linux/drivers/infiniband/hw/qedr/main.c
<<
>>
Prefs
   1/* QLogic qedr NIC Driver
   2 * Copyright (c) 2015-2016  QLogic Corporation
   3 *
   4 * This software is available to you under a choice of one of two
   5 * licenses.  You may choose to be licensed under the terms of the GNU
   6 * General Public License (GPL) Version 2, available from the file
   7 * COPYING in the main directory of this source tree, or the
   8 * OpenIB.org BSD license below:
   9 *
  10 *     Redistribution and use in source and binary forms, with or
  11 *     without modification, are permitted provided that the following
  12 *     conditions are met:
  13 *
  14 *      - Redistributions of source code must retain the above
  15 *        copyright notice, this list of conditions and the following
  16 *        disclaimer.
  17 *
  18 *      - Redistributions in binary form must reproduce the above
  19 *        copyright notice, this list of conditions and the following
  20 *        disclaimer in the documentation and /or other materials
  21 *        provided with the distribution.
  22 *
  23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30 * SOFTWARE.
  31 */
  32#include <linux/module.h>
  33#include <rdma/ib_verbs.h>
  34#include <rdma/ib_addr.h>
  35#include <rdma/ib_user_verbs.h>
  36#include <rdma/iw_cm.h>
  37#include <rdma/ib_mad.h>
  38#include <linux/netdevice.h>
  39#include <linux/iommu.h>
  40#include <linux/pci.h>
  41#include <net/addrconf.h>
  42
  43#include <linux/qed/qed_chain.h>
  44#include <linux/qed/qed_if.h>
  45#include "qedr.h"
  46#include "verbs.h"
  47#include <rdma/qedr-abi.h>
  48#include "qedr_iw_cm.h"
  49
  50MODULE_DESCRIPTION("QLogic 40G/100G ROCE Driver");
  51MODULE_AUTHOR("QLogic Corporation");
  52MODULE_LICENSE("Dual BSD/GPL");
  53
  54#define QEDR_WQ_MULTIPLIER_DFT  (3)
  55
  56static void qedr_ib_dispatch_event(struct qedr_dev *dev, u32 port_num,
  57                                   enum ib_event_type type)
  58{
  59        struct ib_event ibev;
  60
  61        ibev.device = &dev->ibdev;
  62        ibev.element.port_num = port_num;
  63        ibev.event = type;
  64
  65        ib_dispatch_event(&ibev);
  66}
  67
  68static enum rdma_link_layer qedr_link_layer(struct ib_device *device,
  69                                            u32 port_num)
  70{
  71        return IB_LINK_LAYER_ETHERNET;
  72}
  73
  74static void qedr_get_dev_fw_str(struct ib_device *ibdev, char *str)
  75{
  76        struct qedr_dev *qedr = get_qedr_dev(ibdev);
  77        u32 fw_ver = (u32)qedr->attr.fw_ver;
  78
  79        snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d.%d",
  80                 (fw_ver >> 24) & 0xFF, (fw_ver >> 16) & 0xFF,
  81                 (fw_ver >> 8) & 0xFF, fw_ver & 0xFF);
  82}
  83
  84static int qedr_roce_port_immutable(struct ib_device *ibdev, u32 port_num,
  85                                    struct ib_port_immutable *immutable)
  86{
  87        struct ib_port_attr attr;
  88        int err;
  89
  90        err = qedr_query_port(ibdev, port_num, &attr);
  91        if (err)
  92                return err;
  93
  94        immutable->pkey_tbl_len = attr.pkey_tbl_len;
  95        immutable->gid_tbl_len = attr.gid_tbl_len;
  96        immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
  97            RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
  98        immutable->max_mad_size = IB_MGMT_MAD_SIZE;
  99
 100        return 0;
 101}
 102
 103static int qedr_iw_port_immutable(struct ib_device *ibdev, u32 port_num,
 104                                  struct ib_port_immutable *immutable)
 105{
 106        struct ib_port_attr attr;
 107        int err;
 108
 109        err = qedr_query_port(ibdev, port_num, &attr);
 110        if (err)
 111                return err;
 112
 113        immutable->gid_tbl_len = 1;
 114        immutable->core_cap_flags = RDMA_CORE_PORT_IWARP;
 115        immutable->max_mad_size = 0;
 116
 117        return 0;
 118}
 119
 120/* QEDR sysfs interface */
 121static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
 122                           char *buf)
 123{
 124        struct qedr_dev *dev =
 125                rdma_device_to_drv_device(device, struct qedr_dev, ibdev);
 126
 127        return sysfs_emit(buf, "0x%x\n", dev->attr.hw_ver);
 128}
 129static DEVICE_ATTR_RO(hw_rev);
 130
 131static ssize_t hca_type_show(struct device *device,
 132                             struct device_attribute *attr, char *buf)
 133{
 134        struct qedr_dev *dev =
 135                rdma_device_to_drv_device(device, struct qedr_dev, ibdev);
 136
 137        return sysfs_emit(buf, "FastLinQ QL%x %s\n", dev->pdev->device,
 138                          rdma_protocol_iwarp(&dev->ibdev, 1) ? "iWARP" :
 139                                                                "RoCE");
 140}
 141static DEVICE_ATTR_RO(hca_type);
 142
 143static struct attribute *qedr_attributes[] = {
 144        &dev_attr_hw_rev.attr,
 145        &dev_attr_hca_type.attr,
 146        NULL
 147};
 148
 149static const struct attribute_group qedr_attr_group = {
 150        .attrs = qedr_attributes,
 151};
 152
