sysfs.c

   1/*
   2 * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
   3 * Copyright (c) 2005 Mellanox Technologies Ltd.  All rights reserved.
   4 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
   5 *
   6 * This software is available to you under a choice of one of two
   7 * licenses.  You may choose to be licensed under the terms of the GNU
   8 * General Public License (GPL) Version 2, available from the file
   9 * COPYING in the main directory of this source tree, or the
  10 * OpenIB.org BSD license below:
  11 *
  12 *     Redistribution and use in source and binary forms, with or
  13 *     without modification, are permitted provided that the following
  14 *     conditions are met:
  15 *
  16 *      - Redistributions of source code must retain the above
  17 *        copyright notice, this list of conditions and the following
  18 *        disclaimer.
  19 *
  20 *      - Redistributions in binary form must reproduce the above
  21 *        copyright notice, this list of conditions and the following
  22 *        disclaimer in the documentation and/or other materials
  23 *        provided with the distribution.
  24 *
  25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  32 * SOFTWARE.
  33 */
  34
  35#include "core_priv.h"
  36
  37#include <linux/slab.h>
  38#include <linux/stat.h>
  39#include <linux/string.h>
  40#include <linux/netdevice.h>
  41#include <linux/ethtool.h>
  42
  43#include <rdma/ib_mad.h>
  44#include <rdma/ib_pma.h>
  45#include <rdma/ib_cache.h>
  46#include <rdma/rdma_counter.h>
  47#include <rdma/ib_sysfs.h>
  48
  49struct port_table_attribute {
  50	struct ib_port_attribute attr;
  51	char			name[8];
  52	int			index;
  53	__be16			attr_id;
  54};
  55
  56struct gid_attr_group {
  57	struct ib_port *port;
  58	struct kobject kobj;
  59	struct attribute_group groups[2];
  60	const struct attribute_group *groups_list[3];
  61	struct port_table_attribute attrs_list[];
  62};
  63
  64struct ib_port {
  65	struct kobject kobj;
  66	struct ib_device *ibdev;
  67	struct gid_attr_group *gid_attr_group;
  68	struct hw_stats_port_data *hw_stats_data;
  69
  70	struct attribute_group groups[3];
  71	const struct attribute_group *groups_list[5];
  72	u32 port_num;
  73	struct port_table_attribute attrs_list[];
  74};
  75
  76struct hw_stats_device_attribute {
  77	struct device_attribute attr;
  78	ssize_t (*show)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
  79			unsigned int index, unsigned int port_num, char *buf);
  80	ssize_t (*store)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
  81			 unsigned int index, unsigned int port_num,
  82			 const char *buf, size_t count);
  83};
  84
  85struct hw_stats_port_attribute {
  86	struct ib_port_attribute attr;
  87	ssize_t (*show)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
  88			unsigned int index, unsigned int port_num, char *buf);
  89	ssize_t (*store)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
  90			 unsigned int index, unsigned int port_num,
  91			 const char *buf, size_t count);
  92};
  93
  94struct hw_stats_device_data {
  95	struct attribute_group group;
  96	struct rdma_hw_stats *stats;
  97	struct hw_stats_device_attribute attrs[];
  98};
  99
 100struct hw_stats_port_data {
 101	struct rdma_hw_stats *stats;
 102	struct hw_stats_port_attribute attrs[];
 103};
 104
 105static ssize_t port_attr_show(struct kobject *kobj,
 106			      struct attribute *attr, char *buf)
 107{
 108	struct ib_port_attribute *port_attr =
 109		container_of(attr, struct ib_port_attribute, attr);
 110	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
 111
 112	if (!port_attr->show)
 113		return -EIO;
 114
 115	return port_attr->show(p->ibdev, p->port_num, port_attr, buf);
 116}
 117
 118static ssize_t port_attr_store(struct kobject *kobj,
 119			       struct attribute *attr,
 120			       const char *buf, size_t count)
 121{
 122	struct ib_port_attribute *port_attr =
 123		container_of(attr, struct ib_port_attribute, attr);
 124	struct ib_port *p = container_of(kobj, struct ib_port, kobj);
 125
 126	if (!port_attr->store)
 127		return -EIO;
 128	return port_attr->store(p->ibdev, p->port_num, port_attr, buf, count);
 129}
 130
 131struct ib_device *ib_port_sysfs_get_ibdev_kobj(struct kobject *kobj,
 132					       u32 *port_num)
 133{
 134	struct ib_port *port = container_of(kobj, struct ib_port, kobj);
 135
 136	*port_num = port->port_num;
 137	return port->ibdev;
 138}
 139EXPORT_SYMBOL(ib_port_sysfs_get_ibdev_kobj);
 140
 141static const struct sysfs_ops port_sysfs_ops = {
 142	.show	= port_attr_show,
 143	.store	= port_attr_store
 144};
 145
 146static ssize_t hw_stat_device_show(struct device *dev,
 147				   struct device_attribute *attr, char *buf)
 148{
 149	struct hw_stats_device_attribute *stat_attr =
 150		container_of(attr, struct hw_stats_device_attribute, attr);
 151	struct ib_device *ibdev = container_of(dev, struct ib_device, dev);
 152
 153	return stat_attr->show(ibdev, ibdev->hw_stats_data->stats,
 154			       stat_attr - ibdev->hw_stats_data->attrs, 0, buf);
 155}
 156
 157static ssize_t hw_stat_device_store(struct device *dev,
 158				    struct device_attribute *attr,
 159				    const char *buf, size_t count)
 160{
 161	struct hw_stats_device_attribute *stat_attr =
 162		container_of(attr, struct hw_stats_device_attribute, attr);
 163	struct ib_device *ibdev = container_of(dev, struct ib_device, dev);
 164
 165	return stat_attr->store(ibdev, ibdev->hw_stats_data->stats,
