xref: /linux/drivers/infiniband/core/sysfs.c (revision a8fe58cec351c25e09c393bf46117c0c47b5a17c)
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 
42 #include <rdma/ib_mad.h>
43 #include <rdma/ib_pma.h>
44 
45 struct ib_port;
46 
47 struct gid_attr_group {
48 	struct ib_port		*port;
49 	struct kobject		kobj;
50 	struct attribute_group	ndev;
51 	struct attribute_group	type;
52 };
53 struct ib_port {
54 	struct kobject         kobj;
55 	struct ib_device      *ibdev;
56 	struct gid_attr_group *gid_attr_group;
57 	struct attribute_group gid_group;
58 	struct attribute_group pkey_group;
59 	u8                     port_num;
60 	struct attribute_group *pma_table;
61 };
62 
63 struct port_attribute {
64 	struct attribute attr;
65 	ssize_t (*show)(struct ib_port *, struct port_attribute *, char *buf);
66 	ssize_t (*store)(struct ib_port *, struct port_attribute *,
67 			 const char *buf, size_t count);
68 };
69 
70 #define PORT_ATTR(_name, _mode, _show, _store) \
71 struct port_attribute port_attr_##_name = __ATTR(_name, _mode, _show, _store)
72 
73 #define PORT_ATTR_RO(_name) \
74 struct port_attribute port_attr_##_name = __ATTR_RO(_name)
75 
76 struct port_table_attribute {
77 	struct port_attribute	attr;
78 	char			name[8];
79 	int			index;
80 	__be16			attr_id;
81 };
82 
83 static ssize_t port_attr_show(struct kobject *kobj,
84 			      struct attribute *attr, char *buf)
85 {
86 	struct port_attribute *port_attr =
87 		container_of(attr, struct port_attribute, attr);
88 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
89 
90 	if (!port_attr->show)
91 		return -EIO;
92 
93 	return port_attr->show(p, port_attr, buf);
94 }
95 
96 static const struct sysfs_ops port_sysfs_ops = {
97 	.show = port_attr_show
98 };
99 
100 static ssize_t gid_attr_show(struct kobject *kobj,
101 			     struct attribute *attr, char *buf)
102 {
103 	struct port_attribute *port_attr =
104 		container_of(attr, struct port_attribute, attr);
105 	struct ib_port *p = container_of(kobj, struct gid_attr_group,
106 					 kobj)->port;
107 
108 	if (!port_attr->show)
109 		return -EIO;
110 
111 	return port_attr->show(p, port_attr, buf);
112 }
113 
114 static const struct sysfs_ops gid_attr_sysfs_ops = {
115 	.show = gid_attr_show
116 };
117 
118 static ssize_t state_show(struct ib_port *p, struct port_attribute *unused,
119 			  char *buf)
120 {
121 	struct ib_port_attr attr;
122 	ssize_t ret;
123 
124 	static const char *state_name[] = {
125 		[IB_PORT_NOP]		= "NOP",
126 		[IB_PORT_DOWN]		= "DOWN",
127 		[IB_PORT_INIT]		= "INIT",
128 		[IB_PORT_ARMED]		= "ARMED",
129 		[IB_PORT_ACTIVE]	= "ACTIVE",
130 		[IB_PORT_ACTIVE_DEFER]	= "ACTIVE_DEFER"
131 	};
132 
133 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
134 	if (ret)
135 		return ret;
136 
137 	return sprintf(buf, "%d: %s\n", attr.state,
138 		       attr.state >= 0 && attr.state < ARRAY_SIZE(state_name) ?
139 		       state_name[attr.state] : "UNKNOWN");
140 }
141 
142 static ssize_t lid_show(struct ib_port *p, struct port_attribute *unused,
143 			char *buf)
144 {
145 	struct ib_port_attr attr;
146 	ssize_t ret;
147 
148 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
149 	if (ret)
150 		return ret;
151 
152 	return sprintf(buf, "0x%x\n", attr.lid);
153 }
154 
155 static ssize_t lid_mask_count_show(struct ib_port *p,
156 				   struct port_attribute *unused,
157 				   char *buf)
158 {
159 	struct ib_port_attr attr;
160 	ssize_t ret;
161 
162 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
163 	if (ret)
164 		return ret;
165 
166 	return sprintf(buf, "%d\n", attr.lmc);
167 }
168 
169 static ssize_t sm_lid_show(struct ib_port *p, struct port_attribute *unused,
170 			   char *buf)
171 {
172 	struct ib_port_attr attr;
173 	ssize_t ret;
174 
175 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
176 	if (ret)
177 		return ret;
178 
179 	return sprintf(buf, "0x%x\n", attr.sm_lid);
180 }
181 
182 static ssize_t sm_sl_show(struct ib_port *p, struct port_attribute *unused,
183 			  char *buf)
184 {
185 	struct ib_port_attr attr;
186 	ssize_t ret;
187 
188 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
189 	if (ret)
190 		return ret;
191 
192 	return sprintf(buf, "%d\n", attr.sm_sl);
193 }
194 
195 static ssize_t cap_mask_show(struct ib_port *p, struct port_attribute *unused,
196 			     char *buf)
197 {
198 	struct ib_port_attr attr;
199 	ssize_t ret;
200 
