ib_sysfs.c revision 331769
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/string.h>
39#include <linux/netdevice.h>
40#include <linux/fs.h>
41#include <linux/printk.h>
42
43#include <rdma/ib_mad.h>
44#include <rdma/ib_pma.h>
45
46struct ib_port;
47
48struct gid_attr_group {
49	struct ib_port		*port;
50	struct kobject		kobj;
51	struct attribute_group	ndev;
52	struct attribute_group	type;
53};
54struct ib_port {
55	struct kobject         kobj;
56	struct ib_device      *ibdev;
57	struct gid_attr_group *gid_attr_group;
58	struct attribute_group gid_group;
59	struct attribute_group pkey_group;
60	struct attribute_group *pma_table;
61	struct attribute_group *hw_stats_ag;
62	struct rdma_hw_stats   *hw_stats;
63	u8                     port_num;
64};
65
66struct port_attribute {
67	struct attribute attr;
68	ssize_t (*show)(struct ib_port *, struct port_attribute *, char *buf);
69	ssize_t (*store)(struct ib_port *, struct port_attribute *,
70			 const char *buf, size_t count);
71};
72
73#define PORT_ATTR(_name, _mode, _show, _store) \
74struct port_attribute port_attr_##_name = __ATTR(_name, _mode, _show, _store)
75
76#define PORT_ATTR_RO(_name) \
77struct port_attribute port_attr_##_name = __ATTR_RO(_name)
78
79struct port_table_attribute {
80	struct port_attribute	attr;
81	char			name[8];
82	int			index;
83	__be16			attr_id;
84};
85
86struct hw_stats_attribute {
87	struct attribute	attr;
88	ssize_t			(*show)(struct kobject *kobj,
89					struct attribute *attr, char *buf);
90	ssize_t			(*store)(struct kobject *kobj,
91					 struct attribute *attr,
92					 const char *buf,
93					 size_t count);
94	int			index;
95	u8			port_num;
96};
97
98static ssize_t port_attr_show(struct kobject *kobj,
99			      struct attribute *attr, char *buf)
100{
101	struct port_attribute *port_attr =
102		container_of(attr, struct port_attribute, attr);
103	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
104
105	if (!port_attr->show)
106		return -EIO;
107
108	return port_attr->show(p, port_attr, buf);
109}
110
111static const struct sysfs_ops port_sysfs_ops = {
112	.show = port_attr_show
113};
114
115static ssize_t gid_attr_show(struct kobject *kobj,
116			     struct attribute *attr, char *buf)
117{
118	struct port_attribute *port_attr =
119		container_of(attr, struct port_attribute, attr);
120	struct ib_port *p = container_of(kobj, struct gid_attr_group,
121					 kobj)->port;
122
123	if (!port_attr->show)
124		return -EIO;
125
126	return port_attr->show(p, port_attr, buf);
127}
128
129static const struct sysfs_ops gid_attr_sysfs_ops = {
130	.show = gid_attr_show
131};
132
133static ssize_t state_show(struct ib_port *p, struct port_attribute *unused,
134			  char *buf)
135{
136	struct ib_port_attr attr;
137	ssize_t ret;
138
139	static const char *state_name[] = {
140		[IB_PORT_NOP]		= "NOP",
141		[IB_PORT_DOWN]		= "DOWN",
142		[IB_PORT_INIT]		= "INIT",
143		[IB_PORT_ARMED]		= "ARMED",
144		[IB_PORT_ACTIVE]	= "ACTIVE",
145		[IB_PORT_ACTIVE_DEFER]	= "ACTIVE_DEFER"
146	};
147
148	ret = ib_query_port(p->ibdev, p->port_num, &attr);
149	if (ret)
150		return ret;
151
152	return sprintf(buf, "%d: %s\n", attr.state,
153		       attr.state >= 0 && attr.state < ARRAY_SIZE(state_name) ?
154		       state_name[attr.state] : "UNKNOWN");
155}
156
157static ssize_t lid_show(struct ib_port *p, struct port_attribute *unused,
158			char *buf)
159{
160	struct ib_port_attr attr;
161	ssize_t ret;
162
163	ret = ib_query_port(p->ibdev, p->port_num, &attr);
164	if (ret)
165		return ret;
166
167	return sprintf(buf, "0x%x\n", attr.lid);
168}
169
170static ssize_t lid_mask_count_show(struct ib_port *p,
171				   struct port_attribute *unused,
172				   char *buf)
173{
174	struct ib_port_attr attr;
175	ssize_t ret;
176
177	ret = ib_query_port(p->ibdev, p->port_num, &attr);
178	if (ret)
179		return ret;
180
181	return sprintf(buf, "%d\n", attr.lmc);
182}
183
184static ssize_t sm_lid_show(struct ib_port *p, struct port_attribute *unused,
185			   char *buf)
186{
187	struct ib_port_attr attr;
188	ssize_t ret;
189
190	ret = ib_query_port(p->ibdev, p->port_num, &attr);
191	if (ret)
192		return ret;
193
194	return sprintf(buf, "0x%x\n", attr.sm_lid);
195}
196
197static ssize_t sm_sl_show(struct ib_port *p, struct port_attribute *unused,
198			  char *buf)
199{
200	struct ib_port_attr attr;
201	ssize_t ret;
202
203	ret = ib_query_port(p->ibdev, p->port_num, &attr);
204	if (ret)
205		return ret;
206
207	return sprintf(buf, "%d\n", attr.sm_sl);
208}
209
210static ssize_t cap_mask_show(struct ib_port *p, struct port_attribute *unused,
211			     char *buf)
212{
213	struct ib_port_attr attr;
214	ssize_t ret;
215
216	ret = ib_query_port(p->ibdev, p->port_num, &attr);
