drivers/infiniband/hw/ocrdma/ocrdma_main.c

   1 /* This file is part of the Emulex RoCE Device Driver for
   2  * RoCE (RDMA over Converged Ethernet) adapters.
   3  * Copyright (C) 2012-2015 Emulex. All rights reserved.
   4  * EMULEX and SLI are trademarks of Emulex.
   5  * www.emulex.com
   6  *
   7  * This software is available to you under a choice of one of two licenses.
   8  * You may choose to be licensed under the terms of the GNU General Public
   9  * License (GPL) Version 2, available from the file COPYING in the main
  10  * directory of this source tree, or the BSD license below:
  11  *
  12  * Redistribution and use in source and binary forms, with or without
  13  * modification, are permitted provided that the following conditions
  14  * are met:
  15  *
  16  * - Redistributions of source code must retain the above copyright notice,
  17  *   this list of conditions and the following disclaimer.
  18  *
  19  * - Redistributions in binary form must reproduce the above copyright
  20  *   notice, this list of conditions and the following disclaimer in
  21  *   the documentation and/or other materials provided with the distribution.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  24  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,THE
  25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  26  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
  27  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  30  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
  31  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
  32  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
  33  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  34  *
  35  * Contact Information:
  36  * linux-drivers@emulex.com
  37  *
  38  * Emulex
  39  * 3333 Susan Street
  40  * Costa Mesa, CA 92626
  41  */
  42
  43 #include <linux/module.h>
  44 #include <linux/idr.h>
  45 #include <rdma/ib_verbs.h>
  46 #include <rdma/ib_user_verbs.h>
  47 #include <rdma/ib_addr.h>
  48 #include <rdma/ib_mad.h>
  49
  50 #include <linux/netdevice.h>
  51 #include <net/addrconf.h>
  52
  53 #include "ocrdma.h"
  54 #include "ocrdma_verbs.h"
  55 #include "ocrdma_ah.h"
  56 #include "be_roce.h"
  57 #include "ocrdma_hw.h"
  58 #include "ocrdma_stats.h"
  59 #include "ocrdma_abi.h"
  60
  61 MODULE_VERSION(OCRDMA_ROCE_DRV_VERSION);
  62 MODULE_DESCRIPTION(OCRDMA_ROCE_DRV_DESC " " OCRDMA_ROCE_DRV_VERSION);
  63 MODULE_AUTHOR("Emulex Corporation");
  64 MODULE_LICENSE("Dual BSD/GPL");
  65
  66 static DEFINE_IDR(ocrdma_dev_id);
  67
  68 void ocrdma_get_guid(struct ocrdma_dev *dev, u8 *guid)
  69 {
  70         u8 mac_addr[6];
  71
  72         memcpy(&mac_addr[0], &dev->nic_info.mac_addr[0], ETH_ALEN);
  73         guid[0] = mac_addr[0] ^ 2;
  74         guid[1] = mac_addr[1];
  75         guid[2] = mac_addr[2];
  76         guid[3] = 0xff;
  77         guid[4] = 0xfe;
  78         guid[5] = mac_addr[3];
  79         guid[6] = mac_addr[4];
  80         guid[7] = mac_addr[5];
  81 }
  82 static enum rdma_link_layer ocrdma_link_layer(struct ib_device *device,
  83                                               u8 port_num)
  84 {
  85         return IB_LINK_LAYER_ETHERNET;
  86 }
  87
  88 static int ocrdma_port_immutable(struct ib_device *ibdev, u8 port_num,
  89                                  struct ib_port_immutable *immutable)
  90 {
  91         struct ib_port_attr attr;
  92         int err;
  93
  94         err = ocrdma_query_port(ibdev, port_num, &attr);
