/*
 * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <linux/log2.h>
#include <linux/etherdevice.h>
#include <net/ip.h>
#include <linux/slab.h>
#include <linux/netdevice.h>

#include <rdma/ib_cache.h>
#include <rdma/ib_pack.h>
#include <rdma/ib_addr.h>
#include <rdma/ib_mad.h>
#include <rdma/uverbs_ioctl.h>

#include <linux/mlx4/driver.h>
#include <linux/mlx4/qp.h>

#include "mlx4_ib.h"
#include <rdma/mlx4-abi.h>

static void mlx4_ib_lock_cqs(struct mlx4_ib_cq *send_cq,
                             struct mlx4_ib_cq *recv_cq);
static void mlx4_ib_unlock_cqs(struct mlx4_ib_cq *send_cq,
                               struct mlx4_ib_cq *recv_cq);
static int _mlx4_ib_modify_wq(struct ib_wq *ibwq, enum ib_wq_state new_state,
                              struct ib_udata *udata);

enum {
        MLX4_IB_ACK_REQ_FREQ    = 8,
};

enum {
        MLX4_IB_DEFAULT_SCHED_QUEUE     = 0x83,
        MLX4_IB_DEFAULT_QP0_SCHED_QUEUE = 0x3f,
        MLX4_IB_LINK_TYPE_IB            = 0,
        MLX4_IB_LINK_TYPE_ETH           = 1
};

enum {
        MLX4_IB_MIN_SQ_STRIDE   = 6,
        MLX4_IB_CACHE_LINE_SIZE = 64,
};

enum {
        MLX4_RAW_QP_MTU         = 7,
        MLX4_RAW_QP_MSGMAX      = 31,
};

#ifndef ETH_ALEN
#define ETH_ALEN        6
#endif

static const __be32 mlx4_ib_opcode[] = {
        [IB_WR_SEND]                            = cpu_to_be32(MLX4_OPCODE_SEND),
        [IB_WR_LSO]                             = cpu_to_be32(MLX4_OPCODE_LSO),
        [IB_WR_SEND_WITH_IMM]                   = cpu_to_be32(MLX4_OPCODE_SEND_IMM),
        [IB_WR_RDMA_WRITE]                      = cpu_to_be32(MLX4_OPCODE_RDMA_WRITE),
        [IB_WR_RDMA_WRITE_WITH_IMM]             = cpu_to_be32(MLX4_OPCODE_RDMA_WRITE_IMM),
        [IB_WR_RDMA_READ]                       = cpu_to_be32(MLX4_OPCODE_RDMA_READ),
        [IB_WR_ATOMIC_CMP_AND_SWP]              = cpu_to_be32(MLX4_OPCODE_ATOMIC_CS),
        [IB_WR_ATOMIC_FETCH_AND_ADD]            = cpu_to_be32(MLX4_OPCODE_ATOMIC_FA),
        [IB_WR_SEND_WITH_INV]                   = cpu_to_be32(MLX4_OPCODE_SEND_INVAL),
        [IB_WR_LOCAL_INV]                       = cpu_to_be32(MLX4_OPCODE_LOCAL_INVAL),
        [IB_WR_REG_MR]                          = cpu_to_be32(MLX4_OPCODE_FMR),
        [IB_WR_MASKED_ATOMIC_CMP_AND_SWP]       = cpu_to_be32(MLX4_OPCODE_MASKED_ATOMIC_CS),
        [IB_WR_MASKED_ATOMIC_FETCH_AND_ADD]     = cpu_to_be32(MLX4_OPCODE_MASKED_ATOMIC_FA),
};

enum mlx4_ib_source_type {
        MLX4_IB_QP_SRC  = 0,
        MLX4_IB_RWQ_SRC = 1,
};

static int is_tunnel_qp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
{
        if (!mlx4_is_master(dev->dev))
                return 0;

        return qp->mqp.qpn >= dev->dev->phys_caps.base_tunnel_sqpn &&
               qp->mqp.qpn < dev->dev->phys_caps.base_tunnel_sqpn +
                8 * MLX4_MFUNC_MAX;
}

static int is_sqp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
{
        int proxy_sqp = 0;
        int real_sqp = 0;
        int i;
        /* PPF or Native -- real SQP */
        real_sqp = ((mlx4_is_master(dev->dev) || !mlx4_is_mfunc(dev->dev)) &&
                    qp->mqp.qpn >= dev->dev->phys_caps.base_sqpn &&
                    qp->mqp.qpn <= dev->dev->phys_caps.base_sqpn + 3);
        if (real_sqp)
                return 1;
        /* VF or PF -- proxy SQP */
        if (mlx4_is_mfunc(dev->dev)) {
                for (i = 0; i < dev->dev->caps.num_ports; i++) {
                        if (qp->mqp.qpn == dev->dev->caps.spec_qps[i].qp0_proxy ||
                            qp->mqp.qpn == dev->dev->caps.spec_qps[i].qp1_proxy) {
                                proxy_sqp = 1;
                                break;
                        }
                }
        }
        if (proxy_sqp)
                return 1;