217	if (ret)
218		return ret;
219
220	return sprintf(buf, "0x%08x\n", attr.port_cap_flags);
221}
222
223static ssize_t rate_show(struct ib_port *p, struct port_attribute *unused,
224			 char *buf)
225{
226	struct ib_port_attr attr;
227	char *speed = "";
228	int rate;		/* in deci-Gb/sec */
229	ssize_t ret;
230
231	ret = ib_query_port(p->ibdev, p->port_num, &attr);
232	if (ret)
233		return ret;
234
235	switch (attr.active_speed) {
236	case IB_SPEED_DDR:
237		speed = " DDR";
238		rate = 50;
239		break;
240	case IB_SPEED_QDR:
241		speed = " QDR";
242		rate = 100;
243		break;
244	case IB_SPEED_FDR10:
245		speed = " FDR10";
246		rate = 100;
247		break;
248	case IB_SPEED_FDR:
249		speed = " FDR";
250		rate = 140;
251		break;
252	case IB_SPEED_EDR:
253		speed = " EDR";
254		rate = 250;
255		break;
256	case IB_SPEED_SDR:
257	default:		/* default to SDR for invalid rates */
258		rate = 25;
259		break;
260	}
261
262	rate *= ib_width_enum_to_int(attr.active_width);
263	if (rate < 0)
264		return -EINVAL;
265
266	return sprintf(buf, "%d%s Gb/sec (%dX%s)\n",
267		       rate / 10, rate % 10 ? ".5" : "",
268		       ib_width_enum_to_int(attr.active_width), speed);
269}
270
271static ssize_t phys_state_show(struct ib_port *p, struct port_attribute *unused,
272			       char *buf)
273{
274	struct ib_port_attr attr;
275
276	ssize_t ret;
277
278	ret = ib_query_port(p->ibdev, p->port_num, &attr);
279	if (ret)
280		return ret;
281
282	switch (attr.phys_state) {
283	case 1:  return sprintf(buf, "1: Sleep\n");
284	case 2:  return sprintf(buf, "2: Polling\n");
285	case 3:  return sprintf(buf, "3: Disabled\n");
286	case 4:  return sprintf(buf, "4: PortConfigurationTraining\n");
287	case 5:  return sprintf(buf, "5: LinkUp\n");
288	case 6:  return sprintf(buf, "6: LinkErrorRecovery\n");
289	case 7:  return sprintf(buf, "7: Phy Test\n");
290	default: return sprintf(buf, "%d: <unknown>\n", attr.phys_state);
291	}
292}
293
294static ssize_t link_layer_show(struct ib_port *p, struct port_attribute *unused,
295			       char *buf)
296{
297	switch (rdma_port_get_link_layer(p->ibdev, p->port_num)) {
298	case IB_LINK_LAYER_INFINIBAND:
299		return sprintf(buf, "%s\n", "InfiniBand");
300	case IB_LINK_LAYER_ETHERNET:
301		return sprintf(buf, "%s\n", "Ethernet");
302	default:
303		return sprintf(buf, "%s\n", "Unknown");
304	}
305}
306
307static PORT_ATTR_RO(state);
308static PORT_ATTR_RO(lid);
309static PORT_ATTR_RO(lid_mask_count);
310static PORT_ATTR_RO(sm_lid);
311static PORT_ATTR_RO(sm_sl);
312static PORT_ATTR_RO(cap_mask);
313static PORT_ATTR_RO(rate);
314static PORT_ATTR_RO(phys_state);
315static PORT_ATTR_RO(link_layer);
316
317static struct attribute *port_default_attrs[] = {
318	&port_attr_state.attr,
319	&port_attr_lid.attr,
320	&port_attr_lid_mask_count.attr,
321	&port_attr_sm_lid.attr,
322	&port_attr_sm_sl.attr,
323	&port_attr_cap_mask.attr,
324	&port_attr_rate.attr,
325	&port_attr_phys_state.attr,
326	&port_attr_link_layer.attr,
327	NULL
328};
329
330static size_t print_ndev(struct ib_gid_attr *gid_attr, char *buf)
331{
332	if (!gid_attr->ndev)
333		return -EINVAL;
334
335	return sprintf(buf, "%s\n", if_name(gid_attr->ndev));
336}
337
338static size_t print_gid_type(struct ib_gid_attr *gid_attr, char *buf)
339{
340	return sprintf(buf, "%s\n", ib_cache_gid_type_str(gid_attr->gid_type));
341}
342
343static ssize_t _show_port_gid_attr(struct ib_port *p,
344				   struct port_attribute *attr,
345				   char *buf,
346				   size_t (*print)(struct ib_gid_attr *gid_attr,
347						   char *buf))
348{
349	struct port_table_attribute *tab_attr =
350		container_of(attr, struct port_table_attribute, attr);
351	union ib_gid gid;
352	struct ib_gid_attr gid_attr = {};
353	ssize_t ret;
354
355	ret = ib_query_gid(p->ibdev, p->port_num, tab_attr->index, &gid,
356			   &gid_attr);
357	if (ret)
358		goto err;
359
360	ret = print(&gid_attr, buf);
361
362err:
363	if (gid_attr.ndev)
364		dev_put(gid_attr.ndev);
365	return ret;
366}
367
368static ssize_t show_port_gid(struct ib_port *p, struct port_attribute *attr,
369			     char *buf)
370{
371	struct port_table_attribute *tab_attr =
372		container_of(attr, struct port_table_attribute, attr);
373	union ib_gid gid;
374	ssize_t ret;
375
376	ret = ib_query_gid(p->ibdev, p->port_num, tab_attr->index, &gid, NULL);
377	if (ret)
378		return ret;
379
380	return sprintf(buf, GID_PRINT_FMT"\n", GID_PRINT_ARGS(gid.raw));
381}
382
383static ssize_t show_port_gid_attr_ndev(struct ib_port *p,
384				       struct port_attribute *attr, char *buf)
385{
386	return _show_port_gid_attr(p, attr, buf, print_ndev);
387}
388
389static ssize_t show_port_gid_attr_gid_type(struct ib_port *p,