  95         if (err)
  96                 return err;
  97
  98         immutable->pkey_tbl_len = attr.pkey_tbl_len;
  99         immutable->gid_tbl_len = attr.gid_tbl_len;
 100         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
 101         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
 102
 103         return 0;
 104 }
 105
 106 static int ocrdma_register_device(struct ocrdma_dev *dev)
 107 {
 108         strlcpy(dev->ibdev.name, "ocrdma%d", IB_DEVICE_NAME_MAX);
 109         ocrdma_get_guid(dev, (u8 *)&dev->ibdev.node_guid);
 110         memcpy(dev->ibdev.node_desc, OCRDMA_NODE_DESC,
 111                sizeof(OCRDMA_NODE_DESC));
 112         dev->ibdev.owner = THIS_MODULE;
 113         dev->ibdev.uverbs_abi_ver = OCRDMA_ABI_VERSION;
 114         dev->ibdev.uverbs_cmd_mask =
 115             OCRDMA_UVERBS(GET_CONTEXT) |
 116             OCRDMA_UVERBS(QUERY_DEVICE) |
 117             OCRDMA_UVERBS(QUERY_PORT) |
 118             OCRDMA_UVERBS(ALLOC_PD) |
 119             OCRDMA_UVERBS(DEALLOC_PD) |
 120             OCRDMA_UVERBS(REG_MR) |
 121             OCRDMA_UVERBS(DEREG_MR) |
 122             OCRDMA_UVERBS(CREATE_COMP_CHANNEL) |
 123             OCRDMA_UVERBS(CREATE_CQ) |
 124             OCRDMA_UVERBS(RESIZE_CQ) |
 125             OCRDMA_UVERBS(DESTROY_CQ) |
 126             OCRDMA_UVERBS(REQ_NOTIFY_CQ) |
 127             OCRDMA_UVERBS(CREATE_QP) |
 128             OCRDMA_UVERBS(MODIFY_QP) |
 129             OCRDMA_UVERBS(QUERY_QP) |
 130             OCRDMA_UVERBS(DESTROY_QP) |
 131             OCRDMA_UVERBS(POLL_CQ) |
 132             OCRDMA_UVERBS(POST_SEND) |
 133             OCRDMA_UVERBS(POST_RECV);
 134
 135         dev->ibdev.uverbs_cmd_mask |=
 136             OCRDMA_UVERBS(CREATE_AH) |
 137              OCRDMA_UVERBS(MODIFY_AH) |
 138              OCRDMA_UVERBS(QUERY_AH) |
 139              OCRDMA_UVERBS(DESTROY_AH);
 140
 141         dev->ibdev.node_type = RDMA_NODE_IB_CA;
 142         dev->ibdev.phys_port_cnt = 1;
 143         dev->ibdev.num_comp_vectors = dev->eq_cnt;
 144
 145         /* mandatory verbs. */
 146         dev->ibdev.query_device = ocrdma_query_device;
 147         dev->ibdev.query_port = ocrdma_query_port;
 148         dev->ibdev.modify_port = ocrdma_modify_port;
 149         dev->ibdev.query_gid = ocrdma_query_gid;
 150         dev->ibdev.get_netdev = ocrdma_get_netdev;
 151         dev->ibdev.add_gid = ocrdma_add_gid;
 152         dev->ibdev.del_gid = ocrdma_del_gid;
 153         dev->ibdev.get_link_layer = ocrdma_link_layer;
 154         dev->ibdev.alloc_pd = ocrdma_alloc_pd;
 155         dev->ibdev.dealloc_pd = ocrdma_dealloc_pd;
 156
 157         dev->ibdev.create_cq = ocrdma_create_cq;
 158         dev->ibdev.destroy_cq = ocrdma_destroy_cq;
 159         dev->ibdev.resize_cq = ocrdma_resize_cq;
 160
 161         dev->ibdev.create_qp = ocrdma_create_qp;
 162         dev->ibdev.modify_qp = ocrdma_modify_qp;
 163         dev->ibdev.query_qp = ocrdma_query_qp;
 164         dev->ibdev.destroy_qp = ocrdma_destroy_qp;
 165
 166         dev->ibdev.query_pkey = ocrdma_query_pkey;
 167         dev->ibdev.create_ah = ocrdma_create_ah;
 168         dev->ibdev.destroy_ah = ocrdma_destroy_ah;
 169         dev->ibdev.query_ah = ocrdma_query_ah;
 170         dev->ibdev.modify_ah = ocrdma_modify_ah;
 171
 172         dev->ibdev.poll_cq = ocrdma_poll_cq;
 173         dev->ibdev.post_send = ocrdma_post_send;
 174         dev->ibdev.post_recv = ocrdma_post_recv;
 175         dev->ibdev.req_notify_cq = ocrdma_arm_cq;
 176