 166				stat_attr - ibdev->hw_stats_data->attrs, 0, buf,
 167				count);
 168}
 169
 170static ssize_t hw_stat_port_show(struct ib_device *ibdev, u32 port_num,
 171				 struct ib_port_attribute *attr, char *buf)
 172{
 173	struct hw_stats_port_attribute *stat_attr =
 174		container_of(attr, struct hw_stats_port_attribute, attr);
 175	struct ib_port *port = ibdev->port_data[port_num].sysfs;
 176
 177	return stat_attr->show(ibdev, port->hw_stats_data->stats,
 178			       stat_attr - port->hw_stats_data->attrs,
 179			       port->port_num, buf);
 180}
 181
 182static ssize_t hw_stat_port_store(struct ib_device *ibdev, u32 port_num,
 183				  struct ib_port_attribute *attr,
 184				  const char *buf, size_t count)
 185{
 186	struct hw_stats_port_attribute *stat_attr =
 187		container_of(attr, struct hw_stats_port_attribute, attr);
 188	struct ib_port *port = ibdev->port_data[port_num].sysfs;
 189
 190	return stat_attr->store(ibdev, port->hw_stats_data->stats,
 191				stat_attr - port->hw_stats_data->attrs,
 192				port->port_num, buf, count);
 193}
 194
 195static ssize_t gid_attr_show(struct kobject *kobj,
 196			     struct attribute *attr, char *buf)
 197{
 198	struct ib_port_attribute *port_attr =
 199		container_of(attr, struct ib_port_attribute, attr);
 200	struct ib_port *p = container_of(kobj, struct gid_attr_group,
 201					 kobj)->port;
 202
 203	if (!port_attr->show)
 204		return -EIO;
 205
 206	return port_attr->show(p->ibdev, p->port_num, port_attr, buf);
 207}
 208
 209static const struct sysfs_ops gid_attr_sysfs_ops = {
 210	.show = gid_attr_show
 211};
 212
 213static ssize_t state_show(struct ib_device *ibdev, u32 port_num,
 214			  struct ib_port_attribute *unused, char *buf)
 215{
 216	struct ib_port_attr attr;
 217	ssize_t ret;
 218
 219	ret = ib_query_port(ibdev, port_num, &attr);
 220	if (ret)
 221		return ret;
 222
 223	return sysfs_emit(buf, "%d: %s\n", attr.state,
 224			  ib_port_state_to_str(attr.state));
 225}
 226
 227static ssize_t lid_show(struct ib_device *ibdev, u32 port_num,
 228			struct ib_port_attribute *unused, char *buf)
 229{
 230	struct ib_port_attr attr;
 231	ssize_t ret;
 232
 233	ret = ib_query_port(ibdev, port_num, &attr);
 234	if (ret)
 235		return ret;
 236
 237	return sysfs_emit(buf, "0x%x\n", attr.lid);
 238}
 239
 240static ssize_t lid_mask_count_show(struct ib_device *ibdev, u32 port_num,
 241				   struct ib_port_attribute *unused, char *buf)
 242{
 243	struct ib_port_attr attr;
 244	ssize_t ret;
 245
 246	ret = ib_query_port(ibdev, port_num, &attr);
 247	if (ret)
 248		return ret;
 249
 250	return sysfs_emit(buf, "%u\n", attr.lmc);
 251}
 252
 253static ssize_t sm_lid_show(struct ib_device *ibdev, u32 port_num,
 254			   struct ib_port_attribute *unused, char *buf)
 255{
 256	struct ib_port_attr attr;
 257	ssize_t ret;
 258
 259	ret = ib_query_port(ibdev, port_num, &attr);
 260	if (ret)
 261		return ret;
 262
 263	return sysfs_emit(buf, "0x%x\n", attr.sm_lid);
 264}
 265
 266static ssize_t sm_sl_show(struct ib_device *ibdev, u32 port_num,
 267			  struct ib_port_attribute *unused, char *buf)
 268{
 269	struct ib_port_attr attr;
 270	ssize_t ret;
 271
 272	ret = ib_query_port(ibdev, port_num, &attr);
 273	if (ret)
 274		return ret;
 275
 276	return sysfs_emit(buf, "%u\n", attr.sm_sl);
 277}
 278
 279static ssize_t cap_mask_show(struct ib_device *ibdev, u32 port_num,
 280			     struct ib_port_attribute *unused, char *buf)
 281{
 282	struct ib_port_attr attr;
 283	ssize_t ret;
 284
 285	ret = ib_query_port(ibdev, port_num, &attr);
 286	if (ret)
 287		return ret;
 288
 289	return sysfs_emit(buf, "0x%08x\n", attr.port_cap_flags);
 290}
 291
 292static ssize_t rate_show(struct ib_device *ibdev, u32 port_num,
 293			 struct ib_port_attribute *unused, char *buf)
 294{
 295	struct ib_port_attr attr;
 296	char *speed = "";
 297	int rate;		/* in deci-Gb/sec */
 298	ssize_t ret;
 299
 300	ret = ib_query_port(ibdev, port_num, &attr);
 301	if (ret)
 302		return ret;
 303
 304	switch (attr.active_speed) {
 305	case IB_SPEED_DDR:
 306		speed = " DDR";
 307		rate = 50;
 308		break;
 309	case IB_SPEED_QDR:
 310		speed = " QDR";
 311		rate = 100;
 312		break;
 313	case IB_SPEED_FDR10:
 314		speed = " FDR10";
 315		rate = 100;
 316		break;
 317	case IB_SPEED_FDR:
 318		speed = " FDR";
 319		rate = 140;
 320		break;
 321	case IB_SPEED_EDR:
 322		speed = " EDR";
 323		rate = 250;
 324		break;
 325	case IB_SPEED_HDR:
 326		speed = " HDR";
 327		rate = 500;
 328		break;
 329	case IB_SPEED_NDR:
 330		speed = " NDR";
 331		rate = 1000;
 332		break;
 333	case IB_SPEED_XDR:
 334		speed = " XDR";
 335		rate = 2000;
 336		break;
 337	case IB_SPEED_SDR:
 338	default:		/* default to SDR for invalid rates */
 339		speed = " SDR";
 340		rate = 25;
 341		break;
 342	}
 343
 344	rate *= ib_width_enum_to_int(attr.active_width);
 345	if (rate < 0)
 346		return -EINVAL;
 347
 348	return sysfs_emit(buf, "%d%s Gb/sec (%dX%s)\n", rate / 10,
 349			  rate % 10 ? ".5" : "",
 350			  ib_width_enum_to_int(attr.active_width), speed);
 351}
 352
 353static const char *phys_state_to_str(enum ib_port_phys_state phys_state)
 354{
 355	static const char *phys_state_str[] = {
 356		"<unknown>",
 357		"Sleep",
 358		"Polling",
 359		"Disabled",
 360		"PortConfigurationTraining",
 361		"LinkUp",
 362		"LinkErrorRecovery",
 363		"Phy Test",
 364	};
 365
 366	if (phys_state < ARRAY_SIZE(phys_state_str))
 367		return phys_state_str[phys_state];
 368	return "<unknown>";
 369}
 370
 371static ssize_t phys_state_show(struct ib_device *ibdev, u32 port_num,
