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 <rdma/ocrdma-abi.h>
  60
  61 MODULE_DESCRIPTION(OCRDMA_ROCE_DRV_DESC " " OCRDMA_ROCE_DRV_VERSION);
  62 MODULE_AUTHOR("Emulex Corporation");
  63 MODULE_LICENSE("Dual BSD/GPL");
  64
  65 void ocrdma_get_guid(struct ocrdma_dev *dev, u8 *guid)
  66 {
  67         u8 mac_addr[6];
  68
  69         memcpy(&mac_addr[0], &dev->nic_info.mac_addr[0], ETH_ALEN);
  70         guid[0] = mac_addr[0] ^ 2;
  71         guid[1] = mac_addr[1];
  72         guid[2] = mac_addr[2];
  73         guid[3] = 0xff;
  74         guid[4] = 0xfe;
  75         guid[5] = mac_addr[3];
  76         guid[6] = mac_addr[4];
  77         guid[7] = mac_addr[5];
  78 }
  79 static enum rdma_link_layer ocrdma_link_layer(struct ib_device *device,
  80                                               u32 port_num)
  81 {
  82         return IB_LINK_LAYER_ETHERNET;
  83 }
  84
  85 static int ocrdma_port_immutable(struct ib_device *ibdev, u32 port_num,
  86                                  struct ib_port_immutable *immutable)
  87 {
  88         struct ib_port_attr attr;
  89         struct ocrdma_dev *dev;
  90         int err;
  91
  92         dev = get_ocrdma_dev(ibdev);
  93         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
  94         if (ocrdma_is_udp_encap_supported(dev))
  95                 immutable->core_cap_flags |= RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
  96
  97         err = ib_query_port(ibdev, port_num, &attr);
  98         if (err)
  99                 return err;
 100
 101         immutable->pkey_tbl_len = attr.pkey_tbl_len;
 102         immutable->gid_tbl_len = attr.gid_tbl_len;
 103         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
 104
 105         return 0;
 106 }
 107
 108 static void get_dev_fw_str(struct ib_device *device, char *str)
 109 {
 110         struct ocrdma_dev *dev = get_ocrdma_dev(device);
 111
 112         snprintf(str, IB_FW_VERSION_NAME_MAX, "%s", &dev->attr.fw_ver[0]);
 113 }
 114
 115 /* OCRDMA sysfs interface */
 116 static ssize_t hw_rev_show(struct device *device,
 117                            struct device_attribute *attr, char *buf)
 118 {
 119         struct ocrdma_dev *dev =
 120                 rdma_device_to_drv_device(device, struct ocrdma_dev, ibdev);
 121
 122         return sysfs_emit(buf, "0x%x\n", dev->nic_info.pdev->vendor);
 123 }
 124 static DEVICE_ATTR_RO(hw_rev);
 125
 126 static ssize_t hca_type_show(struct device *device,
 127                              struct device_attribute *attr, char *buf)
 128 {
 129         struct ocrdma_dev *dev =
 130                 rdma_device_to_drv_device(device, struct ocrdma_dev, ibdev);
 131
 132         return sysfs_emit(buf, "%s\n", &dev->model_number[0]);
 133 }
 134 static DEVICE_ATTR_RO(hca_type);
 135
 136 static struct attribute *ocrdma_attributes[] = {
 137         &dev_attr_hw_rev.attr,
 138         &dev_attr_hca_type.attr,
 139         NULL
 140 };
 141
 142 static const struct attribute_group ocrdma_attr_group = {
 143         .attrs = ocrdma_attributes,
 144 };
 145
 146 static const struct ib_device_ops ocrdma_dev_ops = {
 147         .owner = THIS_MODULE,
 148         .driver_id = RDMA_DRIVER_OCRDMA,
 149         .uverbs_abi_ver = OCRDMA_ABI_VERSION,
 150
 151         .alloc_mr = ocrdma_alloc_mr,
 152         .alloc_pd = ocrdma_alloc_pd,
 153         .alloc_ucontext = ocrdma_alloc_ucontext,
 154         .create_ah = ocrdma_create_ah,
 155         .create_cq = ocrdma_create_cq,
 156         .create_qp = ocrdma_create_qp,
