drivers/infiniband/core/user_mad.c

   1 /*
   2  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
   3  * Copyright (c) 2005 Voltaire, Inc. 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  * $Id: user_mad.c 2814 2005-07-06 19:14:09Z halr $
  35  */
  36
  37 #include <linux/module.h>
  38 #include <linux/init.h>
  39 #include <linux/device.h>
  40 #include <linux/err.h>
  41 #include <linux/fs.h>
  42 #include <linux/cdev.h>
  43 #include <linux/pci.h>
  44 #include <linux/dma-mapping.h>
  45 #include <linux/poll.h>
  46 #include <linux/rwsem.h>
  47 #include <linux/kref.h>
  48
  49 #include <asm/uaccess.h>
  50 #include <asm/semaphore.h>
  51
  52 #include <rdma/ib_mad.h>
  53 #include <rdma/ib_user_mad.h>
  54
  55 MODULE_AUTHOR("Roland Dreier");
  56 MODULE_DESCRIPTION("InfiniBand userspace MAD packet access");
  57 MODULE_LICENSE("Dual BSD/GPL");
  58
  59 enum {
  60         IB_UMAD_MAX_PORTS  = 64,
  61         IB_UMAD_MAX_AGENTS = 32,
  62
  63         IB_UMAD_MAJOR      = 231,
  64         IB_UMAD_MINOR_BASE = 0
  65 };
  66
  67 struct ib_umad_port {
  68         int                    devnum;
  69         struct cdev            dev;
  70         struct class_device    class_dev;
  71
  72         int                    sm_devnum;
  73         struct cdev            sm_dev;
  74         struct class_device    sm_class_dev;
  75         struct semaphore       sm_sem;
  76
  77         struct ib_device      *ib_dev;
  78         struct ib_umad_device *umad_dev;
  79         u8                     port_num;
  80 };
  81
  82 struct ib_umad_device {
  83         int                  start_port, end_port;
  84         struct kref          ref;
  85         struct ib_umad_port  port[0];
  86 };
  87
  88 struct ib_umad_file {
  89         struct ib_umad_port *port;
  90         spinlock_t           recv_lock;
  91         struct list_head     recv_list;
  92         wait_queue_head_t    recv_wait;
  93         struct rw_semaphore  agent_mutex;
  94         struct ib_mad_agent *agent[IB_UMAD_MAX_AGENTS];
  95         struct ib_mr        *mr[IB_UMAD_MAX_AGENTS];
  96 };
  97
  98 struct ib_umad_packet {
  99         struct ib_ah      *ah;
 100         struct ib_mad_send_buf *msg;
 101         struct list_head   list;
 102         int                length;
 103         DECLARE_PCI_UNMAP_ADDR(mapping)
 104         struct ib_user_mad mad;
 105 };
 106
 107 static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
 108 static spinlock_t map_lock;
 109 static DECLARE_BITMAP(dev_map, IB_UMAD_MAX_PORTS * 2);
 110
 111 static void ib_umad_add_one(struct ib_device *device);
 112 static void ib_umad_remove_one(struct ib_device *device);
 113
 114 static int queue_packet(struct ib_umad_file *file,
 115                         struct ib_mad_agent *agent,
 116                         struct ib_umad_packet *packet)
 117 {
 118         int ret = 1;
 119
 120         down_read(&file->agent_mutex);
 121         for (packet->mad.hdr.id = 0;
 122              packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
 123              packet->mad.hdr.id++)
 124                 if (agent == file->agent[packet->mad.hdr.id]) {
 125                         spin_lock_irq(&file->recv_lock);
 126                         list_add_tail(&packet->list, &file->recv_list);
 127                         spin_unlock_irq(&file->recv_lock);
 128                         wake_up_interruptible(&file->recv_wait);
 129                         ret = 0;
 130                         break;
 131                 }
 132
 133         up_read(&file->agent_mutex);
 134
 135         return ret;
 136 }
 137
 138 static void send_handler(struct ib_mad_agent *agent,
 139                          struct ib_mad_send_wc *send_wc)
 140 {
 141         struct ib_umad_file *file = agent->context;
 142         struct ib_umad_packet *timeout, *packet =
 143                 (void *) (unsigned long) send_wc->wr_id;
 144
 145         ib_destroy_ah(packet->msg->send_wr.wr.ud.ah);
 146         ib_free_send_mad(packet->msg);
 147