390					   struct port_attribute *attr,
391					   char *buf)
392{
393	return _show_port_gid_attr(p, attr, buf, print_gid_type);
394}
395
396static ssize_t show_port_pkey(struct ib_port *p, struct port_attribute *attr,
397			      char *buf)
398{
399	struct port_table_attribute *tab_attr =
400		container_of(attr, struct port_table_attribute, attr);
401	u16 pkey;
402	ssize_t ret;
403
404	ret = ib_query_pkey(p->ibdev, p->port_num, tab_attr->index, &pkey);
405	if (ret)
406		return ret;
407
408	return sprintf(buf, "0x%04x\n", pkey);
409}
410
411#define PORT_PMA_ATTR(_name, _counter, _width, _offset)			\
412struct port_table_attribute port_pma_attr_##_name = {			\
413	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
414	.index = (_offset) | ((_width) << 16) | ((_counter) << 24),	\
415	.attr_id = IB_PMA_PORT_COUNTERS ,				\
416}
417
418#define PORT_PMA_ATTR_EXT(_name, _width, _offset)			\
419struct port_table_attribute port_pma_attr_ext_##_name = {		\
420	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
421	.index = (_offset) | ((_width) << 16),				\
422	.attr_id = IB_PMA_PORT_COUNTERS_EXT ,				\
423}
424
425/*
426 * Get a Perfmgmt MAD block of data.
427 * Returns error code or the number of bytes retrieved.
428 */
429static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
430		void *data, int offset, size_t size)
431{
432	struct ib_mad *in_mad;
433	struct ib_mad *out_mad;
434	size_t mad_size = sizeof(*out_mad);
435	u16 out_mad_pkey_index = 0;
436	ssize_t ret;
437
438	if (!dev->process_mad)
439		return -ENOSYS;
440
441	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
442	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
443	if (!in_mad || !out_mad) {
444		ret = -ENOMEM;
445		goto out;
446	}
447
448	in_mad->mad_hdr.base_version  = 1;
449	in_mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_PERF_MGMT;
450	in_mad->mad_hdr.class_version = 1;
451	in_mad->mad_hdr.method        = IB_MGMT_METHOD_GET;
452	in_mad->mad_hdr.attr_id       = attr;
453
454	if (attr != IB_PMA_CLASS_PORT_INFO)
455		in_mad->data[41] = port_num;	/* PortSelect field */
456
457	if ((dev->process_mad(dev, IB_MAD_IGNORE_MKEY,
458		 port_num, NULL, NULL,
459		 (const struct ib_mad_hdr *)in_mad, mad_size,
460		 (struct ib_mad_hdr *)out_mad, &mad_size,
461		 &out_mad_pkey_index) &
462	     (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
463	    (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
464		ret = -EINVAL;
465		goto out;
466	}
467	memcpy(data, out_mad->data + offset, size);
468	ret = size;
469out:
470	kfree(in_mad);
471	kfree(out_mad);
472	return ret;
473}
474
475static ssize_t show_pma_counter(struct ib_port *p, struct port_attribute *attr,
476				char *buf)
477{
478	struct port_table_attribute *tab_attr =
479		container_of(attr, struct port_table_attribute, attr);
480	int offset = tab_attr->index & 0xffff;
481	int width  = (tab_attr->index >> 16) & 0xff;
482	ssize_t ret;
483	u8 data[8];
484
485	ret = get_perf_mad(p->ibdev, p->port_num, tab_attr->attr_id, &data,
486			40 + offset / 8, sizeof(data));
487	if (ret < 0)
488		return sprintf(buf, "N/A (no PMA)\n");
489
490	switch (width) {
491	case 4:
492		ret = sprintf(buf, "%u\n", (*data >>
493					    (4 - (offset % 8))) & 0xf);
494		break;
495	case 8:
496		ret = sprintf(buf, "%u\n", *data);
497		break;
498	case 16:
499		ret = sprintf(buf, "%u\n",
500			      be16_to_cpup((__be16 *)data));
501		break;
502	case 32:
503		ret = sprintf(buf, "%u\n",
504			      be32_to_cpup((__be32 *)data));
505		break;
506	case 64:
507		ret = sprintf(buf, "%llu\n",
508			      (unsigned long long)be64_to_cpup((__be64 *)data));
509		break;
510
511	default:
512		ret = 0;
513	}
514
515	return ret;
516}
517
518static PORT_PMA_ATTR(symbol_error		    ,  0, 16,  32);
519static PORT_PMA_ATTR(link_error_recovery	    ,  1,  8,  48);
520static PORT_PMA_ATTR(link_downed		    ,  2,  8,  56);
521static PORT_PMA_ATTR(port_rcv_errors		    ,  3, 16,  64);
522static PORT_PMA_ATTR(port_rcv_remote_physical_errors,  4, 16,  80);
523static PORT_PMA_ATTR(port_rcv_switch_relay_errors   ,  5, 16,  96);
524static PORT_PMA_ATTR(port_xmit_discards		    ,  6, 16, 112);
525static PORT_PMA_ATTR(port_xmit_constraint_errors    ,  7,  8, 128);
526static PORT_PMA_ATTR(port_rcv_constraint_errors	    ,  8,  8, 136);
527static PORT_PMA_ATTR(local_link_integrity_errors    ,  9,  4, 152);
528static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10,  4, 156);
529static PORT_PMA_ATTR(VL15_dropped		    , 11, 16, 176);
530static PORT_PMA_ATTR(port_xmit_data		    , 12, 32, 192);
531static PORT_PMA_ATTR(port_rcv_data		    , 13, 32, 224);
532static PORT_PMA_ATTR(port_xmit_packets		    , 14, 32, 256);