 177         dev->ibdev.get_dma_mr = ocrdma_get_dma_mr;
 178         dev->ibdev.reg_phys_mr = ocrdma_reg_kernel_mr;
 179         dev->ibdev.dereg_mr = ocrdma_dereg_mr;
 180         dev->ibdev.reg_user_mr = ocrdma_reg_user_mr;
 181
 182         dev->ibdev.alloc_mr = ocrdma_alloc_mr;
 183         dev->ibdev.map_mr_sg = ocrdma_map_mr_sg;
 184
 185         /* mandatory to support user space verbs consumer. */
 186         dev->ibdev.alloc_ucontext = ocrdma_alloc_ucontext;
 187         dev->ibdev.dealloc_ucontext = ocrdma_dealloc_ucontext;
 188         dev->ibdev.mmap = ocrdma_mmap;
 189         dev->ibdev.dma_device = &dev->nic_info.pdev->dev;
 190
 191         dev->ibdev.process_mad = ocrdma_process_mad;
 192         dev->ibdev.get_port_immutable = ocrdma_port_immutable;
 193
 194         if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
 195                 dev->ibdev.uverbs_cmd_mask |=
 196                      OCRDMA_UVERBS(CREATE_SRQ) |
 197                      OCRDMA_UVERBS(MODIFY_SRQ) |
 198                      OCRDMA_UVERBS(QUERY_SRQ) |
 199                      OCRDMA_UVERBS(DESTROY_SRQ) |
 200                      OCRDMA_UVERBS(POST_SRQ_RECV);
 201
 202                 dev->ibdev.create_srq = ocrdma_create_srq;
 203                 dev->ibdev.modify_srq = ocrdma_modify_srq;
 204                 dev->ibdev.query_srq = ocrdma_query_srq;
 205                 dev->ibdev.destroy_srq = ocrdma_destroy_srq;
 206                 dev->ibdev.post_srq_recv = ocrdma_post_srq_recv;
 207         }
 208         return ib_register_device(&dev->ibdev, NULL);
 209 }
 210
 211 static int ocrdma_alloc_resources(struct ocrdma_dev *dev)
 212 {
 213         mutex_init(&dev->dev_lock);
 214         dev->cq_tbl = kzalloc(sizeof(struct ocrdma_cq *) *
 215                               OCRDMA_MAX_CQ, GFP_KERNEL);
 216         if (!dev->cq_tbl)
 217                 goto alloc_err;
 218
 219         if (dev->attr.max_qp) {
 220                 dev->qp_tbl = kzalloc(sizeof(struct ocrdma_qp *) *
 221                                       OCRDMA_MAX_QP, GFP_KERNEL);
 222                 if (!dev->qp_tbl)
 223                         goto alloc_err;
 224         }
 225
 226         dev->stag_arr = kzalloc(sizeof(u64) * OCRDMA_MAX_STAG, GFP_KERNEL);
 227         if (dev->stag_arr == NULL)
 228                 goto alloc_err;
 229
 230         ocrdma_alloc_pd_pool(dev);
 231
 232         spin_lock_init(&dev->av_tbl.lock);
 233         spin_lock_init(&dev->flush_q_lock);
 234         return 0;
 235 alloc_err:
 236         pr_err("%s(%d) error.\n", __func__, dev->id);
 237         return -ENOMEM;
 238 }
 239
 240 static void ocrdma_free_resources(struct ocrdma_dev *dev)
 241 {
 242         kfree(dev->stag_arr);
 243         kfree(dev->qp_tbl);
 244         kfree(dev->cq_tbl);
 245 }
 246
 247 /* OCRDMA sysfs interface */
 248 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
 249                         char *buf)
 250 {
 251         struct ocrdma_dev *dev = dev_get_drvdata(device);
 252
 253         return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->nic_info.pdev->vendor);
 254 }
 255
 256 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
 257                         char *buf)
 258 {
 259         struct ocrdma_dev *dev = dev_get_drvdata(device);
 260
 261         return scnprintf(buf, PAGE_SIZE, "%s\n", &dev->attr.fw_ver[0]);
 262 }
 263
 264 static ssize_t show_hca_type(struct device *device,
 265                              struct device_attribute *attr, char *buf)
 266 {
