ib_sysfs.c revision 331772
1/*-
2 * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
3 *
4 * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
5 * Copyright (c) 2005 Mellanox Technologies Ltd.  All rights reserved.
6 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
7 *
8 * This software is available to you under a choice of one of two
9 * licenses.  You may choose to be licensed under the terms of the GNU
10 * General Public License (GPL) Version 2, available from the file
11 * COPYING in the main directory of this source tree, or the
12 * OpenIB.org BSD license below:
13 *
14 *     Redistribution and use in source and binary forms, with or
15 *     without modification, are permitted provided that the following
16 *     conditions are met:
17 *
18 *      - Redistributions of source code must retain the above
19 *        copyright notice, this list of conditions and the following
20 *        disclaimer.
21 *
22 *      - Redistributions in binary form must reproduce the above
23 *        copyright notice, this list of conditions and the following
24 *        disclaimer in the documentation and/or other materials
25 *        provided with the distribution.
26 *
27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 * SOFTWARE.
35 *
36 * $FreeBSD: stable/11/sys/ofed/drivers/infiniband/core/ib_sysfs.c 331772 2018-03-30 18:17:33Z hselasky $
37 */
38
39#include "core_priv.h"
40
41#include <linux/slab.h>
42#include <linux/string.h>
43#include <linux/netdevice.h>
44#include <linux/fs.h>
45#include <linux/printk.h>
46
47#include <rdma/ib_mad.h>
48#include <rdma/ib_pma.h>
49
50struct ib_port;
51
52struct gid_attr_group {
53	struct ib_port		*port;
54	struct kobject		kobj;
55	struct attribute_group	ndev;
56	struct attribute_group	type;
57};
58struct ib_port {
59	struct kobject         kobj;
60	struct ib_device      *ibdev;
61	struct gid_attr_group *gid_attr_group;
62	struct attribute_group gid_group;
63	struct attribute_group pkey_group;
64	struct attribute_group *pma_table;
65	struct attribute_group *hw_stats_ag;
66	struct rdma_hw_stats   *hw_stats;
67	u8                     port_num;
68};
69
70struct port_attribute {
71	struct attribute attr;
72	ssize_t (*show)(struct ib_port *, struct port_attribute *, char *buf);
73	ssize_t (*store)(struct ib_port *, struct port_attribute *,
74			 const char *buf, size_t count);
75};
76
77#define PORT_ATTR(_name, _mode, _show, _store) \
78struct port_attribute port_attr_##_name = __ATTR(_name, _mode, _show, _store)
79
80#define PORT_ATTR_RO(_name) \
81struct port_attribute port_attr_##_name = __ATTR_RO(_name)
82
83struct port_table_attribute {
84	struct port_attribute	attr;
85	char			name[8];
86	int			index;
87	__be16			attr_id;
88};
89
90struct hw_stats_attribute {
91	struct attribute	attr;
92	ssize_t			(*show)(struct kobject *kobj,
93					struct attribute *attr, char *buf);
94	ssize_t			(*store)(struct kobject *kobj,
95					 struct attribute *attr,
96					 const char *buf,
97					 size_t count);
98	int			index;
99	u8			port_num;
100};
101
102static ssize_t port_attr_show(struct kobject *kobj,
103			      struct attribute *attr, char *buf)
104{
105	struct port_attribute *port_attr =
106		container_of(attr, struct port_attribute, attr);
107	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
108
109	if (!port_attr->show)
110		return -EIO;
111
112	return port_attr->show(p, port_attr, buf);
113}
114
115static const struct sysfs_ops port_sysfs_ops = {
116	.show = port_attr_show
117};
118
119static ssize_t gid_attr_show(struct kobject *kobj,
120			     struct attribute *attr, char *buf)
121{
122	struct port_attribute *port_attr =
123		container_of(attr, struct port_attribute, attr);
124	struct ib_port *p = container_of(kobj, struct gid_attr_group,
125					 kobj)->port;
126
127	if (!port_attr->show)
128		return -EIO;
129
130	return port_attr->show(p, port_attr, buf);
131}
132
133static const struct sysfs_ops gid_attr_sysfs_ops = {
134	.show = gid_attr_show
135};
136
137static ssize_t state_show(struct ib_port *p, struct port_attribute *unused,
138			  char *buf)
139{
140	struct ib_port_attr attr;
141	ssize_t ret;
142
143	static const char *state_name[] = {
144		[IB_PORT_NOP]		= "NOP",
145		[IB_PORT_DOWN]		= "DOWN",
146		[IB_PORT_INIT]		= "INIT",
147		[IB_PORT_ARMED]		= "ARMED",
148		[IB_PORT_ACTIVE]	= "ACTIVE",
149		[IB_PORT_ACTIVE_DEFER]	= "ACTIVE_DEFER"
150	};
151
152	ret = ib_query_port(p->ibdev, p->port_num, &attr);
153	if (ret)
154		return ret;
155
156	return sprintf(buf, "%d: %s\n", attr.state,
157		       attr.state >= 0 && attr.state < ARRAY_SIZE(state_name) ?