 372			       struct ib_port_attribute *unused, char *buf)
 373{
 374	struct ib_port_attr attr;
 375
 376	ssize_t ret;
 377
 378	ret = ib_query_port(ibdev, port_num, &attr);
 379	if (ret)
 380		return ret;
 381
 382	return sysfs_emit(buf, "%u: %s\n", attr.phys_state,
 383			  phys_state_to_str(attr.phys_state));
 384}
 385
 386static ssize_t link_layer_show(struct ib_device *ibdev, u32 port_num,
 387			       struct ib_port_attribute *unused, char *buf)
 388{
 389	const char *output;
 390
 391	switch (rdma_port_get_link_layer(ibdev, port_num)) {
 392	case IB_LINK_LAYER_INFINIBAND:
 393		output = "InfiniBand";
 394		break;
 395	case IB_LINK_LAYER_ETHERNET:
 396		output = "Ethernet";
 397		break;
 398	default:
 399		output = "Unknown";
 400		break;
 401	}
 402
 403	return sysfs_emit(buf, "%s\n", output);
 404}
 405
 406static IB_PORT_ATTR_RO(state);
 407static IB_PORT_ATTR_RO(lid);
 408static IB_PORT_ATTR_RO(lid_mask_count);
 409static IB_PORT_ATTR_RO(sm_lid);
 410static IB_PORT_ATTR_RO(sm_sl);
 411static IB_PORT_ATTR_RO(cap_mask);
 412static IB_PORT_ATTR_RO(rate);
 413static IB_PORT_ATTR_RO(phys_state);
 414static IB_PORT_ATTR_RO(link_layer);
 415
 416static struct attribute *port_default_attrs[] = {
 417	&ib_port_attr_state.attr,
 418	&ib_port_attr_lid.attr,
 419	&ib_port_attr_lid_mask_count.attr,
 420	&ib_port_attr_sm_lid.attr,
 421	&ib_port_attr_sm_sl.attr,
 422	&ib_port_attr_cap_mask.attr,
 423	&ib_port_attr_rate.attr,
 424	&ib_port_attr_phys_state.attr,
 425	&ib_port_attr_link_layer.attr,
 426	NULL
 427};
 428ATTRIBUTE_GROUPS(port_default);
 429
 430static ssize_t print_ndev(const struct ib_gid_attr *gid_attr, char *buf)
 431{
 432	struct net_device *ndev;
 433	int ret = -EINVAL;
 434
 435	rcu_read_lock();
 436	ndev = rcu_dereference(gid_attr->ndev);
 437	if (ndev)
 438		ret = sysfs_emit(buf, "%s\n", ndev->name);
 439	rcu_read_unlock();
 440	return ret;
 441}
 442
 443static ssize_t print_gid_type(const struct ib_gid_attr *gid_attr, char *buf)
 444{
 445	return sysfs_emit(buf, "%s\n",
 446			  ib_cache_gid_type_str(gid_attr->gid_type));
 447}
 448
 449static ssize_t _show_port_gid_attr(
 450	struct ib_device *ibdev, u32 port_num, struct ib_port_attribute *attr,
 451	char *buf,
 452	ssize_t (*print)(const struct ib_gid_attr *gid_attr, char *buf))
 453{
 454	struct port_table_attribute *tab_attr =
 455		container_of(attr, struct port_table_attribute, attr);
 456	const struct ib_gid_attr *gid_attr;
 457	ssize_t ret;
 458
 459	gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
 460	if (IS_ERR(gid_attr))
 461		/* -EINVAL is returned for user space compatibility reasons. */
 462		return -EINVAL;
 463
 464	ret = print(gid_attr, buf);
 465	rdma_put_gid_attr(gid_attr);
 466	return ret;
 467}
 468
 469static ssize_t show_port_gid(struct ib_device *ibdev, u32 port_num,
 470			     struct ib_port_attribute *attr, char *buf)
 471{
 472	struct port_table_attribute *tab_attr =
 473		container_of(attr, struct port_table_attribute, attr);
 474	const struct ib_gid_attr *gid_attr;
 475	int len;
 476
 477	gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
 478	if (IS_ERR(gid_attr)) {
 479		const union ib_gid zgid = {};
 480
 481		/* If reading GID fails, it is likely due to GID entry being
 482		 * empty (invalid) or reserved GID in the table.  User space
 483		 * expects to read GID table entries as long as it given index
 484		 * is within GID table size.  Administrative/debugging tool
 485		 * fails to query rest of the GID entries if it hits error
 486		 * while querying a GID of the given index.  To avoid user
 487		 * space throwing such error on fail to read gid, return zero
 488		 * GID as before. This maintains backward compatibility.
 489		 */
 490		return sysfs_emit(buf, "%pI6\n", zgid.raw);
 491	}
 492
 493	len = sysfs_emit(buf, "%pI6\n", gid_attr->gid.raw);
 494	rdma_put_gid_attr(gid_attr);
 495	return len;
 496}
 497
 498static ssize_t show_port_gid_attr_ndev(struct ib_device *ibdev, u32 port_num,
 499				       struct ib_port_attribute *attr,
 500				       char *buf)
 501{
 502	return _show_port_gid_attr(ibdev, port_num, attr, buf, print_ndev);
 503}
 504
 505static ssize_t show_port_gid_attr_gid_type(struct ib_device *ibdev,
 506					   u32 port_num,
 507					   struct ib_port_attribute *attr,
 508					   char *buf)
 509{
 510	return _show_port_gid_attr(ibdev, port_num, attr, buf, print_gid_type);
 511}
 512
 513static ssize_t show_port_pkey(struct ib_device *ibdev, u32 port_num,
 514			      struct ib_port_attribute *attr, char *buf)
 515{
 516	struct port_table_attribute *tab_attr =
 517		container_of(attr, struct port_table_attribute, attr);
 518	u16 pkey;
 519	int ret;
 520
 521	ret = ib_query_pkey(ibdev, port_num, tab_attr->index, &pkey);
 522	if (ret)
 523		return ret;
 524
 525	return sysfs_emit(buf, "0x%04x\n", pkey);
 526}
 527
 528#define PORT_PMA_ATTR(_name, _counter, _width, _offset)			\
 529struct port_table_attribute port_pma_attr_##_name = {			\
 530	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
 531	.index = (_offset) | ((_width) << 16) | ((_counter) << 24),	\
 532	.attr_id = IB_PMA_PORT_COUNTERS,				\
 533}
 534
 535#define PORT_PMA_ATTR_EXT(_name, _width, _offset)			\
 536struct port_table_attribute port_pma_attr_ext_##_name = {		\
 537	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
 538	.index = (_offset) | ((_width) << 16),				\
 539	.attr_id = IB_PMA_PORT_COUNTERS_EXT,				\
 540}
 541
 542/*
 543 * Get a Perfmgmt MAD block of data.
 544 * Returns error code or the number of bytes retrieved.