 267         struct ocrdma_dev *dev = dev_get_drvdata(device);
 268
 269         return scnprintf(buf, PAGE_SIZE, "%s\n", &dev->model_number[0]);
 270 }
 271
 272 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
 273 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
 274 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
 275
 276 static struct device_attribute *ocrdma_attributes[] = {
 277         &dev_attr_hw_rev,
 278         &dev_attr_fw_ver,
 279         &dev_attr_hca_type
 280 };
 281
 282 static void ocrdma_remove_sysfiles(struct ocrdma_dev *dev)
 283 {
 284         int i;
 285
 286         for (i = 0; i < ARRAY_SIZE(ocrdma_attributes); i++)
 287                 device_remove_file(&dev->ibdev.dev, ocrdma_attributes[i]);
 288 }
 289
 290 static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info)
 291 {
 292         int status = 0, i;
 293         struct ocrdma_dev *dev;
 294
 295         dev = (struct ocrdma_dev *)ib_alloc_device(sizeof(struct ocrdma_dev));
 296         if (!dev) {
 297                 pr_err("Unable to allocate ib device\n");
 298                 return NULL;
 299         }
 300         dev->mbx_cmd = kzalloc(sizeof(struct ocrdma_mqe_emb_cmd), GFP_KERNEL);
 301         if (!dev->mbx_cmd)
 302                 goto idr_err;
 303
 304         memcpy(&dev->nic_info, dev_info, sizeof(*dev_info));
 305         dev->id = idr_alloc(&ocrdma_dev_id, NULL, 0, 0, GFP_KERNEL);
 306         if (dev->id < 0)
 307                 goto idr_err;
 308
 309         status = ocrdma_init_hw(dev);
 310         if (status)
 311                 goto init_err;
 312
 313         status = ocrdma_alloc_resources(dev);
 314         if (status)
 315                 goto alloc_err;
 316
 317         ocrdma_init_service_level(dev);
 318         status = ocrdma_register_device(dev);
 319         if (status)
 320                 goto alloc_err;
 321
 322         for (i = 0; i < ARRAY_SIZE(ocrdma_attributes); i++)
 323                 if (device_create_file(&dev->ibdev.dev, ocrdma_attributes[i]))
 324                         goto sysfs_err;
 325         /* Init stats */
 326         ocrdma_add_port_stats(dev);
 327         /* Interrupt Moderation */
 328         INIT_DELAYED_WORK(&dev->eqd_work, ocrdma_eqd_set_task);
 329         schedule_delayed_work(&dev->eqd_work, msecs_to_jiffies(1000));
 330
 331         pr_info("%s %s: %s \"%s\" port %d\n",
 332                 dev_name(&dev->nic_info.pdev->dev), hca_name(dev),
 333                 port_speed_string(dev), dev->model_number,
 334                 dev->hba_port_num);
 335         pr_info("%s ocrdma%d driver loaded successfully\n",
 336                 dev_name(&dev->nic_info.pdev->dev), dev->id);
 337         return dev;
 338
 339 sysfs_err:
 340         ocrdma_remove_sysfiles(dev);
 341 alloc_err:
 342         ocrdma_free_resources(dev);
 343         ocrdma_cleanup_hw(dev);
 344 init_err:
 345         idr_remove(&ocrdma_dev_id, dev->id);
 346 idr_err:
 347         kfree(dev->mbx_cmd);
 348         ib_dealloc_device(&dev->ibdev);
 349         pr_err("%s() leaving. ret=%d\n", __func__, status);
 350         return NULL;
 351 }
 352
 353 static void ocrdma_remove_free(struct ocrdma_dev *dev)
 354 {
 355
 356         idr_remove(&ocrdma_dev_id, dev->id);
 357         kfree(dev->mbx_cmd);
 358         ib_dealloc_device(&dev->ibdev);
 359 }
 360
 361 static void ocrdma_remove(struct ocrdma_dev *dev)
 362 {
 363         /* first unregister with stack to stop all the active traffic
 364          * of the registered clients.
