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	static const char *state_name[] = {
 220		[IB_PORT_NOP]		= "NOP",
 221		[IB_PORT_DOWN]		= "DOWN",
 222		[IB_PORT_INIT]		= "INIT",
 223		[IB_PORT_ARMED]		= "ARMED",
 224		[IB_PORT_ACTIVE]	= "ACTIVE",
 225		[IB_PORT_ACTIVE_DEFER]	= "ACTIVE_DEFER"
 226	};
 227
 228	ret = ib_query_port(ibdev, port_num, &attr);
 229	if (ret)
 230		return ret;
 231
 232	return sysfs_emit(buf, "%d: %s\n", attr.state,
 233			  attr.state >= 0 &&
 234					  attr.state < ARRAY_SIZE(state_name) ?
 235				  state_name[attr.state] :
 236				  "UNKNOWN");
 237}
 238
 239static ssize_t lid_show(struct ib_device *ibdev, u32 port_num,
 240			struct ib_port_attribute *unused, char *buf)
 241{
 242	struct ib_port_attr attr;
 243	ssize_t ret;
 244
 245	ret = ib_query_port(ibdev, port_num, &attr);
 246	if (ret)
 247		return ret;
 248
 249	return sysfs_emit(buf, "0x%x\n", attr.lid);
 250}
 251
 252static ssize_t lid_mask_count_show(struct ib_device *ibdev, u32 port_num,
 253				   struct ib_port_attribute *unused, char *buf)
 254{
 255	struct ib_port_attr attr;
 256	ssize_t ret;
 257
 258	ret = ib_query_port(ibdev, port_num, &attr);
 259	if (ret)
 260		return ret;
 261
 262	return sysfs_emit(buf, "%u\n", attr.lmc);
 263}
 264
 265static ssize_t sm_lid_show(struct ib_device *ibdev, u32 port_num,
 266			   struct ib_port_attribute *unused, char *buf)
 267{
 268	struct ib_port_attr attr;
 269	ssize_t ret;
 270
 271	ret = ib_query_port(ibdev, port_num, &attr);
 272	if (ret)
 273		return ret;
 274
 275	return sysfs_emit(buf, "0x%x\n", attr.sm_lid);
 276}
 277
 278static ssize_t sm_sl_show(struct ib_device *ibdev, u32 port_num,
 279			  struct ib_port_attribute *unused, char *buf)
 280{
 281	struct ib_port_attr attr;
 282	ssize_t ret;
 283
 284	ret = ib_query_port(ibdev, port_num, &attr);
 285	if (ret)
 286		return ret;
 287
 288	return sysfs_emit(buf, "%u\n", attr.sm_sl);
 289}
 290
 291static ssize_t cap_mask_show(struct ib_device *ibdev, u32 port_num,
 292			     struct ib_port_attribute *unused, char *buf)
 293{
 294	struct ib_port_attr attr;
 295	ssize_t ret;
 296
 297	ret = ib_query_port(ibdev, port_num, &attr);
 298	if (ret)
 299		return ret;
 300
 301	return sysfs_emit(buf, "0x%08x\n", attr.port_cap_flags);
 302}
 303
 304static ssize_t rate_show(struct ib_device *ibdev, u32 port_num,
 305			 struct ib_port_attribute *unused, char *buf)
 306{
 307	struct ib_port_attr attr;
 308	char *speed = "";
 309	int rate;		/* in deci-Gb/sec */
 310	ssize_t ret;
 311
 312	ret = ib_query_port(ibdev, port_num, &attr);
 313	if (ret)
 314		return ret;
 315
 316	switch (attr.active_speed) {
 317	case IB_SPEED_DDR:
 318		speed = " DDR";
 319		rate = 50;
 320		break;
 321	case IB_SPEED_QDR:
 322		speed = " QDR";
 323		rate = 100;
 324		break;
 325	case IB_SPEED_FDR10:
 326		speed = " FDR10";
 327		rate = 100;
 328		break;
 329	case IB_SPEED_FDR:
 330		speed = " FDR";
 331		rate = 140;
 332		break;
 333	case IB_SPEED_EDR:
 334		speed = " EDR";
 335		rate = 250;
 336		break;
 337	case IB_SPEED_HDR:
 338		speed = " HDR";
 339		rate = 500;
 340		break;
 341	case IB_SPEED_NDR:
 342		speed = " NDR";
 343		rate = 1000;
 344		break;
 345	case IB_SPEED_XDR:
 346		speed = " XDR";
 347		rate = 2000;
 348		break;
 349	case IB_SPEED_SDR:
 350	default:		/* default to SDR for invalid rates */
 351		speed = " SDR";
 352		rate = 25;
 353		break;
 354	}
 355
 356	rate *= ib_width_enum_to_int(attr.active_width);
 357	if (rate < 0)
 358		return -EINVAL;
 359
 360	return sysfs_emit(buf, "%d%s Gb/sec (%dX%s)\n", rate / 10,
 361			  rate % 10 ? ".5" : "",
 362			  ib_width_enum_to_int(attr.active_width), speed);
 363}
 364
 365static const char *phys_state_to_str(enum ib_port_phys_state phys_state)
 366{
 367	static const char *phys_state_str[] = {
 368		"<unknown>",
 369		"Sleep",
 370		"Polling",
 371		"Disabled",
 372		"PortConfigurationTraining",
 373		"LinkUp",
 374		"LinkErrorRecovery",
 375		"Phy Test",
 376	};
 377
 378	if (phys_state < ARRAY_SIZE(phys_state_str))
 379		return phys_state_str[phys_state];
 380	return "<unknown>";
 381}
 382
 383static ssize_t phys_state_show(struct ib_device *ibdev, u32 port_num,
 384			       struct ib_port_attribute *unused, char *buf)
 385{
 386	struct ib_port_attr attr;
 387
 388	ssize_t ret;
 389
 390	ret = ib_query_port(ibdev, port_num, &attr);
 391	if (ret)
 392		return ret;
 393
 394	return sysfs_emit(buf, "%u: %s\n", attr.phys_state,
 395			  phys_state_to_str(attr.phys_state));
 396}
 397
 398static ssize_t link_layer_show(struct ib_device *ibdev, u32 port_num,
 399			       struct ib_port_attribute *unused, char *buf)
 400{
 401	const char *output;
 402
 403	switch (rdma_port_get_link_layer(ibdev, port_num)) {
 404	case IB_LINK_LAYER_INFINIBAND:
 405		output = "InfiniBand";
 406		break;
 407	case IB_LINK_LAYER_ETHERNET:
 408		output = "Ethernet";
 409		break;
 410	default:
 411		output = "Unknown";
 412		break;
 413	}
 414
 415	return sysfs_emit(buf, "%s\n", output);
 416}
 417
 418static IB_PORT_ATTR_RO(state);
 419static IB_PORT_ATTR_RO(lid);
 420static IB_PORT_ATTR_RO(lid_mask_count);
 421static IB_PORT_ATTR_RO(sm_lid);
 422static IB_PORT_ATTR_RO(sm_sl);
 423static IB_PORT_ATTR_RO(cap_mask);
 424static IB_PORT_ATTR_RO(rate);
 425static IB_PORT_ATTR_RO(phys_state);
 426static IB_PORT_ATTR_RO(link_layer);
 427
 428static struct attribute *port_default_attrs[] = {
 429	&ib_port_attr_state.attr,
 430	&ib_port_attr_lid.attr,
 431	&ib_port_attr_lid_mask_count.attr,
 432	&ib_port_attr_sm_lid.attr,
 433	&ib_port_attr_sm_sl.attr,
 434	&ib_port_attr_cap_mask.attr,
 435	&ib_port_attr_rate.attr,
 436	&ib_port_attr_phys_state.attr,
 437	&ib_port_attr_link_layer.attr,
 438	NULL
 439};
 440ATTRIBUTE_GROUPS(port_default);
 441
 442static ssize_t print_ndev(const struct ib_gid_attr *gid_attr, char *buf)
 443{
 444	struct net_device *ndev;
 445	int ret = -EINVAL;
 446
 447	rcu_read_lock();
 448	ndev = rcu_dereference(gid_attr->ndev);
 449	if (ndev)
 450		ret = sysfs_emit(buf, "%s\n", ndev->name);
 451	rcu_read_unlock();
 452	return ret;
 453}
 454
 455static ssize_t print_gid_type(const struct ib_gid_attr *gid_attr, char *buf)
 456{
 457	return sysfs_emit(buf, "%s\n",
 458			  ib_cache_gid_type_str(gid_attr->gid_type));
 459}
 460
 461static ssize_t _show_port_gid_attr(
 462	struct ib_device *ibdev, u32 port_num, struct ib_port_attribute *attr,
 463	char *buf,
 464	ssize_t (*print)(const struct ib_gid_attr *gid_attr, char *buf))
 465{
 466	struct port_table_attribute *tab_attr =
 467		container_of(attr, struct port_table_attribute, attr);
 468	const struct ib_gid_attr *gid_attr;
 469	ssize_t ret;
 470
 471	gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
 472	if (IS_ERR(gid_attr))
 473		/* -EINVAL is returned for user space compatibility reasons. */
 474		return -EINVAL;
 475
 476	ret = print(gid_attr, buf);
 477	rdma_put_gid_attr(gid_attr);
 478	return ret;
 479}
 480
 481static ssize_t show_port_gid(struct ib_device *ibdev, u32 port_num,
 482			     struct ib_port_attribute *attr, char *buf)
 483{
 484	struct port_table_attribute *tab_attr =
 485		container_of(attr, struct port_table_attribute, attr);
 486	const struct ib_gid_attr *gid_attr;
 487	int len;
 488
 489	gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
 490	if (IS_ERR(gid_attr)) {
 491		const union ib_gid zgid = {};
 492
 493		/* If reading GID fails, it is likely due to GID entry being
 494		 * empty (invalid) or reserved GID in the table.  User space
 495		 * expects to read GID table entries as long as it given index
 496		 * is within GID table size.  Administrative/debugging tool
 497		 * fails to query rest of the GID entries if it hits error
 498		 * while querying a GID of the given index.  To avoid user
 499		 * space throwing such error on fail to read gid, return zero
 500		 * GID as before. This maintains backward compatibility.
 501		 */
 502		return sysfs_emit(buf, "%pI6\n", zgid.raw);
 503	}
 504
 505	len = sysfs_emit(buf, "%pI6\n", gid_attr->gid.raw);
 506	rdma_put_gid_attr(gid_attr);
 507	return len;
 508}
 509
 510static ssize_t show_port_gid_attr_ndev(struct ib_device *ibdev, u32 port_num,
 511				       struct ib_port_attribute *attr,
 512				       char *buf)
 513{
 514	return _show_port_gid_attr(ibdev, port_num, attr, buf, print_ndev);
 515}
 516
 517static ssize_t show_port_gid_attr_gid_type(struct ib_device *ibdev,
 518					   u32 port_num,
 519					   struct ib_port_attribute *attr,
 520					   char *buf)
 521{
 522	return _show_port_gid_attr(ibdev, port_num, attr, buf, print_gid_type);
 523}
 524
 525static ssize_t show_port_pkey(struct ib_device *ibdev, u32 port_num,
 526			      struct ib_port_attribute *attr, char *buf)
 527{
 528	struct port_table_attribute *tab_attr =
 529		container_of(attr, struct port_table_attribute, attr);
 530	u16 pkey;
 531	int ret;
 532
 533	ret = ib_query_pkey(ibdev, port_num, tab_attr->index, &pkey);
 534	if (ret)
 535		return ret;
 536
 537	return sysfs_emit(buf, "0x%04x\n", pkey);
 538}
 539
 540#define PORT_PMA_ATTR(_name, _counter, _width, _offset)			\
 541struct port_table_attribute port_pma_attr_##_name = {			\
 542	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
 543	.index = (_offset) | ((_width) << 16) | ((_counter) << 24),	\
 544	.attr_id = IB_PMA_PORT_COUNTERS,				\
 545}
 546
 547#define PORT_PMA_ATTR_EXT(_name, _width, _offset)			\
 548struct port_table_attribute port_pma_attr_ext_##_name = {		\
 549	.attr  = __ATTR(_name, S_IRUGO, show_pma_counter, NULL),	\
 550	.index = (_offset) | ((_width) << 16),				\
 551	.attr_id = IB_PMA_PORT_COUNTERS_EXT,				\
 552}
 553
 554/*
 555 * Get a Perfmgmt MAD block of data.