        return !!(qp->flags & MLX4_IB_ROCE_V2_GSI_QP);
}

/* used for INIT/CLOSE port logic */
static int is_qp0(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
{
        int proxy_qp0 = 0;
        int real_qp0 = 0;
        int i;
        /* PPF or Native -- real QP0 */
        real_qp0 = ((mlx4_is_master(dev->dev) || !mlx4_is_mfunc(dev->dev)) &&
                    qp->mqp.qpn >= dev->dev->phys_caps.base_sqpn &&
                    qp->mqp.qpn <= dev->dev->phys_caps.base_sqpn + 1);
        if (real_qp0)
                return 1;
        /* VF or PF -- proxy QP0 */
        if (mlx4_is_mfunc(dev->dev)) {
                for (i = 0; i < dev->dev->caps.num_ports; i++) {
                        if (qp->mqp.qpn == dev->dev->caps.spec_qps[i].qp0_proxy) {
                                proxy_qp0 = 1;
                                break;
                        }
                }
        }
        return proxy_qp0;
}

static void *get_wqe(struct mlx4_ib_qp *qp, int offset)
{
        return mlx4_buf_offset(&qp->buf, offset);
}

static void *get_recv_wqe(struct mlx4_ib_qp *qp, int n)
{
        return get_wqe(qp, qp->rq.offset + (n << qp->rq.wqe_shift));
}

static void *get_send_wqe(struct mlx4_ib_qp *qp, int n)
{
        return get_wqe(qp, qp->sq.offset + (n << qp->sq.wqe_shift));
}

/*
 * Stamp a SQ WQE so that it is invalid if prefetched by marking the
 * first four bytes of every 64 byte chunk with 0xffffffff, except for
 * the very first chunk of the WQE.
 */
static void stamp_send_wqe(struct mlx4_ib_qp *qp, int n)
{
        __be32 *wqe;
        int i;
        int s;
        void *buf;
        struct mlx4_wqe_ctrl_seg *ctrl;

        buf = get_send_wqe(qp, n & (qp->sq.wqe_cnt - 1));
        ctrl = (struct mlx4_wqe_ctrl_seg *)buf;
        s = (ctrl->qpn_vlan.fence_size & 0x3f) << 4;
        for (i = 64; i < s; i += 64) {
                wqe = buf + i;
                *wqe = cpu_to_be32(0xffffffff);
        }
}

static void mlx4_ib_qp_event(struct mlx4_qp *qp, enum mlx4_event type)
{
        struct ib_event event;
        struct ib_qp *ibqp = &to_mibqp(qp)->ibqp;

        if (type == MLX4_EVENT_TYPE_PATH_MIG)
                to_mibqp(qp)->port = to_mibqp(qp)->alt_port;

        if (ibqp->event_handler) {
                event.device     = ibqp->device;
                event.element.qp = ibqp;
                switch (type) {
                case MLX4_EVENT_TYPE_PATH_MIG:
                        event.event = IB_EVENT_PATH_MIG;
                        break;
                case MLX4_EVENT_TYPE_COMM_EST:
                        event.event = IB_EVENT_COMM_EST;
                        break;
                case MLX4_EVENT_TYPE_SQ_DRAINED:
                        event.event = IB_EVENT_SQ_DRAINED;
                        break;
                case MLX4_EVENT_TYPE_SRQ_QP_LAST_WQE:
                        event.event = IB_EVENT_QP_LAST_WQE_REACHED;
                        break;
                case MLX4_EVENT_TYPE_WQ_CATAS_ERROR:
                        event.event = IB_EVENT_QP_FATAL;
                        break;
                case MLX4_EVENT_TYPE_PATH_MIG_FAILED:
                        event.event = IB_EVENT_PATH_MIG_ERR;
                        break;
                case MLX4_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
                        event.event = IB_EVENT_QP_REQ_ERR;
                        break;
                case MLX4_EVENT_TYPE_WQ_ACCESS_ERROR:
                        event.event = IB_EVENT_QP_ACCESS_ERR;
                        break;
                default:
                        pr_warn("Unexpected event type %d "
                               "on QP %06x\n", type, qp->qpn);
                        return;
                }

                ibqp->event_handler(&event, ibqp->qp_context);
        }
}

static void mlx4_ib_wq_event(struct mlx4_qp *qp, enum mlx4_event type)
{
        pr_warn_ratelimited("Unexpected event type %d on WQ 0x%06x. Events are not supported for WQs\n",
                            type, qp->qpn);
}