 157         .create_user_ah = ocrdma_create_ah,
 158         .dealloc_pd = ocrdma_dealloc_pd,
 159         .dealloc_ucontext = ocrdma_dealloc_ucontext,
 160         .dereg_mr = ocrdma_dereg_mr,
 161         .destroy_ah = ocrdma_destroy_ah,
 162         .destroy_cq = ocrdma_destroy_cq,
 163         .destroy_qp = ocrdma_destroy_qp,
 164         .device_group = &ocrdma_attr_group,
 165         .get_dev_fw_str = get_dev_fw_str,
 166         .get_dma_mr = ocrdma_get_dma_mr,
 167         .get_link_layer = ocrdma_link_layer,
 168         .get_port_immutable = ocrdma_port_immutable,
 169         .map_mr_sg = ocrdma_map_mr_sg,
 170         .mmap = ocrdma_mmap,
 171         .modify_qp = ocrdma_modify_qp,
 172         .poll_cq = ocrdma_poll_cq,
 173         .post_recv = ocrdma_post_recv,
 174         .post_send = ocrdma_post_send,
 175         .process_mad = ocrdma_process_mad,
 176         .query_ah = ocrdma_query_ah,
 177         .query_device = ocrdma_query_device,
 178         .query_pkey = ocrdma_query_pkey,
 179         .query_port = ocrdma_query_port,
 180         .query_qp = ocrdma_query_qp,
 181         .reg_user_mr = ocrdma_reg_user_mr,
 182         .req_notify_cq = ocrdma_arm_cq,
 183         .resize_cq = ocrdma_resize_cq,
 184
 185         INIT_RDMA_OBJ_SIZE(ib_ah, ocrdma_ah, ibah),
 186         INIT_RDMA_OBJ_SIZE(ib_cq, ocrdma_cq, ibcq),
 187         INIT_RDMA_OBJ_SIZE(ib_pd, ocrdma_pd, ibpd),
 188         INIT_RDMA_OBJ_SIZE(ib_ucontext, ocrdma_ucontext, ibucontext),
 189 };
 190
 191 static const struct ib_device_ops ocrdma_dev_srq_ops = {
 192         .create_srq = ocrdma_create_srq,
 193         .destroy_srq = ocrdma_destroy_srq,
 194         .modify_srq = ocrdma_modify_srq,
 195         .post_srq_recv = ocrdma_post_srq_recv,
 196         .query_srq = ocrdma_query_srq,
 197
 198         INIT_RDMA_OBJ_SIZE(ib_srq, ocrdma_srq, ibsrq),
 199 };
 200
 201 static int ocrdma_register_device(struct ocrdma_dev *dev)
 202 {
 203         int ret;
 204
 205         ocrdma_get_guid(dev, (u8 *)&dev->ibdev.node_guid);
 206         BUILD_BUG_ON(sizeof(OCRDMA_NODE_DESC) > IB_DEVICE_NODE_DESC_MAX);
 207         memcpy(dev->ibdev.node_desc, OCRDMA_NODE_DESC,
 208                sizeof(OCRDMA_NODE_DESC));
 209
 210         dev->ibdev.node_type = RDMA_NODE_IB_CA;
 211         dev->ibdev.phys_port_cnt = 1;
 212         dev->ibdev.num_comp_vectors = dev->eq_cnt;
 213
 214         /* mandatory to support user space verbs consumer. */
 215         dev->ibdev.dev.parent = &dev->nic_info.pdev->dev;
 216
 217         ib_set_device_ops(&dev->ibdev, &ocrdma_dev_ops);
 218
 219         if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R)
 220                 ib_set_device_ops(&dev->ibdev, &ocrdma_dev_srq_ops);
 221
 222         ret = ib_device_set_netdev(&dev->ibdev, dev->nic_info.netdev, 1);
 223         if (ret)
 224                 return ret;
 225
 226         dma_set_max_seg_size(&dev->nic_info.pdev->dev, UINT_MAX);
 227         return ib_register_device(&dev->ibdev, "ocrdma%d",
 228                                   &dev->nic_info.pdev->dev);
 229 }
 230
 231 static int ocrdma_alloc_resources(struct ocrdma_dev *dev)
 232 {
 233         mutex_init(&dev->dev_lock);
 234         dev->cq_tbl = kcalloc(OCRDMA_MAX_CQ, sizeof(struct ocrdma_cq *),
 235                               GFP_KERNEL);
 236         if (!dev->cq_tbl)
 237                 goto alloc_err;
 238
 239         if (dev->attr.max_qp) {
 240                 dev->qp_tbl = kcalloc(OCRDMA_MAX_QP,
 241                                       sizeof(struct ocrdma_qp *),