158		       state_name[attr.state] : "UNKNOWN");
159}
160
161static ssize_t lid_show(struct ib_port *p, struct port_attribute *unused,
162			char *buf)
163{
164	struct ib_port_attr attr;
165	ssize_t ret;
166
167	ret = ib_query_port(p->ibdev, p->port_num, &attr);
168	if (ret)
169		return ret;
170
171	return sprintf(buf, "0x%x\n", attr.lid);
172}
173
174static ssize_t lid_mask_count_show(struct ib_port *p,
175				   struct port_attribute *unused,
176				   char *buf)
177{
178	struct ib_port_attr attr;
179	ssize_t ret;
180
181	ret = ib_query_port(p->ibdev, p->port_num, &attr);
182	if (ret)
183		return ret;
184
185	return sprintf(buf, "%d\n", attr.lmc);
186}
187
188static ssize_t sm_lid_show(struct ib_port *p, struct port_attribute *unused,
189			   char *buf)
190{
191	struct ib_port_attr attr;
192	ssize_t ret;
193
194	ret = ib_query_port(p->ibdev, p->port_num, &attr);
195	if (ret)
196		return ret;
197
198	return sprintf(buf, "0x%x\n", attr.sm_lid);
199}
200
201static ssize_t sm_sl_show(struct ib_port *p, struct port_attribute *unused,
202			  char *buf)
203{
204	struct ib_port_attr attr;
205	ssize_t ret;
206
207	ret = ib_query_port(p->ibdev, p->port_num, &attr);
208	if (ret)
209		return ret;
210
211	return sprintf(buf, "%d\n", attr.sm_sl);
212}
213
214static ssize_t cap_mask_show(struct ib_port *p, struct port_attribute *unused,
215			     char *buf)
216{
217	struct ib_port_attr attr;
218	ssize_t ret;
219
220	ret = ib_query_port(p->ibdev, p->port_num, &attr);
221	if (ret)
222		return ret;
223
224	return sprintf(buf, "0x%08x\n", attr.port_cap_flags);
225}
226
227static ssize_t rate_show(struct ib_port *p, struct port_attribute *unused,
228			 char *buf)
229{
230	struct ib_port_attr attr;
231	char *speed = "";
232	int rate;		/* in deci-Gb/sec */
233	ssize_t ret;
234
235	ret = ib_query_port(p->ibdev, p->port_num, &attr);
236	if (ret)
237		return ret;
238
239	switch (attr.active_speed) {
240	case IB_SPEED_DDR:
241		speed = " DDR";
242		rate = 50;
243		break;
244	case IB_SPEED_QDR:
245		speed = " QDR";
246		rate = 100;
247		break;
248	case IB_SPEED_FDR10:
249		speed = " FDR10";
250		rate = 100;
251		break;
252	case IB_SPEED_FDR:
253		speed = " FDR";
254		rate = 140;
255		break;
256	case IB_SPEED_EDR:
257		speed = " EDR";
258		rate = 250;
259		break;
260	case IB_SPEED_SDR:
261	default:		/* default to SDR for invalid rates */
262		rate = 25;
263		break;
264	}
265
266	rate *= ib_width_enum_to_int(attr.active_width);
267	if (rate < 0)
268		return -EINVAL;
269
270	return sprintf(buf, "%d%s Gb/sec (%dX%s)\n",
271		       rate / 10, rate % 10 ? ".5" : "",
272		       ib_width_enum_to_int(attr.active_width), speed);
273}
274
275static ssize_t phys_state_show(struct ib_port *p, struct port_attribute *unused,
276			       char *buf)
277{
278	struct ib_port_attr attr;
279
280	ssize_t ret;
281
282	ret = ib_query_port(p->ibdev, p->port_num, &attr);
283	if (ret)
284		return ret;
285
286	switch (attr.phys_state) {
287	case 1:  return sprintf(buf, "1: Sleep\n");
288	case 2:  return sprintf(buf, "2: Polling\n");
289	case 3:  return sprintf(buf, "3: Disabled\n");
290	case 4:  return sprintf(buf, "4: PortConfigurationTraining\n");
291	case 5:  return sprintf(buf, "5: LinkUp\n");
292	case 6:  return sprintf(buf, "6: LinkErrorRecovery\n");
293	case 7:  return sprintf(buf, "7: Phy Test\n");
294	default: return sprintf(buf, "%d: <unknown>\n", attr.phys_state);
295	}
296}
297
298static ssize_t link_layer_show(struct ib_port *p, struct port_attribute *unused,
299			       char *buf)
300{
301	switch (rdma_port_get_link_layer(p->ibdev, p->port_num)) {
302	case IB_LINK_LAYER_INFINIBAND:
303		return sprintf(buf, "%s\n", "InfiniBand");
304	case IB_LINK_LAYER_ETHERNET:
305		return sprintf(buf, "%s\n", "Ethernet");
306	default:
307		return sprintf(buf, "%s\n", "Unknown");
308	}
309}
310
311static PORT_ATTR_RO(state);
312static PORT_ATTR_RO(lid);
313static PORT_ATTR_RO(lid_mask_count);
314static PORT_ATTR_RO(sm_lid);
315static PORT_ATTR_RO(sm_sl);
316static PORT_ATTR_RO(cap_mask);
317static PORT_ATTR_RO(rate);
318static PORT_ATTR_RO(phys_state);
319static PORT_ATTR_RO(link_layer);
320
321static struct attribute *port_default_attrs[] = {
322	&port_attr_state.attr,
323	&port_attr_lid.attr,
324	&port_attr_lid_mask_count.attr,
325	&port_attr_sm_lid.attr,
326	&port_attr_sm_sl.attr,
327	&port_attr_cap_mask.attr,
328	&port_attr_rate.attr,
329	&port_attr_phys_state.attr,
330	&port_attr_link_layer.attr,
331	NULL
332};
333
334static size_t print_ndev(struct ib_gid_attr *gid_attr, char *buf)
335{
336	if (!gid_attr->ndev)
337		return -EINVAL;
338
339	return sprintf(buf, "%s\n", if_name(gid_attr->ndev));
340}
341
342static size_t print_gid_type(struct ib_gid_attr *gid_attr, char *buf)
343{
344	return sprintf(buf, "%s\n", ib_cache_gid_type_str(gid_attr->gid_type));
345}
346
347static ssize_t _show_port_gid_attr(struct ib_port *p,