static int send_wqe_overhead(enum mlx4_ib_qp_type type, u32 flags)
{
        /*
         * UD WQEs must have a datagram segment.
         * RC and UC WQEs might have a remote address segment.
         * MLX WQEs need two extra inline data segments (for the UD
         * header and space for the ICRC).
         */
        switch (type) {
        case MLX4_IB_QPT_UD:
                return sizeof (struct mlx4_wqe_ctrl_seg) +
                        sizeof (struct mlx4_wqe_datagram_seg) +
                        ((flags & MLX4_IB_QP_LSO) ? MLX4_IB_LSO_HEADER_SPARE : 0);
        case MLX4_IB_QPT_PROXY_SMI_OWNER:
        case MLX4_IB_QPT_PROXY_SMI:
        case MLX4_IB_QPT_PROXY_GSI:
                return sizeof (struct mlx4_wqe_ctrl_seg) +
                        sizeof (struct mlx4_wqe_datagram_seg) + 64;
        case MLX4_IB_QPT_TUN_SMI_OWNER:
        case MLX4_IB_QPT_TUN_GSI:
                return sizeof (struct mlx4_wqe_ctrl_seg) +
                        sizeof (struct mlx4_wqe_datagram_seg);

        case MLX4_IB_QPT_UC:
                return sizeof (struct mlx4_wqe_ctrl_seg) +
                        sizeof (struct mlx4_wqe_raddr_seg);
        case MLX4_IB_QPT_RC:
                return sizeof (struct mlx4_wqe_ctrl_seg) +
                        sizeof (struct mlx4_wqe_masked_atomic_seg) +
                        sizeof (struct mlx4_wqe_raddr_seg);
        case MLX4_IB_QPT_SMI:
        case MLX4_IB_QPT_GSI:
                return sizeof (struct mlx4_wqe_ctrl_seg) +
                        ALIGN(MLX4_IB_UD_HEADER_SIZE +
                              DIV_ROUND_UP(MLX4_IB_UD_HEADER_SIZE,
                                           MLX4_INLINE_ALIGN) *
                              sizeof (struct mlx4_wqe_inline_seg),
                              sizeof (struct mlx4_wqe_data_seg)) +
                        ALIGN(4 +
                              sizeof (struct mlx4_wqe_inline_seg),
                              sizeof (struct mlx4_wqe_data_seg));
        default:
                return sizeof (struct mlx4_wqe_ctrl_seg);
        }
}

static int set_rq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
                       bool is_user, bool has_rq, struct mlx4_ib_qp *qp,
                       u32 inl_recv_sz)
{
        /* Sanity check RQ size before proceeding */
        if (cap->max_recv_wr > dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE ||
            cap->max_recv_sge > min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg))
                return -EINVAL;

        if (!has_rq) {
                if (cap->max_recv_wr || inl_recv_sz)
                        return -EINVAL;

                qp->rq.wqe_cnt = qp->rq.max_gs = 0;
        } else {
                u32 max_inl_recv_sz = dev->dev->caps.max_rq_sg *
                        sizeof(struct mlx4_wqe_data_seg);
                u32 wqe_size;

                /* HW requires >= 1 RQ entry with >= 1 gather entry */
                if (is_user && (!cap->max_recv_wr || !cap->max_recv_sge ||
                                inl_recv_sz > max_inl_recv_sz))
                        return -EINVAL;

                qp->rq.wqe_cnt   = roundup_pow_of_two(max(1U, cap->max_recv_wr));
                qp->rq.max_gs    = roundup_pow_of_two(max(1U, cap->max_recv_sge));
                wqe_size = qp->rq.max_gs * sizeof(struct mlx4_wqe_data_seg);
                qp->rq.wqe_shift = ilog2(max_t(u32, wqe_size, inl_recv_sz));
        }

        /* leave userspace return values as they were, so as not to break ABI */
        if (is_user) {
                cap->max_recv_wr  = qp->rq.max_post = qp->rq.wqe_cnt;
                cap->max_recv_sge = qp->rq.max_gs;
        } else {
                cap->max_recv_wr  = qp->rq.max_post =
                        min(dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE, qp->rq.wqe_cnt);
                cap->max_recv_sge = min(qp->rq.max_gs,
                                        min(dev->dev->caps.max_sq_sg,
                                            dev->dev->caps.max_rq_sg));
        }

        return 0;
}

static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
                              enum mlx4_ib_qp_type type, struct mlx4_ib_qp *qp)
{
        int s;

        /* Sanity check SQ size before proceeding */
        if (cap->max_send_wr  > (dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE) ||
            cap->max_send_sge > min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg) ||
            cap->max_inline_data + send_wqe_overhead(type, qp->flags) +
            sizeof (struct mlx4_wqe_inline_seg) > dev->dev->caps.max_sq_desc_sz)
                return -EINVAL;

        /*
         * For MLX transport we need 2 extra S/G entries:
         * one for the header and one for the checksum at the end
         */
        if ((type == MLX4_IB_QPT_SMI || type == MLX4_IB_QPT_GSI ||
             type & (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER)) &&
            cap->max_send_sge + 2 > dev->dev->caps.max_sq_sg)
                return -EINVAL;

        s = max(cap->max_send_sge * sizeof (struct mlx4_wqe_data_seg),
                cap->max_inline_data + sizeof (struct mlx4_wqe_inline_seg)) +
                send_wqe_overhead(type, qp->flags);

        if (s > dev->dev->caps.max_sq_desc_sz)
                return -EINVAL;

        qp->sq.wqe_shift = ilog2(roundup_pow_of_two(s));