201 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
202 	if (ret)
203 		return ret;
204 
205 	return sprintf(buf, "0x%08x\n", attr.port_cap_flags);
206 }
207 
208 static ssize_t rate_show(struct ib_port *p, struct port_attribute *unused,
209 			 char *buf)
210 {
211 	struct ib_port_attr attr;
212 	char *speed = "";
213 	int rate;		/* in deci-Gb/sec */
214 	ssize_t ret;
215 
216 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
217 	if (ret)
218 		return ret;
219 
220 	switch (attr.active_speed) {
221 	case IB_SPEED_DDR:
222 		speed = " DDR";
223 		rate = 50;
224 		break;
225 	case IB_SPEED_QDR:
226 		speed = " QDR";
227 		rate = 100;
228 		break;
229 	case IB_SPEED_FDR10:
230 		speed = " FDR10";
231 		rate = 100;
232 		break;
233 	case IB_SPEED_FDR:
234 		speed = " FDR";
235 		rate = 140;
236 		break;
237 	case IB_SPEED_EDR:
238 		speed = " EDR";
239 		rate = 250;
240 		break;
241 	case IB_SPEED_SDR:
242 	default:		/* default to SDR for invalid rates */
243 		rate = 25;
244 		break;
245 	}
246 
247 	rate *= ib_width_enum_to_int(attr.active_width);
248 	if (rate < 0)
249 		return -EINVAL;
250 
251 	return sprintf(buf, "%d%s Gb/sec (%dX%s)\n",
252 		       rate / 10, rate % 10 ? ".5" : "",
253 		       ib_width_enum_to_int(attr.active_width), speed);
254 }
255 
256 static ssize_t phys_state_show(struct ib_port *p, struct port_attribute *unused,
257 			       char *buf)
258 {
259 	struct ib_port_attr attr;
260 
261 	ssize_t ret;
262 
263 	ret = ib_query_port(p->ibdev, p->port_num, &attr);
264 	if (ret)
265 		return ret;
266 
267 	switch (attr.phys_state) {
268 	case 1:  return sprintf(buf, "1: Sleep\n");
269 	case 2:  return sprintf(buf, "2: Polling\n");
270 	case 3:  return sprintf(buf, "3: Disabled\n");
271 	case 4:  return sprintf(buf, "4: PortConfigurationTraining\n");
272 	case 5:  return sprintf(buf, "5: LinkUp\n");
273 	case 6:  return sprintf(buf, "6: LinkErrorRecovery\n");
274 	case 7:  return sprintf(buf, "7: Phy Test\n");
275 	default: return sprintf(buf, "%d: <unknown>\n", attr.phys_state);
276 	}
277 }
278 
279 static ssize_t link_layer_show(struct ib_port *p, struct port_attribute *unused,
280 			       char *buf)
281 {
282 	switch (rdma_port_get_link_layer(p->ibdev, p->port_num)) {
283 	case IB_LINK_LAYER_INFINIBAND:
284 		return sprintf(buf, "%s\n", "InfiniBand");
285 	case IB_LINK_LAYER_ETHERNET:
286 		return sprintf(buf, "%s\n", "Ethernet");
287 	default:
288 		return sprintf(buf, "%s\n", "Unknown");
289 	}
290 }
291 
292 static PORT_ATTR_RO(state);
293 static PORT_ATTR_RO(lid);
294 static PORT_ATTR_RO(lid_mask_count);
295 static PORT_ATTR_RO(sm_lid);
296 static PORT_ATTR_RO(sm_sl);
297 static PORT_ATTR_RO(cap_mask);
298 static PORT_ATTR_RO(rate);
299 static PORT_ATTR_RO(phys_state);
300 static PORT_ATTR_RO(link_layer);
301 
302 static struct attribute *port_default_attrs[] = {
303 	&port_attr_state.attr,
304 	&port_attr_lid.attr,
305 	&port_attr_lid_mask_count.attr,
306 	&port_attr_sm_lid.attr,
307 	&port_attr_sm_sl.attr,
308 	&port_attr_cap_mask.attr,
309 	&port_attr_rate.attr,
310 	&port_attr_phys_state.attr,
311 	&port_attr_link_layer.attr,
312 	NULL
313 };
314 
315 static size_t print_ndev(struct ib_gid_attr *gid_attr, char *buf)
316 {
317 	if (!gid_attr->ndev)
318 		return -EINVAL;
319 
320 	return sprintf(buf, "%s\n", gid_attr->ndev->name);
321 }
322 
323 static size_t print_gid_type(struct ib_gid_attr *gid_attr, char *buf)
324 {
325 	return sprintf(buf, "%s\n", ib_cache_gid_type_str(gid_attr->gid_type));
326 }
327 
328 static ssize_t _show_port_gid_attr(struct ib_port *p,
329 				   struct port_attribute *attr,
330 				   char *buf,
331 				   size_t (*print)(struct ib_gid_attr *gid_attr,
332 						   char *buf))
333 {
334 	struct port_table_attribute *tab_attr =
335 		container_of(attr, struct port_table_attribute, attr);
336 	union ib_gid gid;
337 	struct ib_gid_attr gid_attr = {};
338 	ssize_t ret;
339 	va_list args;
340 
341 	ret = ib_query_gid(p->ibdev, p->port_num, tab_attr->index, &gid,
342 			   &gid_attr);
343 	if (ret)
344 		goto err;
345 
346 	ret = print(&gid_attr, buf);
347 
348 err:
349 	if (gid_attr.ndev)
350 		dev_put(gid_attr.ndev);
351 	va_end(args);
352 	return ret;
353 }
354 
355 static ssize_t show_port_gid(struct ib_port *p, struct port_attribute *attr,
356 			     char *buf)
357 {
358 	struct port_table_attribute *tab_attr =
359 		container_of(attr, struct port_table_attribute, attr);
360 	union ib_gid gid;