 153static const struct ib_device_ops qedr_iw_dev_ops = {
 154        .get_port_immutable = qedr_iw_port_immutable,
 155        .iw_accept = qedr_iw_accept,
 156        .iw_add_ref = qedr_iw_qp_add_ref,
 157        .iw_connect = qedr_iw_connect,
 158        .iw_create_listen = qedr_iw_create_listen,
 159        .iw_destroy_listen = qedr_iw_destroy_listen,
 160        .iw_get_qp = qedr_iw_get_qp,
 161        .iw_reject = qedr_iw_reject,
 162        .iw_rem_ref = qedr_iw_qp_rem_ref,
 163        .query_gid = qedr_iw_query_gid,
 164};
 165
 166static int qedr_iw_register_device(struct qedr_dev *dev)
 167{
 168        dev->ibdev.node_type = RDMA_NODE_RNIC;
 169
 170        ib_set_device_ops(&dev->ibdev, &qedr_iw_dev_ops);
 171
 172        memcpy(dev->ibdev.iw_ifname,
 173               dev->ndev->name, sizeof(dev->ibdev.iw_ifname));
 174
 175        return 0;
 176}
 177
 178static const struct ib_device_ops qedr_roce_dev_ops = {
 179        .alloc_xrcd = qedr_alloc_xrcd,
 180        .dealloc_xrcd = qedr_dealloc_xrcd,
 181        .get_port_immutable = qedr_roce_port_immutable,
 182        .query_pkey = qedr_query_pkey,
 183};
 184
 185static void qedr_roce_register_device(struct qedr_dev *dev)
 186{
 187        dev->ibdev.node_type = RDMA_NODE_IB_CA;
 188
 189        ib_set_device_ops(&dev->ibdev, &qedr_roce_dev_ops);
 190}
 191
 192static const struct ib_device_ops qedr_dev_ops = {
 193        .owner = THIS_MODULE,
 194        .driver_id = RDMA_DRIVER_QEDR,
 195        .uverbs_abi_ver = QEDR_ABI_VERSION,
 196
 197        .alloc_mr = qedr_alloc_mr,
 198        .alloc_pd = qedr_alloc_pd,
 199        .alloc_ucontext = qedr_alloc_ucontext,
 200        .create_ah = qedr_create_ah,
 201        .create_cq = qedr_create_cq,
 202        .create_qp = qedr_create_qp,
 203        .create_srq = qedr_create_srq,
 204        .dealloc_pd = qedr_dealloc_pd,
 205        .dealloc_ucontext = qedr_dealloc_ucontext,
 206        .dereg_mr = qedr_dereg_mr,
 207        .destroy_ah = qedr_destroy_ah,
 208        .destroy_cq = qedr_destroy_cq,
 209        .destroy_qp = qedr_destroy_qp,
 210        .destroy_srq = qedr_destroy_srq,
 211        .device_group = &qedr_attr_group,
 212        .get_dev_fw_str = qedr_get_dev_fw_str,
 213        .get_dma_mr = qedr_get_dma_mr,
 214        .get_link_layer = qedr_link_layer,
 215        .map_mr_sg = qedr_map_mr_sg,
 216        .mmap = qedr_mmap,
 217        .mmap_free = qedr_mmap_free,
 218        .modify_qp = qedr_modify_qp,
 219        .modify_srq = qedr_modify_srq,
 220        .poll_cq = qedr_poll_cq,
 221        .post_recv = qedr_post_recv,
 222        .post_send = qedr_post_send,
 223        .post_srq_recv = qedr_post_srq_recv,
 224        .process_mad = qedr_process_mad,
 225        .query_device = qedr_query_device,
 226        .query_port = qedr_query_port,
 227        .query_qp = qedr_query_qp,
 228        .query_srq = qedr_query_srq,
 229        .reg_user_mr = qedr_reg_user_mr,
 230        .req_notify_cq = qedr_arm_cq,
 231
 232        INIT_RDMA_OBJ_SIZE(ib_ah, qedr_ah, ibah),
 233        INIT_RDMA_OBJ_SIZE(ib_cq, qedr_cq, ibcq),
 234        INIT_RDMA_OBJ_SIZE(ib_pd, qedr_pd, ibpd),
 235        INIT_RDMA_OBJ_SIZE(ib_qp, qedr_qp, ibqp),
 236        INIT_RDMA_OBJ_SIZE(ib_srq, qedr_srq, ibsrq),
 237        INIT_RDMA_OBJ_SIZE(ib_xrcd, qedr_xrcd, ibxrcd),
 238        INIT_RDMA_OBJ_SIZE(ib_ucontext, qedr_ucontext, ibucontext),
 239};
 240
 241static int qedr_register_device(struct qedr_dev *dev)
 242{
 243        int rc;
 244
 245        dev->ibdev.node_guid = dev->attr.node_guid;
 246        memcpy(dev->ibdev.node_desc, QEDR_NODE_DESC, sizeof(QEDR_NODE_DESC));
 247
 248        if (IS_IWARP(dev)) {
 249                rc = qedr_iw_register_device(dev);
 250                if (rc)
 251                        return rc;
 252        } else {
 253                qedr_roce_register_device(dev);
 254        }
 255
 256        dev->ibdev.phys_port_cnt = 1;
 257        dev->ibdev.num_comp_vectors = dev->num_cnq;
 258        dev->ibdev.dev.parent = &dev->pdev->dev;
 259
 260        ib_set_device_ops(&dev->ibdev, &qedr_dev_ops);
 261
 262        rc = ib_device_set_netdev(&dev->ibdev, dev->ndev, 1);
 263        if (rc)
 264                return rc;
 265
 266        dma_set_max_seg_size(&dev->pdev->dev, UINT_MAX);
 267        return ib_register_device(&dev->ibdev, "qedr%d", &dev->pdev->dev);
 268}
 269
 270/* This function allocates fast-path status block memory */
 271static int qedr_alloc_mem_sb(struct qedr_dev *dev,
 272                             struct qed_sb_info *sb_info, u16 sb_id)
 273{
 274        struct status_block *sb_virt;
 275        dma_addr_t sb_phys;
 276        int rc;
 277
 278        sb_virt = dma_alloc_coherent(&dev->pdev->dev,
 279                                     sizeof(*sb_virt), &sb_phys, GFP_KERNEL);
 280        if (!sb_virt)
 281                return -ENOMEM;
 282
 283        rc = dev->ops->common->sb_init(dev->cdev, sb_info,
 284                                       sb_virt, sb_phys, sb_id,
 285                                       QED_SB_TYPE_CNQ);
 286        if (rc) {
 287                pr_err("Status block initialization failed\n");