 556 * Returns error code or the number of bytes retrieved.
 557 */
 558static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
 559		void *data, int offset, size_t size)
 560{
 561	struct ib_mad *in_mad;
 562	struct ib_mad *out_mad;
 563	size_t mad_size = sizeof(*out_mad);
 564	u16 out_mad_pkey_index = 0;
 565	ssize_t ret;
 566
 567	if (!dev->ops.process_mad)
 568		return -ENOSYS;
 569
 570	in_mad = kzalloc(sizeof(*in_mad), GFP_KERNEL);
 571	out_mad = kzalloc(sizeof(*out_mad), GFP_KERNEL);
 572	if (!in_mad || !out_mad) {
 573		ret = -ENOMEM;
 574		goto out;
 575	}
 576
 577	in_mad->mad_hdr.base_version  = 1;
 578	in_mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_PERF_MGMT;
 579	in_mad->mad_hdr.class_version = 1;
 580	in_mad->mad_hdr.method        = IB_MGMT_METHOD_GET;
 581	in_mad->mad_hdr.attr_id       = attr;
 582
 583	if (attr != IB_PMA_CLASS_PORT_INFO)
 584		in_mad->data[41] = port_num;	/* PortSelect field */
 585
 586	if ((dev->ops.process_mad(dev, IB_MAD_IGNORE_MKEY, port_num, NULL, NULL,
 587				  in_mad, out_mad, &mad_size,
 588				  &out_mad_pkey_index) &
 589	     (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
 590	    (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
 591		ret = -EINVAL;
 592		goto out;
 593	}
 594	memcpy(data, out_mad->data + offset, size);
 595	ret = size;
 596out:
 597	kfree(in_mad);
 598	kfree(out_mad);
 599	return ret;
 600}
 601
 602static ssize_t show_pma_counter(struct ib_device *ibdev, u32 port_num,
 603				struct ib_port_attribute *attr, char *buf)
 604{
 605	struct port_table_attribute *tab_attr =
 606		container_of(attr, struct port_table_attribute, attr);
 607	int offset = tab_attr->index & 0xffff;
 608	int width  = (tab_attr->index >> 16) & 0xff;
 609	int ret;
 610	u8 data[8];
 611	int len;
 612
 613	ret = get_perf_mad(ibdev, port_num, tab_attr->attr_id, &data,
 614			40 + offset / 8, sizeof(data));
 615	if (ret < 0)
 616		return ret;
 617
 618	switch (width) {
 619	case 4:
 620		len = sysfs_emit(buf, "%d\n",
 621				 (*data >> (4 - (offset % 8))) & 0xf);
 622		break;
 623	case 8:
 624		len = sysfs_emit(buf, "%u\n", *data);
 625		break;
 626	case 16:
 627		len = sysfs_emit(buf, "%u\n", be16_to_cpup((__be16 *)data));
 628		break;
 629	case 32:
 630		len = sysfs_emit(buf, "%u\n", be32_to_cpup((__be32 *)data));
 631		break;
 632	case 64:
 633		len = sysfs_emit(buf, "%llu\n", be64_to_cpup((__be64 *)data));
 634		break;
 635	default:
 636		len = 0;
 637		break;
 638	}
 639
 640	return len;
 641}
 642
 643static PORT_PMA_ATTR(symbol_error		    ,  0, 16,  32);
 644static PORT_PMA_ATTR(link_error_recovery	    ,  1,  8,  48);
 645static PORT_PMA_ATTR(link_downed		    ,  2,  8,  56);
 646static PORT_PMA_ATTR(port_rcv_errors		    ,  3, 16,  64);
 647static PORT_PMA_ATTR(port_rcv_remote_physical_errors,  4, 16,  80);
 648static PORT_PMA_ATTR(port_rcv_switch_relay_errors   ,  5, 16,  96);
 649static PORT_PMA_ATTR(port_xmit_discards		    ,  6, 16, 112);
 650static PORT_PMA_ATTR(port_xmit_constraint_errors    ,  7,  8, 128);
 651static PORT_PMA_ATTR(port_rcv_constraint_errors	    ,  8,  8, 136);
 652static PORT_PMA_ATTR(local_link_integrity_errors    ,  9,  4, 152);
 653static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10,  4, 156);
 654static PORT_PMA_ATTR(VL15_dropped		    , 11, 16, 176);
 655static PORT_PMA_ATTR(port_xmit_data		    , 12, 32, 192);
 656static PORT_PMA_ATTR(port_rcv_data		    , 13, 32, 224);
 657static PORT_PMA_ATTR(port_xmit_packets		    , 14, 32, 256);
 658static PORT_PMA_ATTR(port_rcv_packets		    , 15, 32, 288);
 659static PORT_PMA_ATTR(port_xmit_wait		    ,  0, 32, 320);
 660
 661/*
 662 * Counters added by extended set
 663 */
 664static PORT_PMA_ATTR_EXT(port_xmit_data		    , 64,  64);
 665static PORT_PMA_ATTR_EXT(port_rcv_data		    , 64, 128);
 666static PORT_PMA_ATTR_EXT(port_xmit_packets	    , 64, 192);
 667static PORT_PMA_ATTR_EXT(port_rcv_packets	    , 64, 256);
 668static PORT_PMA_ATTR_EXT(unicast_xmit_packets	    , 64, 320);
 669static PORT_PMA_ATTR_EXT(unicast_rcv_packets	    , 64, 384);
 670static PORT_PMA_ATTR_EXT(multicast_xmit_packets	    , 64, 448);
 671static PORT_PMA_ATTR_EXT(multicast_rcv_packets	    , 64, 512);
 672
 673static struct attribute *pma_attrs[] = {
 674	&port_pma_attr_symbol_error.attr.attr,
 675	&port_pma_attr_link_error_recovery.attr.attr,
 676	&port_pma_attr_link_downed.attr.attr,