361 	ssize_t ret;
362 
363 	ret = ib_query_gid(p->ibdev, p->port_num, tab_attr->index, &gid, NULL);
364 	if (ret)
365 		return ret;
366 
367 	return sprintf(buf, "%pI6\n", gid.raw);
368 }
369 
370 static ssize_t show_port_gid_attr_ndev(struct ib_port *p,
371 				       struct port_attribute *attr, char *buf)
372 {
373 	return _show_port_gid_attr(p, attr, buf, print_ndev);
374 }
375 
376 static ssize_t show_port_gid_attr_gid_type(struct ib_port *p,
377 					   struct port_attribute *attr,
378 					   char *buf)
379 {
380 	return _show_port_gid_attr(p, attr, buf, print_gid_type);
381 }
382 
383 static ssize_t show_port_pkey(struct ib_port *p, struct port_attribute *attr,
384 			      char *buf)
385 {
386 	struct port_table_attribute *tab_attr =
387 		container_of(attr, struct port_table_attribute, attr);
388 	u16 pkey;
389 	ssize_t ret;
390 
391 	ret = ib_query_pkey(p->ibdev, p->port_num, tab_attr->index, &pkey);
392 	if (ret)
393 		return ret;
394 
395 	return sprintf(buf, "0x%04x\n", pkey);
396 }
397 
398 #define PORT_PMA_ATTR(_name, _counter, _width, _offset)			\
399 struct port_table_attribute port_pma_attr_##_name = {			\
400 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
401 	.index = (_offset) | ((_width) << 16) | ((_counter) << 24),	\
402 	.attr_id = IB_PMA_PORT_COUNTERS ,				\
403 }
404 
405 #define PORT_PMA_ATTR_EXT(_name, _width, _offset)			\
406 struct port_table_attribute port_pma_attr_ext_##_name = {		\
407 	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
408 	.index = (_offset) | ((_width) << 16),				\
409 	.attr_id = IB_PMA_PORT_COUNTERS_EXT ,				\
410 }
411 
412 /*
413  * Get a Perfmgmt MAD block of data.
414  * Returns error code or the number of bytes retrieved.
415  */
416 static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
417 		void *data, int offset, size_t size)
418 {
419 	struct ib_mad *in_mad;
420 	struct ib_mad *out_mad;
421 	size_t mad_size = sizeof(*out_mad);
422 	u16 out_mad_pkey_index = 0;
423 	ssize_t ret;
424 
425 	if (!dev->process_mad)
426 		return -ENOSYS;
427 
428 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
429 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
430 	if (!in_mad || !out_mad) {
431 		ret = -ENOMEM;
432 		goto out;
433 	}
434 
435 	in_mad->mad_hdr.base_version  = 1;
436 	in_mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_PERF_MGMT;
437 	in_mad->mad_hdr.class_version = 1;
438 	in_mad->mad_hdr.method        = IB_MGMT_METHOD_GET;
439 	in_mad->mad_hdr.attr_id       = attr;
440 
441 	if (attr != IB_PMA_CLASS_PORT_INFO)
442 		in_mad->data[41] = port_num;	/* PortSelect field */
443 
444 	if ((dev->process_mad(dev, IB_MAD_IGNORE_MKEY,
445 		 port_num, NULL, NULL,
446 		 (const struct ib_mad_hdr *)in_mad, mad_size,
447 		 (struct ib_mad_hdr *)out_mad, &mad_size,
448 		 &out_mad_pkey_index) &
449 	     (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
450 	    (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
451 		ret = -EINVAL;
452 		goto out;
453 	}
454 	memcpy(data, out_mad->data + offset, size);
455 	ret = size;
456 out:
457 	kfree(in_mad);
458 	kfree(out_mad);
459 	return ret;
460 }
461 
462 static ssize_t show_pma_counter(struct ib_port *p, struct port_attribute *attr,
463 				char *buf)
464 {
465 	struct port_table_attribute *tab_attr =
466 		container_of(attr, struct port_table_attribute, attr);
467 	int offset = tab_attr->index & 0xffff;
468 	int width  = (tab_attr->index >> 16) & 0xff;
469 	ssize_t ret;
470 	u8 data[8];
471 
472 	ret = get_perf_mad(p->ibdev, p->port_num, tab_attr->attr_id, &data,
473 			40 + offset / 8, sizeof(data));
474 	if (ret < 0)
475 		return sprintf(buf, "N/A (no PMA)\n");
476 
477 	switch (width) {
478 	case 4:
479 		ret = sprintf(buf, "%u\n", (*data >>
480 					    (4 - (offset % 8))) & 0xf);
481 		break;
482 	case 8:
483 		ret = sprintf(buf, "%u\n", *data);
484 		break;
485 	case 16:
486 		ret = sprintf(buf, "%u\n",
487 			      be16_to_cpup((__be16 *)data));
488 		break;
489 	case 32:
490 		ret = sprintf(buf, "%u\n",
491 			      be32_to_cpup((__be32 *)data));
492 		break;
493 	case 64:
494 		ret = sprintf(buf, "%llu\n",
495 				be64_to_cpup((__be64 *)data));
496 		break;
497 
498 	default:
499 		ret = 0;
500 	}
501 
502 	return ret;
503 }
504 
505 static PORT_PMA_ATTR(symbol_error		    ,  0, 16,  32);
506 static PORT_PMA_ATTR(link_error_recovery	    ,  1,  8,  48);