 288                dma_free_coherent(&dev->pdev->dev, sizeof(*sb_virt),
 289                                  sb_virt, sb_phys);
 290                return rc;
 291        }
 292
 293        return 0;
 294}
 295
 296static void qedr_free_mem_sb(struct qedr_dev *dev,
 297                             struct qed_sb_info *sb_info, int sb_id)
 298{
 299        if (sb_info->sb_virt) {
 300                dev->ops->common->sb_release(dev->cdev, sb_info, sb_id,
 301                                             QED_SB_TYPE_CNQ);
 302                dma_free_coherent(&dev->pdev->dev, sizeof(*sb_info->sb_virt),
 303                                  (void *)sb_info->sb_virt, sb_info->sb_phys);
 304        }
 305}
 306
 307static void qedr_free_resources(struct qedr_dev *dev)
 308{
 309        int i;
 310
 311        if (IS_IWARP(dev))
 312                destroy_workqueue(dev->iwarp_wq);
 313
 314        for (i = 0; i < dev->num_cnq; i++) {
 315                qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
 316                dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
 317        }
 318
 319        kfree(dev->cnq_array);
 320        kfree(dev->sb_array);
 321        kfree(dev->sgid_tbl);
 322}
 323
 324static int qedr_alloc_resources(struct qedr_dev *dev)
 325{
 326        struct qed_chain_init_params params = {
 327                .mode           = QED_CHAIN_MODE_PBL,
 328                .intended_use   = QED_CHAIN_USE_TO_CONSUME,
 329                .cnt_type       = QED_CHAIN_CNT_TYPE_U16,
 330                .elem_size      = sizeof(struct regpair *),
 331        };
 332        struct qedr_cnq *cnq;
 333        __le16 *cons_pi;
 334        int i, rc;
 335
 336        dev->sgid_tbl = kcalloc(QEDR_MAX_SGID, sizeof(union ib_gid),
 337                                GFP_KERNEL);
 338        if (!dev->sgid_tbl)
 339                return -ENOMEM;
 340
 341        spin_lock_init(&dev->sgid_lock);
 342        xa_init_flags(&dev->srqs, XA_FLAGS_LOCK_IRQ);
 343
 344        if (IS_IWARP(dev)) {
 345                xa_init(&dev->qps);
 346                dev->iwarp_wq = create_singlethread_workqueue("qedr_iwarpq");
 347        }
 348
 349        /* Allocate Status blocks for CNQ */
 350        dev->sb_array = kcalloc(dev->num_cnq, sizeof(*dev->sb_array),
 351                                GFP_KERNEL);
 352        if (!dev->sb_array) {
 353                rc = -ENOMEM;
 354                goto err1;
 355        }
 356
 357        dev->cnq_array = kcalloc(dev->num_cnq,
 358                                 sizeof(*dev->cnq_array), GFP_KERNEL);
 359        if (!dev->cnq_array) {
 360                rc = -ENOMEM;
 361                goto err2;
 362        }
 363
 364        dev->sb_start = dev->ops->rdma_get_start_sb(dev->cdev);
 365
 366        /* Allocate CNQ PBLs */
 367        params.num_elems = min_t(u32, QED_RDMA_MAX_CNQ_SIZE,
 368                                 QEDR_ROCE_MAX_CNQ_SIZE);
 369
 370        for (i = 0; i < dev->num_cnq; i++) {
 371                cnq = &dev->cnq_array[i];
 372
 373                rc = qedr_alloc_mem_sb(dev, &dev->sb_array[i],
 374                                       dev->sb_start + i);
 375                if (rc)
 376                        goto err3;
 377
 378                rc = dev->ops->common->chain_alloc(dev->cdev, &cnq->pbl,
 379                                                   &params);
 380                if (rc)
 381                        goto err4;
 382
 383                cnq->dev = dev;
 384                cnq->sb = &dev->sb_array[i];
 385                cons_pi = dev->sb_array[i].sb_virt->pi_array;
 386                cnq->hw_cons_ptr = &cons_pi[QED_ROCE_PROTOCOL_INDEX];
 387                cnq->index = i;
 388                sprintf(cnq->name, "qedr%d@pci:%s", i, pci_name(dev->pdev));
 389
 390                DP_DEBUG(dev, QEDR_MSG_INIT, "cnq[%d].cons=%d\n",
 391                         i, qed_chain_get_cons_idx(&cnq->pbl));
 392        }
 393
 394        return 0;
 395err4:
 396        qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
 397err3:
 398        for (--i; i >= 0; i--) {
 399                dev->ops->common->chain_free(dev->cdev, &dev->cnq_array[i].pbl);
 400                qedr_free_mem_sb(dev, &dev->sb_array[i], dev->sb_start + i);
 401        }
 402        kfree(dev->cnq_array);
 403err2:
 404        kfree(dev->sb_array);
 405err1:
 406        kfree(dev->sgid_tbl);
 407        return rc;
 408}
 409
 410static void qedr_pci_set_atomic(struct qedr_dev *dev, struct pci_dev *pdev)
 411{
 412        int rc = pci_enable_atomic_ops_to_root(pdev,
 413                                               PCI_EXP_DEVCAP2_ATOMIC_COMP64);
 414
 415        if (rc) {
 416                dev->atomic_cap = IB_ATOMIC_NONE;
 417                DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability disabled\n");
 418        } else {
 419                dev->atomic_cap = IB_ATOMIC_GLOB;
 420                DP_DEBUG(dev, QEDR_MSG_INIT, "Atomic capability enabled\n");
 421        }
 422}
 423
 424static const struct qed_rdma_ops *qed_ops;
 425
 426#define HILO_U64(hi, lo)                ((((u64)(hi)) << 32) + (lo))
 427