533static PORT_PMA_ATTR(port_rcv_packets		    , 15, 32, 288);
534static PORT_PMA_ATTR(port_xmit_wait		    ,  0, 32, 320);
535
536/*
537 * Counters added by extended set
538 */
539static PORT_PMA_ATTR_EXT(port_xmit_data		    , 64,  64);
540static PORT_PMA_ATTR_EXT(port_rcv_data		    , 64, 128);
541static PORT_PMA_ATTR_EXT(port_xmit_packets	    , 64, 192);
542static PORT_PMA_ATTR_EXT(port_rcv_packets	    , 64, 256);
543static PORT_PMA_ATTR_EXT(unicast_xmit_packets	    , 64, 320);
544static PORT_PMA_ATTR_EXT(unicast_rcv_packets	    , 64, 384);
545static PORT_PMA_ATTR_EXT(multicast_xmit_packets	    , 64, 448);
546static PORT_PMA_ATTR_EXT(multicast_rcv_packets	    , 64, 512);
547
548static struct attribute *pma_attrs[] = {
549	&port_pma_attr_symbol_error.attr.attr,
550	&port_pma_attr_link_error_recovery.attr.attr,
551	&port_pma_attr_link_downed.attr.attr,
552	&port_pma_attr_port_rcv_errors.attr.attr,
553	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
554	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
555	&port_pma_attr_port_xmit_discards.attr.attr,
556	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
557	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
558	&port_pma_attr_local_link_integrity_errors.attr.attr,
559	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
560	&port_pma_attr_VL15_dropped.attr.attr,
561	&port_pma_attr_port_xmit_data.attr.attr,
562	&port_pma_attr_port_rcv_data.attr.attr,
563	&port_pma_attr_port_xmit_packets.attr.attr,
564	&port_pma_attr_port_rcv_packets.attr.attr,
565	&port_pma_attr_port_xmit_wait.attr.attr,
566	NULL
567};
568
569static struct attribute *pma_attrs_ext[] = {
570	&port_pma_attr_symbol_error.attr.attr,
571	&port_pma_attr_link_error_recovery.attr.attr,
572	&port_pma_attr_link_downed.attr.attr,
573	&port_pma_attr_port_rcv_errors.attr.attr,
574	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
575	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
576	&port_pma_attr_port_xmit_discards.attr.attr,
577	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
578	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
579	&port_pma_attr_local_link_integrity_errors.attr.attr,
580	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
581	&port_pma_attr_VL15_dropped.attr.attr,
582	&port_pma_attr_ext_port_xmit_data.attr.attr,
583	&port_pma_attr_ext_port_rcv_data.attr.attr,
584	&port_pma_attr_ext_port_xmit_packets.attr.attr,
585	&port_pma_attr_port_xmit_wait.attr.attr,
586	&port_pma_attr_ext_port_rcv_packets.attr.attr,
587	&port_pma_attr_ext_unicast_rcv_packets.attr.attr,
588	&port_pma_attr_ext_unicast_xmit_packets.attr.attr,
589	&port_pma_attr_ext_multicast_rcv_packets.attr.attr,
590	&port_pma_attr_ext_multicast_xmit_packets.attr.attr,
591	NULL
592};
593
594static struct attribute *pma_attrs_noietf[] = {
595	&port_pma_attr_symbol_error.attr.attr,
596	&port_pma_attr_link_error_recovery.attr.attr,
597	&port_pma_attr_link_downed.attr.attr,
598	&port_pma_attr_port_rcv_errors.attr.attr,
599	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
600	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
601	&port_pma_attr_port_xmit_discards.attr.attr,
602	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
603	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
604	&port_pma_attr_local_link_integrity_errors.attr.attr,
605	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
606	&port_pma_attr_VL15_dropped.attr.attr,
607	&port_pma_attr_ext_port_xmit_data.attr.attr,
608	&port_pma_attr_ext_port_rcv_data.attr.attr,
609	&port_pma_attr_ext_port_xmit_packets.attr.attr,
610	&port_pma_attr_ext_port_rcv_packets.attr.attr,
611	&port_pma_attr_port_xmit_wait.attr.attr,
612	NULL
613};
614
615static struct attribute_group pma_group = {
616	.name  = "counters",
617	.attrs  = pma_attrs
618};
619
620static struct attribute_group pma_group_ext = {
621	.name  = "counters",
622	.attrs  = pma_attrs_ext
623};
624
625static struct attribute_group pma_group_noietf = {
626	.name  = "counters",
627	.attrs  = pma_attrs_noietf
628};
629
630static void ib_port_release(struct kobject *kobj)
631{
632	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
633	struct attribute *a;
634	int i;
635
636	if (p->gid_group.attrs) {
637		for (i = 0; (a = p->gid_group.attrs[i]); ++i)
638			kfree(a);
639
640		kfree(p->gid_group.attrs);
641	}
642
643	if (p->pkey_group.attrs) {
644		for (i = 0; (a = p->pkey_group.attrs[i]); ++i)
645			kfree(a);
646
647		kfree(p->pkey_group.attrs);
648	}
649
650	kfree(p);
651}
652
653static void ib_port_gid_attr_release(struct kobject *kobj)
654{
655	struct gid_attr_group *g = container_of(kobj, struct gid_attr_group,
656						kobj);
657	struct attribute *a;