507 static PORT_PMA_ATTR(link_downed		    ,  2,  8,  56);
508 static PORT_PMA_ATTR(port_rcv_errors		    ,  3, 16,  64);
509 static PORT_PMA_ATTR(port_rcv_remote_physical_errors,  4, 16,  80);
510 static PORT_PMA_ATTR(port_rcv_switch_relay_errors   ,  5, 16,  96);
511 static PORT_PMA_ATTR(port_xmit_discards		    ,  6, 16, 112);
512 static PORT_PMA_ATTR(port_xmit_constraint_errors    ,  7,  8, 128);
513 static PORT_PMA_ATTR(port_rcv_constraint_errors	    ,  8,  8, 136);
514 static PORT_PMA_ATTR(local_link_integrity_errors    ,  9,  4, 152);
515 static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10,  4, 156);
516 static PORT_PMA_ATTR(VL15_dropped		    , 11, 16, 176);
517 static PORT_PMA_ATTR(port_xmit_data		    , 12, 32, 192);
518 static PORT_PMA_ATTR(port_rcv_data		    , 13, 32, 224);
519 static PORT_PMA_ATTR(port_xmit_packets		    , 14, 32, 256);
520 static PORT_PMA_ATTR(port_rcv_packets		    , 15, 32, 288);
521 
522 /*
523  * Counters added by extended set
524  */
525 static PORT_PMA_ATTR_EXT(port_xmit_data		    , 64,  64);
526 static PORT_PMA_ATTR_EXT(port_rcv_data		    , 64, 128);
527 static PORT_PMA_ATTR_EXT(port_xmit_packets	    , 64, 192);
528 static PORT_PMA_ATTR_EXT(port_rcv_packets	    , 64, 256);
529 static PORT_PMA_ATTR_EXT(unicast_xmit_packets	    , 64, 320);
530 static PORT_PMA_ATTR_EXT(unicast_rcv_packets	    , 64, 384);
531 static PORT_PMA_ATTR_EXT(multicast_xmit_packets	    , 64, 448);
532 static PORT_PMA_ATTR_EXT(multicast_rcv_packets	    , 64, 512);
533 
534 static struct attribute *pma_attrs[] = {
535 	&port_pma_attr_symbol_error.attr.attr,
536 	&port_pma_attr_link_error_recovery.attr.attr,
537 	&port_pma_attr_link_downed.attr.attr,
538 	&port_pma_attr_port_rcv_errors.attr.attr,
539 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
540 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
541 	&port_pma_attr_port_xmit_discards.attr.attr,
542 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
543 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
544 	&port_pma_attr_local_link_integrity_errors.attr.attr,
545 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
546 	&port_pma_attr_VL15_dropped.attr.attr,
547 	&port_pma_attr_port_xmit_data.attr.attr,
548 	&port_pma_attr_port_rcv_data.attr.attr,
549 	&port_pma_attr_port_xmit_packets.attr.attr,
550 	&port_pma_attr_port_rcv_packets.attr.attr,
551 	NULL
552 };
553 
554 static struct attribute *pma_attrs_ext[] = {
555 	&port_pma_attr_symbol_error.attr.attr,
556 	&port_pma_attr_link_error_recovery.attr.attr,
557 	&port_pma_attr_link_downed.attr.attr,
558 	&port_pma_attr_port_rcv_errors.attr.attr,
559 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
560 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
561 	&port_pma_attr_port_xmit_discards.attr.attr,
562 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
563 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
564 	&port_pma_attr_local_link_integrity_errors.attr.attr,
565 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
566 	&port_pma_attr_VL15_dropped.attr.attr,
567 	&port_pma_attr_ext_port_xmit_data.attr.attr,
568 	&port_pma_attr_ext_port_rcv_data.attr.attr,
569 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
570 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
571 	&port_pma_attr_ext_unicast_rcv_packets.attr.attr,
572 	&port_pma_attr_ext_unicast_xmit_packets.attr.attr,
573 	&port_pma_attr_ext_multicast_rcv_packets.attr.attr,
574 	&port_pma_attr_ext_multicast_xmit_packets.attr.attr,
575 	NULL
576 };
577 
578 static struct attribute *pma_attrs_noietf[] = {
579 	&port_pma_attr_symbol_error.attr.attr,
580 	&port_pma_attr_link_error_recovery.attr.attr,
581 	&port_pma_attr_link_downed.attr.attr,
582 	&port_pma_attr_port_rcv_errors.attr.attr,
583 	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
584 	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
585 	&port_pma_attr_port_xmit_discards.attr.attr,
586 	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
587 	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
588 	&port_pma_attr_local_link_integrity_errors.attr.attr,
589 	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
590 	&port_pma_attr_VL15_dropped.attr.attr,
591 	&port_pma_attr_ext_port_xmit_data.attr.attr,
592 	&port_pma_attr_ext_port_rcv_data.attr.attr,
593 	&port_pma_attr_ext_port_xmit_packets.attr.attr,
594 	&port_pma_attr_ext_port_rcv_packets.attr.attr,
595 	NULL
596 };