 148         if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
 149                 timeout = kmalloc(sizeof *timeout + sizeof (struct ib_mad_hdr),
 150                                   GFP_KERNEL);
 151                 if (!timeout)
 152                         goto out;
 153
 154                 memset(timeout, 0, sizeof *timeout + sizeof (struct ib_mad_hdr));
 155
 156                 timeout->length = sizeof (struct ib_mad_hdr);
 157                 timeout->mad.hdr.id = packet->mad.hdr.id;
 158                 timeout->mad.hdr.status = ETIMEDOUT;
 159                 memcpy(timeout->mad.data, packet->mad.data,
 160                        sizeof (struct ib_mad_hdr));
 161
 162                 if (!queue_packet(file, agent, timeout))
 163                                 return;
 164         }
 165 out:
 166         kfree(packet);
 167 }
 168
 169 static void recv_handler(struct ib_mad_agent *agent,
 170                          struct ib_mad_recv_wc *mad_recv_wc)
 171 {
 172         struct ib_umad_file *file = agent->context;
 173         struct ib_umad_packet *packet;
 174         int length;
 175
 176         if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
 177                 goto out;
 178
 179         length = mad_recv_wc->mad_len;
 180         packet = kmalloc(sizeof *packet + length, GFP_KERNEL);
 181         if (!packet)
 182                 goto out;
 183
 184         memset(packet, 0, sizeof *packet + length);
 185         packet->length = length;
 186
 187         ib_coalesce_recv_mad(mad_recv_wc, packet->mad.data);
 188
 189         packet->mad.hdr.status    = 0;
 190         packet->mad.hdr.length    = length + sizeof (struct ib_user_mad);
 191         packet->mad.hdr.qpn       = cpu_to_be32(mad_recv_wc->wc->src_qp);
 192         packet->mad.hdr.lid       = cpu_to_be16(mad_recv_wc->wc->slid);
 193         packet->mad.hdr.sl        = mad_recv_wc->wc->sl;
 194         packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
 195         packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
 196         if (packet->mad.hdr.grh_present) {
 197                 /* XXX parse GRH */
 198                 packet->mad.hdr.gid_index       = 0;
 199                 packet->mad.hdr.hop_limit       = 0;
 200                 packet->mad.hdr.traffic_class   = 0;
 201                 memset(packet->mad.hdr.gid, 0, 16);
 202                 packet->mad.hdr.flow_label      = 0;
 203         }
 204
 205         if (queue_packet(file, agent, packet))
 206                 kfree(packet);
 207
 208 out:
 209         ib_free_recv_mad(mad_recv_wc);
 210 }
 211
 212 static ssize_t ib_umad_read(struct file *filp, char __user *buf,
 213                             size_t count, loff_t *pos)
 214 {
 215         struct ib_umad_file *file = filp->private_data;
 216         struct ib_umad_packet *packet;
 217         ssize_t ret;
 218
 219         if (count < sizeof (struct ib_user_mad) + sizeof (struct ib_mad))
 220                 return -EINVAL;
 221
 222         spin_lock_irq(&file->recv_lock);
 223
 224         while (list_empty(&file->recv_list)) {
 225                 spin_unlock_irq(&file->recv_lock);
 226
 227                 if (filp->f_flags & O_NONBLOCK)
 228                         return -EAGAIN;
 229
 230                 if (wait_event_interruptible(file->recv_wait,
 231                                              !list_empty(&file->recv_list)))
 232                         return -ERESTARTSYS;
 233
 234                 spin_lock_irq(&file->recv_lock);
 235         }
 236
 237         packet = list_entry(file->recv_list.next, struct ib_umad_packet, list);
 238         list_del(&packet->list);
 239
 240         spin_unlock_irq(&file->recv_lock);
 241
 242         if (count < packet->length + sizeof (struct ib_user_mad)) {
 243                 /* Return length needed (and first RMPP segment) if too small */
 244                 if (copy_to_user(buf, &packet->mad,
 245                                  sizeof (struct ib_user_mad) + sizeof (struct ib_mad)))
 246                         ret = -EFAULT;