 677	&port_pma_attr_port_rcv_errors.attr.attr,
 678	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
 679	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
 680	&port_pma_attr_port_xmit_discards.attr.attr,
 681	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
 682	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
 683	&port_pma_attr_local_link_integrity_errors.attr.attr,
 684	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
 685	&port_pma_attr_VL15_dropped.attr.attr,
 686	&port_pma_attr_port_xmit_data.attr.attr,
 687	&port_pma_attr_port_rcv_data.attr.attr,
 688	&port_pma_attr_port_xmit_packets.attr.attr,
 689	&port_pma_attr_port_rcv_packets.attr.attr,
 690	&port_pma_attr_port_xmit_wait.attr.attr,
 691	NULL
 692};
 693
 694static struct attribute *pma_attrs_ext[] = {
 695	&port_pma_attr_symbol_error.attr.attr,
 696	&port_pma_attr_link_error_recovery.attr.attr,
 697	&port_pma_attr_link_downed.attr.attr,
 698	&port_pma_attr_port_rcv_errors.attr.attr,
 699	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
 700	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
 701	&port_pma_attr_port_xmit_discards.attr.attr,
 702	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
 703	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
 704	&port_pma_attr_local_link_integrity_errors.attr.attr,
 705	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
 706	&port_pma_attr_VL15_dropped.attr.attr,
 707	&port_pma_attr_ext_port_xmit_data.attr.attr,
 708	&port_pma_attr_ext_port_rcv_data.attr.attr,
 709	&port_pma_attr_ext_port_xmit_packets.attr.attr,
 710	&port_pma_attr_port_xmit_wait.attr.attr,
 711	&port_pma_attr_ext_port_rcv_packets.attr.attr,
 712	&port_pma_attr_ext_unicast_rcv_packets.attr.attr,
 713	&port_pma_attr_ext_unicast_xmit_packets.attr.attr,
 714	&port_pma_attr_ext_multicast_rcv_packets.attr.attr,
 715	&port_pma_attr_ext_multicast_xmit_packets.attr.attr,
 716	NULL
 717};
 718
 719static struct attribute *pma_attrs_noietf[] = {
 720	&port_pma_attr_symbol_error.attr.attr,
 721	&port_pma_attr_link_error_recovery.attr.attr,
 722	&port_pma_attr_link_downed.attr.attr,
 723	&port_pma_attr_port_rcv_errors.attr.attr,
 724	&port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
 725	&port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
 726	&port_pma_attr_port_xmit_discards.attr.attr,
 727	&port_pma_attr_port_xmit_constraint_errors.attr.attr,
 728	&port_pma_attr_port_rcv_constraint_errors.attr.attr,
 729	&port_pma_attr_local_link_integrity_errors.attr.attr,
 730	&port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
 731	&port_pma_attr_VL15_dropped.attr.attr,
 732	&port_pma_attr_ext_port_xmit_data.attr.attr,
 733	&port_pma_attr_ext_port_rcv_data.attr.attr,
 734	&port_pma_attr_ext_port_xmit_packets.attr.attr,
 735	&port_pma_attr_ext_port_rcv_packets.attr.attr,
 736	&port_pma_attr_port_xmit_wait.attr.attr,
 737	NULL
 738};
 739
 740static const struct attribute_group pma_group = {
 741	.name  = "counters",
 742	.attrs  = pma_attrs
 743};
 744
 745static const struct attribute_group pma_group_ext = {
 746	.name  = "counters",
 747	.attrs  = pma_attrs_ext
 748};
 749
 750static const struct attribute_group pma_group_noietf = {
 751	.name  = "counters",
 752	.attrs  = pma_attrs_noietf
 753};
 754
 755static void ib_port_release(struct kobject *kobj)
 756{
 757	struct ib_port *port = container_of(kobj, struct ib_port, kobj);
 758	int i;
 759
 760	for (i = 0; i != ARRAY_SIZE(port->groups); i++)
 761		kfree(port->groups[i].attrs);
 762	if (port->hw_stats_data)
 763		rdma_free_hw_stats_struct(port->hw_stats_data->stats);
 764	kfree(port->hw_stats_data);
 765	kvfree(port);
 766}
 767
 768static void ib_port_gid_attr_release(struct kobject *kobj)
 769{
 770	struct gid_attr_group *gid_attr_group =
 771		container_of(kobj, struct gid_attr_group, kobj);
 772	int i;
 773
 774	for (i = 0; i != ARRAY_SIZE(gid_attr_group->groups); i++)
 775		kfree(gid_attr_group->groups[i].attrs);
 776	kfree(gid_attr_group);
 777}
 778
 779static struct kobj_type port_type = {
 780	.release       = ib_port_release,
 781	.sysfs_ops     = &port_sysfs_ops,
 782	.default_groups = port_default_groups,
 783};
 784
 785static struct kobj_type gid_attr_type = {
 786	.sysfs_ops      = &gid_attr_sysfs_ops,
 787	.release        = ib_port_gid_attr_release
 788};
 789
 790/*
 791 * Figure out which counter table to use depending on
 792 * the device capabilities.