 428static irqreturn_t qedr_irq_handler(int irq, void *handle)
 429{
 430        u16 hw_comp_cons, sw_comp_cons;
 431        struct qedr_cnq *cnq = handle;
 432        struct regpair *cq_handle;
 433        struct qedr_cq *cq;
 434
 435        qed_sb_ack(cnq->sb, IGU_INT_DISABLE, 0);
 436
 437        qed_sb_update_sb_idx(cnq->sb);
 438
 439        hw_comp_cons = le16_to_cpu(*cnq->hw_cons_ptr);
 440        sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
 441
 442        /* Align protocol-index and chain reads */
 443        rmb();
 444
 445        while (sw_comp_cons != hw_comp_cons) {
 446                cq_handle = (struct regpair *)qed_chain_consume(&cnq->pbl);
 447                cq = (struct qedr_cq *)(uintptr_t)HILO_U64(cq_handle->hi,
 448                                cq_handle->lo);
 449
 450                if (cq == NULL) {
 451                        DP_ERR(cnq->dev,
 452                               "Received NULL CQ cq_handle->hi=%d cq_handle->lo=%d sw_comp_cons=%d hw_comp_cons=%d\n",
 453                               cq_handle->hi, cq_handle->lo, sw_comp_cons,
 454                               hw_comp_cons);
 455
 456                        break;
 457                }
 458
 459                if (cq->sig != QEDR_CQ_MAGIC_NUMBER) {
 460                        DP_ERR(cnq->dev,
 461                               "Problem with cq signature, cq_handle->hi=%d ch_handle->lo=%d cq=%p\n",
 462                               cq_handle->hi, cq_handle->lo, cq);
 463                        break;
 464                }
 465
 466                cq->arm_flags = 0;
 467
 468                if (!cq->destroyed && cq->ibcq.comp_handler)
 469                        (*cq->ibcq.comp_handler)
 470                                (&cq->ibcq, cq->ibcq.cq_context);
 471
 472                /* The CQ's CNQ notification counter is checked before
 473                 * destroying the CQ in a busy-wait loop that waits for all of
 474                 * the CQ's CNQ interrupts to be processed. It is increased
 475                 * here, only after the completion handler, to ensure that the
 476                 * the handler is not running when the CQ is destroyed.
 477                 */
 478                cq->cnq_notif++;
 479
 480                sw_comp_cons = qed_chain_get_cons_idx(&cnq->pbl);
 481
 482                cnq->n_comp++;
 483        }
 484
 485        qed_ops->rdma_cnq_prod_update(cnq->dev->rdma_ctx, cnq->index,
 486                                      sw_comp_cons);
 487
 488        qed_sb_ack(cnq->sb, IGU_INT_ENABLE, 1);
 489
 490        return IRQ_HANDLED;
 491}
 492
 493static void qedr_sync_free_irqs(struct qedr_dev *dev)
 494{
 495        u32 vector;
 496        u16 idx;
 497        int i;
 498
 499        for (i = 0; i < dev->int_info.used_cnt; i++) {
 500                if (dev->int_info.msix_cnt) {
 501                        idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
 502                        vector = dev->int_info.msix[idx].vector;
 503                        synchronize_irq(vector);
 504                        free_irq(vector, &dev->cnq_array[i]);
 505                }
 506        }
 507
 508        dev->int_info.used_cnt = 0;
 509}
 510
 511static int qedr_req_msix_irqs(struct qedr_dev *dev)
 512{
 513        int i, rc = 0;
 514        u16 idx;
 515
 516        if (dev->num_cnq > dev->int_info.msix_cnt) {
 517                DP_ERR(dev,
 518                       "Interrupt mismatch: %d CNQ queues > %d MSI-x vectors\n",
 519                       dev->num_cnq, dev->int_info.msix_cnt);
 520                return -EINVAL;
 521        }
 522
 523        for (i = 0; i < dev->num_cnq; i++) {
 524                idx = i * dev->num_hwfns + dev->affin_hwfn_idx;
 525                rc = request_irq(dev->int_info.msix[idx].vector,
 526                                 qedr_irq_handler, 0, dev->cnq_array[i].name,
 527                                 &dev->cnq_array[i]);
 528                if (rc) {
 529                        DP_ERR(dev, "Request cnq %d irq failed\n", i);
 530                        qedr_sync_free_irqs(dev);
 531                } else {
 532                        DP_DEBUG(dev, QEDR_MSG_INIT,
 533                                 "Requested cnq irq for %s [entry %d]. Cookie is at %p\n",
 534                                 dev->cnq_array[i].name, i,
 535                                 &dev->cnq_array[i]);
 536                        dev->int_info.used_cnt++;
 537                }
 538        }
 539
 540        return rc;
 541}
 542
 543static int qedr_setup_irqs(struct qedr_dev *dev)
 544{
 545        int rc;
 546
 547        DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs\n");
 548
 549        /* Learn Interrupt configuration */
 550        rc = dev->ops->rdma_set_rdma_int(dev->cdev, dev->num_cnq);
 551        if (rc < 0)
 552                return rc;
 553
 554        rc = dev->ops->rdma_get_rdma_int(dev->cdev, &dev->int_info);
 555        if (rc) {
 556                DP_DEBUG(dev, QEDR_MSG_INIT, "get_rdma_int failed\n");
 557                return rc;
 558        }
 559
 560        if (dev->int_info.msix_cnt) {