348				   struct port_attribute *attr,
349				   char *buf,
350				   size_t (*print)(struct ib_gid_attr *gid_attr,
351						   char *buf))
352{
353	struct port_table_attribute *tab_attr =
354		container_of(attr, struct port_table_attribute, attr);
355	union ib_gid gid;
356	struct ib_gid_attr gid_attr = {};
357	ssize_t ret;
358
359	ret = ib_query_gid(p->ibdev, p->port_num, tab_attr->index, &gid,
360			   &gid_attr);
361	if (ret)
362		goto err;
363
364	ret = print(&gid_attr, buf);
365
366err:
367	if (gid_attr.ndev)
368		dev_put(gid_attr.ndev);
369	return ret;
370}
371
372static ssize_t show_port_gid(struct ib_port *p, struct port_attribute *attr,
373			     char *buf)
374{
375	struct port_table_attribute *tab_attr =
376		container_of(attr, struct port_table_attribute, attr);
377	union ib_gid gid;
378	ssize_t ret;
379
380	ret = ib_query_gid(p->ibdev, p->port_num, tab_attr->index, &gid, NULL);
381	if (ret)
382		return ret;
383
384	return sprintf(buf, GID_PRINT_FMT"\n", GID_PRINT_ARGS(gid.raw));
385}
386
387static ssize_t show_port_gid_attr_ndev(struct ib_port *p,
388				       struct port_attribute *attr, char *buf)
389{
390	return _show_port_gid_attr(p, attr, buf, print_ndev);
391}
392
393static ssize_t show_port_gid_attr_gid_type(struct ib_port *p,
394					   struct port_attribute *attr,
395					   char *buf)
396{
397	return _show_port_gid_attr(p, attr, buf, print_gid_type);
398}
399
400static ssize_t show_port_pkey(struct ib_port *p, struct port_attribute *attr,
401			      char *buf)
402{
403	struct port_table_attribute *tab_attr =
404		container_of(attr, struct port_table_attribute, attr);
405	u16 pkey;
406	ssize_t ret;
407
408	ret = ib_query_pkey(p->ibdev, p->port_num, tab_attr->index, &pkey);
409	if (ret)
410		return ret;
411
412	return sprintf(buf, "0x%04x\n", pkey);
413}
414
415#define PORT_PMA_ATTR(_name, _counter, _width, _offset)			\
416struct port_table_attribute port_pma_attr_##_name = {			\
417	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
418	.index = (_offset) | ((_width) << 16) | ((_counter) << 24),	\
419	.attr_id = IB_PMA_PORT_COUNTERS ,				\
420}
421
422#define PORT_PMA_ATTR_EXT(_name, _width, _offset)			\
423struct port_table_attribute port_pma_attr_ext_##_name = {		\
424	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
425	.index = (_offset) | ((_width) << 16),				\
426	.attr_id = IB_PMA_PORT_COUNTERS_EXT ,				\
427}
428
429/*
430 * Get a Perfmgmt MAD block of data.
431 * Returns error code or the number of bytes retrieved.
432 */
433static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
434		void *data, int offset, size_t size)
435{
436	struct ib_mad *in_mad;
437	struct ib_mad *out_mad;
438	size_t mad_size = sizeof(*out_mad);
439	u16 out_mad_pkey_index = 0;
440	ssize_t ret;
441
442	if (!dev->process_mad)
443		return -ENOSYS;
444
445	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
446	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
447	if (!in_mad || !out_mad) {
448		ret = -ENOMEM;
449		goto out;
450	}
451
452	in_mad->mad_hdr.base_version  = 1;
453	in_mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_PERF_MGMT;
454	in_mad->mad_hdr.class_version = 1;
455	in_mad->mad_hdr.method        = IB_MGMT_METHOD_GET;
456	in_mad->mad_hdr.attr_id       = attr;
457
458	if (attr != IB_PMA_CLASS_PORT_INFO)
459		in_mad->data[41] = port_num;	/* PortSelect field */
460
461	if ((dev->process_mad(dev, IB_MAD_IGNORE_MKEY,
462		 port_num, NULL, NULL,
463		 (const struct ib_mad_hdr *)in_mad, mad_size,
464		 (struct ib_mad_hdr *)out_mad, &mad_size,
465		 &out_mad_pkey_index) &
466	     (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
467	    (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
468		ret = -EINVAL;
469		goto out;
470	}
471	memcpy(data, out_mad->data + offset, size);
472	ret = size;
473out:
474	kfree(in_mad);
475	kfree(out_mad);
476	return ret;
477}
478
479static ssize_t show_pma_counter(struct ib_port *p, struct port_attribute *attr,
480				char *buf)
481{
482	struct port_table_attribute *tab_attr =
483		container_of(attr, struct port_table_attribute, attr);
484	int offset = tab_attr->index & 0xffff;
485	int width  = (tab_attr->index >> 16) & 0xff;
486	ssize_t ret;
487	u8 data[8];
488
489	ret = get_perf_mad(p->ibdev, p->port_num, tab_attr->attr_id, &data,
490			40 + offset / 8, sizeof(data));
491	if (ret < 0)
492		return sprintf(buf, "N/A (no PMA)\n");
493
494	switch (width) {
495	case 4:
496		ret = sprintf(buf, "%u\n", (*data >>
497					    (4 - (offset % 8))) & 0xf);
498		break;
499	case 8:
500		ret = sprintf(buf, "%u\n", *data);
501		break;
502	case 16:
503		ret = sprintf(buf, "%u\n",
504			      be16_to_cpup((__be16 *)data));
505		break;
506	case 32:
507		ret = sprintf(buf, "%u\n",
508			      be32_to_cpup((__be32 *)data));
509		break;
510	case 64:
511		ret = sprintf(buf, "%llu\n",
512			      (unsigned long long)be64_to_cpup((__be64 *)data));
513		break;