 545 */
 546static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
 547		void *data, int offset, size_t size)
 548{
 549	struct ib_mad *in_mad;
 550	struct ib_mad *out_mad;
 551	size_t mad_size = sizeof(*out_mad);
 552	u16 out_mad_pkey_index = 0;
 553	ssize_t ret;
 554
 555	if (!dev->ops.process_mad)
 556		return -ENOSYS;
 557
 558	in_mad = kzalloc(sizeof(*in_mad), GFP_KERNEL);
 559	out_mad = kzalloc(sizeof(*out_mad), GFP_KERNEL);
 560	if (!in_mad || !out_mad) {
 561		ret = -ENOMEM;
 562		goto out;
 563	}
 564
 565	in_mad->mad_hdr.base_version  = 1;
 566	in_mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_PERF_MGMT;
 567	in_mad->mad_hdr.class_version = 1;
 568	in_mad->mad_hdr.method        = IB_MGMT_METHOD_GET;
 569	in_mad->mad_hdr.attr_id       = attr;
 570
 571	if (attr != IB_PMA_CLASS_PORT_INFO)
 572		in_mad->data[41] = port_num;	/* PortSelect field */
 573
 574	if ((dev->ops.process_mad(dev, IB_MAD_IGNORE_MKEY, port_num, NULL, NULL,
 575				  in_mad, out_mad, &mad_size,
 576				  &out_mad_pkey_index) &
 577	     (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
 578	    (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
 579		ret = -EINVAL;
 580		goto out;
 581	}
 582	memcpy(data, out_mad->data + offset, size);
 583	ret = size;
 584out:
 585	kfree(in_mad);
 586	kfree(out_mad);
 587	return ret;
 588}
 589
 590static ssize_t show_pma_counter(struct ib_device *ibdev, u32 port_num,
 591				struct ib_port_attribute *attr, char *buf)
 592{
 593	struct port_table_attribute *tab_attr =
 594		container_of(attr, struct port_table_attribute, attr);
 595	int offset = tab_attr->index & 0xffff;
 596	int width  = (tab_attr->index >> 16) & 0xff;
 597	int ret;
 598	u8 data[8];
 599	int len;
 600
 601	ret = get_perf_mad(ibdev, port_num, tab_attr->attr_id, &data,
 602			40 + offset / 8, sizeof(data));
 603	if (ret < 0)
 604		return ret;
 605
 606	switch (width) {
 607	case 4:
 608		len = sysfs_emit(buf, "%d\n",
 609				 (*data >> (4 - (offset % 8))) & 0xf);
 610		break;
 611	case 8:
 612		len = sysfs_emit(buf, "%u\n", *data);
 613		break;
 614	case 16:
 615		len = sysfs_emit(buf, "%u\n", be16_to_cpup((__be16 *)data));
 616		break;
 617	case 32:
 618		len = sysfs_emit(buf, "%u\n", be32_to_cpup((__be32 *)data));
 619		break;
 620	case 64:
 621		len = sysfs_emit(buf, "%llu\n", be64_to_cpup((__be64 *)data));
 622		break;
 623	default:
 624		len = 0;
 625		break;
 626	}
 627
 628	return len;
 629}
 630
 631static PORT_PMA_ATTR(symbol_error		    ,  0, 16,  32);
 632static PORT_PMA_ATTR(link_error_recovery	    ,  1,  8,  48);
 633static PORT_PMA_ATTR(link_downed		    ,  2,  8,  56);
 634static PORT_PMA_ATTR(port_rcv_errors		    ,  3, 16,  64);
 635static PORT_PMA_ATTR(port_rcv_remote_physical_errors,  4, 16,  80);
 636static PORT_PMA_ATTR(port_rcv_switch_relay_errors   ,  5, 16,  96);
 637static PORT_PMA_ATTR(port_xmit_discards		    ,  6, 16, 112);
 638static PORT_PMA_ATTR(port_xmit_constraint_errors    ,  7,  8, 128);
 639static PORT_PMA_ATTR(port_rcv_constraint_errors	    ,  8,  8, 136);
 640static PORT_PMA_ATTR(local_link_integrity_errors    ,  9,  4, 152);
 641static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10,  4, 156);
 642static PORT_PMA_ATTR(VL15_dropped		    , 11, 16, 176);
 643static PORT_PMA_ATTR(port_xmit_data		    , 12, 32, 192);
 644static PORT_PMA_ATTR(port_rcv_data		    , 13, 32, 224);
 645static PORT_PMA_ATTR(port_xmit_packets		    , 14, 32, 256);
 646static PORT_PMA_ATTR(port_rcv_packets		    , 15, 32, 288);
 647static PORT_PMA_ATTR(port_xmit_wait		    ,  0, 32, 320);
 648
 649/*
 650 * Counters added by extended set
 651 */
 652static PORT_PMA_ATTR_EXT(port_xmit_data		    , 64,  64);
 653static PORT_PMA_ATTR_EXT(port_rcv_data		    , 64, 128);
 654static PORT_PMA_ATTR_EXT(port_xmit_packets	    , 64, 192);
 655static PORT_PMA_ATTR_EXT(port_rcv_packets	    , 64, 256);
 656static PORT_PMA_ATTR_EXT(unicast_xmit_packets	    , 64, 320);
 657static PORT_PMA_ATTR_EXT(unicast_rcv_packets	    , 64, 384);
 658static PORT_PMA_ATTR_EXT(multicast_xmit_packets	    , 64, 448);
 659static PORT_PMA_ATTR_EXT(multicast_rcv_packets	    , 64, 512);
 660
 661static struct attribute *pma_attrs[] = {
 662	&port_pma_attr_symbol_error.attr.attr,
 663	&port_pma_attr_link_error_recovery.attr.attr,
 664	&port_pma_attr_link_downed.attr.attr,
 665	&port_pma_attr_port_rcv_errors.attr.attr,
 666	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
 667	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
 668	&port_pma_attr_port_xmit_discards.attr.attr,
 669	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
 670	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
 671	&port_pma_attr_local_link_integrity_errors.attr.attr,
 672	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
 673	&port_pma_attr_VL15_dropped.attr.attr,
 674	&port_pma_attr_port_xmit_data.attr.attr,
 675	&port_pma_attr_port_rcv_data.attr.attr,
 676	&port_pma_attr_port_xmit_packets.attr.attr,
 677	&port_pma_attr_port_rcv_packets.attr.attr,
 678	&port_pma_attr_port_xmit_wait.attr.attr,
 679	NULL
 680};
 681
 682static struct attribute *pma_attrs_ext[] = {
 683	&port_pma_attr_symbol_error.attr.attr,
 684	&port_pma_attr_link_error_recovery.attr.attr,
 685	&port_pma_attr_link_downed.attr.attr,
 686	&port_pma_attr_port_rcv_errors.attr.attr,
 687	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
 688	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
 689	&port_pma_attr_port_xmit_discards.attr.attr,
 690	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