597 
598 static struct attribute_group pma_group = {
599 	.name  = "counters",
600 	.attrs  = pma_attrs
601 };
602 
603 static struct attribute_group pma_group_ext = {
604 	.name  = "counters",
605 	.attrs  = pma_attrs_ext
606 };
607 
608 static struct attribute_group pma_group_noietf = {
609 	.name  = "counters",
610 	.attrs  = pma_attrs_noietf
611 };
612 
613 static void ib_port_release(struct kobject *kobj)
614 {
615 	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
616 	struct attribute *a;
617 	int i;
618 
619 	if (p->gid_group.attrs) {
620 		for (i = 0; (a = p->gid_group.attrs[i]); ++i)
621 			kfree(a);
622 
623 		kfree(p->gid_group.attrs);
624 	}
625 
626 	if (p->pkey_group.attrs) {
627 		for (i = 0; (a = p->pkey_group.attrs[i]); ++i)
628 			kfree(a);
629 
630 		kfree(p->pkey_group.attrs);
631 	}
632 
633 	kfree(p);
634 }
635 
636 static void ib_port_gid_attr_release(struct kobject *kobj)
637 {
638 	struct gid_attr_group *g = container_of(kobj, struct gid_attr_group,
639 						kobj);
640 	struct attribute *a;
641 	int i;
642 
643 	if (g->ndev.attrs) {
644 		for (i = 0; (a = g->ndev.attrs[i]); ++i)
645 			kfree(a);
646 
647 		kfree(g->ndev.attrs);
648 	}
649 
650 	if (g->type.attrs) {
651 		for (i = 0; (a = g->type.attrs[i]); ++i)
652 			kfree(a);
653 
654 		kfree(g->type.attrs);
655 	}
656 
657 	kfree(g);
658 }
659 
660 static struct kobj_type port_type = {
661 	.release       = ib_port_release,
662 	.sysfs_ops     = &port_sysfs_ops,
663 	.default_attrs = port_default_attrs
664 };
665 
666 static struct kobj_type gid_attr_type = {
667 	.sysfs_ops      = &gid_attr_sysfs_ops,
668 	.release        = ib_port_gid_attr_release
669 };
670 
671 static struct attribute **
672 alloc_group_attrs(ssize_t (*show)(struct ib_port *,
673 				  struct port_attribute *, char *buf),
674 		  int len)
675 {
676 	struct attribute **tab_attr;
677 	struct port_table_attribute *element;
678 	int i;
679 
680 	tab_attr = kcalloc(1 + len, sizeof(struct attribute *), GFP_KERNEL);
681 	if (!tab_attr)
682 		return NULL;
683 
684 	for (i = 0; i < len; i++) {
685 		element = kzalloc(sizeof(struct port_table_attribute),
686 				  GFP_KERNEL);
687 		if (!element)
688 			goto err;
689 
690 		if (snprintf(element->name, sizeof(element->name),
691 			     "%d", i) >= sizeof(element->name)) {
692 			kfree(element);
693 			goto err;
694 		}
695 
696 		element->attr.attr.name  = element->name;
697 		element->attr.attr.mode  = S_IRUGO;
698 		element->attr.show       = show;
699 		element->index		 = i;
700 		sysfs_attr_init(&element->attr.attr);
701 
702 		tab_attr[i] = &element->attr.attr;
703 	}
704 
705 	return tab_attr;
706 
707 err:
708 	while (--i >= 0)
709 		kfree(tab_attr[i]);
710 	kfree(tab_attr);
711 	return NULL;
712 }
713 
714 /*
715  * Figure out which counter table to use depending on
716  * the device capabilities.
717  */
718 static struct attribute_group *get_counter_table(struct ib_device *dev,
719 						 int port_num)
720 {
721 	struct ib_class_port_info cpi;
722 
723 	if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
724 				&cpi, 40, sizeof(cpi)) >= 0) {
725 
726 		if (cpi.capability_mask && IB_PMA_CLASS_CAP_EXT_WIDTH)
727 			/* We have extended counters */
728 			return &pma_group_ext;
729 
730 		if (cpi.capability_mask && IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
731 			/* But not the IETF ones */
732 			return &pma_group_noietf;
733 	}
734 
735 	/* Fall back to normal counters */
736 	return &pma_group;
737 }
738 
739 static int add_port(struct ib_device *device, int port_num,
740 		    int (*port_callback)(struct ib_device *,
741 					 u8, struct kobject *))
742 {
743 	struct ib_port *p;
744 	struct ib_port_attr attr;
745 	int i;
746 	int ret;
747 
748 	ret = ib_query_port(device, port_num, &attr);
749 	if (ret)
750 		return ret;
751 
752 	p = kzalloc(sizeof *p, GFP_KERNEL);
753 	if (!p)
754 		return -ENOMEM;
755 
756 	p->ibdev      = device;
757 	p->port_num   = port_num;
758 
759 	ret = kobject_init_and_add(&p->kobj, &port_type,
760 				   device->ports_parent,
761 				   "%d", port_num);
762 	if (ret) {
763 		kfree(p);
764 		return ret;
765 	}
766 
767 	p->gid_attr_group = kzalloc(sizeof(*p->gid_attr_group), GFP_KERNEL);
768 	if (!p->gid_attr_group) {
769 		ret = -ENOMEM;
770 		goto err_put;
771 	}
772 
773 	p->gid_attr_group->port = p;
774 	ret = kobject_init_and_add(&p->gid_attr_group->kobj, &gid_attr_type,