658	int i;
659
660	if (g->ndev.attrs) {
661		for (i = 0; (a = g->ndev.attrs[i]); ++i)
662			kfree(a);
663
664		kfree(g->ndev.attrs);
665	}
666
667	if (g->type.attrs) {
668		for (i = 0; (a = g->type.attrs[i]); ++i)
669			kfree(a);
670
671		kfree(g->type.attrs);
672	}
673
674	kfree(g);
675}
676
677static struct kobj_type port_type = {
678	.release       = ib_port_release,
679	.sysfs_ops     = &port_sysfs_ops,
680	.default_attrs = port_default_attrs
681};
682
683static struct kobj_type gid_attr_type = {
684	.sysfs_ops      = &gid_attr_sysfs_ops,
685	.release        = ib_port_gid_attr_release
686};
687
688static struct attribute **
689alloc_group_attrs(ssize_t (*show)(struct ib_port *,
690				  struct port_attribute *, char *buf),
691		  int len)
692{
693	struct attribute **tab_attr;
694	struct port_table_attribute *element;
695	int i;
696
697	tab_attr = kcalloc(1 + len, sizeof(struct attribute *), GFP_KERNEL);
698	if (!tab_attr)
699		return NULL;
700
701	for (i = 0; i < len; i++) {
702		element = kzalloc(sizeof(struct port_table_attribute),
703				  GFP_KERNEL);
704		if (!element)
705			goto err;
706
707		if (snprintf(element->name, sizeof(element->name),
708			     "%d", i) >= sizeof(element->name)) {
709			kfree(element);
710			goto err;
711		}
712
713		element->attr.attr.name  = element->name;
714		element->attr.attr.mode  = S_IRUGO;
715		element->attr.show       = show;
716		element->index		 = i;
717		sysfs_attr_init(&element->attr.attr);
718
719		tab_attr[i] = &element->attr.attr;
720	}
721
722	return tab_attr;
723
724err:
725	while (--i >= 0)
726		kfree(tab_attr[i]);
727	kfree(tab_attr);
728	return NULL;
729}
730
731/*
732 * Figure out which counter table to use depending on
733 * the device capabilities.
734 */
735static struct attribute_group *get_counter_table(struct ib_device *dev,
736						 int port_num)
737{
738	struct ib_class_port_info cpi;
739
740	if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
741				&cpi, 40, sizeof(cpi)) >= 0) {
742		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
743			/* We have extended counters */
744			return &pma_group_ext;
745
746		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
747			/* But not the IETF ones */
748			return &pma_group_noietf;
749	}
750
751	/* Fall back to normal counters */
752	return &pma_group;
753}
754
755static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
756			   u8 port_num, int index)
757{
758	int ret;
759
760	if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
761		return 0;
762	ret = dev->get_hw_stats(dev, stats, port_num, index);
763	if (ret < 0)
764		return ret;
765	if (ret == stats->num_counters)
766		stats->timestamp = jiffies;
767
768	return 0;
769}
770
771static ssize_t print_hw_stat(struct rdma_hw_stats *stats, int index, char *buf)
772{
773	return sprintf(buf, "%llu\n", (unsigned long long)stats->value[index]);
774}
775
776static ssize_t show_hw_stats(struct kobject *kobj, struct attribute *attr,
777			     char *buf)
778{
779	struct ib_device *dev;
780	struct ib_port *port;
781	struct hw_stats_attribute *hsa;
782	struct rdma_hw_stats *stats;
783	int ret;
784
785	hsa = container_of(attr, struct hw_stats_attribute, attr);
786	if (!hsa->port_num) {
787		dev = container_of((struct device *)kobj,
788				   struct ib_device, dev);
789		stats = dev->hw_stats;
790	} else {
791		port = container_of(kobj, struct ib_port, kobj);
792		dev = port->ibdev;
793		stats = port->hw_stats;
794	}
795	ret = update_hw_stats(dev, stats, hsa->port_num, hsa->index);
796	if (ret)
797		return ret;
798	return print_hw_stat(stats, hsa->index, buf);
799}
800
801static ssize_t show_stats_lifespan(struct kobject *kobj,
802				   struct attribute *attr,
803				   char *buf)
804{
805	struct hw_stats_attribute *hsa;
806	int msecs;
807
808	hsa = container_of(attr, struct hw_stats_attribute, attr);
809	if (!hsa->port_num) {
810		struct ib_device *dev = container_of((struct device *)kobj,
811						     struct ib_device, dev);
812		msecs = jiffies_to_msecs(dev->hw_stats->lifespan);
813	} else {
814		struct ib_port *p = container_of(kobj, struct ib_port, kobj);
815		msecs = jiffies_to_msecs(p->hw_stats->lifespan);
816	}
817	return sprintf(buf, "%d\n", msecs);
818}
819
820static ssize_t set_stats_lifespan(struct kobject *kobj,
821				  struct attribute *attr,
822				  const char *buf, size_t count)
823{
824	struct hw_stats_attribute *hsa;
825	int msecs;
826	int jiffies;
827	int ret;
828
829	ret = kstrtoint(buf, 10, &msecs);
830	if (ret)
831		return ret;
832	if (msecs < 0 || msecs > 10000)
833		return -EINVAL;
834	jiffies = msecs_to_jiffies(msecs);
835	hsa = container_of(attr, struct hw_stats_attribute, attr);
836	if (!hsa->port_num) {