 691	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
 692	&port_pma_attr_local_link_integrity_errors.attr.attr,
 693	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
 694	&port_pma_attr_VL15_dropped.attr.attr,
 695	&port_pma_attr_ext_port_xmit_data.attr.attr,
 696	&port_pma_attr_ext_port_rcv_data.attr.attr,
 697	&port_pma_attr_ext_port_xmit_packets.attr.attr,
 698	&port_pma_attr_port_xmit_wait.attr.attr,
 699	&port_pma_attr_ext_port_rcv_packets.attr.attr,
 700	&port_pma_attr_ext_unicast_rcv_packets.attr.attr,
 701	&port_pma_attr_ext_unicast_xmit_packets.attr.attr,
 702	&port_pma_attr_ext_multicast_rcv_packets.attr.attr,
 703	&port_pma_attr_ext_multicast_xmit_packets.attr.attr,
 704	NULL
 705};
 706
 707static struct attribute *pma_attrs_noietf[] = {
 708	&port_pma_attr_symbol_error.attr.attr,
 709	&port_pma_attr_link_error_recovery.attr.attr,
 710	&port_pma_attr_link_downed.attr.attr,
 711	&port_pma_attr_port_rcv_errors.attr.attr,
 712	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
 713	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
 714	&port_pma_attr_port_xmit_discards.attr.attr,
 715	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
 716	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
 717	&port_pma_attr_local_link_integrity_errors.attr.attr,
 718	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
 719	&port_pma_attr_VL15_dropped.attr.attr,
 720	&port_pma_attr_ext_port_xmit_data.attr.attr,
 721	&port_pma_attr_ext_port_rcv_data.attr.attr,
 722	&port_pma_attr_ext_port_xmit_packets.attr.attr,
 723	&port_pma_attr_ext_port_rcv_packets.attr.attr,
 724	&port_pma_attr_port_xmit_wait.attr.attr,
 725	NULL
 726};
 727
 728static const struct attribute_group pma_group = {
 729	.name  = "counters",
 730	.attrs  = pma_attrs
 731};
 732
 733static const struct attribute_group pma_group_ext = {
 734	.name  = "counters",
 735	.attrs  = pma_attrs_ext
 736};
 737
 738static const struct attribute_group pma_group_noietf = {
 739	.name  = "counters",
 740	.attrs  = pma_attrs_noietf
 741};
 742
 743static void ib_port_release(struct kobject *kobj)
 744{
 745	struct ib_port *port = container_of(kobj, struct ib_port, kobj);
 746	int i;
 747
 748	for (i = 0; i != ARRAY_SIZE(port->groups); i++)
 749		kfree(port->groups[i].attrs);
 750	if (port->hw_stats_data)
 751		rdma_free_hw_stats_struct(port->hw_stats_data->stats);
 752	kfree(port->hw_stats_data);
 753	kvfree(port);
 754}
 755
 756static void ib_port_gid_attr_release(struct kobject *kobj)
 757{
 758	struct gid_attr_group *gid_attr_group =
 759		container_of(kobj, struct gid_attr_group, kobj);
 760	int i;
 761
 762	for (i = 0; i != ARRAY_SIZE(gid_attr_group->groups); i++)
 763		kfree(gid_attr_group->groups[i].attrs);
 764	kfree(gid_attr_group);
 765}
 766
 767static struct kobj_type port_type = {
 768	.release       = ib_port_release,
 769	.sysfs_ops     = &port_sysfs_ops,
 770	.default_groups = port_default_groups,
 771};
 772
 773static struct kobj_type gid_attr_type = {
 774	.sysfs_ops      = &gid_attr_sysfs_ops,
 775	.release        = ib_port_gid_attr_release
 776};
 777
 778/*
 779 * Figure out which counter table to use depending on
 780 * the device capabilities.
 781 */
 782static const struct attribute_group *get_counter_table(struct ib_device *dev,
 783						       int port_num)
 784{
 785	struct ib_class_port_info cpi;
 786
 787	if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
 788				&cpi, 40, sizeof(cpi)) >= 0) {
 789		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
 790			/* We have extended counters */
 791			return &pma_group_ext;
 792
 793		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
 794			/* But not the IETF ones */
 795			return &pma_group_noietf;
 796	}
 797
 798	/* Fall back to normal counters */
 799	return &pma_group;
 800}
 801
 802static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
 803			   u32 port_num, int index)
 804{
 805	int ret;
 806
 807	if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
 808		return 0;
 809	ret = dev->ops.get_hw_stats(dev, stats, port_num, index);
 810	if (ret < 0)
 811		return ret;
 812	if (ret == stats->num_counters)
 813		stats->timestamp = jiffies;
 814
 815	return 0;
 816}
 817
 818static int print_hw_stat(struct ib_device *dev, int port_num,
 819			 struct rdma_hw_stats *stats, int index, char *buf)
 820{
 821	u64 v = rdma_counter_get_hwstat_value(dev, port_num, index);
 822
 823	return sysfs_emit(buf, "%llu\n", stats->value[index] + v);
 824}
 825
 826static ssize_t show_hw_stats(struct ib_device *ibdev,
 827			     struct rdma_hw_stats *stats, unsigned int index,
 828			     unsigned int port_num, char *buf)
 829{
 830	int ret;
 831
 832	mutex_lock(&stats->lock);
 833	ret = update_hw_stats(ibdev, stats, port_num, index);
 834	if (ret)
 835		goto unlock;
 836	ret = print_hw_stat(ibdev, port_num, stats, index, buf);
 837unlock:
 838	mutex_unlock(&stats->lock);
 839
 840	return ret;
 841}
 842
 843static ssize_t show_stats_lifespan(struct ib_device *ibdev,
 844				   struct rdma_hw_stats *stats,
 845				   unsigned int index, unsigned int port_num,
 846				   char *buf)
 847{
 848	int msecs;
 849
 850	mutex_lock(&stats->lock);
 851	msecs = jiffies_to_msecs(stats->lifespan);
 852	mutex_unlock(&stats->lock);
 853
 854	return sysfs_emit(buf, "%d\n", msecs);
 855}
 856
 857static ssize_t set_stats_lifespan(struct ib_device *ibdev,
 858				   struct rdma_hw_stats *stats,
 859				   unsigned int index, unsigned int port_num,
 860				   const char *buf, size_t count)
 861{
 862	int msecs;
 863	int jiffies;