775 				   &p->kobj, "gid_attrs");
776 	if (ret) {
777 		kfree(p->gid_attr_group);
778 		goto err_put;
779 	}
780 
781 	p->pma_table = get_counter_table(device, port_num);
782 	ret = sysfs_create_group(&p->kobj, p->pma_table);
783 	if (ret)
784 		goto err_put_gid_attrs;
785 
786 	p->gid_group.name  = "gids";
787 	p->gid_group.attrs = alloc_group_attrs(show_port_gid, attr.gid_tbl_len);
788 	if (!p->gid_group.attrs) {
789 		ret = -ENOMEM;
790 		goto err_remove_pma;
791 	}
792 
793 	ret = sysfs_create_group(&p->kobj, &p->gid_group);
794 	if (ret)
795 		goto err_free_gid;
796 
797 	p->gid_attr_group->ndev.name = "ndevs";
798 	p->gid_attr_group->ndev.attrs = alloc_group_attrs(show_port_gid_attr_ndev,
799 							  attr.gid_tbl_len);
800 	if (!p->gid_attr_group->ndev.attrs) {
801 		ret = -ENOMEM;
802 		goto err_remove_gid;
803 	}
804 
805 	ret = sysfs_create_group(&p->gid_attr_group->kobj,
806 				 &p->gid_attr_group->ndev);
807 	if (ret)
808 		goto err_free_gid_ndev;
809 
810 	p->gid_attr_group->type.name = "types";
811 	p->gid_attr_group->type.attrs = alloc_group_attrs(show_port_gid_attr_gid_type,
812 							  attr.gid_tbl_len);
813 	if (!p->gid_attr_group->type.attrs) {
814 		ret = -ENOMEM;
815 		goto err_remove_gid_ndev;
816 	}
817 
818 	ret = sysfs_create_group(&p->gid_attr_group->kobj,
819 				 &p->gid_attr_group->type);
820 	if (ret)
821 		goto err_free_gid_type;
822 
823 	p->pkey_group.name  = "pkeys";
824 	p->pkey_group.attrs = alloc_group_attrs(show_port_pkey,
825 						attr.pkey_tbl_len);
826 	if (!p->pkey_group.attrs) {
827 		ret = -ENOMEM;
828 		goto err_remove_gid_type;
829 	}
830 
831 	ret = sysfs_create_group(&p->kobj, &p->pkey_group);
832 	if (ret)
833 		goto err_free_pkey;
834 
835 	if (port_callback) {
836 		ret = port_callback(device, port_num, &p->kobj);
837 		if (ret)
838 			goto err_remove_pkey;
839 	}
840 
841 	list_add_tail(&p->kobj.entry, &device->port_list);
842 
843 	kobject_uevent(&p->kobj, KOBJ_ADD);
844 	return 0;
845 
846 err_remove_pkey:
847 	sysfs_remove_group(&p->kobj, &p->pkey_group);
848 
849 err_free_pkey:
850 	for (i = 0; i < attr.pkey_tbl_len; ++i)
851 		kfree(p->pkey_group.attrs[i]);
852 
853 	kfree(p->pkey_group.attrs);
854 	p->pkey_group.attrs = NULL;
855 
856 err_remove_gid_type:
857 	sysfs_remove_group(&p->gid_attr_group->kobj,
858 			   &p->gid_attr_group->type);
859 
860 err_free_gid_type:
861 	for (i = 0; i < attr.gid_tbl_len; ++i)
862 		kfree(p->gid_attr_group->type.attrs[i]);
863 
864 	kfree(p->gid_attr_group->type.attrs);
865 	p->gid_attr_group->type.attrs = NULL;
866 
867 err_remove_gid_ndev:
868 	sysfs_remove_group(&p->gid_attr_group->kobj,
869 			   &p->gid_attr_group->ndev);
870 
871 err_free_gid_ndev:
872 	for (i = 0; i < attr.gid_tbl_len; ++i)
873 		kfree(p->gid_attr_group->ndev.attrs[i]);
874 
875 	kfree(p->gid_attr_group->ndev.attrs);
876 	p->gid_attr_group->ndev.attrs = NULL;
877 
878 err_remove_gid:
879 	sysfs_remove_group(&p->kobj, &p->gid_group);
880 
881 err_free_gid:
882 	for (i = 0; i < attr.gid_tbl_len; ++i)
883 		kfree(p->gid_group.attrs[i]);
884 
885 	kfree(p->gid_group.attrs);
886 	p->gid_group.attrs = NULL;
887 
888 err_remove_pma:
889 	sysfs_remove_group(&p->kobj, p->pma_table);
890 
891 err_put_gid_attrs:
892 	kobject_put(&p->gid_attr_group->kobj);
893 
894 err_put:
895 	kobject_put(&p->kobj);
896 	return ret;
897 }
898 
899 static ssize_t show_node_type(struct device *device,
900 			      struct device_attribute *attr, char *buf)
901 {
902 	struct ib_device *dev = container_of(device, struct ib_device, dev);
903 
904 	switch (dev->node_type) {
905 	case RDMA_NODE_IB_CA:	  return sprintf(buf, "%d: CA\n", dev->node_type);
906 	case RDMA_NODE_RNIC:	  return sprintf(buf, "%d: RNIC\n", dev->node_type);
907 	case RDMA_NODE_USNIC:	  return sprintf(buf, "%d: usNIC\n", dev->node_type);
908 	case RDMA_NODE_USNIC_UDP: return sprintf(buf, "%d: usNIC UDP\n", dev->node_type);
909 	case RDMA_NODE_IB_SWITCH: return sprintf(buf, "%d: switch\n", dev->node_type);
910 	case RDMA_NODE_IB_ROUTER: return sprintf(buf, "%d: router\n", dev->node_type);
911 	default:		  return sprintf(buf, "%d: <unknown>\n", dev->node_type);
912 	}
913 }
914 
915 static ssize_t show_sys_image_guid(struct device *device,
916 				   struct device_attribute *dev_attr, char *buf)
917 {
918 	struct ib_device *dev = container_of(device, struct ib_device, dev);
919 
920 	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