 247                 else
 248                         ret = -ENOSPC;
 249         } else if (copy_to_user(buf, &packet->mad,
 250                               packet->length + sizeof (struct ib_user_mad)))
 251                 ret = -EFAULT;
 252         else
 253                 ret = packet->length + sizeof (struct ib_user_mad);
 254         if (ret < 0) {
 255                 /* Requeue packet */
 256                 spin_lock_irq(&file->recv_lock);
 257                 list_add(&packet->list, &file->recv_list);
 258                 spin_unlock_irq(&file->recv_lock);
 259         } else
 260                 kfree(packet);
 261         return ret;
 262 }
 263
 264 static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
 265                              size_t count, loff_t *pos)
 266 {
 267         struct ib_umad_file *file = filp->private_data;
 268         struct ib_umad_packet *packet;
 269         struct ib_mad_agent *agent;
 270         struct ib_ah_attr ah_attr;
 271         struct ib_send_wr *bad_wr;
 272         struct ib_rmpp_mad *rmpp_mad;
 273         u8 method;
 274         __be64 *tid;
 275         int ret, length, hdr_len, data_len, rmpp_hdr_size;
 276         int rmpp_active = 0;
 277
 278         if (count < sizeof (struct ib_user_mad))
 279                 return -EINVAL;
 280
 281         length = count - sizeof (struct ib_user_mad);
 282         packet = kmalloc(sizeof *packet + sizeof(struct ib_mad_hdr) +
 283                          sizeof (struct ib_rmpp_hdr), GFP_KERNEL);
 284         if (!packet)
 285                 return -ENOMEM;
 286
 287         if (copy_from_user(&packet->mad, buf,
 288                             sizeof (struct ib_user_mad) +
 289                             sizeof (struct ib_mad_hdr) +
 290                             sizeof (struct ib_rmpp_hdr))) {
 291                 ret = -EFAULT;
 292                 goto err;
 293         }
 294
 295         if (packet->mad.hdr.id < 0 ||
 296             packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
 297                 ret = -EINVAL;
 298                 goto err;
 299         }
 300
 301         packet->length = length;
 302
 303         down_read(&file->agent_mutex);
 304
 305         agent = file->agent[packet->mad.hdr.id];
 306         if (!agent) {
 307                 ret = -EINVAL;
 308                 goto err_up;
 309         }
 310
 311         memset(&ah_attr, 0, sizeof ah_attr);
 312         ah_attr.dlid          = be16_to_cpu(packet->mad.hdr.lid);
 313         ah_attr.sl            = packet->mad.hdr.sl;
 314         ah_attr.src_path_bits = packet->mad.hdr.path_bits;
 315         ah_attr.port_num      = file->port->port_num;
 316         if (packet->mad.hdr.grh_present) {
 317                 ah_attr.ah_flags = IB_AH_GRH;
 318                 memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
 319                 ah_attr.grh.flow_label     = be32_to_cpu(packet->mad.hdr.flow_label);
 320                 ah_attr.grh.hop_limit      = packet->mad.hdr.hop_limit;
 321                 ah_attr.grh.traffic_class  = packet->mad.hdr.traffic_class;
 322         }
 323
 324         packet->ah = ib_create_ah(agent->qp->pd, &ah_attr);
 325         if (IS_ERR(packet->ah)) {
 326                 ret = PTR_ERR(packet->ah);
 327                 goto err_up;
 328         }
 329
 330         rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
 331         if (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE) {
 332                 /* RMPP active */
 333                 if (!agent->rmpp_version) {
 334                         ret = -EINVAL;
 335                         goto err_ah;
 336                 }
 337
 338                 /* Validate that the management class can support RMPP */
 339                 if (rmpp_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_ADM) {
 340                         hdr_len = offsetof(struct ib_sa_mad, data);
 341                         data_len = length - hdr_len;
 342                 } else if ((rmpp_mad->mad_hdr.mgmt_class >= IB_MGMT_CLASS_VENDOR_RANGE2_START) &&