 242                                       GFP_KERNEL);
 243                 if (!dev->qp_tbl)
 244                         goto alloc_err;
 245         }
 246
 247         dev->stag_arr = kcalloc(OCRDMA_MAX_STAG, sizeof(u64), GFP_KERNEL);
 248         if (dev->stag_arr == NULL)
 249                 goto alloc_err;
 250
 251         ocrdma_alloc_pd_pool(dev);
 252
 253         if (!ocrdma_alloc_stats_resources(dev)) {
 254                 pr_err("%s: stats resource allocation failed\n", __func__);
 255                 goto alloc_err;
 256         }
 257
 258         spin_lock_init(&dev->av_tbl.lock);
 259         spin_lock_init(&dev->flush_q_lock);
 260         return 0;
 261 alloc_err:
 262         pr_err("%s(%d) error.\n", __func__, dev->id);
 263         return -ENOMEM;
 264 }
 265
 266 static void ocrdma_free_resources(struct ocrdma_dev *dev)
 267 {
 268         ocrdma_release_stats_resources(dev);
 269         kfree(dev->stag_arr);
 270         kfree(dev->qp_tbl);
 271         kfree(dev->cq_tbl);
 272 }
 273
 274 static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info)
 275 {
 276         int status = 0;
 277         u8 lstate = 0;
 278         struct ocrdma_dev *dev;
 279
 280         dev = ib_alloc_device(ocrdma_dev, ibdev);
 281         if (!dev) {
 282                 pr_err("Unable to allocate ib device\n");
 283                 return NULL;
 284         }
 285
 286         dev->mbx_cmd = kzalloc(sizeof(struct ocrdma_mqe_emb_cmd), GFP_KERNEL);
 287         if (!dev->mbx_cmd)
 288                 goto init_err;
 289
 290         memcpy(&dev->nic_info, dev_info, sizeof(*dev_info));
 291         dev->id = PCI_FUNC(dev->nic_info.pdev->devfn);
 292         status = ocrdma_init_hw(dev);
 293         if (status)
 294                 goto init_err;
 295
 296         status = ocrdma_alloc_resources(dev);
 297         if (status)
 298                 goto alloc_err;
 299
 300         ocrdma_init_service_level(dev);
 301         status = ocrdma_register_device(dev);
 302         if (status)
 303                 goto alloc_err;
 304
 305         /* Query Link state and update */
 306         status = ocrdma_mbx_get_link_speed(dev, NULL, &lstate);
 307         if (!status)
 308                 ocrdma_update_link_state(dev, lstate);
 309
 310         /* Init stats */
 311         ocrdma_add_port_stats(dev);
 312         /* Interrupt Moderation */
 313         INIT_DELAYED_WORK(&dev->eqd_work, ocrdma_eqd_set_task);
 314         schedule_delayed_work(&dev->eqd_work, msecs_to_jiffies(1000));
 315
 316         pr_info("%s %s: %s \"%s\" port %d\n",
 317                 dev_name(&dev->nic_info.pdev->dev), hca_name(dev),
 318                 port_speed_string(dev), dev->model_number,
 319                 dev->hba_port_num);
 320         pr_info("%s ocrdma%d driver loaded successfully\n",
 321                 dev_name(&dev->nic_info.pdev->dev), dev->id);
 322         return dev;
 323
 324 alloc_err:
 325         ocrdma_free_resources(dev);
 326         ocrdma_cleanup_hw(dev);
 327 init_err:
 328         kfree(dev->mbx_cmd);
 329         ib_dealloc_device(&dev->ibdev);
 330         pr_err("%s() leaving. ret=%d\n", __func__, status);
 331         return NULL;
 332 }
 333
 334 static void ocrdma_remove_free(struct ocrdma_dev *dev)
 335 {
 336
 337         kfree(dev->mbx_cmd);
 338         ib_dealloc_device(&dev->ibdev);
 339 }
 340
 341 static void ocrdma_remove(struct ocrdma_dev *dev)
 342 {
 343         /* first unregister with stack to stop all the active traffic
 344          * of the registered clients.