 864	int ret;
 865
 866	ret = kstrtoint(buf, 10, &msecs);
 867	if (ret)
 868		return ret;
 869	if (msecs < 0 || msecs > 10000)
 870		return -EINVAL;
 871	jiffies = msecs_to_jiffies(msecs);
 872
 873	mutex_lock(&stats->lock);
 874	stats->lifespan = jiffies;
 875	mutex_unlock(&stats->lock);
 876
 877	return count;
 878}
 879
 880static struct hw_stats_device_data *
 881alloc_hw_stats_device(struct ib_device *ibdev)
 882{
 883	struct hw_stats_device_data *data;
 884	struct rdma_hw_stats *stats;
 885
 886	if (!ibdev->ops.alloc_hw_device_stats)
 887		return ERR_PTR(-EOPNOTSUPP);
 888	stats = ibdev->ops.alloc_hw_device_stats(ibdev);
 889	if (!stats)
 890		return ERR_PTR(-ENOMEM);
 891	if (!stats->descs || stats->num_counters <= 0)
 892		goto err_free_stats;
 893
 894	/*
 895	 * Two extra attribue elements here, one for the lifespan entry and
 896	 * one to NULL terminate the list for the sysfs core code
 897	 */
 898	data = kzalloc(struct_size(data, attrs, size_add(stats->num_counters, 1)),
 899		       GFP_KERNEL);
 900	if (!data)
 901		goto err_free_stats;
 902	data->group.attrs = kcalloc(stats->num_counters + 2,
 903				    sizeof(*data->group.attrs), GFP_KERNEL);
 904	if (!data->group.attrs)
 905		goto err_free_data;
 906
 907	data->group.name = "hw_counters";
 908	data->stats = stats;
 909	return data;
 910
 911err_free_data:
 912	kfree(data);
 913err_free_stats:
 914	rdma_free_hw_stats_struct(stats);
 915	return ERR_PTR(-ENOMEM);
 916}
 917
 918void ib_device_release_hw_stats(struct hw_stats_device_data *data)
 919{
 920	kfree(data->group.attrs);
 921	rdma_free_hw_stats_struct(data->stats);
 922	kfree(data);
 923}
 924
 925int ib_setup_device_attrs(struct ib_device *ibdev)
 926{
 927	struct hw_stats_device_attribute *attr;
 928	struct hw_stats_device_data *data;
 929	bool opstat_skipped = false;
 930	int i, ret, pos = 0;
 931
 932	data = alloc_hw_stats_device(ibdev);
 933	if (IS_ERR(data)) {
 934		if (PTR_ERR(data) == -EOPNOTSUPP)
 935			return 0;
 936		return PTR_ERR(data);
 937	}
 938	ibdev->hw_stats_data = data;
 939
 940	ret = ibdev->ops.get_hw_stats(ibdev, data->stats, 0,
 941				      data->stats->num_counters);
 942	if (ret != data->stats->num_counters) {
 943		if (WARN_ON(ret >= 0))
 944			return -EINVAL;
 945		return ret;
 946	}
 947
 948	data->stats->timestamp = jiffies;
 949
 950	for (i = 0; i < data->stats->num_counters; i++) {
 951		if (data->stats->descs[i].flags & IB_STAT_FLAG_OPTIONAL) {
 952			opstat_skipped = true;
 953			continue;
 954		}
 955
 956		WARN_ON(opstat_skipped);
 957		attr = &data->attrs[pos];
 958		sysfs_attr_init(&attr->attr.attr);
 959		attr->attr.attr.name = data->stats->descs[i].name;
 960		attr->attr.attr.mode = 0444;
 961		attr->attr.show = hw_stat_device_show;
 962		attr->show = show_hw_stats;
 963		data->group.attrs[pos] = &attr->attr.attr;
 964		pos++;
 965	}
 966
 967	attr = &data->attrs[pos];
 968	sysfs_attr_init(&attr->attr.attr);
 969	attr->attr.attr.name = "lifespan";
 970	attr->attr.attr.mode = 0644;
 971	attr->attr.show = hw_stat_device_show;
 972	attr->show = show_stats_lifespan;
 973	attr->attr.store = hw_stat_device_store;
 974	attr->store = set_stats_lifespan;
 975	data->group.attrs[pos] = &attr->attr.attr;
 976	for (i = 0; i != ARRAY_SIZE(ibdev->groups); i++)
 977		if (!ibdev->groups[i]) {
 978			ibdev->groups[i] = &data->group;
 979			ibdev->hw_stats_attr_index = i;
 980			return 0;
 981		}
 982	WARN(true, "struct ib_device->groups is too small");
 983	return -EINVAL;
 984}
 985
 986static struct hw_stats_port_data *
 987alloc_hw_stats_port(struct ib_port *port, struct attribute_group *group)
 988{
 989	struct ib_device *ibdev = port->ibdev;
 990	struct hw_stats_port_data *data;
 991	struct rdma_hw_stats *stats;
 992
 993	if (!ibdev->ops.alloc_hw_port_stats)
 994		return ERR_PTR(-EOPNOTSUPP);
 995	stats = ibdev->ops.alloc_hw_port_stats(port->ibdev, port->port_num);
 996	if (!stats)
 997		return ERR_PTR(-ENOMEM);
 998	if (!stats->descs || stats->num_counters <= 0)
 999		goto err_free_stats;
1000
1001	/*
1002	 * Two extra attribue elements here, one for the lifespan entry and
1003	 * one to NULL terminate the list for the sysfs core code
1004	 */
1005	data = kzalloc(struct_size(data, attrs, size_add(stats->num_counters, 1)),
1006		       GFP_KERNEL);
1007	if (!data)
1008		goto err_free_stats;
1009	group->attrs = kcalloc(stats->num_counters + 2,
1010				    sizeof(*group->attrs), GFP_KERNEL);
1011	if (!group->attrs)
1012		goto err_free_data;
1013
1014	group->name = "hw_counters";
1015	data->stats = stats;
1016	return data;
1017
1018err_free_data:
1019	kfree(data);
1020err_free_stats:
1021	rdma_free_hw_stats_struct(stats);
1022	return ERR_PTR(-ENOMEM);
1023}
1024
1025static int setup_hw_port_stats(struct ib_port *port,
1026			       struct attribute_group *group)
1027{
1028	struct hw_stats_port_attribute *attr;
1029	struct hw_stats_port_data *data;
1030	bool opstat_skipped = false;
1031	int i, ret, pos = 0;
1032
1033	data = alloc_hw_stats_port(port, group);
1034	if (IS_ERR(data))
1035		return PTR_ERR(data);
1036
1037	ret = port->ibdev->ops.get_hw_stats(port->ibdev, data->stats,
1038					    port->port_num,
1039					    data->stats->num_counters);
1040	if (ret != data->stats->num_counters) {
1041		if (WARN_ON(ret >= 0))
1042			return -EINVAL;
1043		return ret;
1044	}
1045
1046	data->stats->timestamp = jiffies;
1047
1048	for (i = 0; i < data->stats->num_counters; i++) {
1049		if (data->stats->descs[i].flags & IB_STAT_FLAG_OPTIONAL) {
1050			opstat_skipped = true;
1051			continue;
1052		}
1053