        /*
         * We need to leave 2 KB + 1 WR of headroom in the SQ to
         * allow HW to prefetch.
         */
        qp->sq_spare_wqes = MLX4_IB_SQ_HEADROOM(qp->sq.wqe_shift);
        qp->sq.wqe_cnt = roundup_pow_of_two(cap->max_send_wr +
                                            qp->sq_spare_wqes);

        qp->sq.max_gs =
                (min(dev->dev->caps.max_sq_desc_sz,
                     (1 << qp->sq.wqe_shift)) -
                 send_wqe_overhead(type, qp->flags)) /
                sizeof (struct mlx4_wqe_data_seg);

        qp->buf_size = (qp->rq.wqe_cnt << qp->rq.wqe_shift) +
                (qp->sq.wqe_cnt << qp->sq.wqe_shift);
        if (qp->rq.wqe_shift > qp->sq.wqe_shift) {
                qp->rq.offset = 0;
                qp->sq.offset = qp->rq.wqe_cnt << qp->rq.wqe_shift;
        } else {
                qp->rq.offset = qp->sq.wqe_cnt << qp->sq.wqe_shift;
                qp->sq.offset = 0;
        }

        cap->max_send_wr  = qp->sq.max_post =
                qp->sq.wqe_cnt - qp->sq_spare_wqes;
        cap->max_send_sge = min(qp->sq.max_gs,
                                min(dev->dev->caps.max_sq_sg,
                                    dev->dev->caps.max_rq_sg));
        /* We don't support inline sends for kernel QPs (yet) */
        cap->max_inline_data = 0;

        return 0;
}

static int set_user_sq_size(struct mlx4_ib_dev *dev,
                            struct mlx4_ib_qp *qp,
                            struct mlx4_ib_create_qp *ucmd)
{
        /* Sanity check SQ size before proceeding */
        if ((1 << ucmd->log_sq_bb_count) > dev->dev->caps.max_wqes       ||
            ucmd->log_sq_stride >
                ilog2(roundup_pow_of_two(dev->dev->caps.max_sq_desc_sz)) ||
            ucmd->log_sq_stride < MLX4_IB_MIN_SQ_STRIDE)
                return -EINVAL;

        qp->sq.wqe_cnt   = 1 << ucmd->log_sq_bb_count;
        qp->sq.wqe_shift = ucmd->log_sq_stride;

        qp->buf_size = (qp->rq.wqe_cnt << qp->rq.wqe_shift) +
                (qp->sq.wqe_cnt << qp->sq.wqe_shift);

        return 0;
}

static int alloc_proxy_bufs(struct ib_device *dev, struct mlx4_ib_qp *qp)
{
        int i;

        qp->sqp_proxy_rcv =
                kmalloc_array(qp->rq.wqe_cnt, sizeof(struct mlx4_ib_buf),
                              GFP_KERNEL);
        if (!qp->sqp_proxy_rcv)
                return -ENOMEM;
        for (i = 0; i < qp->rq.wqe_cnt; i++) {
                qp->sqp_proxy_rcv[i].addr =
                        kmalloc(sizeof (struct mlx4_ib_proxy_sqp_hdr),
                                GFP_KERNEL);
                if (!qp->sqp_proxy_rcv[i].addr)
                        goto err;
                qp->sqp_proxy_rcv[i].map =
                        ib_dma_map_single(dev, qp->sqp_proxy_rcv[i].addr,
                                          sizeof (struct mlx4_ib_proxy_sqp_hdr),
                                          DMA_FROM_DEVICE);
                if (ib_dma_mapping_error(dev, qp->sqp_proxy_rcv[i].map)) {
                        kfree(qp->sqp_proxy_rcv[i].addr);
                        goto err;
                }
        }
        return 0;

err:
        while (i > 0) {
                --i;
                ib_dma_unmap_single(dev, qp->sqp_proxy_rcv[i].map,
                                    sizeof (struct mlx4_ib_proxy_sqp_hdr),
                                    DMA_FROM_DEVICE);
                kfree(qp->sqp_proxy_rcv[i].addr);
        }
        kfree(qp->sqp_proxy_rcv);
        qp->sqp_proxy_rcv = NULL;
        return -ENOMEM;
}

static void free_proxy_bufs(struct ib_device *dev, struct mlx4_ib_qp *qp)
{
        int i;

        for (i = 0; i < qp->rq.wqe_cnt; i++) {
                ib_dma_unmap_single(dev, qp->sqp_proxy_rcv[i].map,
                                    sizeof (struct mlx4_ib_proxy_sqp_hdr),
                                    DMA_FROM_DEVICE);
                kfree(qp->sqp_proxy_rcv[i].addr);
        }
        kfree(qp->sqp_proxy_rcv);
}