 343                             (rmpp_mad->mad_hdr.mgmt_class <= IB_MGMT_CLASS_VENDOR_RANGE2_END)) {
 344                                 hdr_len = offsetof(struct ib_vendor_mad, data);
 345                                 data_len = length - hdr_len;
 346                 } else {
 347                         ret = -EINVAL;
 348                         goto err_ah;
 349                 }
 350                 rmpp_active = 1;
 351         } else {
 352                 if (length > sizeof (struct ib_mad)) {
 353                         ret = -EINVAL;
 354                         goto err_ah;
 355                 }
 356                 hdr_len = offsetof(struct ib_mad, data);
 357                 data_len = length - hdr_len;
 358         }
 359
 360         packet->msg = ib_create_send_mad(agent,
 361                                          be32_to_cpu(packet->mad.hdr.qpn),
 362                                          0, packet->ah, rmpp_active,
 363                                          hdr_len, data_len,
 364                                          GFP_KERNEL);
 365         if (IS_ERR(packet->msg)) {
 366                 ret = PTR_ERR(packet->msg);
 367                 goto err_ah;
 368         }
 369
 370         packet->msg->send_wr.wr.ud.timeout_ms  = packet->mad.hdr.timeout_ms;
 371         packet->msg->send_wr.wr.ud.retries = packet->mad.hdr.retries;
 372
 373         /* Override send WR WRID initialized in ib_create_send_mad */
 374         packet->msg->send_wr.wr_id = (unsigned long) packet;
 375
 376         if (!rmpp_active) {
 377                 /* Copy message from user into send buffer */
 378                 if (copy_from_user(packet->msg->mad,
 379                                    buf + sizeof (struct ib_user_mad), length)) {
 380                         ret = -EFAULT;
 381                         goto err_msg;
 382                 }
 383         } else {
 384                 rmpp_hdr_size = sizeof (struct ib_mad_hdr) +
 385                                 sizeof (struct ib_rmpp_hdr);
 386
 387                 /* Only copy MAD headers (RMPP header in place) */
 388                 memcpy(packet->msg->mad, packet->mad.data,
 389                        sizeof (struct ib_mad_hdr));
 390
 391                 /* Now, copy rest of message from user into send buffer */
 392                 if (copy_from_user(((struct ib_rmpp_mad *) packet->msg->mad)->data,
 393                                    buf + sizeof (struct ib_user_mad) + rmpp_hdr_size,
 394                                    length - rmpp_hdr_size)) {
 395                         ret = -EFAULT;
 396                         goto err_msg;
 397                 }
 398         }
 399
 400         /*
 401          * If userspace is generating a request that will generate a
 402          * response, we need to make sure the high-order part of the
 403          * transaction ID matches the agent being used to send the
 404          * MAD.
 405          */
 406         method = packet->msg->mad->mad_hdr.method;
 407
 408         if (!(method & IB_MGMT_METHOD_RESP)       &&
 409             method != IB_MGMT_METHOD_TRAP_REPRESS &&
 410             method != IB_MGMT_METHOD_SEND) {
 411                 tid = &packet->msg->mad->mad_hdr.tid;
 412                 *tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
 413                                    (be64_to_cpup(tid) & 0xffffffff));
 414         }
 415
 416         ret = ib_post_send_mad(agent, &packet->msg->send_wr, &bad_wr);
 417         if (ret)
 418                 goto err_msg;
 419
 420         up_read(&file->agent_mutex);
 421
 422         return sizeof (struct ib_user_mad_hdr) + packet->length;
 423
 424 err_msg:
 425         ib_free_send_mad(packet->msg);
 426
 427 err_ah:
 428         ib_destroy_ah(packet->ah);
 429
 430 err_up:
 431         up_read(&file->agent_mutex);
 432
 433 err:
 434         kfree(packet);
 435         return ret;
 436 }
 437
 438 static unsigned int ib_umad_poll(struct file *filp, struct poll_table_struct *wait)
 439 {