837		struct ib_device *dev = container_of((struct device *)kobj,
838						     struct ib_device, dev);
839		dev->hw_stats->lifespan = jiffies;
840	} else {
841		struct ib_port *p = container_of(kobj, struct ib_port, kobj);
842		p->hw_stats->lifespan = jiffies;
843	}
844	return count;
845}
846
847static void free_hsag(struct kobject *kobj, struct attribute_group *attr_group)
848{
849	struct attribute **attr;
850
851	sysfs_remove_group(kobj, attr_group);
852
853	for (attr = attr_group->attrs; *attr; attr++)
854		kfree(*attr);
855	kfree(attr_group);
856}
857
858static struct attribute *alloc_hsa(int index, u8 port_num, const char *name)
859{
860	struct hw_stats_attribute *hsa;
861
862	hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
863	if (!hsa)
864		return NULL;
865
866	hsa->attr.name = __DECONST(char *, name);
867	hsa->attr.mode = S_IRUGO;
868	hsa->show = show_hw_stats;
869	hsa->store = NULL;
870	hsa->index = index;
871	hsa->port_num = port_num;
872
873	return &hsa->attr;
874}
875
876static struct attribute *alloc_hsa_lifespan(char *name, u8 port_num)
877{
878	struct hw_stats_attribute *hsa;
879
880	hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
881	if (!hsa)
882		return NULL;
883
884	hsa->attr.name = name;
885	hsa->attr.mode = S_IWUSR | S_IRUGO;
886	hsa->show = show_stats_lifespan;
887	hsa->store = set_stats_lifespan;
888	hsa->index = 0;
889	hsa->port_num = port_num;
890
891	return &hsa->attr;
892}
893
894static void setup_hw_stats(struct ib_device *device, struct ib_port *port,
895			   u8 port_num)
896{
897	struct attribute_group *hsag;
898	struct rdma_hw_stats *stats;
899	int i, ret;
900
901	stats = device->alloc_hw_stats(device, port_num);
902
903	if (!stats)
904		return;
905
906	if (!stats->names || stats->num_counters <= 0)
907		goto err_free_stats;
908
909	/*
910	 * Two extra attribue elements here, one for the lifespan entry and
911	 * one to NULL terminate the list for the sysfs core code
912	 */
913	hsag = kzalloc(sizeof(*hsag) +
914		       sizeof(void *) * (stats->num_counters + 2),
915		       GFP_KERNEL);
916	if (!hsag)
917		goto err_free_stats;
918
919	ret = device->get_hw_stats(device, stats, port_num,
920				   stats->num_counters);
921	if (ret != stats->num_counters)
922		goto err_free_hsag;
923
924	stats->timestamp = jiffies;
925
926	hsag->name = "hw_counters";
927	hsag->attrs = (void *)((char *)hsag + sizeof(*hsag));
928
929	for (i = 0; i < stats->num_counters; i++) {
930		hsag->attrs[i] = alloc_hsa(i, port_num, stats->names[i]);
931		if (!hsag->attrs[i])
932			goto err;
933		sysfs_attr_init(hsag->attrs[i]);
934	}
935
936	/* treat an error here as non-fatal */
937	hsag->attrs[i] = alloc_hsa_lifespan("lifespan", port_num);
938	if (hsag->attrs[i])
939		sysfs_attr_init(hsag->attrs[i]);
940
941	if (port) {
942		struct kobject *kobj = &port->kobj;
943		ret = sysfs_create_group(kobj, hsag);
944		if (ret)
945			goto err;
946		port->hw_stats_ag = hsag;
947		port->hw_stats = stats;
948	} else {
949		struct kobject *kobj = &device->dev.kobj;
950		ret = sysfs_create_group(kobj, hsag);
951		if (ret)
952			goto err;
953		device->hw_stats_ag = hsag;
954		device->hw_stats = stats;
955	}
956
957	return;
958
959err:
960	for (; i >= 0; i--)
961		kfree(hsag->attrs[i]);
962err_free_hsag:
963	kfree(hsag);
964err_free_stats:
965	kfree(stats);
966	return;
967}
968
969static int add_port(struct ib_device *device, int port_num,
970		    int (*port_callback)(struct ib_device *,
971					 u8, struct kobject *))
972{
973	struct ib_port *p;
974	struct ib_port_attr attr;
975	int i;
976	int ret;
977
978	ret = ib_query_port(device, port_num, &attr);
979	if (ret)
980		return ret;
981
982	p = kzalloc(sizeof *p, GFP_KERNEL);
983	if (!p)
984		return -ENOMEM;
985
986	p->ibdev      = device;
987	p->port_num   = port_num;
988
989	ret = kobject_init_and_add(&p->kobj, &port_type,
990				   device->ports_parent,
991				   "%d", port_num);
992	if (ret) {
993		kfree(p);
994		return ret;
995	}
996
997	p->gid_attr_group = kzalloc(sizeof(*p->gid_attr_group), GFP_KERNEL);
998	if (!p->gid_attr_group) {
999		ret = -ENOMEM;
1000		goto err_put;
1001	}
1002
1003	p->gid_attr_group->port = p;
1004	ret = kobject_init_and_add(&p->gid_attr_group->kobj, &gid_attr_type,
1005				   &p->kobj, "gid_attrs");
1006	if (ret) {
1007		kfree(p->gid_attr_group);
1008		goto err_put;
1009	}
1010
1011	p->pma_table = get_counter_table(device, port_num);
1012	ret = sysfs_create_group(&p->kobj, p->pma_table);
1013	if (ret)
1014		goto err_put_gid_attrs;
1015
1016	p->gid_group.name  = "gids";
1017	p->gid_group.attrs = alloc_group_attrs(show_port_gid, attr.gid_tbl_len);
1018	if (!p->gid_group.attrs) {
1019		ret = -ENOMEM;