 561                DP_DEBUG(dev, QEDR_MSG_INIT, "rdma msix_cnt = %d\n",
 562                         dev->int_info.msix_cnt);
 563                rc = qedr_req_msix_irqs(dev);
 564                if (rc)
 565                        return rc;
 566        }
 567
 568        DP_DEBUG(dev, QEDR_MSG_INIT, "qedr_setup_irqs succeeded\n");
 569
 570        return 0;
 571}
 572
 573static int qedr_set_device_attr(struct qedr_dev *dev)
 574{
 575        struct qed_rdma_device *qed_attr;
 576        struct qedr_device_attr *attr;
 577        u32 page_size;
 578
 579        /* Part 1 - query core capabilities */
 580        qed_attr = dev->ops->rdma_query_device(dev->rdma_ctx);
 581
 582        /* Part 2 - check capabilities */
 583        page_size = ~qed_attr->page_size_caps + 1;
 584        if (page_size > PAGE_SIZE) {
 585                DP_ERR(dev,
 586                       "Kernel PAGE_SIZE is %ld which is smaller than minimum page size (%d) required by qedr\n",
 587                       PAGE_SIZE, page_size);
 588                return -ENODEV;
 589        }
 590
 591        /* Part 3 - copy and update capabilities */
 592        attr = &dev->attr;
 593        attr->vendor_id = qed_attr->vendor_id;
 594        attr->vendor_part_id = qed_attr->vendor_part_id;
 595        attr->hw_ver = qed_attr->hw_ver;
 596        attr->fw_ver = qed_attr->fw_ver;
 597        attr->node_guid = qed_attr->node_guid;
 598        attr->sys_image_guid = qed_attr->sys_image_guid;
 599        attr->max_cnq = qed_attr->max_cnq;
 600        attr->max_sge = qed_attr->max_sge;
 601        attr->max_inline = qed_attr->max_inline;
 602        attr->max_sqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_SQE);
 603        attr->max_rqe = min_t(u32, qed_attr->max_wqe, QEDR_MAX_RQE);
 604        attr->max_qp_resp_rd_atomic_resc = qed_attr->max_qp_resp_rd_atomic_resc;
 605        attr->max_qp_req_rd_atomic_resc = qed_attr->max_qp_req_rd_atomic_resc;
 606        attr->max_dev_resp_rd_atomic_resc =
 607            qed_attr->max_dev_resp_rd_atomic_resc;
 608        attr->max_cq = qed_attr->max_cq;
 609        attr->max_qp = qed_attr->max_qp;
 610        attr->max_mr = qed_attr->max_mr;
 611        attr->max_mr_size = qed_attr->max_mr_size;
 612        attr->max_cqe = min_t(u64, qed_attr->max_cqe, QEDR_MAX_CQES);
 613        attr->max_mw = qed_attr->max_mw;
 614        attr->max_mr_mw_fmr_pbl = qed_attr->max_mr_mw_fmr_pbl;
 615        attr->max_mr_mw_fmr_size = qed_attr->max_mr_mw_fmr_size;
 616        attr->max_pd = qed_attr->max_pd;
 617        attr->max_ah = qed_attr->max_ah;
 618        attr->max_pkey = qed_attr->max_pkey;
 619        attr->max_srq = qed_attr->max_srq;
 620        attr->max_srq_wr = qed_attr->max_srq_wr;
 621        attr->dev_caps = qed_attr->dev_caps;
 622        attr->page_size_caps = qed_attr->page_size_caps;
 623        attr->dev_ack_delay = qed_attr->dev_ack_delay;
 624        attr->reserved_lkey = qed_attr->reserved_lkey;
 625        attr->bad_pkey_counter = qed_attr->bad_pkey_counter;
 626        attr->max_stats_queues = qed_attr->max_stats_queues;
 627
 628        return 0;
 629}
 630
 631static void qedr_unaffiliated_event(void *context, u8 event_code)
 632{
 633        pr_err("unaffiliated event not implemented yet\n");
 634}
 635
 636static void qedr_affiliated_event(void *context, u8 e_code, void *fw_handle)
 637{
 638#define EVENT_TYPE_NOT_DEFINED  0
 639#define EVENT_TYPE_CQ           1
 640#define EVENT_TYPE_QP           2
 641#define EVENT_TYPE_SRQ          3
 642        struct qedr_dev *dev = (struct qedr_dev *)context;
 643        struct regpair *async_handle = (struct regpair *)fw_handle;
 644        u64 roce_handle64 = ((u64) async_handle->hi << 32) + async_handle->lo;
 645        u8 event_type = EVENT_TYPE_NOT_DEFINED;
 646        struct ib_event event;
 647        struct ib_srq *ibsrq;
 648        struct qedr_srq *srq;
 649        unsigned long flags;
 650        struct ib_cq *ibcq;
 651        struct ib_qp *ibqp;
 652        struct qedr_cq *cq;
 653        struct qedr_qp *qp;
 654        u16 srq_id;
 655
 656        if (IS_ROCE(dev)) {
 657                switch (e_code) {
 658                case ROCE_ASYNC_EVENT_CQ_OVERFLOW_ERR:
 659                        event.event = IB_EVENT_CQ_ERR;
 660                        event_type = EVENT_TYPE_CQ;
 661                        break;
 662                case ROCE_ASYNC_EVENT_SQ_DRAINED:
 663                        event.event = IB_EVENT_SQ_DRAINED;
 664                        event_type = EVENT_TYPE_QP;
 665                        break;
 666                case ROCE_ASYNC_EVENT_QP_CATASTROPHIC_ERR:
 667                        event.event = IB_EVENT_QP_FATAL;
 668                        event_type = EVENT_TYPE_QP;
 669                        break;
 670                case ROCE_ASYNC_EVENT_LOCAL_INVALID_REQUEST_ERR:
 671                        event.event = IB_EVENT_QP_REQ_ERR;
 672                        event_type = EVENT_TYPE_QP;
 673                        break;
 674                case ROCE_ASYNC_EVENT_LOCAL_ACCESS_ERR:
 675                        event.event = IB_EVENT_QP_ACCESS_ERR;
 676                        event_type = EVENT_TYPE_QP;