514
515	default:
516		ret = 0;
517	}
518
519	return ret;
520}
521
522static PORT_PMA_ATTR(symbol_error		    ,  0, 16,  32);
523static PORT_PMA_ATTR(link_error_recovery	    ,  1,  8,  48);
524static PORT_PMA_ATTR(link_downed		    ,  2,  8,  56);
525static PORT_PMA_ATTR(port_rcv_errors		    ,  3, 16,  64);
526static PORT_PMA_ATTR(port_rcv_remote_physical_errors,  4, 16,  80);
527static PORT_PMA_ATTR(port_rcv_switch_relay_errors   ,  5, 16,  96);
528static PORT_PMA_ATTR(port_xmit_discards		    ,  6, 16, 112);
529static PORT_PMA_ATTR(port_xmit_constraint_errors    ,  7,  8, 128);
530static PORT_PMA_ATTR(port_rcv_constraint_errors	    ,  8,  8, 136);
531static PORT_PMA_ATTR(local_link_integrity_errors    ,  9,  4, 152);
532static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10,  4, 156);
533static PORT_PMA_ATTR(VL15_dropped		    , 11, 16, 176);
534static PORT_PMA_ATTR(port_xmit_data		    , 12, 32, 192);
535static PORT_PMA_ATTR(port_rcv_data		    , 13, 32, 224);
536static PORT_PMA_ATTR(port_xmit_packets		    , 14, 32, 256);
537static PORT_PMA_ATTR(port_rcv_packets		    , 15, 32, 288);
538static PORT_PMA_ATTR(port_xmit_wait		    ,  0, 32, 320);
539
540/*
541 * Counters added by extended set
542 */
543static PORT_PMA_ATTR_EXT(port_xmit_data		    , 64,  64);
544static PORT_PMA_ATTR_EXT(port_rcv_data		    , 64, 128);
545static PORT_PMA_ATTR_EXT(port_xmit_packets	    , 64, 192);
546static PORT_PMA_ATTR_EXT(port_rcv_packets	    , 64, 256);
547static PORT_PMA_ATTR_EXT(unicast_xmit_packets	    , 64, 320);
548static PORT_PMA_ATTR_EXT(unicast_rcv_packets	    , 64, 384);
549static PORT_PMA_ATTR_EXT(multicast_xmit_packets	    , 64, 448);
550static PORT_PMA_ATTR_EXT(multicast_rcv_packets	    , 64, 512);
551
552static struct attribute *pma_attrs[] = {
553	&port_pma_attr_symbol_error.attr.attr,
554	&port_pma_attr_link_error_recovery.attr.attr,
555	&port_pma_attr_link_downed.attr.attr,
556	&port_pma_attr_port_rcv_errors.attr.attr,
557	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
558	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
559	&port_pma_attr_port_xmit_discards.attr.attr,
560	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
561	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
562	&port_pma_attr_local_link_integrity_errors.attr.attr,
563	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
564	&port_pma_attr_VL15_dropped.attr.attr,
565	&port_pma_attr_port_xmit_data.attr.attr,
566	&port_pma_attr_port_rcv_data.attr.attr,
567	&port_pma_attr_port_xmit_packets.attr.attr,
568	&port_pma_attr_port_rcv_packets.attr.attr,
569	&port_pma_attr_port_xmit_wait.attr.attr,
570	NULL
571};
572
573static struct attribute *pma_attrs_ext[] = {
574	&port_pma_attr_symbol_error.attr.attr,
575	&port_pma_attr_link_error_recovery.attr.attr,
576	&port_pma_attr_link_downed.attr.attr,
577	&port_pma_attr_port_rcv_errors.attr.attr,
578	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
579	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
580	&port_pma_attr_port_xmit_discards.attr.attr,
581	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
582	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
583	&port_pma_attr_local_link_integrity_errors.attr.attr,
584	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
585	&port_pma_attr_VL15_dropped.attr.attr,
586	&port_pma_attr_ext_port_xmit_data.attr.attr,
587	&port_pma_attr_ext_port_rcv_data.attr.attr,
588	&port_pma_attr_ext_port_xmit_packets.attr.attr,
589	&port_pma_attr_port_xmit_wait.attr.attr,
590	&port_pma_attr_ext_port_rcv_packets.attr.attr,
591	&port_pma_attr_ext_unicast_rcv_packets.attr.attr,
592	&port_pma_attr_ext_unicast_xmit_packets.attr.attr,
593	&port_pma_attr_ext_multicast_rcv_packets.attr.attr,
594	&port_pma_attr_ext_multicast_xmit_packets.attr.attr,
595	NULL
596};
597
598static struct attribute *pma_attrs_noietf[] = {
599	&port_pma_attr_symbol_error.attr.attr,
600	&port_pma_attr_link_error_recovery.attr.attr,
601	&port_pma_attr_link_downed.attr.attr,
602	&port_pma_attr_port_rcv_errors.attr.attr,
603	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
604	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
605	&port_pma_attr_port_xmit_discards.attr.attr,
606	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
607	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
608	&port_pma_attr_local_link_integrity_errors.attr.attr,
609	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
610	&port_pma_attr_VL15_dropped.attr.attr,
611	&port_pma_attr_ext_port_xmit_data.attr.attr,
612	&port_pma_attr_ext_port_rcv_data.attr.attr,
613	&port_pma_attr_ext_port_xmit_packets.attr.attr,
614	&port_pma_attr_ext_port_rcv_packets.attr.attr,
615	&port_pma_attr_port_xmit_wait.attr.attr,
616	NULL
617};
618
619static struct attribute_group pma_group = {
620	.name  = "counters",
621	.attrs  = pma_attrs
622};
623
624static struct attribute_group pma_group_ext = {