1054		WARN_ON(opstat_skipped);
1055		attr = &data->attrs[pos];
1056		sysfs_attr_init(&attr->attr.attr);
1057		attr->attr.attr.name = data->stats->descs[i].name;
1058		attr->attr.attr.mode = 0444;
1059		attr->attr.show = hw_stat_port_show;
1060		attr->show = show_hw_stats;
1061		group->attrs[pos] = &attr->attr.attr;
1062		pos++;
1063	}
1064
1065	attr = &data->attrs[pos];
1066	sysfs_attr_init(&attr->attr.attr);
1067	attr->attr.attr.name = "lifespan";
1068	attr->attr.attr.mode = 0644;
1069	attr->attr.show = hw_stat_port_show;
1070	attr->show = show_stats_lifespan;
1071	attr->attr.store = hw_stat_port_store;
1072	attr->store = set_stats_lifespan;
1073	group->attrs[pos] = &attr->attr.attr;
1074
1075	port->hw_stats_data = data;
1076	return 0;
1077}
1078
1079struct rdma_hw_stats *ib_get_hw_stats_port(struct ib_device *ibdev,
1080					   u32 port_num)
1081{
1082	if (!ibdev->port_data || !rdma_is_port_valid(ibdev, port_num) ||
1083	    !ibdev->port_data[port_num].sysfs->hw_stats_data)
1084		return NULL;
1085	return ibdev->port_data[port_num].sysfs->hw_stats_data->stats;
1086}
1087
1088static int
1089alloc_port_table_group(const char *name, struct attribute_group *group,
1090		       struct port_table_attribute *attrs, size_t num,
1091		       ssize_t (*show)(struct ib_device *ibdev, u32 port_num,
1092				       struct ib_port_attribute *, char *buf))
1093{
1094	struct attribute **attr_list;
1095	int i;
1096
1097	attr_list = kcalloc(num + 1, sizeof(*attr_list), GFP_KERNEL);
1098	if (!attr_list)
1099		return -ENOMEM;
1100
1101	for (i = 0; i < num; i++) {
1102		struct port_table_attribute *element = &attrs[i];
1103
1104		if (snprintf(element->name, sizeof(element->name), "%d", i) >=
1105		    sizeof(element->name))
1106			goto err;
1107
1108		sysfs_attr_init(&element->attr.attr);
1109		element->attr.attr.name = element->name;
1110		element->attr.attr.mode = 0444;
1111		element->attr.show = show;
1112		element->index = i;
1113
1114		attr_list[i] = &element->attr.attr;
1115	}
1116	group->name = name;
1117	group->attrs = attr_list;
1118	return 0;
1119err:
1120	kfree(attr_list);
1121	return -EINVAL;
1122}
1123
1124/*
1125 * Create the sysfs:
1126 *  ibp0s9/ports/XX/gid_attrs/{ndevs,types}/YYY
1127 * YYY is the gid table index in decimal
1128 */
1129static int setup_gid_attrs(struct ib_port *port,
1130			   const struct ib_port_attr *attr)
1131{
1132	struct gid_attr_group *gid_attr_group;
1133	int ret;
1134
1135	gid_attr_group = kzalloc(struct_size(gid_attr_group, attrs_list,
1136					     size_mul(attr->gid_tbl_len, 2)),
1137				 GFP_KERNEL);
1138	if (!gid_attr_group)
1139		return -ENOMEM;
1140	gid_attr_group->port = port;
1141	kobject_init(&gid_attr_group->kobj, &gid_attr_type);
1142
1143	ret = alloc_port_table_group("ndevs", &gid_attr_group->groups[0],
1144				     gid_attr_group->attrs_list,
1145				     attr->gid_tbl_len,
1146				     show_port_gid_attr_ndev);
1147	if (ret)
1148		goto err_put;
1149	gid_attr_group->groups_list[0] = &gid_attr_group->groups[0];
1150
1151	ret = alloc_port_table_group(
1152		"types", &gid_attr_group->groups[1],
1153		gid_attr_group->attrs_list + attr->gid_tbl_len,
1154		attr->gid_tbl_len, show_port_gid_attr_gid_type);
1155	if (ret)
1156		goto err_put;
1157	gid_attr_group->groups_list[1] = &gid_attr_group->groups[1];
1158
1159	ret = kobject_add(&gid_attr_group->kobj, &port->kobj, "gid_attrs");
1160	if (ret)
1161		goto err_put;
1162	ret = sysfs_create_groups(&gid_attr_group->kobj,
1163				  gid_attr_group->groups_list);
1164	if (ret)
1165		goto err_del;
1166	port->gid_attr_group = gid_attr_group;
1167	return 0;
1168
1169err_del:
1170	kobject_del(&gid_attr_group->kobj);
1171err_put:
1172	kobject_put(&gid_attr_group->kobj);
1173	return ret;
1174}
1175
1176static void destroy_gid_attrs(struct ib_port *port)
1177{
1178	struct gid_attr_group *gid_attr_group = port->gid_attr_group;
1179
1180	if (!gid_attr_group)
1181		return;
1182	sysfs_remove_groups(&gid_attr_group->kobj, gid_attr_group->groups_list);
1183	kobject_del(&gid_attr_group->kobj);
1184	kobject_put(&gid_attr_group->kobj);
1185}
1186
1187/*
1188 * Create the sysfs:
1189 *  ibp0s9/ports/XX/{gids,pkeys,counters}/YYY
1190 */
1191static struct ib_port *setup_port(struct ib_core_device *coredev, int port_num,
1192				  const struct ib_port_attr *attr)
1193{
1194	struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1195	bool is_full_dev = &device->coredev == coredev;
1196	const struct attribute_group **cur_group;
1197	struct ib_port *p;
1198	int ret;
1199
1200	p = kvzalloc(struct_size(p, attrs_list,
1201				size_add(attr->gid_tbl_len, attr->pkey_tbl_len)),
1202		     GFP_KERNEL);
1203	if (!p)
1204		return ERR_PTR(-ENOMEM);
1205	p->ibdev = device;
1206	p->port_num = port_num;
1207	kobject_init(&p->kobj, &port_type);
1208
1209	if (device->port_data && is_full_dev)
1210		device->port_data[port_num].sysfs = p;
1211
1212	cur_group = p->groups_list;
1213	ret = alloc_port_table_group("gids", &p->groups[0], p->attrs_list,
1214				     attr->gid_tbl_len, show_port_gid);
1215	if (ret)
1216		goto err_put;
1217	*cur_group++ = &p->groups[0];
1218
1219	if (attr->pkey_tbl_len) {
1220		ret = alloc_port_table_group("pkeys", &p->groups[1],
1221					     p->attrs_list + attr->gid_tbl_len,
1222					     attr->pkey_tbl_len, show_port_pkey);
1223		if (ret)
1224			goto err_put;
1225		*cur_group++ = &p->groups[1];
1226	}
1227
1228	/*
1229	 * If port == 0, it means hw_counters are per device and not per
1230	 * port, so holder should be device. Therefore skip per port
1231	 * counter initialization.