static bool qp_has_rq(struct ib_qp_init_attr *attr)
{
        if (attr->qp_type == IB_QPT_XRC_INI || attr->qp_type == IB_QPT_XRC_TGT)
                return false;

        return !attr->srq;
}

static int qp0_enabled_vf(struct mlx4_dev *dev, int qpn)
{
        int i;
        for (i = 0; i < dev->caps.num_ports; i++) {
                if (qpn == dev->caps.spec_qps[i].qp0_proxy)
                        return !!dev->caps.spec_qps[i].qp0_qkey;
        }
        return 0;
}

static void mlx4_ib_free_qp_counter(struct mlx4_ib_dev *dev,
                                    struct mlx4_ib_qp *qp)
{
        mutex_lock(&dev->counters_table[qp->port - 1].mutex);
        mlx4_counter_free(dev->dev, qp->counter_index->index);
        list_del(&qp->counter_index->list);
        mutex_unlock(&dev->counters_table[qp->port - 1].mutex);

        kfree(qp->counter_index);
        qp->counter_index = NULL;
}

static int set_qp_rss(struct mlx4_ib_dev *dev, struct mlx4_ib_rss *rss_ctx,
                      struct ib_qp_init_attr *init_attr,
                      struct mlx4_ib_create_qp_rss *ucmd)
{
        rss_ctx->base_qpn_tbl_sz = init_attr->rwq_ind_tbl->ind_tbl[0]->wq_num |
                (init_attr->rwq_ind_tbl->log_ind_tbl_size << 24);

        if ((ucmd->rx_hash_function == MLX4_IB_RX_HASH_FUNC_TOEPLITZ) &&
            (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS_TOP)) {
                memcpy(rss_ctx->rss_key, ucmd->rx_hash_key,
                       MLX4_EN_RSS_KEY_SIZE);
        } else {
                pr_debug("RX Hash function is not supported\n");
                return (-EOPNOTSUPP);
        }

        if (ucmd->rx_hash_fields_mask & ~(MLX4_IB_RX_HASH_SRC_IPV4      |
                                          MLX4_IB_RX_HASH_DST_IPV4      |
                                          MLX4_IB_RX_HASH_SRC_IPV6      |
                                          MLX4_IB_RX_HASH_DST_IPV6      |
                                          MLX4_IB_RX_HASH_SRC_PORT_TCP  |
                                          MLX4_IB_RX_HASH_DST_PORT_TCP  |
                                          MLX4_IB_RX_HASH_SRC_PORT_UDP  |
                                          MLX4_IB_RX_HASH_DST_PORT_UDP  |
                                          MLX4_IB_RX_HASH_INNER)) {
                pr_debug("RX Hash fields_mask has unsupported mask (0x%llx)\n",
                         ucmd->rx_hash_fields_mask);
                return (-EOPNOTSUPP);
        }

        if ((ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_SRC_IPV4) &&
            (ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_DST_IPV4)) {
                rss_ctx->flags = MLX4_RSS_IPV4;
        } else if ((ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_SRC_IPV4) ||
                   (ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_DST_IPV4)) {
                pr_debug("RX Hash fields_mask is not supported - both IPv4 SRC and DST must be set\n");
                return (-EOPNOTSUPP);
        }

        if ((ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_SRC_IPV6) &&
            (ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_DST_IPV6)) {
                rss_ctx->flags |= MLX4_RSS_IPV6;
        } else if ((ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_SRC_IPV6) ||
                   (ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_DST_IPV6)) {
                pr_debug("RX Hash fields_mask is not supported - both IPv6 SRC and DST must be set\n");
                return (-EOPNOTSUPP);
        }

        if ((ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_SRC_PORT_UDP) &&
            (ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_DST_PORT_UDP)) {
                if (!(dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UDP_RSS)) {
                        pr_debug("RX Hash fields_mask for UDP is not supported\n");
                        return (-EOPNOTSUPP);
                }

                if (rss_ctx->flags & MLX4_RSS_IPV4)
                        rss_ctx->flags |= MLX4_RSS_UDP_IPV4;
                if (rss_ctx->flags & MLX4_RSS_IPV6)
                        rss_ctx->flags |= MLX4_RSS_UDP_IPV6;
                if (!(rss_ctx->flags & (MLX4_RSS_IPV6 | MLX4_RSS_IPV4))) {
                        pr_debug("RX Hash fields_mask is not supported - UDP must be set with IPv4 or IPv6\n");
                        return (-EOPNOTSUPP);
                }
        } else if ((ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_SRC_PORT_UDP) ||
                   (ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_DST_PORT_UDP)) {
                pr_debug("RX Hash fields_mask is not supported - both UDP SRC and DST must be set\n");
                return (-EOPNOTSUPP);
        }