1020		goto err_remove_pma;
1021	}
1022
1023	ret = sysfs_create_group(&p->kobj, &p->gid_group);
1024	if (ret)
1025		goto err_free_gid;
1026
1027	p->gid_attr_group->ndev.name = "ndevs";
1028	p->gid_attr_group->ndev.attrs = alloc_group_attrs(show_port_gid_attr_ndev,
1029							  attr.gid_tbl_len);
1030	if (!p->gid_attr_group->ndev.attrs) {
1031		ret = -ENOMEM;
1032		goto err_remove_gid;
1033	}
1034
1035	ret = sysfs_create_group(&p->gid_attr_group->kobj,
1036				 &p->gid_attr_group->ndev);
1037	if (ret)
1038		goto err_free_gid_ndev;
1039
1040	p->gid_attr_group->type.name = "types";
1041	p->gid_attr_group->type.attrs = alloc_group_attrs(show_port_gid_attr_gid_type,
1042							  attr.gid_tbl_len);
1043	if (!p->gid_attr_group->type.attrs) {
1044		ret = -ENOMEM;
1045		goto err_remove_gid_ndev;
1046	}
1047
1048	ret = sysfs_create_group(&p->gid_attr_group->kobj,
1049				 &p->gid_attr_group->type);
1050	if (ret)
1051		goto err_free_gid_type;
1052
1053	p->pkey_group.name  = "pkeys";
1054	p->pkey_group.attrs = alloc_group_attrs(show_port_pkey,
1055						attr.pkey_tbl_len);
1056	if (!p->pkey_group.attrs) {
1057		ret = -ENOMEM;
1058		goto err_remove_gid_type;
1059	}
1060
1061	ret = sysfs_create_group(&p->kobj, &p->pkey_group);
1062	if (ret)
1063		goto err_free_pkey;
1064
1065	if (port_callback) {
1066		ret = port_callback(device, port_num, &p->kobj);
1067		if (ret)
1068			goto err_remove_pkey;
1069	}
1070
1071	/*
1072	 * If port == 0, it means we have only one port and the parent
1073	 * device, not this port device, should be the holder of the
1074	 * hw_counters
1075	 */
1076	if (device->alloc_hw_stats && port_num)
1077		setup_hw_stats(device, p, port_num);
1078
1079	list_add_tail(&p->kobj.entry, &device->port_list);
1080
1081	return 0;
1082
1083err_remove_pkey:
1084	sysfs_remove_group(&p->kobj, &p->pkey_group);
1085
1086err_free_pkey:
1087	for (i = 0; i < attr.pkey_tbl_len; ++i)
1088		kfree(p->pkey_group.attrs[i]);
1089
1090	kfree(p->pkey_group.attrs);
1091	p->pkey_group.attrs = NULL;
1092
1093err_remove_gid_type:
1094	sysfs_remove_group(&p->gid_attr_group->kobj,
1095			   &p->gid_attr_group->type);
1096
1097err_free_gid_type:
1098	for (i = 0; i < attr.gid_tbl_len; ++i)
1099		kfree(p->gid_attr_group->type.attrs[i]);
1100
1101	kfree(p->gid_attr_group->type.attrs);
1102	p->gid_attr_group->type.attrs = NULL;
1103
1104err_remove_gid_ndev:
1105	sysfs_remove_group(&p->gid_attr_group->kobj,
1106			   &p->gid_attr_group->ndev);
1107
1108err_free_gid_ndev:
1109	for (i = 0; i < attr.gid_tbl_len; ++i)
1110		kfree(p->gid_attr_group->ndev.attrs[i]);
1111
1112	kfree(p->gid_attr_group->ndev.attrs);
1113	p->gid_attr_group->ndev.attrs = NULL;
1114
1115err_remove_gid:
1116	sysfs_remove_group(&p->kobj, &p->gid_group);
1117
1118err_free_gid:
1119	for (i = 0; i < attr.gid_tbl_len; ++i)
1120		kfree(p->gid_group.attrs[i]);
1121
1122	kfree(p->gid_group.attrs);
1123	p->gid_group.attrs = NULL;
1124
1125err_remove_pma:
1126	sysfs_remove_group(&p->kobj, p->pma_table);
1127
1128err_put_gid_attrs:
1129	kobject_put(&p->gid_attr_group->kobj);
1130
1131err_put:
1132	kobject_put(&p->kobj);
1133	return ret;
1134}
1135
1136static ssize_t show_node_type(struct device *device,
1137			      struct device_attribute *attr, char *buf)
1138{
1139	struct ib_device *dev = container_of(device, struct ib_device, dev);
1140
1141	switch (dev->node_type) {
1142	case RDMA_NODE_IB_CA:	  return sprintf(buf, "%d: CA\n", dev->node_type);
1143	case RDMA_NODE_RNIC:	  return sprintf(buf, "%d: RNIC\n", dev->node_type);
1144	case RDMA_NODE_USNIC:	  return sprintf(buf, "%d: usNIC\n", dev->node_type);
1145	case RDMA_NODE_USNIC_UDP: return sprintf(buf, "%d: usNIC UDP\n", dev->node_type);
1146	case RDMA_NODE_IB_SWITCH: return sprintf(buf, "%d: switch\n", dev->node_type);
1147	case RDMA_NODE_IB_ROUTER: return sprintf(buf, "%d: router\n", dev->node_type);
1148	default:		  return sprintf(buf, "%d: <unknown>\n", dev->node_type);
1149	}
1150}
1151
1152static ssize_t show_sys_image_guid(struct device *device,
1153				   struct device_attribute *dev_attr, char *buf)
1154{
1155	struct ib_device *dev = container_of(device, struct ib_device, dev);
1156
1157	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1158		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[0]),
1159		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[1]),
1160		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[2]),
1161		       be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[3]));
1162}
1163
1164static ssize_t show_node_guid(struct device *device,
1165			      struct device_attribute *attr, char *buf)
1166{
1167	struct ib_device *dev = container_of(device, struct ib_device, dev);
1168