 677                        break;
 678                case ROCE_ASYNC_EVENT_SRQ_LIMIT:
 679                        event.event = IB_EVENT_SRQ_LIMIT_REACHED;
 680                        event_type = EVENT_TYPE_SRQ;
 681                        break;
 682                case ROCE_ASYNC_EVENT_SRQ_EMPTY:
 683                        event.event = IB_EVENT_SRQ_ERR;
 684                        event_type = EVENT_TYPE_SRQ;
 685                        break;
 686                case ROCE_ASYNC_EVENT_XRC_DOMAIN_ERR:
 687                        event.event = IB_EVENT_QP_ACCESS_ERR;
 688                        event_type = EVENT_TYPE_QP;
 689                        break;
 690                case ROCE_ASYNC_EVENT_INVALID_XRCETH_ERR:
 691                        event.event = IB_EVENT_QP_ACCESS_ERR;
 692                        event_type = EVENT_TYPE_QP;
 693                        break;
 694                case ROCE_ASYNC_EVENT_XRC_SRQ_CATASTROPHIC_ERR:
 695                        event.event = IB_EVENT_CQ_ERR;
 696                        event_type = EVENT_TYPE_CQ;
 697                        break;
 698                default:
 699                        DP_ERR(dev, "unsupported event %d on handle=%llx\n",
 700                               e_code, roce_handle64);
 701                }
 702        } else {
 703                switch (e_code) {
 704                case QED_IWARP_EVENT_SRQ_LIMIT:
 705                        event.event = IB_EVENT_SRQ_LIMIT_REACHED;
 706                        event_type = EVENT_TYPE_SRQ;
 707                        break;
 708                case QED_IWARP_EVENT_SRQ_EMPTY:
 709                        event.event = IB_EVENT_SRQ_ERR;
 710                        event_type = EVENT_TYPE_SRQ;
 711                        break;
 712                default:
 713                DP_ERR(dev, "unsupported event %d on handle=%llx\n", e_code,
 714                       roce_handle64);
 715                }
 716        }
 717        switch (event_type) {
 718        case EVENT_TYPE_CQ:
 719                cq = (struct qedr_cq *)(uintptr_t)roce_handle64;
 720                if (cq) {
 721                        ibcq = &cq->ibcq;
 722                        if (ibcq->event_handler) {
 723                                event.device = ibcq->device;
 724                                event.element.cq = ibcq;
 725                                ibcq->event_handler(&event, ibcq->cq_context);
 726                        }
 727                } else {
 728                        WARN(1,
 729                             "Error: CQ event with NULL pointer ibcq. Handle=%llx\n",
 730                             roce_handle64);
 731                }
 732                DP_ERR(dev, "CQ event %d on handle %p\n", e_code, cq);
 733                break;
 734        case EVENT_TYPE_QP:
 735                qp = (struct qedr_qp *)(uintptr_t)roce_handle64;
 736                if (qp) {
 737                        ibqp = &qp->ibqp;
 738                        if (ibqp->event_handler) {
 739                                event.device = ibqp->device;
 740                                event.element.qp = ibqp;
 741                                ibqp->event_handler(&event, ibqp->qp_context);
 742                        }
 743                } else {
 744                        WARN(1,
 745                             "Error: QP event with NULL pointer ibqp. Handle=%llx\n",
 746                             roce_handle64);
 747                }
 748                DP_ERR(dev, "QP event %d on handle %p\n", e_code, qp);
 749                break;
 750        case EVENT_TYPE_SRQ:
 751                srq_id = (u16)roce_handle64;
 752                xa_lock_irqsave(&dev->srqs, flags);
 753                srq = xa_load(&dev->srqs, srq_id);
 754                if (srq) {
 755                        ibsrq = &srq->ibsrq;
 756                        if (ibsrq->event_handler) {
 757                                event.device = ibsrq->device;
 758                                event.element.srq = ibsrq;
 759                                ibsrq->event_handler(&event,
 760                                                     ibsrq->srq_context);
 761                        }
 762                } else {
 763                        DP_NOTICE(dev,
 764                                  "SRQ event with NULL pointer ibsrq. Handle=%llx\n",
 765                                  roce_handle64);
 766                }
 767                xa_unlock_irqrestore(&dev->srqs, flags);
 768                DP_NOTICE(dev, "SRQ event %d on handle %p\n", e_code, srq);
 769                break;
 770        default:
 771                break;
 772        }
 773}
 774
 775static int qedr_init_hw(struct qedr_dev *dev)
 776{
 777        struct qed_rdma_add_user_out_params out_params;
 778        struct qed_rdma_start_in_params *in_params;
 779        struct qed_rdma_cnq_params *cur_pbl;
 780        struct qed_rdma_events events;
 781        dma_addr_t p_phys_table;
 782        u32 page_cnt;
 783        int rc = 0;
 784        int i;
 785
 786        in_params =  kzalloc(sizeof(*in_params), GFP_KERNEL);