 440         struct ib_umad_file *file = filp->private_data;
 441
 442         /* we will always be able to post a MAD send */
 443         unsigned int mask = POLLOUT | POLLWRNORM;
 444
 445         poll_wait(filp, &file->recv_wait, wait);
 446
 447         if (!list_empty(&file->recv_list))
 448                 mask |= POLLIN | POLLRDNORM;
 449
 450         return mask;
 451 }
 452
 453 static int ib_umad_reg_agent(struct ib_umad_file *file, unsigned long arg)
 454 {
 455         struct ib_user_mad_reg_req ureq;
 456         struct ib_mad_reg_req req;
 457         struct ib_mad_agent *agent;
 458         int agent_id;
 459         int ret;
 460
 461         down_write(&file->agent_mutex);
 462
 463         if (copy_from_user(&ureq, (void __user *) arg, sizeof ureq)) {
 464                 ret = -EFAULT;
 465                 goto out;
 466         }
 467
 468         if (ureq.qpn != 0 && ureq.qpn != 1) {
 469                 ret = -EINVAL;
 470                 goto out;
 471         }
 472
 473         for (agent_id = 0; agent_id < IB_UMAD_MAX_AGENTS; ++agent_id)
 474                 if (!file->agent[agent_id])
 475                         goto found;
 476
 477         ret = -ENOMEM;
 478         goto out;
 479
 480 found:
 481         if (ureq.mgmt_class) {
 482                 req.mgmt_class         = ureq.mgmt_class;
 483                 req.mgmt_class_version = ureq.mgmt_class_version;
 484                 memcpy(req.method_mask, ureq.method_mask, sizeof req.method_mask);
 485                 memcpy(req.oui,         ureq.oui,         sizeof req.oui);
 486         }
 487
 488         agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
 489                                       ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
 490                                       ureq.mgmt_class ? &req : NULL,
 491                                       ureq.rmpp_version,
 492                                       send_handler, recv_handler, file);
 493         if (IS_ERR(agent)) {
 494                 ret = PTR_ERR(agent);
 495                 goto out;
 496         }
 497
 498         file->agent[agent_id] = agent;
 499
 500         file->mr[agent_id] = ib_get_dma_mr(agent->qp->pd, IB_ACCESS_LOCAL_WRITE);
 501         if (IS_ERR(file->mr[agent_id])) {
 502                 ret = -ENOMEM;
 503                 goto err;
 504         }
 505
 506         if (put_user(agent_id,
 507                      (u32 __user *) (arg + offsetof(struct ib_user_mad_reg_req, id)))) {
 508                 ret = -EFAULT;
 509                 goto err_mr;
 510         }
 511
 512         ret = 0;
 513         goto out;
 514
 515 err_mr:
 516         ib_dereg_mr(file->mr[agent_id]);
 517
 518 err:
 519         file->agent[agent_id] = NULL;
 520         ib_unregister_mad_agent(agent);
 521
 522 out:
 523         up_write(&file->agent_mutex);
 524         return ret;
 525 }
 526
 527 static int ib_umad_unreg_agent(struct ib_umad_file *file, unsigned long arg)
 528 {
 529         u32 id;
 530         int ret = 0;
 531
 532         down_write(&file->agent_mutex);
 533
 534         if (get_user(id, (u32 __user *) arg)) {
 535                 ret = -EFAULT;
 536                 goto out;
 537         }
 538
 539         if (id < 0 || id >= IB_UMAD_MAX_AGENTS || !file->agent[id]) {
 540                 ret = -EINVAL;
 541                 goto out;
 542         }
 543
 544         ib_dereg_mr(file->mr[id]);
 545         ib_unregister_mad_agent(file->agent[id]);
 546         file->agent[id] = NULL;
 547
 548 out:
 549         up_write(&file->agent_mutex);
 550         return ret;
 551 }
 552
 553 static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
 554                           unsigned long arg)
 555 {
 556         switch (cmd) {
 557         case IB_USER_MAD_REGISTER_AGENT:
 558                 return ib_umad_reg_agent(filp->private_data, arg);
 559         case IB_USER_MAD_UNREGISTER_AGENT:
 560                 return ib_umad_unreg_agent(filp->private_data, arg);
 561         default:
 562                 return -ENOIOCTLCMD;