 345          */
 346         cancel_delayed_work_sync(&dev->eqd_work);
 347         ib_unregister_device(&dev->ibdev);
 348
 349         ocrdma_rem_port_stats(dev);
 350         ocrdma_free_resources(dev);
 351         ocrdma_cleanup_hw(dev);
 352         ocrdma_remove_free(dev);
 353 }
 354
 355 static int ocrdma_dispatch_port_active(struct ocrdma_dev *dev)
 356 {
 357         struct ib_event port_event;
 358
 359         port_event.event = IB_EVENT_PORT_ACTIVE;
 360         port_event.element.port_num = 1;
 361         port_event.device = &dev->ibdev;
 362         ib_dispatch_event(&port_event);
 363         return 0;
 364 }
 365
 366 static int ocrdma_dispatch_port_error(struct ocrdma_dev *dev)
 367 {
 368         struct ib_event err_event;
 369
 370         err_event.event = IB_EVENT_PORT_ERR;
 371         err_event.element.port_num = 1;
 372         err_event.device = &dev->ibdev;
 373         ib_dispatch_event(&err_event);
 374         return 0;
 375 }
 376
 377 static void ocrdma_shutdown(struct ocrdma_dev *dev)
 378 {
 379         ocrdma_dispatch_port_error(dev);
 380         ocrdma_remove(dev);
 381 }
 382
 383 /* event handling via NIC driver ensures that all the NIC specific
 384  * initialization done before RoCE driver notifies
 385  * event to stack.
 386  */
 387 static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event)
 388 {
 389         switch (event) {
 390         case BE_DEV_SHUTDOWN:
 391                 ocrdma_shutdown(dev);
 392                 break;
 393         default:
 394                 break;
 395         }
 396 }
 397
 398 void ocrdma_update_link_state(struct ocrdma_dev *dev, u8 lstate)
 399 {
 400         if (!(dev->flags & OCRDMA_FLAGS_LINK_STATUS_INIT)) {
 401                 dev->flags |= OCRDMA_FLAGS_LINK_STATUS_INIT;
 402                 if (!lstate)
 403                         return;
 404         }
 405
 406         if (!lstate)
 407                 ocrdma_dispatch_port_error(dev);
 408         else
 409                 ocrdma_dispatch_port_active(dev);
 410 }
 411
 412 static struct ocrdma_driver ocrdma_drv = {
 413         .name                   = "ocrdma_driver",
 414         .add                    = ocrdma_add,
 415         .remove                 = ocrdma_remove,
 416         .state_change_handler   = ocrdma_event_handler,
 417         .be_abi_version         = OCRDMA_BE_ROCE_ABI_VERSION,
 418 };
 419
 420 static int __init ocrdma_init_module(void)
 421 {
 422         int status;
 423
 424         ocrdma_init_debugfs();
 425
 426         status = be_roce_register_driver(&ocrdma_drv);
 427         if (status)
 428                 goto err_be_reg;
 429
 430         return 0;
 431
 432 err_be_reg:
 433
 434         return status;
 435 }
 436
 437 static void __exit ocrdma_exit_module(void)
 438 {
 439         be_roce_unregister_driver(&ocrdma_drv);
 440         ocrdma_rem_debugfs();
 441 }
 442
 443 module_init(ocrdma_init_module);
 444 module_exit(ocrdma_exit_module);