 793 */
 794static const struct attribute_group *get_counter_table(struct ib_device *dev,
 795						       int port_num)
 796{
 797	struct ib_class_port_info cpi;
 798
 799	if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
 800				&cpi, 40, sizeof(cpi)) >= 0) {
 801		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
 802			/* We have extended counters */
 803			return &pma_group_ext;
 804
 805		if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
 806			/* But not the IETF ones */
 807			return &pma_group_noietf;
 808	}
 809
 810	/* Fall back to normal counters */
 811	return &pma_group;
 812}
 813
 814static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
 815			   u32 port_num, int index)
 816{
 817	int ret;
 818
 819	if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
 820		return 0;
 821	ret = dev->ops.get_hw_stats(dev, stats, port_num, index);
 822	if (ret < 0)
 823		return ret;
 824	if (ret == stats->num_counters)
 825		stats->timestamp = jiffies;
 826
 827	return 0;
 828}
 829
 830static int print_hw_stat(struct ib_device *dev, int port_num,
 831			 struct rdma_hw_stats *stats, int index, char *buf)
 832{
 833	u64 v = rdma_counter_get_hwstat_value(dev, port_num, index);
 834
 835	return sysfs_emit(buf, "%llu\n", stats->value[index] + v);
 836}
 837
 838static ssize_t show_hw_stats(struct ib_device *ibdev,
 839			     struct rdma_hw_stats *stats, unsigned int index,
 840			     unsigned int port_num, char *buf)
 841{
 842	int ret;
 843
 844	mutex_lock(&stats->lock);
 845	ret = update_hw_stats(ibdev, stats, port_num, index);
 846	if (ret)
 847		goto unlock;
 848	ret = print_hw_stat(ibdev, port_num, stats, index, buf);
 849unlock:
 850	mutex_unlock(&stats->lock);
 851
 852	return ret;
 853}
 854
 855static ssize_t show_stats_lifespan(struct ib_device *ibdev,
 856				   struct rdma_hw_stats *stats,
 857				   unsigned int index, unsigned int port_num,
 858				   char *buf)
 859{
 860	int msecs;
 861
 862	mutex_lock(&stats->lock);
 863	msecs = jiffies_to_msecs(stats->lifespan);
 864	mutex_unlock(&stats->lock);
 865
 866	return sysfs_emit(buf, "%d\n", msecs);
 867}
 868
 869static ssize_t set_stats_lifespan(struct ib_device *ibdev,
 870				   struct rdma_hw_stats *stats,
 871				   unsigned int index, unsigned int port_num,
 872				   const char *buf, size_t count)
 873{
 874	int msecs;
 875	int jiffies;
 876	int ret;
 877
 878	ret = kstrtoint(buf, 10, &msecs);
 879	if (ret)
 880		return ret;
 881	if (msecs < 0 || msecs > 10000)
 882		return -EINVAL;
 883	jiffies = msecs_to_jiffies(msecs);
 884
 885	mutex_lock(&stats->lock);
 886	stats->lifespan = jiffies;
 887	mutex_unlock(&stats->lock);
 888
 889	return count;
 890}
 891
 892static struct hw_stats_device_data *
 893alloc_hw_stats_device(struct ib_device *ibdev)
 894{
 895	struct hw_stats_device_data *data;
 896	struct rdma_hw_stats *stats;
 897
 898	if (!ibdev->ops.alloc_hw_device_stats)
 899		return ERR_PTR(-EOPNOTSUPP);
 900	stats = ibdev->ops.alloc_hw_device_stats(ibdev);
 901	if (!stats)
 902		return ERR_PTR(-ENOMEM);
 903	if (!stats->descs || stats->num_counters <= 0)
 904		goto err_free_stats;
 905
 906	/*
 907	 * Two extra attribue elements here, one for the lifespan entry and
 908	 * one to NULL terminate the list for the sysfs core code
 909	 */
 910	data = kzalloc(struct_size(data, attrs, size_add(stats->num_counters, 1)),
 911		       GFP_KERNEL);
 912	if (!data)
 913		goto err_free_stats;
 914	data->group.attrs = kcalloc(stats->num_counters + 2,
 915				    sizeof(*data->group.attrs), GFP_KERNEL);
 916	if (!data->group.attrs)
 917		goto err_free_data;
 918
 919	data->group.name = "hw_counters";
 920	data->stats = stats;
 921	return data;
 922
 923err_free_data:
 924	kfree(data);
 925err_free_stats:
 926	rdma_free_hw_stats_struct(stats);
 927	return ERR_PTR(-ENOMEM);
 928}
 929
 930void ib_device_release_hw_stats(struct hw_stats_device_data *data)
 931{
 932	kfree(data->group.attrs);
 933	rdma_free_hw_stats_struct(data->stats);
 934	kfree(data);
 935}
 936
 937int ib_setup_device_attrs(struct ib_device *ibdev)
 938{
 939	struct hw_stats_device_attribute *attr;
 940	struct hw_stats_device_data *data;
 941	bool opstat_skipped = false;
 942	int i, ret, pos = 0;
 943
 944	data = alloc_hw_stats_device(ibdev);
 945	if (IS_ERR(data)) {
 946		if (PTR_ERR(data) == -EOPNOTSUPP)
 947			return 0;
 948		return PTR_ERR(data);
 949	}
 950	ibdev->hw_stats_data = data;
 951
 952	ret = ibdev->ops.get_hw_stats(ibdev, data->stats, 0,
 953				      data->stats->num_counters);
 954	if (ret != data->stats->num_counters) {
 955		if (WARN_ON(ret >= 0))
 956			return -EINVAL;
 957		return ret;
 958	}
 959
 960	data->stats->timestamp = jiffies;
 961
 962	for (i = 0; i < data->stats->num_counters; i++) {
 963		if (data->stats->descs[i].flags & IB_STAT_FLAG_OPTIONAL) {
 964			opstat_skipped = true;
 965			continue;
 966		}
 967
 968		WARN_ON(opstat_skipped);
 969		attr = &data->attrs[pos];
 970		sysfs_attr_init(&attr->attr.attr);
 971		attr->attr.attr.name = data->stats->descs[i].name;
 972		attr->attr.attr.mode = 0444;
 973		attr->attr.show = hw_stat_device_show;