        if ((ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_SRC_PORT_TCP) &&
            (ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_DST_PORT_TCP)) {
                if (rss_ctx->flags & MLX4_RSS_IPV4)
                        rss_ctx->flags |= MLX4_RSS_TCP_IPV4;
                if (rss_ctx->flags & MLX4_RSS_IPV6)
                        rss_ctx->flags |= MLX4_RSS_TCP_IPV6;
                if (!(rss_ctx->flags & (MLX4_RSS_IPV6 | MLX4_RSS_IPV4))) {
                        pr_debug("RX Hash fields_mask is not supported - TCP must be set with IPv4 or IPv6\n");
                        return (-EOPNOTSUPP);
                }
        } else if ((ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_SRC_PORT_TCP) ||
                   (ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_DST_PORT_TCP)) {
                pr_debug("RX Hash fields_mask is not supported - both TCP SRC and DST must be set\n");
                return (-EOPNOTSUPP);
        }

        if (ucmd->rx_hash_fields_mask & MLX4_IB_RX_HASH_INNER) {
                if (dev->dev->caps.tunnel_offload_mode ==
                    MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
                        /*
                         * Hash according to inner headers if exist, otherwise
                         * according to outer headers.
                         */
                        rss_ctx->flags |= MLX4_RSS_BY_INNER_HEADERS_IPONLY;
                } else {
                        pr_debug("RSS Hash for inner headers isn't supported\n");
                        return (-EOPNOTSUPP);
                }
        }

        return 0;
}

static int create_qp_rss(struct mlx4_ib_dev *dev,
                         struct ib_qp_init_attr *init_attr,
                         struct mlx4_ib_create_qp_rss *ucmd,
                         struct mlx4_ib_qp *qp)
{
        int qpn;
        int err;

        qp->mqp.usage = MLX4_RES_USAGE_USER_VERBS;

        err = mlx4_qp_reserve_range(dev->dev, 1, 1, &qpn, 0, qp->mqp.usage);
        if (err)
                return err;

        err = mlx4_qp_alloc(dev->dev, qpn, &qp->mqp);
        if (err)
                goto err_qpn;

        INIT_LIST_HEAD(&qp->gid_list);
        INIT_LIST_HEAD(&qp->steering_rules);

        qp->mlx4_ib_qp_type = MLX4_IB_QPT_RAW_PACKET;
        qp->state = IB_QPS_RESET;

        /* Set dummy send resources to be compatible with HV and PRM */
        qp->sq_no_prefetch = 1;
        qp->sq.wqe_cnt = 1;
        qp->sq.wqe_shift = MLX4_IB_MIN_SQ_STRIDE;
        qp->buf_size = qp->sq.wqe_cnt << MLX4_IB_MIN_SQ_STRIDE;
        qp->mtt = (to_mqp(
                   (struct ib_qp *)init_attr->rwq_ind_tbl->ind_tbl[0]))->mtt;

        qp->rss_ctx = kzalloc(sizeof(*qp->rss_ctx), GFP_KERNEL);
        if (!qp->rss_ctx) {
                err = -ENOMEM;
                goto err_qp_alloc;
        }

        err = set_qp_rss(dev, qp->rss_ctx, init_attr, ucmd);
        if (err)
                goto err;

        return 0;

err:
        kfree(qp->rss_ctx);

err_qp_alloc:
        mlx4_qp_remove(dev->dev, &qp->mqp);
        mlx4_qp_free(dev->dev, &qp->mqp);

err_qpn:
        mlx4_qp_release_range(dev->dev, qpn, 1);
        return err;
}

static int _mlx4_ib_create_qp_rss(struct ib_pd *pd, struct mlx4_ib_qp *qp,
                                  struct ib_qp_init_attr *init_attr,
                                  struct ib_udata *udata)
{
        struct mlx4_ib_create_qp_rss ucmd = {};
        size_t required_cmd_sz;
        int err;

        if (!udata) {
                pr_debug("RSS QP with NULL udata\n");
                return -EINVAL;
        }

        if (udata->outlen)
                return -EOPNOTSUPP;

        required_cmd_sz = offsetof(typeof(ucmd), reserved1) +
                                        sizeof(ucmd.reserved1);
        if (udata->inlen < required_cmd_sz) {
                pr_debug("invalid inlen\n");
                return -EINVAL;
        }

        if (ib_copy_from_udata(&ucmd, udata, min(sizeof(ucmd), udata->inlen))) {
                pr_debug("copy failed\n");
                return -EFAULT;
        }

        if (memchr_inv(ucmd.reserved, 0, sizeof(ucmd.reserved)))
                return -EOPNOTSUPP;

        if (ucmd.comp_mask || ucmd.reserved1)
                return -EOPNOTSUPP;

        if (udata->inlen > sizeof(ucmd) &&
            !ib_is_udata_cleared(udata, sizeof(ucmd),
                                 udata->inlen - sizeof(ucmd))) {
                pr_debug("inlen is not supported\n");
                return -EOPNOTSUPP;
        }