 787        if (!in_params) {
 788                rc = -ENOMEM;
 789                goto out;
 790        }
 791
 792        in_params->desired_cnq = dev->num_cnq;
 793        for (i = 0; i < dev->num_cnq; i++) {
 794                cur_pbl = &in_params->cnq_pbl_list[i];
 795
 796                page_cnt = qed_chain_get_page_cnt(&dev->cnq_array[i].pbl);
 797                cur_pbl->num_pbl_pages = page_cnt;
 798
 799                p_phys_table = qed_chain_get_pbl_phys(&dev->cnq_array[i].pbl);
 800                cur_pbl->pbl_ptr = (u64)p_phys_table;
 801        }
 802
 803        events.affiliated_event = qedr_affiliated_event;
 804        events.unaffiliated_event = qedr_unaffiliated_event;
 805        events.context = dev;
 806
 807        in_params->events = &events;
 808        in_params->cq_mode = QED_RDMA_CQ_MODE_32_BITS;
 809        in_params->max_mtu = dev->ndev->mtu;
 810        dev->iwarp_max_mtu = dev->ndev->mtu;
 811        ether_addr_copy(&in_params->mac_addr[0], dev->ndev->dev_addr);
 812
 813        rc = dev->ops->rdma_init(dev->cdev, in_params);
 814        if (rc)
 815                goto out;
 816
 817        rc = dev->ops->rdma_add_user(dev->rdma_ctx, &out_params);
 818        if (rc)
 819                goto out;
 820
 821        dev->db_addr = out_params.dpi_addr;
 822        dev->db_phys_addr = out_params.dpi_phys_addr;
 823        dev->db_size = out_params.dpi_size;
 824        dev->dpi = out_params.dpi;
 825
 826        rc = qedr_set_device_attr(dev);
 827out:
 828        kfree(in_params);
 829        if (rc)
 830                DP_ERR(dev, "Init HW Failed rc = %d\n", rc);
 831
 832        return rc;
 833}
 834
 835static void qedr_stop_hw(struct qedr_dev *dev)
 836{
 837        dev->ops->rdma_remove_user(dev->rdma_ctx, dev->dpi);
 838        dev->ops->rdma_stop(dev->rdma_ctx);
 839}
 840
 841static struct qedr_dev *qedr_add(struct qed_dev *cdev, struct pci_dev *pdev,
 842                                 struct net_device *ndev)
 843{
 844        struct qed_dev_rdma_info dev_info;
 845        struct qedr_dev *dev;
 846        int rc = 0;
 847
 848        dev = ib_alloc_device(qedr_dev, ibdev);
 849        if (!dev) {
 850                pr_err("Unable to allocate ib device\n");
 851                return NULL;
 852        }
 853
 854        DP_DEBUG(dev, QEDR_MSG_INIT, "qedr add device called\n");
 855
 856        dev->pdev = pdev;
 857        dev->ndev = ndev;
 858        dev->cdev = cdev;
 859
 860        qed_ops = qed_get_rdma_ops();
 861        if (!qed_ops) {
 862                DP_ERR(dev, "Failed to get qed roce operations\n");
 863                goto init_err;
 864        }
 865
 866        dev->ops = qed_ops;
 867        rc = qed_ops->fill_dev_info(cdev, &dev_info);
 868        if (rc)
 869                goto init_err;
 870
 871        dev->user_dpm_enabled = dev_info.user_dpm_enabled;
 872        dev->rdma_type = dev_info.rdma_type;
 873        dev->num_hwfns = dev_info.common.num_hwfns;
 874
 875        if (IS_IWARP(dev) && QEDR_IS_CMT(dev)) {
 876                rc = dev->ops->iwarp_set_engine_affin(cdev, false);
 877                if (rc) {
 878                        DP_ERR(dev, "iWARP is disabled over a 100g device Enabling it may impact L2 performance. To enable it run devlink dev param set <dev> name iwarp_cmt value true cmode runtime\n");
 879                        goto init_err;
 880                }
 881        }
 882        dev->affin_hwfn_idx = dev->ops->common->get_affin_hwfn_idx(cdev);
 883
 884        dev->rdma_ctx = dev->ops->rdma_get_rdma_ctx(cdev);
 885
 886        dev->num_cnq = dev->ops->rdma_get_min_cnq_msix(cdev);
 887        if (!dev->num_cnq) {
 888                DP_ERR(dev, "Failed. At least one CNQ is required.\n");
 889                rc = -ENOMEM;
 890                goto init_err;
 891        }
 892
 893        dev->wq_multiplier = QEDR_WQ_MULTIPLIER_DFT;
 894
 895        qedr_pci_set_atomic(dev, pdev);
 896
 897        rc = qedr_alloc_resources(dev);
 898        if (rc)
 899                goto init_err;
 900
 901        rc = qedr_init_hw(dev);
 902        if (rc)
 903                goto alloc_err;
 904
 905        rc = qedr_setup_irqs(dev);
 906        if (rc)
 907                goto irq_err;
 908
 909        rc = qedr_register_device(dev);
 910        if (rc) {
 911                DP_ERR(dev, "Unable to allocate register device\n");
 912                goto reg_err;
 913        }
 914
 915        if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
 916                qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
 917
 918        DP_DEBUG(dev, QEDR_MSG_INIT, "qedr driver loaded successfully\n");
 919        return dev;
 920
 921reg_err:
 922        qedr_sync_free_irqs(dev);
 923irq_err:
 924        qedr_stop_hw(dev);
 925alloc_err:
 926        qedr_free_resources(dev);
 927init_err:
 928        ib_dealloc_device(&dev->ibdev);
 929        DP_ERR(dev, "qedr driver load failed rc=%d\n", rc);
 930
 931        return NULL;
 932}
 933
 934static void qedr_remove(struct qedr_dev *dev)
 935{
 936        /* First unregister with stack to stop all the active traffic
 937         * of the registered clients.