 974		attr->show = show_hw_stats;
 975		data->group.attrs[pos] = &attr->attr.attr;
 976		pos++;
 977	}
 978
 979	attr = &data->attrs[pos];
 980	sysfs_attr_init(&attr->attr.attr);
 981	attr->attr.attr.name = "lifespan";
 982	attr->attr.attr.mode = 0644;
 983	attr->attr.show = hw_stat_device_show;
 984	attr->show = show_stats_lifespan;
 985	attr->attr.store = hw_stat_device_store;
 986	attr->store = set_stats_lifespan;
 987	data->group.attrs[pos] = &attr->attr.attr;
 988	for (i = 0; i != ARRAY_SIZE(ibdev->groups); i++)
 989		if (!ibdev->groups[i]) {
 990			ibdev->groups[i] = &data->group;
 991			return 0;
 992		}
 993	WARN(true, "struct ib_device->groups is too small");
 994	return -EINVAL;
 995}
 996
 997static struct hw_stats_port_data *
 998alloc_hw_stats_port(struct ib_port *port, struct attribute_group *group)
 999{
1000	struct ib_device *ibdev = port->ibdev;
1001	struct hw_stats_port_data *data;
1002	struct rdma_hw_stats *stats;
1003
1004	if (!ibdev->ops.alloc_hw_port_stats)
1005		return ERR_PTR(-EOPNOTSUPP);
1006	stats = ibdev->ops.alloc_hw_port_stats(port->ibdev, port->port_num);
1007	if (!stats)
1008		return ERR_PTR(-ENOMEM);
1009	if (!stats->descs || stats->num_counters <= 0)
1010		goto err_free_stats;
1011
1012	/*
1013	 * Two extra attribue elements here, one for the lifespan entry and
1014	 * one to NULL terminate the list for the sysfs core code
1015	 */
1016	data = kzalloc(struct_size(data, attrs, size_add(stats->num_counters, 1)),
1017		       GFP_KERNEL);
1018	if (!data)
1019		goto err_free_stats;
1020	group->attrs = kcalloc(stats->num_counters + 2,
1021				    sizeof(*group->attrs), GFP_KERNEL);
1022	if (!group->attrs)
1023		goto err_free_data;
1024
1025	group->name = "hw_counters";
1026	data->stats = stats;
1027	return data;
1028
1029err_free_data:
1030	kfree(data);
1031err_free_stats:
1032	rdma_free_hw_stats_struct(stats);
1033	return ERR_PTR(-ENOMEM);
1034}
1035
1036static int setup_hw_port_stats(struct ib_port *port,
1037			       struct attribute_group *group)
1038{
1039	struct hw_stats_port_attribute *attr;
1040	struct hw_stats_port_data *data;
1041	bool opstat_skipped = false;
1042	int i, ret, pos = 0;
1043
1044	data = alloc_hw_stats_port(port, group);
1045	if (IS_ERR(data))
1046		return PTR_ERR(data);
1047
1048	ret = port->ibdev->ops.get_hw_stats(port->ibdev, data->stats,
1049					    port->port_num,
1050					    data->stats->num_counters);
1051	if (ret != data->stats->num_counters) {
1052		if (WARN_ON(ret >= 0))
1053			return -EINVAL;
1054		return ret;
1055	}
1056
1057	data->stats->timestamp = jiffies;
1058
1059	for (i = 0; i < data->stats->num_counters; i++) {
1060		if (data->stats->descs[i].flags & IB_STAT_FLAG_OPTIONAL) {
1061			opstat_skipped = true;
1062			continue;
1063		}
1064
1065		WARN_ON(opstat_skipped);
1066		attr = &data->attrs[pos];
1067		sysfs_attr_init(&attr->attr.attr);
1068		attr->attr.attr.name = data->stats->descs[i].name;
1069		attr->attr.attr.mode = 0444;
1070		attr->attr.show = hw_stat_port_show;
1071		attr->show = show_hw_stats;
1072		group->attrs[pos] = &attr->attr.attr;
1073		pos++;
1074	}
1075
1076	attr = &data->attrs[pos];
1077	sysfs_attr_init(&attr->attr.attr);
1078	attr->attr.attr.name = "lifespan";
1079	attr->attr.attr.mode = 0644;
1080	attr->attr.show = hw_stat_port_show;
1081	attr->show = show_stats_lifespan;
1082	attr->attr.store = hw_stat_port_store;
1083	attr->store = set_stats_lifespan;
1084	group->attrs[pos] = &attr->attr.attr;
1085
1086	port->hw_stats_data = data;
1087	return 0;
1088}
1089
1090struct rdma_hw_stats *ib_get_hw_stats_port(struct ib_device *ibdev,
1091					   u32 port_num)
1092{
1093	if (!ibdev->port_data || !rdma_is_port_valid(ibdev, port_num) ||
1094	    !ibdev->port_data[port_num].sysfs->hw_stats_data)
1095		return NULL;
1096	return ibdev->port_data[port_num].sysfs->hw_stats_data->stats;
1097}
1098
1099static int
1100alloc_port_table_group(const char *name, struct attribute_group *group,
1101		       struct port_table_attribute *attrs, size_t num,
1102		       ssize_t (*show)(struct ib_device *ibdev, u32 port_num,
1103				       struct ib_port_attribute *, char *buf))
1104{
1105	struct attribute **attr_list;
1106	int i;
1107
1108	attr_list = kcalloc(num + 1, sizeof(*attr_list), GFP_KERNEL);
1109	if (!attr_list)
1110		return -ENOMEM;
1111
1112	for (i = 0; i < num; i++) {
1113		struct port_table_attribute *element = &attrs[i];
1114
1115		if (snprintf(element->name, sizeof(element->name), "%d", i) >=
1116		    sizeof(element->name))
1117			goto err;
1118
1119		sysfs_attr_init(&element->attr.attr);
1120		element->attr.attr.name = element->name;
1121		element->attr.attr.mode = 0444;
1122		element->attr.show = show;
1123		element->index = i;
1124
1125		attr_list[i] = &element->attr.attr;
1126	}
1127	group->name = name;
1128	group->attrs = attr_list;
1129	return 0;
1130err:
1131	kfree(attr_list);
1132	return -EINVAL;
1133}
1134
1135/*
1136 * Create the sysfs:
1137 *  ibp0s9/ports/XX/gid_attrs/{ndevs,types}/YYY
1138 * YYY is the gid table index in decimal
1139 */
1140static int setup_gid_attrs(struct ib_port *port,