 563         }
 564 }
 565
 566 static int ib_umad_open(struct inode *inode, struct file *filp)
 567 {
 568         struct ib_umad_port *port =
 569                 container_of(inode->i_cdev, struct ib_umad_port, dev);
 570         struct ib_umad_file *file;
 571
 572         file = kmalloc(sizeof *file, GFP_KERNEL);
 573         if (!file)
 574                 return -ENOMEM;
 575
 576         memset(file, 0, sizeof *file);
 577
 578         spin_lock_init(&file->recv_lock);
 579         init_rwsem(&file->agent_mutex);
 580         INIT_LIST_HEAD(&file->recv_list);
 581         init_waitqueue_head(&file->recv_wait);
 582
 583         file->port = port;
 584         filp->private_data = file;
 585
 586         return 0;
 587 }
 588
 589 static int ib_umad_close(struct inode *inode, struct file *filp)
 590 {
 591         struct ib_umad_file *file = filp->private_data;
 592         struct ib_umad_packet *packet, *tmp;
 593         int i;
 594
 595         for (i = 0; i < IB_UMAD_MAX_AGENTS; ++i)
 596                 if (file->agent[i]) {
 597                         ib_dereg_mr(file->mr[i]);
 598                         ib_unregister_mad_agent(file->agent[i]);
 599                 }
 600
 601         list_for_each_entry_safe(packet, tmp, &file->recv_list, list)
 602                 kfree(packet);
 603
 604         kfree(file);
 605
 606         return 0;
 607 }
 608
 609 static struct file_operations umad_fops = {
 610         .owner          = THIS_MODULE,
 611         .read           = ib_umad_read,
 612         .write          = ib_umad_write,
 613         .poll           = ib_umad_poll,
 614         .unlocked_ioctl = ib_umad_ioctl,
 615         .compat_ioctl   = ib_umad_ioctl,
 616         .open           = ib_umad_open,
 617         .release        = ib_umad_close
 618 };
 619
 620 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
 621 {
 622         struct ib_umad_port *port =
 623                 container_of(inode->i_cdev, struct ib_umad_port, sm_dev);
 624         struct ib_port_modify props = {
 625                 .set_port_cap_mask = IB_PORT_SM
 626         };
 627         int ret;
 628
 629         if (filp->f_flags & O_NONBLOCK) {
 630                 if (down_trylock(&port->sm_sem))
 631                         return -EAGAIN;
 632         } else {
 633                 if (down_interruptible(&port->sm_sem))
 634                         return -ERESTARTSYS;
 635         }
 636
 637         ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
 638         if (ret) {
 639                 up(&port->sm_sem);
 640                 return ret;
 641         }
 642
 643         filp->private_data = port;
 644
 645         return 0;
 646 }
 647
 648 static int ib_umad_sm_close(struct inode *inode, struct file *filp)
 649 {
 650         struct ib_umad_port *port = filp->private_data;
 651         struct ib_port_modify props = {
 652                 .clr_port_cap_mask = IB_PORT_SM
 653         };
 654         int ret;
 655
 656         ret = ib_modify_port(port->ib_dev, port->port_num, 0, &props);
 657         up(&port->sm_sem);
 658
 659         return ret;
 660 }
 661
 662 static struct file_operations umad_sm_fops = {
 663         .owner   = THIS_MODULE,
 664         .open    = ib_umad_sm_open,
 665         .release = ib_umad_sm_close
 666 };
 667
 668 static struct ib_client umad_client = {
 669         .name   = "umad",
 670         .add    = ib_umad_add_one,
 671         .remove = ib_umad_remove_one
 672 };
 673
 674 static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
 675 {
 676         struct ib_umad_port *port = class_get_devdata(class_dev);
 677
 678         return sprintf(buf, "%s\n", port->ib_dev->name);
 679 }
 680 static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
 681
 682 static ssize_t show_port(struct class_device *class_dev, char *buf)
 683 {
 684         struct ib_umad_port *port = class_get_devdata(class_dev);
 685
 686         return sprintf(buf, "%d\n", port->port_num);
 687 }
 688 static CLASS_DEVICE_ATTR(port, S_IRUGO, show_port, NULL);