 938         */
 939        ib_unregister_device(&dev->ibdev);
 940
 941        qedr_stop_hw(dev);
 942        qedr_sync_free_irqs(dev);
 943        qedr_free_resources(dev);
 944
 945        if (IS_IWARP(dev) && QEDR_IS_CMT(dev))
 946                dev->ops->iwarp_set_engine_affin(dev->cdev, true);
 947
 948        ib_dealloc_device(&dev->ibdev);
 949}
 950
 951static void qedr_close(struct qedr_dev *dev)
 952{
 953        if (test_and_clear_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
 954                qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ERR);
 955}
 956
 957static void qedr_shutdown(struct qedr_dev *dev)
 958{
 959        qedr_close(dev);
 960        qedr_remove(dev);
 961}
 962
 963static void qedr_open(struct qedr_dev *dev)
 964{
 965        if (!test_and_set_bit(QEDR_ENET_STATE_BIT, &dev->enet_state))
 966                qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_PORT_ACTIVE);
 967}
 968
 969static void qedr_mac_address_change(struct qedr_dev *dev)
 970{
 971        union ib_gid *sgid = &dev->sgid_tbl[0];
 972        u8 guid[8], mac_addr[6];
 973        int rc;
 974
 975        /* Update SGID */
 976        ether_addr_copy(&mac_addr[0], dev->ndev->dev_addr);
 977        guid[0] = mac_addr[0] ^ 2;
 978        guid[1] = mac_addr[1];
 979        guid[2] = mac_addr[2];
 980        guid[3] = 0xff;
 981        guid[4] = 0xfe;
 982        guid[5] = mac_addr[3];
 983        guid[6] = mac_addr[4];
 984        guid[7] = mac_addr[5];
 985        sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
 986        memcpy(&sgid->raw[8], guid, sizeof(guid));
 987
 988        /* Update LL2 */
 989        rc = dev->ops->ll2_set_mac_filter(dev->cdev,
 990                                          dev->gsi_ll2_mac_address,
 991                                          dev->ndev->dev_addr);
 992
 993        ether_addr_copy(dev->gsi_ll2_mac_address, dev->ndev->dev_addr);
 994
 995        qedr_ib_dispatch_event(dev, QEDR_PORT, IB_EVENT_GID_CHANGE);
 996
 997        if (rc)
 998                DP_ERR(dev, "Error updating mac filter\n");
 999}
1000
1001/* event handling via NIC driver ensures that all the NIC specific
1002 * initialization done before RoCE driver notifies
1003 * event to stack.
1004 */
1005static void qedr_notify(struct qedr_dev *dev, enum qede_rdma_event event)
1006{
1007        switch (event) {
1008        case QEDE_UP:
1009                qedr_open(dev);
1010                break;
1011        case QEDE_DOWN:
1012                qedr_close(dev);
1013                break;
1014        case QEDE_CLOSE:
1015                qedr_shutdown(dev);
1016                break;
1017        case QEDE_CHANGE_ADDR:
1018                qedr_mac_address_change(dev);
1019                break;
1020        case QEDE_CHANGE_MTU:
1021                if (rdma_protocol_iwarp(&dev->ibdev, 1))
1022                        if (dev->ndev->mtu != dev->iwarp_max_mtu)
1023                                DP_NOTICE(dev,
1024                                          "Mtu was changed from %d to %d. This will not take affect for iWARP until qedr is reloaded\n",
1025                                          dev->iwarp_max_mtu, dev->ndev->mtu);
1026                break;
1027        default:
1028                pr_err("Event not supported\n");
1029        }
1030}
1031
1032static struct qedr_driver qedr_drv = {
1033        .name = "qedr_driver",
1034        .add = qedr_add,
1035        .remove = qedr_remove,
1036        .notify = qedr_notify,
1037};
1038
1039static int __init qedr_init_module(void)
1040{
1041        return qede_rdma_register_driver(&qedr_drv);
1042}
1043
1044static void __exit qedr_exit_module(void)
1045{
1046        qede_rdma_unregister_driver(&qedr_drv);
1047}
1048
1049module_init(qedr_init_module);
1050module_exit(qedr_exit_module);
1051