625	.name  = "counters",
626	.attrs  = pma_attrs_ext
627};
628
629static struct attribute_group pma_group_noietf = {
630	.name  = "counters",
631	.attrs  = pma_attrs_noietf
632};
633
634static void ib_port_release(struct kobject *kobj)
635{
636	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
637	struct attribute *a;
638	int i;
639
640	if (p->gid_group.attrs) {
641		for (i = 0; (a = p->gid_group.attrs[i]); ++i)
642			kfree(a);
643
644		kfree(p->gid_group.attrs);
645	}
646
647	if (p->pkey_group.attrs) {
648		for (i = 0; (a = p->pkey_group.attrs[i]); ++i)
649			kfree(a);
650
651		kfree(p->pkey_group.attrs);
652	}
653
654	kfree(p);
655}
656
657static void ib_port_gid_attr_release(struct kobject *kobj)
658{
659	struct gid_attr_group *g = container_of(kobj, struct gid_attr_group,
660						kobj);
661	struct attribute *a;
662	int i;
663
664	if (g->ndev.attrs) {
665		for (i = 0; (a = g->ndev.attrs[i]); ++i)
666			kfree(a);
667
668		kfree(g->ndev.attrs);
669	}
670
671	if (g->type.attrs) {
672		for (i = 0; (a = g->type.attrs[i]); ++i)
673			kfree(a);
674
675		kfree(g->type.attrs);
676	}
677
678	kfree(g);
679}
680
681static struct kobj_type port_type = {
682	.release       = ib_port_release,
683	.sysfs_ops     = &port_sysfs_ops,
684	.default_attrs = port_default_attrs
685};
686
687static struct kobj_type gid_attr_type = {
688	.sysfs_ops      = &gid_attr_sysfs_ops,
689	.release        = ib_port_gid_attr_release
690};
691
692static struct attribute **
693alloc_group_attrs(ssize_t (*show)(struct ib_port *,
694				  struct port_attribute *, char *buf),
695		  int len)
696{
697	struct attribute **tab_attr;
698	struct port_table_attribute *element;
699	int i;
700
701	tab_attr = kcalloc(1 + len, sizeof(struct attribute *), GFP_KERNEL);
702	if (!tab_attr)
703		return NULL;
704
705	for (i = 0; i < len; i++) {
706		element = kzalloc(sizeof(struct port_table_attribute),
707				  GFP_KERNEL);
708		if (!element)
709			goto err;
710
711		if (snprintf(element->name, sizeof(element->name),
712			     "%d", i) >= sizeof(element->name)) {
713			kfree(element);
714			goto err;
715		}
716
717		element->attr.attr.name  = element->name;
718		element->attr.attr.mode  = S_IRUGO;
719		element->attr.show       = show;
720		element->index		 = i;
721		sysfs_attr_init(&element->attr.attr);
722
723		tab_attr[i] = &element->attr.attr;
724	}
725
726	return tab_attr;
727
728err:
729	while (--i >= 0)
730		kfree(tab_attr[i]);
731	kfree(tab_attr);
732	return NULL;
733}
734
735/*
736 * Figure out which counter table to use depending on
737 * the device capabilities.
738 */
739static struct attribute_group *get_counter_table(struct ib_device *dev,
740						 int port_num)
741{
742	struct ib_class_port_info cpi;
743
744	if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
745				&cpi, 40, sizeof(cpi)) >= 0) {
746		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
747			/* We have extended counters */
748			return &pma_group_ext;
749
750		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
751			/* But not the IETF ones */
752			return &pma_group_noietf;
753	}
754
755	/* Fall back to normal counters */
756	return &pma_group;
757}
758
759static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
760			   u8 port_num, int index)
761{
762	int ret;
763
764	if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
765		return 0;
766	ret = dev->get_hw_stats(dev, stats, port_num, index);
767	if (ret < 0)
768		return ret;
769	if (ret == stats->num_counters)
770		stats->timestamp = jiffies;
771
772	return 0;
773}
774
775static ssize_t print_hw_stat(struct rdma_hw_stats *stats, int index, char *buf)
776{
777	return sprintf(buf, "%llu\n", (unsigned long long)stats->value[index]);
778}
779
780static ssize_t show_hw_stats(struct kobject *kobj, struct attribute *attr,
781			     char *buf)
782{
783	struct ib_device *dev;
784	struct ib_port *port;
785	struct hw_stats_attribute *hsa;
786	struct rdma_hw_stats *stats;
787	int ret;
788
789	hsa = container_of(attr, struct hw_stats_attribute, attr);
790	if (!hsa->port_num) {
791		dev = container_of((struct device *)kobj,
792				   struct ib_device, dev);
793		stats = dev->hw_stats;
794	} else {
795		port = container_of(kobj, struct ib_port, kobj);
796		dev = port->ibdev;
797		stats = port->hw_stats;
798	}
799	ret = update_hw_stats(dev, stats, hsa->port_num, hsa->index);
800	if (ret)
801		return ret;
802	return print_hw_stat(stats, hsa->index, buf);
803}
804
805static ssize_t show_stats_lifespan(struct kobject *kobj,
806				   struct attribute *attr,
807				   char *buf)
808{
809	struct hw_stats_attribute *hsa;
810	int msecs;
811
812	hsa = container_of(attr, struct hw_stats_attribute, attr);
813	if (!hsa->port_num) {
814		struct ib_device *dev = container_of((struct device *)kobj,
815						     struct ib_device, dev);
816		msecs = jiffies_to_msecs(dev->hw_stats->lifespan);
817	} else {