        if (init_attr->qp_type != IB_QPT_RAW_PACKET) {
                pr_debug("RSS QP with unsupported QP type %d\n",
                         init_attr->qp_type);
                return -EOPNOTSUPP;
        }

        if (init_attr->create_flags) {
                pr_debug("RSS QP doesn't support create flags\n");
                return -EOPNOTSUPP;
        }

        if (init_attr->send_cq || init_attr->cap.max_send_wr) {
                pr_debug("RSS QP with unsupported send attributes\n");
                return -EOPNOTSUPP;
        }

        qp->pri.vid = 0xFFFF;
        qp->alt.vid = 0xFFFF;

        err = create_qp_rss(to_mdev(pd->device), init_attr, &ucmd, qp);
        if (err)
                return err;

        qp->ibqp.qp_num = qp->mqp.qpn;
        return 0;
}

/*
 * This function allocates a WQN from a range which is consecutive and aligned
 * to its size. In case the range is full, then it creates a new range and
 * allocates WQN from it. The new range will be used for following allocations.
 */
static int mlx4_ib_alloc_wqn(struct mlx4_ib_ucontext *context,
                             struct mlx4_ib_qp *qp, int range_size, int *wqn)
{
        struct mlx4_ib_dev *dev = to_mdev(context->ibucontext.device);
        struct mlx4_wqn_range *range;
        int err = 0;

        mutex_lock(&context->wqn_ranges_mutex);

        range = list_first_entry_or_null(&context->wqn_ranges_list,
                                         struct mlx4_wqn_range, list);

        if (!range || (range->refcount == range->size) || range->dirty) {
                range = kzalloc(sizeof(*range), GFP_KERNEL);
                if (!range) {
                        err = -ENOMEM;
                        goto out;
                }

                err = mlx4_qp_reserve_range(dev->dev, range_size,
                                            range_size, &range->base_wqn, 0,
                                            qp->mqp.usage);
                if (err) {
                        kfree(range);
                        goto out;
                }

                range->size = range_size;
                list_add(&range->list, &context->wqn_ranges_list);
        } else if (range_size != 1) {
                /*
                 * Requesting a new range (>1) when last range is still open, is
                 * not valid.
                 */
                err = -EINVAL;
                goto out;
        }

        qp->wqn_range = range;

        *wqn = range->base_wqn + range->refcount;

        range->refcount++;

out:
        mutex_unlock(&context->wqn_ranges_mutex);

        return err;
}

static void mlx4_ib_release_wqn(struct mlx4_ib_ucontext *context,
                                struct mlx4_ib_qp *qp, bool dirty_release)
{
        struct mlx4_ib_dev *dev = to_mdev(context->ibucontext.device);
        struct mlx4_wqn_range *range;

        mutex_lock(&context->wqn_ranges_mutex);

        range = qp->wqn_range;

        range->refcount--;
        if (!range->refcount) {
                mlx4_qp_release_range(dev->dev, range->base_wqn,
                                      range->size);
                list_del(&range->list);
                kfree(range);
        } else if (dirty_release) {
        /*
         * A range which one of its WQNs is destroyed, won't be able to be
         * reused for further WQN allocations.
         * The next created WQ will allocate a new range.
         */
                range->dirty = true;
        }

        mutex_unlock(&context->wqn_ranges_mutex);
}

static int create_rq(struct ib_pd *pd, struct ib_qp_init_attr *init_attr,
                     struct ib_udata *udata, struct mlx4_ib_qp *qp)
{
        struct mlx4_ib_dev *dev = to_mdev(pd->device);
        int qpn;
        int err;
        struct mlx4_ib_ucontext *context = rdma_udata_to_drv_context(
                udata, struct mlx4_ib_ucontext, ibucontext);
        struct mlx4_ib_cq *mcq;
        unsigned long flags;
        int range_size;
        struct mlx4_ib_create_wq wq;
        size_t copy_len;
        int shift;
        int n;

        qp->mlx4_ib_qp_type = MLX4_IB_QPT_RAW_PACKET;

        spin_lock_init(&qp->sq.lock);
        spin_lock_init(&qp->rq.lock);
        INIT_LIST_HEAD(&qp->gid_list);
        INIT_LIST_HEAD(&qp->steering_rules);

        qp->state = IB_QPS_RESET;

        copy_len = min(sizeof(struct mlx4_ib_create_wq), udata->inlen);

        if (ib_copy_from_udata(&wq, udata, copy_len)) {
                err = -EFAULT;
                goto err;
        }

        if (wq.comp_mask || wq.reserved[0] || wq.reserved[1] ||
            wq.reserved[2]) {
                pr_debug("user command isn't supported\n");
                err = -EOPNOTSUPP;
                goto err;
        }

        if (wq.log_range_size > ilog2(dev->dev->caps.max_rss_tbl_sz)) {
                pr_debug("WQN range size must be equal or smaller than %d\n",
                         dev->dev->caps.max_rss_tbl_sz);
                err = -EOPNOTSUPP;
                goto err;
        }
        range_size = 1 << wq.log_range_size;