1232	 */
1233	if (port_num && is_full_dev) {
1234		ret = setup_hw_port_stats(p, &p->groups[2]);
1235		if (ret && ret != -EOPNOTSUPP)
1236			goto err_put;
1237		if (!ret)
1238			*cur_group++ = &p->groups[2];
1239	}
1240
1241	if (device->ops.process_mad && is_full_dev)
1242		*cur_group++ = get_counter_table(device, port_num);
1243
1244	ret = kobject_add(&p->kobj, coredev->ports_kobj, "%d", port_num);
1245	if (ret)
1246		goto err_put;
1247	ret = sysfs_create_groups(&p->kobj, p->groups_list);
1248	if (ret)
1249		goto err_del;
1250	if (is_full_dev) {
1251		ret = sysfs_create_groups(&p->kobj, device->ops.port_groups);
1252		if (ret)
1253			goto err_groups;
1254	}
1255
1256	list_add_tail(&p->kobj.entry, &coredev->port_list);
1257	return p;
1258
1259err_groups:
1260	sysfs_remove_groups(&p->kobj, p->groups_list);
1261err_del:
1262	kobject_del(&p->kobj);
1263err_put:
1264	if (device->port_data && is_full_dev)
1265		device->port_data[port_num].sysfs = NULL;
1266	kobject_put(&p->kobj);
1267	return ERR_PTR(ret);
1268}
1269
1270static void destroy_port(struct ib_core_device *coredev, struct ib_port *port)
1271{
1272	bool is_full_dev = &port->ibdev->coredev == coredev;
1273
1274	list_del(&port->kobj.entry);
1275	if (is_full_dev)
1276		sysfs_remove_groups(&port->kobj, port->ibdev->ops.port_groups);
1277
1278	sysfs_remove_groups(&port->kobj, port->groups_list);
1279	kobject_del(&port->kobj);
1280
1281	if (port->ibdev->port_data &&
1282	    port->ibdev->port_data[port->port_num].sysfs == port)
1283		port->ibdev->port_data[port->port_num].sysfs = NULL;
1284
1285	kobject_put(&port->kobj);
1286}
1287
1288static const char *node_type_string(int node_type)
1289{
1290	switch (node_type) {
1291	case RDMA_NODE_IB_CA:
1292		return "CA";
1293	case RDMA_NODE_IB_SWITCH:
1294		return "switch";
1295	case RDMA_NODE_IB_ROUTER:
1296		return "router";
1297	case RDMA_NODE_RNIC:
1298		return "RNIC";
1299	case RDMA_NODE_USNIC:
1300		return "usNIC";
1301	case RDMA_NODE_USNIC_UDP:
1302		return "usNIC UDP";
1303	case RDMA_NODE_UNSPECIFIED:
1304		return "unspecified";
1305	}
1306	return "<unknown>";
1307}
1308
1309static ssize_t node_type_show(struct device *device,
1310			      struct device_attribute *attr, char *buf)
1311{
1312	struct ib_device *dev = rdma_device_to_ibdev(device);
1313
1314	return sysfs_emit(buf, "%u: %s\n", dev->node_type,
1315			  node_type_string(dev->node_type));
1316}
1317static DEVICE_ATTR_RO(node_type);
1318
1319static ssize_t sys_image_guid_show(struct device *device,
1320				   struct device_attribute *dev_attr, char *buf)
1321{
1322	struct ib_device *dev = rdma_device_to_ibdev(device);
1323	__be16 *guid = (__be16 *)&dev->attrs.sys_image_guid;
1324
1325	return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1326			  be16_to_cpu(guid[0]),
1327			  be16_to_cpu(guid[1]),
1328			  be16_to_cpu(guid[2]),
1329			  be16_to_cpu(guid[3]));
1330}
1331static DEVICE_ATTR_RO(sys_image_guid);
1332
1333static ssize_t node_guid_show(struct device *device,
1334			      struct device_attribute *attr, char *buf)
1335{
1336	struct ib_device *dev = rdma_device_to_ibdev(device);
1337	__be16 *node_guid = (__be16 *)&dev->node_guid;
1338
1339	return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1340			  be16_to_cpu(node_guid[0]),
1341			  be16_to_cpu(node_guid[1]),
1342			  be16_to_cpu(node_guid[2]),
1343			  be16_to_cpu(node_guid[3]));
1344}
1345static DEVICE_ATTR_RO(node_guid);
1346
1347static ssize_t node_desc_show(struct device *device,
1348			      struct device_attribute *attr, char *buf)
1349{
1350	struct ib_device *dev = rdma_device_to_ibdev(device);
1351
1352	return sysfs_emit(buf, "%.64s\n", dev->node_desc);
1353}
1354
1355static ssize_t node_desc_store(struct device *device,
1356			       struct device_attribute *attr,
1357			       const char *buf, size_t count)
1358{
1359	struct ib_device *dev = rdma_device_to_ibdev(device);
1360	struct ib_device_modify desc = {};
1361	int ret;
1362
1363	if (!dev->ops.modify_device)
1364		return -EOPNOTSUPP;
1365
1366	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1367	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1368	if (ret)
1369		return ret;
1370
1371	return count;
1372}
1373static DEVICE_ATTR_RW(node_desc);
1374
1375static ssize_t fw_ver_show(struct device *device, struct device_attribute *attr,
1376			   char *buf)
1377{
1378	struct ib_device *dev = rdma_device_to_ibdev(device);
1379	char version[IB_FW_VERSION_NAME_MAX] = {};
1380
1381	ib_get_device_fw_str(dev, version);
1382
1383	return sysfs_emit(buf, "%s\n", version);
1384}
1385static DEVICE_ATTR_RO(fw_ver);
1386
1387static struct attribute *ib_dev_attrs[] = {
1388	&dev_attr_node_type.attr,
1389	&dev_attr_node_guid.attr,
1390	&dev_attr_sys_image_guid.attr,
1391	&dev_attr_fw_ver.attr,
1392	&dev_attr_node_desc.attr,
1393	NULL,
1394};
1395
1396const struct attribute_group ib_dev_attr_group = {
1397	.attrs = ib_dev_attrs,
1398};
1399
1400void ib_free_port_attrs(struct ib_core_device *coredev)
1401{
1402	struct kobject *p, *t;
1403
1404	list_for_each_entry_safe(p, t, &coredev->port_list, entry) {
1405		struct ib_port *port = container_of(p, struct ib_port, kobj);
1406
1407		destroy_gid_attrs(port);
1408		destroy_port(coredev, port);
1409	}
1410
1411	kobject_put(coredev->ports_kobj);
1412}
1413
1414int ib_setup_port_attrs(struct ib_core_device *coredev)
1415{
1416	struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1417	u32 port_num;
1418	int ret;
1419
1420	coredev->ports_kobj = kobject_create_and_add("ports",
1421						     &coredev->dev.kobj);
1422	if (!coredev->ports_kobj)
1423		return -ENOMEM;
1424
1425	rdma_for_each_port (device, port_num) {
1426		struct ib_port_attr attr;
1427		struct ib_port *port;
1428
1429		ret = ib_query_port(device, port_num, &attr);
1430		if (ret)
1431			goto err_put;
1432
1433		port = setup_port(coredev, port_num, &attr);
1434		if (IS_ERR(port)) {
1435			ret = PTR_ERR(port);
1436			goto err_put;
1437		}
1438
1439		ret = setup_gid_attrs(port, &attr);
1440		if (ret)
1441			goto err_put;
1442	}
1443	return 0;
1444
1445err_put:
1446	ib_free_port_attrs(coredev);
1447	return ret;
1448}
1449
1450/**
1451 * ib_port_register_client_groups - Add an ib_client's attributes to the port
1452 *
1453 * @ibdev: IB device to add counters
1454 * @port_num: valid port number
1455 * @groups: Group list of attributes
1456 *
1457 * Do not use. Only for legacy sysfs compatibility.
1458 */
1459int ib_port_register_client_groups(struct ib_device *ibdev, u32 port_num,
1460				   const struct attribute_group **groups)
1461{
1462	return sysfs_create_groups(&ibdev->port_data[port_num].sysfs->kobj,
1463				   groups);
1464}
1465EXPORT_SYMBOL(ib_port_register_client_groups);
1466
1467void ib_port_unregister_client_groups(struct ib_device *ibdev, u32 port_num,
1468				      const struct attribute_group **groups)
1469{
1470	return sysfs_remove_groups(&ibdev->port_data[port_num].sysfs->kobj,
1471				   groups);
1472}
1473EXPORT_SYMBOL(ib_port_unregister_client_groups);