818		struct ib_port *p = container_of(kobj, struct ib_port, kobj);
819		msecs = jiffies_to_msecs(p->hw_stats->lifespan);
820	}
821	return sprintf(buf, "%d\n", msecs);
822}
823
824static ssize_t set_stats_lifespan(struct kobject *kobj,
825				  struct attribute *attr,
826				  const char *buf, size_t count)
827{
828	struct hw_stats_attribute *hsa;
829	int msecs;
830	int jiffies;
831	int ret;
832
833	ret = kstrtoint(buf, 10, &msecs);
834	if (ret)
835		return ret;
836	if (msecs < 0 || msecs > 10000)
837		return -EINVAL;
838	jiffies = msecs_to_jiffies(msecs);
839	hsa = container_of(attr, struct hw_stats_attribute, attr);
840	if (!hsa->port_num) {
841		struct ib_device *dev = container_of((struct device *)kobj,
842						     struct ib_device, dev);
843		dev->hw_stats->lifespan = jiffies;
844	} else {
845		struct ib_port *p = container_of(kobj, struct ib_port, kobj);
846		p->hw_stats->lifespan = jiffies;
847	}
848	return count;
849}
850
851static void free_hsag(struct kobject *kobj, struct attribute_group *attr_group)
852{
853	struct attribute **attr;
854
855	sysfs_remove_group(kobj, attr_group);
856
857	for (attr = attr_group->attrs; *attr; attr++)
858		kfree(*attr);
859	kfree(attr_group);
860}
861
862static struct attribute *alloc_hsa(int index, u8 port_num, const char *name)
863{
864	struct hw_stats_attribute *hsa;
865
866	hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
867	if (!hsa)
868		return NULL;
869
870	hsa->attr.name = __DECONST(char *, name);
871	hsa->attr.mode = S_IRUGO;
872	hsa->show = show_hw_stats;
873	hsa->store = NULL;
874	hsa->index = index;
875	hsa->port_num = port_num;
876
877	return &hsa->attr;
878}
879
880static struct attribute *alloc_hsa_lifespan(char *name, u8 port_num)
881{
882	struct hw_stats_attribute *hsa;
883
884	hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
885	if (!hsa)
886		return NULL;
887
888	hsa->attr.name = name;
889	hsa->attr.mode = S_IWUSR | S_IRUGO;
890	hsa->show = show_stats_lifespan;
891	hsa->store = set_stats_lifespan;
892	hsa->index = 0;
893	hsa->port_num = port_num;
894
895	return &hsa->attr;
896}
897
898static void setup_hw_stats(struct ib_device *device, struct ib_port *port,
899			   u8 port_num)
900{
901	struct attribute_group *hsag;
902	struct rdma_hw_stats *stats;
903	int i, ret;
904
905	stats = device->alloc_hw_stats(device, port_num);
906
907	if (!stats)
908		return;
909
910	if (!stats->names || stats->num_counters <= 0)
911		goto err_free_stats;
912
913	/*
914	 * Two extra attribue elements here, one for the lifespan entry and
915	 * one to NULL terminate the list for the sysfs core code
916	 */
917	hsag = kzalloc(sizeof(*hsag) +
918		       sizeof(void *) * (stats->num_counters + 2),
919		       GFP_KERNEL);
920	if (!hsag)
921		goto err_free_stats;
922
923	ret = device->get_hw_stats(device, stats, port_num,
924				   stats->num_counters);
925	if (ret != stats->num_counters)
926		goto err_free_hsag;
927
928	stats->timestamp = jiffies;
929
930	hsag->name = "hw_counters";
931	hsag->attrs = (void *)((char *)hsag + sizeof(*hsag));
932
933	for (i = 0; i < stats->num_counters; i++) {
934		hsag->attrs[i] = alloc_hsa(i, port_num, stats->names[i]);
935		if (!hsag->attrs[i])
936			goto err;
937		sysfs_attr_init(hsag->attrs[i]);
938	}
939
940	/* treat an error here as non-fatal */
941	hsag->attrs[i] = alloc_hsa_lifespan("lifespan", port_num);
942	if (hsag->attrs[i])
943		sysfs_attr_init(hsag->attrs[i]);
944
945	if (port) {
946		struct kobject *kobj = &port->kobj;
947		ret = sysfs_create_group(kobj, hsag);
948		if (ret)
949			goto err;
950		port->hw_stats_ag = hsag;
951		port->hw_stats = stats;
952	} else {
953		struct kobject *kobj = &device->dev.kobj;
954		ret = sysfs_create_group(kobj, hsag);
955		if (ret)
956			goto err;
957		device->hw_stats_ag = hsag;
958		device->hw_stats = stats;
959	}
960
961	return;
962
963err:
964	for (; i >= 0; i--)
965		kfree(hsag->attrs[i]);
966err_free_hsag:
967	kfree(hsag);
968err_free_stats:
969	kfree(stats);
970	return;
971}
972
973static int add_port(struct ib_device *device, int port_num,
974		    int (*port_callback)(struct ib_device *,
975					 u8, struct kobject *))
976{
977	struct ib_port *p;
978	struct ib_port_attr attr;
979	int i;
980	int ret;
981
982	ret = ib_query_port(device, port_num, &attr);
983	if (ret)
984		return ret;
985
986	p = kzalloc(sizeof *p, GFP_KERNEL);
987	if (!p)
988		return -ENOMEM;
989
990	p->ibdev      = device;
991	p->port_num   = port_num;
992
993	ret = kobject_init_and_add(&p->kobj, &port_type,
994				   device->ports_parent,
995				   "%d", port_num);
996	if (ret) {
997		kfree(p);
998		return ret;
999	}
1000
1001	p->gid_attr_group = kzalloc(sizeof(*p->gid_attr_group), GFP_KERNEL);
1002	if (!p->gid_attr_group) {
1003		ret = -ENOMEM;
1004		goto err_put;
1005	}
1006
1007	p->gid_attr_group->port = p;
1008	ret = kobject_init_and_add(&p->gid_attr_group->kobj, &gid_attr_type,
