289 			  be16_to_cpu(guid[3]));
1290 }
1291 static DEVICE_ATTR_RO(sys_image_guid);
1292 
1293 static ssize_t node_guid_show(struct device *device,
1294 			      struct device_attribute *attr, char *buf)
1295 {
1296 	struct ib_device *dev = rdma_device_to_ibdev(device);
1297 	__be16 *node_guid = (__be16 *)&dev->node_guid;
1298 
1299 	return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1300 			  be16_to_cpu(node_guid[0]),
1301 			  be16_to_cpu(node_guid[1]),
1302 			  be16_to_cpu(node_guid[2]),
1303 			  be16_to_cpu(node_guid[3]));
1304 }
1305 static DEVICE_ATTR_RO(node_guid);
1306 
1307 static ssize_t node_desc_show(struct device *device,
1308 			      struct device_attribute *attr, char *buf)
1309 {
1310 	struct ib_device *dev = rdma_device_to_ibdev(device);
1311 
1312 	return sysfs_emit(buf, "%.64s\n", dev->node_desc);
1313 }
1314 
1315 static ssize_t node_desc_store(struct device *device,
1316 			       struct device_attribute *attr,
1317 			       const char *buf, size_t count)
1318 {
1319 	struct ib_device *dev = rdma_device_to_ibdev(device);
1320 	struct ib_device_modify desc = {};
1321 	int ret;
1322 
1323 	if (!dev->ops.modify_device)
1324 		return -EOPNOTSUPP;
1325 
1326 	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1327 	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1328 	if (ret)
1329 		return ret;
1330 
1331 	return count;
1332 }
1333 static DEVICE_ATTR_RW(node_desc);
1334 
1335 static ssize_t fw_ver_show(struct device *device, struct device_attribute *attr,
1336 			   char *buf)
1337 {
1338 	struct ib_device *dev = rdma_device_to_ibdev(device);
1339 	char version[IB_FW_VERSION_NAME_MAX] = {};
1340 
1341 	ib_get_device_fw_str(dev, version);
1342 
1343 	return sysfs_emit(buf, "%s\n", version);
1344 }
1345 static DEVICE_ATTR_RO(fw_ver);
1346 
1347 static struct attribute *ib_dev_attrs[] = {
1348 	&dev_attr_node_type.attr,
1349 	&dev_attr_node_guid.attr,
1350 	&dev_attr_sys_image_guid.attr,
1351 	&dev_attr_fw_ver.attr,
1352 	&dev_attr_node_desc.attr,
1353 	NULL,
1354 };
1355 
1356 const struct attribute_group ib_dev_attr_group = {
1357 	.attrs = ib_dev_attrs,
1358 };
1359 
1360 void ib_free_port_attrs(struct ib_core_device *coredev)
1361 {
1362 	struct kobject *p, *t;
1363 
1364 	list_for_each_entry_safe(p, t, &coredev->port_list, entry) {
1365 		struct ib_port *port = container_of(p, struct ib_port, kobj);
1366 
1367 		destroy_gid_attrs(port);
1368 		destroy_port(coredev, port);
1369 	}
1370 
1371 	kobject_put(coredev->ports_kobj);
1372 }
1373 
1374 int ib_setup_port_attrs(struct ib_core_device *coredev)
1375 {
1376 	struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1377 	u32 port_num;
1378 	int ret;
1379 
1380 	coredev->ports_kobj = kobject_create_and_add("ports",
1381 						     &coredev->dev.kobj);
1382 	if (!coredev->ports_kobj)
1383 		return -ENOMEM;
1384 
1385 	rdma_for_each_port (device, port_num) {
1386 		struct ib_port_attr attr;
1387 		struct ib_port *port;
1388 
1389 		ret = ib_query_port(device, port_num, &attr);
1390 		if (ret)
1391 			goto err_put;
1392 
1393 		port = setup_port(coredev, port_num, &attr);
1394 		if (IS_ERR(port)) {
1395 			ret = PTR_ERR(port);
1396 			goto err_put;
1397 		}
1398 
1399 		ret = setup_gid_attrs(port, &attr);
1400 		if (ret)
1401 			goto err_put;
1402 	}
1403 	return 0;
1404 
1405 err_put:
1406 	ib_free_port_attrs(coredev);
1407 	return ret;
1408 }
1409 
1410 /**
1411  * ib_port_register_client_groups - Add an ib_client's attributes to the port
1412  *
1413  * @ibdev: IB device to add counters
1414  * @port_num: valid port number
1415  * @groups: Group list of attributes
1416  *
1417  * Do not use. Only for legacy sysfs compatibility.
1418  */
1419 int ib_port_register_client_groups(struct ib_device *ibdev, u32 port_num,
1420 				   const struct attribute_group **groups)
1421 {
1422 	return sysfs_create_groups(&ibdev->port_data[port_num].sysfs->kobj,
1423 				   groups);
1424 }
1425 EXPORT_SYMBOL(ib_port_register_client_groups);
1426 
1427 void ib_port_unregister_client_groups(struct ib_device *ibdev, u32 port_num,
1428 				      const struct attribute_group **groups)
1429 {
1430 	return sysfs_remove_groups(&ibdev->port_data[port_num].sysfs->kobj,
1431 				   groups);
1432 }
1433 EXPORT_SYMBOL(ib_port_unregister_client_groups);
1434