1169	return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1170		       be16_to_cpu(((__be16 *) &dev->node_guid)[0]),
1171		       be16_to_cpu(((__be16 *) &dev->node_guid)[1]),
1172		       be16_to_cpu(((__be16 *) &dev->node_guid)[2]),
1173		       be16_to_cpu(((__be16 *) &dev->node_guid)[3]));
1174}
1175
1176static ssize_t show_node_desc(struct device *device,
1177			      struct device_attribute *attr, char *buf)
1178{
1179	struct ib_device *dev = container_of(device, struct ib_device, dev);
1180
1181	return sprintf(buf, "%.64s\n", dev->node_desc);
1182}
1183
1184static ssize_t set_node_desc(struct device *device,
1185			     struct device_attribute *attr,
1186			     const char *buf, size_t count)
1187{
1188	struct ib_device *dev = container_of(device, struct ib_device, dev);
1189	struct ib_device_modify desc = {};
1190	int ret;
1191
1192	if (!dev->modify_device)
1193		return -EIO;
1194
1195	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1196	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1197	if (ret)
1198		return ret;
1199
1200	return count;
1201}
1202
1203static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1204			   char *buf)
1205{
1206	struct ib_device *dev = container_of(device, struct ib_device, dev);
1207
1208	ib_get_device_fw_str(dev, buf, PAGE_SIZE);
1209	strlcat(buf, "\n", PAGE_SIZE);
1210	return strlen(buf);
1211}
1212
1213static DEVICE_ATTR(node_type, S_IRUGO, show_node_type, NULL);
1214static DEVICE_ATTR(sys_image_guid, S_IRUGO, show_sys_image_guid, NULL);
1215static DEVICE_ATTR(node_guid, S_IRUGO, show_node_guid, NULL);
1216static DEVICE_ATTR(node_desc, S_IRUGO | S_IWUSR, show_node_desc, set_node_desc);
1217static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1218
1219static struct device_attribute *ib_class_attributes[] = {
1220	&dev_attr_node_type,
1221	&dev_attr_sys_image_guid,
1222	&dev_attr_node_guid,
1223	&dev_attr_node_desc,
1224	&dev_attr_fw_ver,
1225};
1226
1227static void free_port_list_attributes(struct ib_device *device)
1228{
1229	struct kobject *p, *t;
1230
1231	list_for_each_entry_safe(p, t, &device->port_list, entry) {
1232		struct ib_port *port = container_of(p, struct ib_port, kobj);
1233		list_del(&p->entry);
1234		if (port->hw_stats) {
1235			kfree(port->hw_stats);
1236			free_hsag(&port->kobj, port->hw_stats_ag);
1237		}
1238		sysfs_remove_group(p, port->pma_table);
1239		sysfs_remove_group(p, &port->pkey_group);
1240		sysfs_remove_group(p, &port->gid_group);
1241		sysfs_remove_group(&port->gid_attr_group->kobj,
1242				   &port->gid_attr_group->ndev);
1243		sysfs_remove_group(&port->gid_attr_group->kobj,
1244				   &port->gid_attr_group->type);
1245		kobject_put(&port->gid_attr_group->kobj);
1246		kobject_put(p);
1247	}
1248
1249	kobject_put(device->ports_parent);
1250}
1251
1252int ib_device_register_sysfs(struct ib_device *device,
1253			     int (*port_callback)(struct ib_device *,
1254						  u8, struct kobject *))
1255{
1256	struct device *class_dev = &device->dev;
1257	int ret;
1258	int i;
1259
1260	device->dev.parent = device->dma_device;
1261	ret = dev_set_name(class_dev, "%s", device->name);
1262	if (ret)
1263		return ret;
1264
1265	ret = device_add(class_dev);
1266	if (ret)
1267		goto err;
1268
1269	for (i = 0; i < ARRAY_SIZE(ib_class_attributes); ++i) {
1270		ret = device_create_file(class_dev, ib_class_attributes[i]);
1271		if (ret)
1272			goto err_unregister;
1273	}
1274
1275	device->ports_parent = kobject_create_and_add("ports",
1276						      &class_dev->kobj);
1277	if (!device->ports_parent) {
1278		ret = -ENOMEM;
1279		goto err_put;
1280	}
1281
1282	if (rdma_cap_ib_switch(device)) {
1283		ret = add_port(device, 0, port_callback);
1284		if (ret)
1285			goto err_put;
1286	} else {
1287		for (i = 1; i <= device->phys_port_cnt; ++i) {
1288			ret = add_port(device, i, port_callback);
1289			if (ret)
1290				goto err_put;
1291		}
1292	}
1293
1294	if (device->alloc_hw_stats)
1295		setup_hw_stats(device, NULL, 0);
1296
1297	return 0;
1298
1299err_put:
1300	free_port_list_attributes(device);
1301
1302err_unregister:
1303	device_unregister(class_dev);
1304
1305err:
1306	return ret;
1307}
1308
1309void ib_device_unregister_sysfs(struct ib_device *device)
1310{
1311	int i;
1312
1313	/* Hold kobject until ib_dealloc_device() */
1314	kobject_get(&device->dev.kobj);
1315
1316	free_port_list_attributes(device);
1317
1318	if (device->hw_stats) {
1319		kfree(device->hw_stats);
1320		free_hsag(&device->dev.kobj, device->hw_stats_ag);
1321	}
1322
1323	for (i = 0; i < ARRAY_SIZE(ib_class_attributes); ++i)
1324		device_remove_file(&device->dev, ib_class_attributes[i]);
1325
1326	device_unregister(&device->dev);
1327}
1328