1009				   &p->kobj, "gid_attrs");
1010	if (ret) {
1011		kfree(p->gid_attr_group);
1012		goto err_put;
1013	}
1014
1015	p->pma_table = get_counter_table(device, port_num);
1016	ret = sysfs_create_group(&p->kobj, p->pma_table);
1017	if (ret)
1018		goto err_put_gid_attrs;
1019
1020	p->gid_group.name  = "gids";
1021	p->gid_group.attrs = alloc_group_attrs(show_port_gid, attr.gid_tbl_len);
1022	if (!p->gid_group.attrs) {
1023		ret = -ENOMEM;
1024		goto err_remove_pma;
1025	}
1026
1027	ret = sysfs_create_group(&p->kobj, &p->gid_group);
1028	if (ret)
1029		goto err_free_gid;
1030
1031	p->gid_attr_group->ndev.name = "ndevs";
1032	p->gid_attr_group->ndev.attrs = alloc_group_attrs(show_port_gid_attr_ndev,
1033							  attr.gid_tbl_len);
1034	if (!p->gid_attr_group->ndev.attrs) {
1035		ret = -ENOMEM;
1036		goto err_remove_gid;
1037	}
1038
1039	ret = sysfs_create_group(&p->gid_attr_group->kobj,
1040				 &p->gid_attr_group->ndev);
1041	if (ret)
1042		goto err_free_gid_ndev;
1043
1044	p->gid_attr_group->type.name = "types";
1045	p->gid_attr_group->type.attrs = alloc_group_attrs(show_port_gid_attr_gid_type,
1046							  attr.gid_tbl_len);
1047	if (!p->gid_attr_group->type.attrs) {
1048		ret = -ENOMEM;
1049		goto err_remove_gid_ndev;
1050	}
1051
1052	ret = sysfs_create_group(&p->gid_attr_group->kobj,
1053				 &p->gid_attr_group->type);
1054	if (ret)
1055		goto err_free_gid_type;
1056
1057	p->pkey_group.name  = "pkeys";
1058	p->pkey_group.attrs = alloc_group_attrs(show_port_pkey,
1059						attr.pkey_tbl_len);
1060	if (!p->pkey_group.attrs) {
1061		ret = -ENOMEM;
1062		goto err_remove_gid_type;
1063	}
1064
1065	ret = sysfs_create_group(&p->kobj, &p->pkey_group);
1066	if (ret)
1067		goto err_free_pkey;
1068
1069	if (port_callback) {
1070		ret = port_callback(device, port_num, &p->kobj);
1071		if (ret)
1072			goto err_remove_pkey;
1073	}
1074
1075	/*
1076	 * If port == 0, it means we have only one port and the parent
1077	 * device, not this port device, should be the holder of the
1078	 * hw_counters
1079	 */
1080	if (device->alloc_hw_stats && port_num)
1081		setup_hw_stats(device, p, port_num);
1082
1083	list_add_tail(&p->kobj.entry, &device->port_list);
1084
1085	return 0;
1086
1087err_remove_pkey:
1088	sysfs_remove_group(&p->kobj, &p->pkey_group);
1089
1090err_free_pkey:
1091	for (i = 0; i < attr.pkey_tbl_len; ++i)
1092		kfree(p->pkey_group.attrs[i]);
1093
1094	kfree(p->pkey_group.attrs);
1095	p->pkey_group.attrs = NULL;
1096
1097err_remove_gid_type:
1098	sysfs_remove_group(&p->gid_attr_group->kobj,
1099			   &p->gid_attr_group->type);
1100
1101err_free_gid_type:
1102	for (i = 0; i < attr.gid_tbl_len; ++i)
1103		kfree(p->gid_attr_group->type.attrs[i]);
1104
1105	kfree(p->gid_attr_group->type.attrs);
1106	p->gid_attr_group->type.attrs = NULL;
1107
1108err_remove_gid_ndev:
1109	sysfs_remove_group(&p->gid_attr_group->kobj,
1110			   &p->gid_attr_group->ndev);
1111
1112err_free_gid_ndev:
1113	for (i = 0; i < attr.gid_tbl_len; ++i)
1114		kfree(p->gid_attr_group->ndev.attrs[i]);
1115
1116	kfree(p->gid_attr_group->ndev.attrs);
1117	p->gid_attr_group->ndev.attrs = NULL;
1118
1119err_remove_gid:
1120	sysfs_remove_group(&p->kobj, &p->gid_group);
1121
1122err_free_gid:
1123	for (i = 0; i < attr.gid_tbl_len; ++i)
1124		kfree(p->gid_group.attrs[i]);
1125
1126	kfree(p->gid_group.attrs);
1127	p->gid_group.attrs = NULL;
1128
1129err_remove_pma:
1130	sysfs_remove_group(&p->kobj, p->pma_table);
1131
1132err_put_gid_attrs:
1133	kobject_put(&p->gid_attr_group->kobj);
1134
1135err_put:
1136	kobject_put(&p->kobj);
1137	return ret;
1138}
1139
1140static ssize_t show_node_type(struct device *device,
1141			      struct device_attribute *attr, char *buf)
1142{
1143	struct ib_device *dev = container_of(device, struct ib_device, dev);
1144
1145	switch (dev->node_type) {
1146	case RDMA_NODE_IB_CA:	  return sprintf(buf, "%d: CA\n", dev->node_type);
1147	case RDMA_NODE_RNIC:	  return sprintf(buf, "%d: RNIC\n", dev->node_type);
1148	case RDMA_NODE_USNIC:	  return sprintf(buf, "%d: usNIC\n", dev->node_type);
1149	case RDMA_NODE_USNIC_UDP: return sprintf(buf, "%d: usNIC UDP\n", dev->node_type);
1150	case RDMA_NODE_IB_SWITCH: return sprintf(buf, "%d: switch\n", dev->node_type);
1151	case RDMA_NODE_IB_ROUTER: return sprintf(buf, "%d: router\n", dev->node_type);
1152	default:		  return sprintf(buf, "%d: <unknown>\n", dev->node_type);
1153	}
1154}
1155
1156static ssize_t show_sys_image_guid(struct device *device,
1157				   struct device_attribute *dev_attr, char *buf)
1158{
1159	struct ib_device *dev = container_of(device, struct ib_device, dev);
1160
1161	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1162		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[0]),
1163		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[1]),
1164		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[2]),
1165		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[3]));