1141			   const struct ib_port_attr *attr)
1142{
1143	struct gid_attr_group *gid_attr_group;
1144	int ret;
1145
1146	gid_attr_group = kzalloc(struct_size(gid_attr_group, attrs_list,
1147					     size_mul(attr->gid_tbl_len, 2)),
1148				 GFP_KERNEL);
1149	if (!gid_attr_group)
1150		return -ENOMEM;
1151	gid_attr_group->port = port;
1152	kobject_init(&gid_attr_group->kobj, &gid_attr_type);
1153
1154	ret = alloc_port_table_group("ndevs", &gid_attr_group->groups[0],
1155				     gid_attr_group->attrs_list,
1156				     attr->gid_tbl_len,
1157				     show_port_gid_attr_ndev);
1158	if (ret)
1159		goto err_put;
1160	gid_attr_group->groups_list[0] = &gid_attr_group->groups[0];
1161
1162	ret = alloc_port_table_group(
1163		"types", &gid_attr_group->groups[1],
1164		gid_attr_group->attrs_list + attr->gid_tbl_len,
1165		attr->gid_tbl_len, show_port_gid_attr_gid_type);
1166	if (ret)
1167		goto err_put;
1168	gid_attr_group->groups_list[1] = &gid_attr_group->groups[1];
1169
1170	ret = kobject_add(&gid_attr_group->kobj, &port->kobj, "gid_attrs");
1171	if (ret)
1172		goto err_put;
1173	ret = sysfs_create_groups(&gid_attr_group->kobj,
1174				  gid_attr_group->groups_list);
1175	if (ret)
1176		goto err_del;
1177	port->gid_attr_group = gid_attr_group;
1178	return 0;
1179
1180err_del:
1181	kobject_del(&gid_attr_group->kobj);
1182err_put:
1183	kobject_put(&gid_attr_group->kobj);
1184	return ret;
1185}
1186
1187static void destroy_gid_attrs(struct ib_port *port)
1188{
1189	struct gid_attr_group *gid_attr_group = port->gid_attr_group;
1190
1191	if (!gid_attr_group)
1192		return;
1193	sysfs_remove_groups(&gid_attr_group->kobj, gid_attr_group->groups_list);
1194	kobject_del(&gid_attr_group->kobj);
1195	kobject_put(&gid_attr_group->kobj);
1196}
1197
1198/*
1199 * Create the sysfs:
1200 *  ibp0s9/ports/XX/{gids,pkeys,counters}/YYY
1201 */
1202static struct ib_port *setup_port(struct ib_core_device *coredev, int port_num,
1203				  const struct ib_port_attr *attr)
1204{
1205	struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1206	bool is_full_dev = &device->coredev == coredev;
1207	const struct attribute_group **cur_group;
1208	struct ib_port *p;
1209	int ret;
1210
1211	p = kvzalloc(struct_size(p, attrs_list,
1212				size_add(attr->gid_tbl_len, attr->pkey_tbl_len)),
1213		     GFP_KERNEL);
1214	if (!p)
1215		return ERR_PTR(-ENOMEM);
1216	p->ibdev = device;
1217	p->port_num = port_num;
1218	kobject_init(&p->kobj, &port_type);
1219
1220	if (device->port_data && is_full_dev)
1221		device->port_data[port_num].sysfs = p;
1222
1223	cur_group = p->groups_list;
1224	ret = alloc_port_table_group("gids", &p->groups[0], p->attrs_list,
1225				     attr->gid_tbl_len, show_port_gid);
1226	if (ret)
1227		goto err_put;
1228	*cur_group++ = &p->groups[0];
1229
1230	if (attr->pkey_tbl_len) {
1231		ret = alloc_port_table_group("pkeys", &p->groups[1],
1232					     p->attrs_list + attr->gid_tbl_len,
1233					     attr->pkey_tbl_len, show_port_pkey);
1234		if (ret)
1235			goto err_put;
1236		*cur_group++ = &p->groups[1];
1237	}
1238
1239	/*
1240	 * If port == 0, it means hw_counters are per device and not per
1241	 * port, so holder should be device. Therefore skip per port
1242	 * counter initialization.
1243	 */
1244	if (port_num && is_full_dev) {
1245		ret = setup_hw_port_stats(p, &p->groups[2]);
1246		if (ret && ret != -EOPNOTSUPP)
1247			goto err_put;
1248		if (!ret)
1249			*cur_group++ = &p->groups[2];
1250	}
1251
1252	if (device->ops.process_mad && is_full_dev)
1253		*cur_group++ = get_counter_table(device, port_num);
1254
1255	ret = kobject_add(&p->kobj, coredev->ports_kobj, "%d", port_num);
1256	if (ret)
1257		goto err_put;
1258	ret = sysfs_create_groups(&p->kobj, p->groups_list);
1259	if (ret)
1260		goto err_del;
1261	if (is_full_dev) {
1262		ret = sysfs_create_groups(&p->kobj, device->ops.port_groups);
1263		if (ret)
1264			goto err_groups;
1265	}
1266
1267	list_add_tail(&p->kobj.entry, &coredev->port_list);
1268	return p;
1269
1270err_groups:
1271	sysfs_remove_groups(&p->kobj, p->groups_list);
1272err_del:
1273	kobject_del(&p->kobj);
1274err_put:
1275	if (device->port_data && is_full_dev)
1276		device->port_data[port_num].sysfs = NULL;
1277	kobject_put(&p->kobj);
1278	return ERR_PTR(ret);
1279}
1280
1281static void destroy_port(struct ib_core_device *coredev, struct ib_port *port)
1282{
1283	bool is_full_dev = &port->ibdev->coredev == coredev;
1284
1285	list_del(&port->kobj.entry);
1286	if (is_full_dev)
1287		sysfs_remove_groups(&port->kobj, port->ibdev->ops.port_groups);
1288
1289	sysfs_remove_groups(&port->kobj, port->groups_list);
1290	kobject_del(&port->kobj);
1291
1292	if (port->ibdev->port_data &&
1293	    port->ibdev->port_data[port->port_num].sysfs == port)
1294		port->ibdev->port_data[port->port_num].sysfs = NULL;
1295
1296	kobject_put(&port->kobj);
1297}
1298
1299static const char *node_type_string(int node_type)
1300{
1301	switch (node_type) {
1302	case RDMA_NODE_IB_CA:
1303		return "CA";
1304	case RDMA_NODE_IB_SWITCH:
1305		return "switch";
1306	case RDMA_NODE_IB_ROUTER:
1307		return "router";
1308	case RDMA_NODE_RNIC:
1309		return "RNIC";
1310	case RDMA_NODE_USNIC:
1311		return "usNIC";
1312	case RDMA_NODE_USNIC_UDP:
1313		return "usNIC UDP";
1314	case RDMA_NODE_UNSPECIFIED:
1315		return "unspecified";
1316	}
1317	return "<unknown>";
1318}
1319
1320static ssize_t node_type_show(struct device *device,
1321			      struct device_attribute *attr, char *buf)
1322{
1323	struct ib_device *dev = rdma_device_to_ibdev(device);
1324
1325	return sysfs_emit(buf, "%u: %s\n", dev->node_type,
1326			  node_type_string(dev->node_type));
1327}
1328static DEVICE_ATTR_RO(node_type);
1329
1330static ssize_t sys_image_guid_show(struct device *device,
1331				   struct device_attribute *dev_attr, char *buf)
1332{
1333	struct ib_device *dev = rdma_device_to_ibdev(device);
1334	__be16 *guid = (__be16 *)&dev->attrs.sys_image_guid;
1335
1336	return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1337			  be16_to_cpu(guid[0]),
1338			  be16_to_cpu(guid[1]),
1339			  be16_to_cpu(guid[2]),
1340			  be16_to_cpu(guid[3]));
1341}
1342static DEVICE_ATTR_RO(sys_image_guid);
1343
1344static ssize_t node_guid_show(struct device *device,
1345			      struct device_attribute *attr, char *buf)
1346{
1347	struct ib_device *dev = rdma_device_to_ibdev(device);
1348	__be16 *node_guid = (__be16 *)&dev->node_guid;
1349
1350	return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1351			  be16_to_cpu(node_guid[0]),
1352			  be16_to_cpu(node_guid[1]),
1353			  be16_to_cpu(node_guid[2]),
1354			  be16_to_cpu(node_guid[3]));
1355}
1356static DEVICE_ATTR_RO(node_guid);
1357
1358static ssize_t node_desc_show(struct device *device,
1359			      struct device_attribute *attr, char *buf)
1360{
1361	struct ib_device *dev = rdma_device_to_ibdev(device);
1362
1363	return sysfs_emit(buf, "%.64s\n", dev->node_desc);
1364}
1365
1366static ssize_t node_desc_store(struct device *device,
1367			       struct device_attribute *attr,
1368			       const char *buf, size_t count)
1369{
1370	struct ib_device *dev = rdma_device_to_ibdev(device);
1371	struct ib_device_modify desc = {};
1372	int ret;
1373
1374	if (!dev->ops.modify_device)
1375		return -EOPNOTSUPP;
1376
1377	memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1378	ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1379	if (ret)
1380		return ret;
1381
1382	return count;
1383}
1384static DEVICE_ATTR_RW(node_desc);
1385
1386static ssize_t fw_ver_show(struct device *device, struct device_attribute *attr,
1387			   char *buf)
1388{
1389	struct ib_device *dev = rdma_device_to_ibdev(device);
1390	char version[IB_FW_VERSION_NAME_MAX] = {};
1391
1392	ib_get_device_fw_str(dev, version);
1393
1394	return sysfs_emit(buf, "%s\n", version);
1395}
1396static DEVICE_ATTR_RO(fw_ver);
1397
1398static struct attribute *ib_dev_attrs[] = {
1399	&dev_attr_node_type.attr,
1400	&dev_attr_node_guid.attr,
1401	&dev_attr_sys_image_guid.attr,
1402	&dev_attr_fw_ver.attr,
1403	&dev_attr_node_desc.attr,
1404	NULL,
1405};
1406
1407const struct attribute_group ib_dev_attr_group = {
1408	.attrs = ib_dev_attrs,
1409};
1410
1411void ib_free_port_attrs(struct ib_core_device *coredev)
1412{
1413	struct kobject *p, *t;
1414
1415	list_for_each_entry_safe(p, t, &coredev->port_list, entry) {
1416		struct ib_port *port = container_of(p, struct ib_port, kobj);
1417
1418		destroy_gid_attrs(port);
1419		destroy_port(coredev, port);
1420	}
1421
1422	kobject_put(coredev->ports_kobj);
1423}
1424
1425int ib_setup_port_attrs(struct ib_core_device *coredev)
1426{
1427	struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1428	u32 port_num;
1429	int ret;
1430
1431	coredev->ports_kobj = kobject_create_and_add("ports",
1432						     &coredev->dev.kobj);
1433	if (!coredev->ports_kobj)
1434		return -ENOMEM;
1435
1436	rdma_for_each_port (device, port_num) {
1437		struct ib_port_attr attr;
1438		struct ib_port *port;
1439
1440		ret = ib_query_port(device, port_num, &attr);
1441		if (ret)
1442			goto err_put;
1443
1444		port = setup_port(coredev, port_num, &attr);
1445		if (IS_ERR(port)) {
1446			ret = PTR_ERR(port);
1447			goto err_put;
1448		}
1449
1450		ret = setup_gid_attrs(port, &attr);
1451		if (ret)
1452			goto err_put;
1453	}
1454	return 0;
1455
1456err_put:
1457	ib_free_port_attrs(coredev);
1458	return ret;
1459}
1460
1461/**
1462 * ib_port_register_client_groups - Add an ib_client's attributes to the port
1463 *
1464 * @ibdev: IB device to add counters
1465 * @port_num: valid port number
1466 * @groups: Group list of attributes
1467 *
1468 * Do not use. Only for legacy sysfs compatibility.
1469 */
1470int ib_port_register_client_groups(struct ib_device *ibdev, u32 port_num,
1471				   const struct attribute_group **groups)
1472{
1473	return sysfs_create_groups(&ibdev->port_data[port_num].sysfs->kobj,
1474				   groups);
1475}
1476EXPORT_SYMBOL(ib_port_register_client_groups);
1477
1478void ib_port_unregister_client_groups(struct ib_device *ibdev, u32 port_num,
1479				      const struct attribute_group **groups)
1480{
1481	return sysfs_remove_groups(&ibdev->port_data[port_num].sysfs->kobj,
1482				   groups);
1483}
1484EXPORT_SYMBOL(ib_port_unregister_client_groups);