921 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[0]),
922 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[1]),
923 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[2]),
924 		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[3]));
925 }
926 
927 static ssize_t show_node_guid(struct device *device,
928 			      struct device_attribute *attr, char *buf)
929 {
930 	struct ib_device *dev = container_of(device, struct ib_device, dev);
931 
932 	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
933 		       be16_to_cpu(((__be16 *) &dev->node_guid)[0]),
934 		       be16_to_cpu(((__be16 *) &dev->node_guid)[1]),
935 		       be16_to_cpu(((__be16 *) &dev->node_guid)[2]),
936 		       be16_to_cpu(((__be16 *) &dev->node_guid)[3]));
937 }
938 
939 static ssize_t show_node_desc(struct device *device,
940 			      struct device_attribute *attr, char *buf)
941 {
942 	struct ib_device *dev = container_of(device, struct ib_device, dev);
943 
944 	return sprintf(buf, "%.64s\n", dev->node_desc);
945 }
946 
947 static ssize_t set_node_desc(struct device *device,
948 			     struct device_attribute *attr,
949 			     const char *buf, size_t count)
950 {
951 	struct ib_device *dev = container_of(device, struct ib_device, dev);
952 	struct ib_device_modify desc = {};
953 	int ret;
954 
955 	if (!dev->modify_device)
956 		return -EIO;
957 
958 	memcpy(desc.node_desc, buf, min_t(int, count, 64));
959 	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
960 	if (ret)
961 		return ret;
962 
963 	return count;
964 }
965 
966 static DEVICE_ATTR(node_type, S_IRUGO, show_node_type, NULL);
967 static DEVICE_ATTR(sys_image_guid, S_IRUGO, show_sys_image_guid, NULL);
968 static DEVICE_ATTR(node_guid, S_IRUGO, show_node_guid, NULL);
969 static DEVICE_ATTR(node_desc, S_IRUGO | S_IWUSR, show_node_desc, set_node_desc);
970 
971 static struct device_attribute *ib_class_attributes[] = {
972 	&dev_attr_node_type,
973 	&dev_attr_sys_image_guid,
974 	&dev_attr_node_guid,
975 	&dev_attr_node_desc
976 };
977 
978 /* Show a given an attribute in the statistics group */
979 static ssize_t show_protocol_stat(const struct device *device,
980 			    struct device_attribute *attr, char *buf,
981 			    unsigned offset)
982 {
983 	struct ib_device *dev = container_of(device, struct ib_device, dev);
984 	union rdma_protocol_stats stats;
985 	ssize_t ret;
986 
987 	ret = dev->get_protocol_stats(dev, &stats);
988 	if (ret)
989 		return ret;
990 
991 	return sprintf(buf, "%llu\n",
992 		       (unsigned long long) ((u64 *) &stats)[offset]);
993 }
994 
995 /* generate a read-only iwarp statistics attribute */
996 #define IW_STATS_ENTRY(name)						\
997 static ssize_t show_##name(struct device *device,			\
998 			   struct device_attribute *attr, char *buf)	\
999 {									\
1000 	return show_protocol_stat(device, attr, buf,			\
1001 				  offsetof(struct iw_protocol_stats, name) / \
1002 				  sizeof (u64));			\
1003 }									\
1004 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
1005 
1006 IW_STATS_ENTRY(ipInReceives);
1007 IW_STATS_ENTRY(ipInHdrErrors);
1008 IW_STATS_ENTRY(ipInTooBigErrors);
1009 IW_STATS_ENTRY(ipInNoRoutes);
1010 IW_STATS_ENTRY(ipInAddrErrors);
1011 IW_STATS_ENTRY(ipInUnknownProtos);
1012 IW_STATS_ENTRY(ipInTruncatedPkts);
1013 IW_STATS_ENTRY(ipInDiscards);
1014 IW_STATS_ENTRY(ipInDelivers);
1015 IW_STATS_ENTRY(ipOutForwDatagrams);
1016 IW_STATS_ENTRY(ipOutRequests);
1017 IW_STATS_ENTRY(ipOutDiscards);
1018 IW_STATS_ENTRY(ipOutNoRoutes);
1019 IW_STATS_ENTRY(ipReasmTimeout);
1020 IW_STATS_ENTRY(ipReasmReqds);
1021 IW_STATS_ENTRY(ipReasmOKs);
1022 IW_STATS_ENTRY(ipReasmFails);
1023 IW_STATS_ENTRY(ipFragOKs);
1024 IW_STATS_ENTRY(ipFragFails);
1025 IW_STATS_ENTRY(ipFragCreates);
1026 IW_STATS_ENTRY(ipInMcastPkts);
1027 IW_STATS_ENTRY(ipOutMcastPkts);
1028 IW_STATS_ENTRY(ipInBcastPkts);
1029 IW_STATS_ENTRY(ipOutBcastPkts);
1030 IW_STATS_ENTRY(tcpRtoAlgorithm);
1031 IW_STATS_ENTRY(tcpRtoMin);
1032 IW_STATS_ENTRY(tcpRtoMax);
1033 IW_STATS_ENTRY(tcpMaxConn);
1034 IW_STATS_ENTRY(tcpActiveOpens);
1035 IW_STATS_ENTRY(tcpPassiveOpens);
1036 IW_STATS_ENTRY(tcpAttemptFails);
1037 IW_STATS_ENTRY(tcpEstabResets);
1038 IW_STATS_ENTRY(tcpCurrEstab);
1039 IW_STATS_ENTRY(tcpInSegs);
1040 IW_STATS_ENTRY(tcpOutSegs);
1041 IW_STATS_ENTRY(tcpRetransSegs);
1042 IW_STATS_ENTRY(tcpInErrs);
1043 IW_STATS_ENTRY(tcpOutRsts);
1044 