 365          */
 366         cancel_delayed_work_sync(&dev->eqd_work);
 367         ocrdma_remove_sysfiles(dev);
 368         ib_unregister_device(&dev->ibdev);
 369
 370         ocrdma_rem_port_stats(dev);
 371         ocrdma_free_resources(dev);
 372         ocrdma_cleanup_hw(dev);
 373         ocrdma_remove_free(dev);
 374 }
 375
 376 static int ocrdma_open(struct ocrdma_dev *dev)
 377 {
 378         struct ib_event port_event;
 379
 380         port_event.event = IB_EVENT_PORT_ACTIVE;
 381         port_event.element.port_num = 1;
 382         port_event.device = &dev->ibdev;
 383         ib_dispatch_event(&port_event);
 384         return 0;
 385 }
 386
 387 static int ocrdma_close(struct ocrdma_dev *dev)
 388 {
 389         int i;
 390         struct ocrdma_qp *qp, **cur_qp;
 391         struct ib_event err_event;
 392         struct ib_qp_attr attrs;
 393         int attr_mask = IB_QP_STATE;
 394
 395         attrs.qp_state = IB_QPS_ERR;
 396         mutex_lock(&dev->dev_lock);
 397         if (dev->qp_tbl) {
 398                 cur_qp = dev->qp_tbl;
 399                 for (i = 0; i < OCRDMA_MAX_QP; i++) {
 400                         qp = cur_qp[i];
 401                         if (qp && qp->ibqp.qp_type != IB_QPT_GSI) {
 402                                 /* change the QP state to ERROR */
 403                                 _ocrdma_modify_qp(&qp->ibqp, &attrs, attr_mask);
 404
 405                                 err_event.event = IB_EVENT_QP_FATAL;
 406                                 err_event.element.qp = &qp->ibqp;
 407                                 err_event.device = &dev->ibdev;
 408                                 ib_dispatch_event(&err_event);
 409                         }
 410                 }
 411         }
 412         mutex_unlock(&dev->dev_lock);
 413
 414         err_event.event = IB_EVENT_PORT_ERR;
 415         err_event.element.port_num = 1;
 416         err_event.device = &dev->ibdev;
 417         ib_dispatch_event(&err_event);
 418         return 0;
 419 }
 420
 421 static void ocrdma_shutdown(struct ocrdma_dev *dev)
 422 {
 423         ocrdma_close(dev);
 424         ocrdma_remove(dev);
 425 }
 426
 427 /* event handling via NIC driver ensures that all the NIC specific
 428  * initialization done before RoCE driver notifies
 429  * event to stack.
 430  */
 431 static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event)
 432 {
 433         switch (event) {
 434         case BE_DEV_UP:
 435                 ocrdma_open(dev);
 436                 break;
 437         case BE_DEV_DOWN:
 438                 ocrdma_close(dev);
 439                 break;
 440         case BE_DEV_SHUTDOWN:
 441                 ocrdma_shutdown(dev);
 442                 break;
 443         }
 444 }
 445
 446 static struct ocrdma_driver ocrdma_drv = {
 447         .name                   = "ocrdma_driver",
 448         .add                    = ocrdma_add,
 449         .remove                 = ocrdma_remove,
 450         .state_change_handler   = ocrdma_event_handler,
 451         .be_abi_version         = OCRDMA_BE_ROCE_ABI_VERSION,
 452 };
 453
 454 static int __init ocrdma_init_module(void)
 455 {
 456         int status;
 457
 458         ocrdma_init_debugfs();
 459
 460         status = be_roce_register_driver(&ocrdma_drv);
 461         if (status)
 462                 goto err_be_reg;
 463
 464         return 0;
 465
 466 err_be_reg:
 467
 468         return status;
 469 }
 470
 471 static void __exit ocrdma_exit_module(void)
 472 {
 473         be_roce_unregister_driver(&ocrdma_drv);
 474         ocrdma_rem_debugfs();
 475         idr_destroy(&ocrdma_dev_id);
 476 }
 477
 478 module_init(ocrdma_init_module);
 479 module_exit(ocrdma_exit_module);