1166}
1167
1168static ssize_t show_node_guid(struct device *device,
1169			      struct device_attribute *attr, char *buf)
1170{
1171	struct ib_device *dev = container_of(device, struct ib_device, dev);
1172
1173	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1174		       be16_to_cpu(((__be16 *) &dev->node_guid)[0]),
1175		       be16_to_cpu(((__be16 *) &dev->node_guid)[1]),
1176		       be16_to_cpu(((__be16 *) &dev->node_guid)[2]),
1177		       be16_to_cpu(((__be16 *) &dev->node_guid)[3]));
1178}
1179
1180static ssize_t show_node_desc(struct device *device,
1181			      struct device_attribute *attr, char *buf)
1182{
1183	struct ib_device *dev = container_of(device, struct ib_device, dev);
1184
1185	return sprintf(buf, "%.64s\n", dev->node_desc);
1186}
1187
1188static ssize_t set_node_desc(struct device *device,
1189			     struct device_attribute *attr,
1190			     const char *buf, size_t count)
1191{
1192	struct ib_device *dev = container_of(device, struct ib_device, dev);
1193	struct ib_device_modify desc = {};
1194	int ret;
1195
1196	if (!dev->modify_device)
1197		return -EIO;
1198
1199	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1200	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1201	if (ret)
1202		return ret;
1203
1204	return count;
1205}
1206
1207static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1208			   char *buf)
1209{
1210	struct ib_device *dev = container_of(device, struct ib_device, dev);
1211
1212	ib_get_device_fw_str(dev, buf, PAGE_SIZE);
1213	strlcat(buf, "\n", PAGE_SIZE);
1214	return strlen(buf);
1215}
1216
1217static DEVICE_ATTR(node_type, S_IRUGO, show_node_type, NULL);
1218static DEVICE_ATTR(sys_image_guid, S_IRUGO, show_sys_image_guid, NULL);
1219static DEVICE_ATTR(node_guid, S_IRUGO, show_node_guid, NULL);
1220static DEVICE_ATTR(node_desc, S_IRUGO | S_IWUSR, show_node_desc, set_node_desc);
1221static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1222
1223static struct device_attribute *ib_class_attributes[] = {
1224	&dev_attr_node_type,
1225	&dev_attr_sys_image_guid,
1226	&dev_attr_node_guid,
1227	&dev_attr_node_desc,
1228	&dev_attr_fw_ver,
1229};
1230
1231static void free_port_list_attributes(struct ib_device *device)
1232{
1233	struct kobject *p, *t;
1234
1235	list_for_each_entry_safe(p, t, &device->port_list, entry) {
1236		struct ib_port *port = container_of(p, struct ib_port, kobj);
1237		list_del(&p->entry);
1238		if (port->hw_stats) {
1239			kfree(port->hw_stats);
1240			free_hsag(&port->kobj, port->hw_stats_ag);
1241		}
1242		sysfs_remove_group(p, port->pma_table);
1243		sysfs_remove_group(p, &port->pkey_group);
1244		sysfs_remove_group(p, &port->gid_group);
1245		sysfs_remove_group(&port->gid_attr_group->kobj,
1246				   &port->gid_attr_group->ndev);
1247		sysfs_remove_group(&port->gid_attr_group->kobj,
1248				   &port->gid_attr_group->type);
1249		kobject_put(&port->gid_attr_group->kobj);
1250		kobject_put(p);
1251	}
1252
1253	kobject_put(device->ports_parent);
1254}
1255
1256int ib_device_register_sysfs(struct ib_device *device,
1257			     int (*port_callback)(struct ib_device *,
1258						  u8, struct kobject *))
1259{
1260	struct device *class_dev = &device->dev;
1261	int ret;
1262	int i;
1263
1264	device->dev.parent = device->dma_device;
1265	ret = dev_set_name(class_dev, "%s", device->name);
1266	if (ret)
1267		return ret;
1268
1269	ret = device_add(class_dev);
1270	if (ret)
1271		goto err;
1272
1273	for (i = 0; i < ARRAY_SIZE(ib_class_attributes); ++i) {
1274		ret = device_create_file(class_dev, ib_class_attributes[i]);
1275		if (ret)
1276			goto err_unregister;
1277	}
1278
1279	device->ports_parent = kobject_create_and_add("ports",
1280						      &class_dev->kobj);
1281	if (!device->ports_parent) {
1282		ret = -ENOMEM;
1283		goto err_put;
1284	}
1285
1286	if (rdma_cap_ib_switch(device)) {
1287		ret = add_port(device, 0, port_callback);
1288		if (ret)
1289			goto err_put;
1290	} else {
1291		for (i = 1; i <= device->phys_port_cnt; ++i) {
1292			ret = add_port(device, i, port_callback);
1293			if (ret)
1294				goto err_put;
1295		}
1296	}
1297
1298	if (device->alloc_hw_stats)
1299		setup_hw_stats(device, NULL, 0);
1300
1301	return 0;
1302
1303err_put:
1304	free_port_list_attributes(device);
1305
1306err_unregister:
1307	device_unregister(class_dev);
1308
1309err:
1310	return ret;
1311}
1312
1313void ib_device_unregister_sysfs(struct ib_device *device)
1314{
1315	int i;
1316
1317	/* Hold kobject until ib_dealloc_device() */
1318	kobject_get(&device->dev.kobj);
1319
1320	free_port_list_attributes(device);
1321
1322	if (device->hw_stats) {
1323		kfree(device->hw_stats);
1324		free_hsag(&device->dev.kobj, device->hw_stats_ag);
1325	}
1326
1327	for (i = 0; i < ARRAY_SIZE(ib_class_attributes); ++i)
1328		device_remove_file(&device->dev, ib_class_attributes[i]);
1329
1330	device_unregister(&device->dev);
1331}
1332