1045 static struct attribute *iw_proto_stats_attrs[] = {
1046 	&dev_attr_ipInReceives.attr,
1047 	&dev_attr_ipInHdrErrors.attr,
1048 	&dev_attr_ipInTooBigErrors.attr,
1049 	&dev_attr_ipInNoRoutes.attr,
1050 	&dev_attr_ipInAddrErrors.attr,
1051 	&dev_attr_ipInUnknownProtos.attr,
1052 	&dev_attr_ipInTruncatedPkts.attr,
1053 	&dev_attr_ipInDiscards.attr,
1054 	&dev_attr_ipInDelivers.attr,
1055 	&dev_attr_ipOutForwDatagrams.attr,
1056 	&dev_attr_ipOutRequests.attr,
1057 	&dev_attr_ipOutDiscards.attr,
1058 	&dev_attr_ipOutNoRoutes.attr,
1059 	&dev_attr_ipReasmTimeout.attr,
1060 	&dev_attr_ipReasmReqds.attr,
1061 	&dev_attr_ipReasmOKs.attr,
1062 	&dev_attr_ipReasmFails.attr,
1063 	&dev_attr_ipFragOKs.attr,
1064 	&dev_attr_ipFragFails.attr,
1065 	&dev_attr_ipFragCreates.attr,
1066 	&dev_attr_ipInMcastPkts.attr,
1067 	&dev_attr_ipOutMcastPkts.attr,
1068 	&dev_attr_ipInBcastPkts.attr,
1069 	&dev_attr_ipOutBcastPkts.attr,
1070 	&dev_attr_tcpRtoAlgorithm.attr,
1071 	&dev_attr_tcpRtoMin.attr,
1072 	&dev_attr_tcpRtoMax.attr,
1073 	&dev_attr_tcpMaxConn.attr,
1074 	&dev_attr_tcpActiveOpens.attr,
1075 	&dev_attr_tcpPassiveOpens.attr,
1076 	&dev_attr_tcpAttemptFails.attr,
1077 	&dev_attr_tcpEstabResets.attr,
1078 	&dev_attr_tcpCurrEstab.attr,
1079 	&dev_attr_tcpInSegs.attr,
1080 	&dev_attr_tcpOutSegs.attr,
1081 	&dev_attr_tcpRetransSegs.attr,
1082 	&dev_attr_tcpInErrs.attr,
1083 	&dev_attr_tcpOutRsts.attr,
1084 	NULL
1085 };
1086 
1087 static struct attribute_group iw_stats_group = {
1088 	.name	= "proto_stats",
1089 	.attrs	= iw_proto_stats_attrs,
1090 };
1091 
1092 static void free_port_list_attributes(struct ib_device *device)
1093 {
1094 	struct kobject *p, *t;
1095 
1096 	list_for_each_entry_safe(p, t, &device->port_list, entry) {
1097 		struct ib_port *port = container_of(p, struct ib_port, kobj);
1098 		list_del(&p->entry);
1099 		sysfs_remove_group(p, port->pma_table);
1100 		sysfs_remove_group(p, &port->pkey_group);
1101 		sysfs_remove_group(p, &port->gid_group);
1102 		sysfs_remove_group(&port->gid_attr_group->kobj,
1103 				   &port->gid_attr_group->ndev);
1104 		sysfs_remove_group(&port->gid_attr_group->kobj,
1105 				   &port->gid_attr_group->type);
1106 		kobject_put(&port->gid_attr_group->kobj);
1107 		kobject_put(p);
1108 	}
1109 
1110 	kobject_put(device->ports_parent);
1111 }
1112 
1113 int ib_device_register_sysfs(struct ib_device *device,
1114 			     int (*port_callback)(struct ib_device *,
1115 						  u8, struct kobject *))
1116 {
1117 	struct device *class_dev = &device->dev;
1118 	int ret;
1119 	int i;
1120 
1121 	device->dev.parent = device->dma_device;
1122 	ret = dev_set_name(class_dev, "%s", device->name);
1123 	if (ret)
1124 		return ret;
1125 
1126 	ret = device_add(class_dev);
1127 	if (ret)
1128 		goto err;
1129 
1130 	for (i = 0; i < ARRAY_SIZE(ib_class_attributes); ++i) {
1131 		ret = device_create_file(class_dev, ib_class_attributes[i]);
1132 		if (ret)
1133 			goto err_unregister;
1134 	}
1135 
1136 	device->ports_parent = kobject_create_and_add("ports",
1137 						      &class_dev->kobj);
1138 	if (!device->ports_parent) {
1139 		ret = -ENOMEM;
1140 		goto err_put;
1141 	}
1142 
1143 	if (rdma_cap_ib_switch(device)) {
1144 		ret = add_port(device, 0, port_callback);
1145 		if (ret)
1146 			goto err_put;
1147 	} else {
1148 		for (i = 1; i <= device->phys_port_cnt; ++i) {
1149 			ret = add_port(device, i, port_callback);
1150 			if (ret)
1151 				goto err_put;
1152 		}
1153 	}
1154 
1155 	if (device->node_type == RDMA_NODE_RNIC && device->get_protocol_stats) {
1156 		ret = sysfs_create_group(&class_dev->kobj, &iw_stats_group);
1157 		if (ret)
1158 			goto err_put;
1159 	}
1160 
1161 	return 0;
1162 
1163 err_put:
1164 	free_port_list_attributes(device);
1165 
1166 err_unregister:
1167 	device_unregister(class_dev);
1168 
1169 err:
1170 	return ret;
1171 }
1172 
1173 void ib_device_unregister_sysfs(struct ib_device *device)
1174 {
1175 	/* Hold kobject until ib_dealloc_device() */
1176 	struct kobject *kobj_dev = kobject_get(&device->dev.kobj);
1177 	int i;
1178 
1179 	if (device->node_type == RDMA_NODE_RNIC && device->get_protocol_stats)
1180 		sysfs_remove_group(kobj_dev, &iw_stats_group);
1181 
1182 	free_port_list_attributes(device);
1183 
1184 	for (i = 0; i < ARRAY_SIZE(ib_class_attributes); ++i)
1185 		device_remove_file(&device->dev, ib_class_attributes[i]);
1186 
1187 	device_unregister(&device->dev);
1188 }
1189