 689
 690 static void ib_umad_release_dev(struct kref *ref)
 691 {
 692         struct ib_umad_device *dev =
 693                 container_of(ref, struct ib_umad_device, ref);
 694
 695         kfree(dev);
 696 }
 697
 698 static void ib_umad_release_port(struct class_device *class_dev)
 699 {
 700         struct ib_umad_port *port = class_get_devdata(class_dev);
 701
 702         if (class_dev == &port->class_dev) {
 703                 cdev_del(&port->dev);
 704                 clear_bit(port->devnum, dev_map);
 705         } else {
 706                 cdev_del(&port->sm_dev);
 707                 clear_bit(port->sm_devnum, dev_map);
 708         }
 709
 710         kref_put(&port->umad_dev->ref, ib_umad_release_dev);
 711 }
 712
 713 static struct class umad_class = {
 714         .name    = "infiniband_mad",
 715         .release = ib_umad_release_port
 716 };
 717
 718 static ssize_t show_abi_version(struct class *class, char *buf)
 719 {
 720         return sprintf(buf, "%d\n", IB_USER_MAD_ABI_VERSION);
 721 }
 722 static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
 723
 724 static int ib_umad_init_port(struct ib_device *device, int port_num,
 725                              struct ib_umad_port *port)
 726 {
 727         spin_lock(&map_lock);
 728         port->devnum = find_first_zero_bit(dev_map, IB_UMAD_MAX_PORTS);
 729         if (port->devnum >= IB_UMAD_MAX_PORTS) {
 730                 spin_unlock(&map_lock);
 731                 return -1;
 732         }
 733         port->sm_devnum = find_next_zero_bit(dev_map, IB_UMAD_MAX_PORTS * 2, IB_UMAD_MAX_PORTS);
 734         if (port->sm_devnum >= IB_UMAD_MAX_PORTS * 2) {
 735                 spin_unlock(&map_lock);
 736                 return -1;
 737         }
 738         set_bit(port->devnum, dev_map);
 739         set_bit(port->sm_devnum, dev_map);
 740         spin_unlock(&map_lock);
 741
 742         port->ib_dev   = device;
 743         port->port_num = port_num;
 744         init_MUTEX(&port->sm_sem);
 745
 746         cdev_init(&port->dev, &umad_fops);
 747         port->dev.owner = THIS_MODULE;
 748         kobject_set_name(&port->dev.kobj, "umad%d", port->devnum);
 749         if (cdev_add(&port->dev, base_dev + port->devnum, 1))
 750                 return -1;
 751
 752         port->class_dev.class = &umad_class;
 753         port->class_dev.dev   = device->dma_device;
 754         port->class_dev.devt  = port->dev.dev;
 755
 756         snprintf(port->class_dev.class_id, BUS_ID_SIZE, "umad%d", port->devnum);
 757
 758         if (class_device_register(&port->class_dev))
 759                 goto err_cdev;
 760
 761         class_set_devdata(&port->class_dev, port);
 762         kref_get(&port->umad_dev->ref);
 763
 764         if (class_device_create_file(&port->class_dev, &class_device_attr_ibdev))
 765                 goto err_class;
 766         if (class_device_create_file(&port->class_dev, &class_device_attr_port))
 767                 goto err_class;
 768
 769         cdev_init(&port->sm_dev, &umad_sm_fops);
 770         port->sm_dev.owner = THIS_MODULE;
 771         kobject_set_name(&port->dev.kobj, "issm%d", port->sm_devnum - IB_UMAD_MAX_PORTS);
 772         if (cdev_add(&port->sm_dev, base_dev + port->sm_devnum, 1))
 773                 return -1;
 774
 775         port->sm_class_dev.class = &umad_class;
 776         port->sm_class_dev.dev   = device->dma_device;
 777         port->sm_class_dev.devt  = port->sm_dev.dev;
 778
 779         snprintf(port->sm_class_dev.class_id, BUS_ID_SIZE, "issm%d", port->sm_devnum - IB_UMAD_MAX_PORTS);
 780
 781         if (class_device_register(&port->sm_class_dev))
 782                 goto err_sm_cdev;
 783
 784         class_set_devdata(&port->sm_class_dev, port);
 785         kref_get(&port->umad_dev->ref);
 786
 787         if (class_device_create_file(&port->sm_class_dev, &class_device_attr_ibdev))
 788                 goto err_sm_class;
 789         if (class_device_create_file(&port->sm_class_dev, &class_device_attr_port))
 790                 goto err_sm_class;
 791
