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