 792         return 0;
 793
 794 err_sm_class:
 795         class_device_unregister(&port->sm_class_dev);
 796
 797 err_sm_cdev:
 798         cdev_del(&port->sm_dev);
 799
 800 err_class:
 801         class_device_unregister(&port->class_dev);
 802
 803 err_cdev:
 804         cdev_del(&port->dev);
 805         clear_bit(port->devnum, dev_map);
 806
 807         return -1;
 808 }
 809
 810 static void ib_umad_add_one(struct ib_device *device)
 811 {
 812         struct ib_umad_device *umad_dev;
 813         int s, e, i;
 814
 815         if (device->node_type == IB_NODE_SWITCH)
 816                 s = e = 0;
 817         else {
 818                 s = 1;
 819                 e = device->phys_port_cnt;
 820         }
 821
 822         umad_dev = kmalloc(sizeof *umad_dev +
 823                            (e - s + 1) * sizeof (struct ib_umad_port),
 824                            GFP_KERNEL);
 825         if (!umad_dev)
 826                 return;
 827
 828         memset(umad_dev, 0, sizeof *umad_dev +
 829                (e - s + 1) * sizeof (struct ib_umad_port));
 830
 831         kref_init(&umad_dev->ref);
 832
 833         umad_dev->start_port = s;
 834         umad_dev->end_port   = e;
 835
 836         for (i = s; i <= e; ++i) {
 837                 umad_dev->port[i - s].umad_dev = umad_dev;
 838
 839                 if (ib_umad_init_port(device, i, &umad_dev->port[i - s]))
 840                         goto err;
 841         }
 842
 843         ib_set_client_data(device, &umad_client, umad_dev);
 844
 845         return;
 846
 847 err:
 848         while (--i >= s) {
 849                 class_device_unregister(&umad_dev->port[i - s].class_dev);
 850                 class_device_unregister(&umad_dev->port[i - s].sm_class_dev);
 851         }
 852
 853         kref_put(&umad_dev->ref, ib_umad_release_dev);
 854 }
 855
 856 static void ib_umad_remove_one(struct ib_device *device)
 857 {
 858         struct ib_umad_device *umad_dev = ib_get_client_data(device, &umad_client);
 859         int i;
 860
 861         if (!umad_dev)
 862                 return;
 863
 864         for (i = 0; i <= umad_dev->end_port - umad_dev->start_port; ++i) {
 865                 class_device_unregister(&umad_dev->port[i].class_dev);
 866                 class_device_unregister(&umad_dev->port[i].sm_class_dev);
 867         }
 868
 869         kref_put(&umad_dev->ref, ib_umad_release_dev);
 870 }
 871
 872 static int __init ib_umad_init(void)
 873 {
 874         int ret;
 875
 876         spin_lock_init(&map_lock);
 877
 878         ret = register_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2,
 879                                      "infiniband_mad");
 880         if (ret) {
 881                 printk(KERN_ERR "user_mad: couldn't register device number\n");
 882                 goto out;
 883         }
 884
 885         ret = class_register(&umad_class);
 886         if (ret) {
 887                 printk(KERN_ERR "user_mad: couldn't create class infiniband_mad\n");
 888                 goto out_chrdev;
 889         }
 890
 891         ret = class_create_file(&umad_class, &class_attr_abi_version);
 892         if (ret) {
 893                 printk(KERN_ERR "user_mad: couldn't create abi_version attribute\n");
 894                 goto out_class;
 895         }
 896
 897         ret = ib_register_client(&umad_client);
 898         if (ret) {
 899                 printk(KERN_ERR "user_mad: couldn't register ib_umad client\n");
 900                 goto out_class;
 901         }
 902
 903         return 0;
 904
 905 out_class:
 906         class_unregister(&umad_class);
 907
 908 out_chrdev:
 909         unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
 910
 911 out:
 912         return ret;
 913 }
 914
 915 static void __exit ib_umad_cleanup(void)
 916 {
 917         ib_unregister_client(&umad_client);
 918         class_unregister(&umad_class);
 919         unregister_chrdev_region(base_dev, IB_UMAD_MAX_PORTS * 2);
 920 }
 921
 922 module_init(ib_umad_init);
 923 module_exit(ib_umad_cleanup);