        if (init_attr->create_flags & IB_QP_CREATE_SCATTER_FCS)
                qp->flags |= MLX4_IB_QP_SCATTER_FCS;

        err = set_rq_size(dev, &init_attr->cap, true, true, qp, qp->inl_recv_sz);
        if (err)
                goto err;

        qp->sq_no_prefetch = 1;
        qp->sq.wqe_cnt = 1;
        qp->sq.wqe_shift = MLX4_IB_MIN_SQ_STRIDE;
        qp->buf_size = (qp->rq.wqe_cnt << qp->rq.wqe_shift) +
                       (qp->sq.wqe_cnt << qp->sq.wqe_shift);

        qp->umem = ib_umem_get(pd->device, wq.buf_addr, qp->buf_size, 0);
        if (IS_ERR(qp->umem)) {
                err = PTR_ERR(qp->umem);
                goto err;
        }

        shift = mlx4_ib_umem_calc_optimal_mtt_size(qp->umem, 0, &n);
        err = mlx4_mtt_init(dev->dev, n, shift, &qp->mtt);

        if (err)
                goto err_buf;

        err = mlx4_ib_umem_write_mtt(dev, &qp->mtt, qp->umem);
        if (err)
                goto err_mtt;

        err = mlx4_ib_db_map_user(udata, wq.db_addr, &qp->db);
        if (err)
                goto err_mtt;
        qp->mqp.usage = MLX4_RES_USAGE_USER_VERBS;

        err = mlx4_ib_alloc_wqn(context, qp, range_size, &qpn);
        if (err)
                goto err_wrid;

        err = mlx4_qp_alloc(dev->dev, qpn, &qp->mqp);
        if (err)
                goto err_qpn;

        /*
         * Hardware wants QPN written in big-endian order (after
         * shifting) for send doorbell.  Precompute this value to save
         * a little bit when posting sends.
         */
        qp->doorbell_qpn = swab32(qp->mqp.qpn << 8);

        qp->mqp.event = mlx4_ib_wq_event;

        spin_lock_irqsave(&dev->reset_flow_resource_lock, flags);
        mlx4_ib_lock_cqs(to_mcq(init_attr->send_cq),
                         to_mcq(init_attr->recv_cq));
        /* Maintain device to QPs access, needed for further handling
         * via reset flow
         */
        list_add_tail(&qp->qps_list, &dev->qp_list);
        /* Maintain CQ to QPs access, needed for further handling
         * via reset flow
         */
        mcq = to_mcq(init_attr->send_cq);
        list_add_tail(&qp->cq_send_list, &mcq->send_qp_list);
        mcq = to_mcq(init_attr->recv_cq);
        list_add_tail(&qp->cq_recv_list, &mcq->recv_qp_list);
        mlx4_ib_unlock_cqs(to_mcq(init_attr->send_cq),
                           to_mcq(init_attr->recv_cq));
        spin_unlock_irqrestore(&dev->reset_flow_resource_lock, flags);
        return 0;

err_qpn:
        mlx4_ib_release_wqn(context, qp, 0);
err_wrid:
        mlx4_ib_db_unmap_user(context, &qp->db);

err_mtt:
        mlx4_mtt_cleanup(dev->dev, &qp->mtt);
err_buf:
        ib_umem_release(qp->umem);
err:
        return err;
}

static int create_qp_common(struct ib_pd *pd, struct ib_qp_init_attr *init_attr,
                            struct ib_udata *udata, int sqpn,
                            struct mlx4_ib_qp *qp)
{
        struct mlx4_ib_dev *dev = to_mdev(pd->device);
        int qpn;
        int err;
        struct mlx4_ib_ucontext *context = rdma_udata_to_drv_context(
                udata, struct mlx4_ib_ucontext, ibucontext);
        enum mlx4_ib_qp_type qp_type = (enum mlx4_ib_qp_type) init_attr->qp_type;
        struct mlx4_ib_cq *mcq;
        unsigned long flags;

        /* When tunneling special qps, we use a plain UD qp */
        if (sqpn) {
                if (mlx4_is_mfunc(dev->dev) &&
                    (!mlx4_is_master(dev->dev) ||
                     !(init_attr->create_flags & MLX4_IB_SRIOV_SQP))) {
                        if (init_attr->qp_type == IB_QPT_GSI)
                                qp_type = MLX4_IB_QPT_PROXY_GSI;
                        else {
                                if (mlx4_is_master(dev->dev) ||
                                    qp0_enabled_vf(dev->dev, sqpn))
                                        qp_type = MLX4_IB_QPT_PROXY_SMI_OWNER;
                                else
                                        qp_type = MLX4_IB_QPT_PROXY_SMI;
                        }
                }
                qpn = sqpn;
                /* add extra sg entry for tunneling */
                init_attr->cap.max_recv_sge++;
        } else if (init_attr->create_flags & MLX4_IB_SRIOV_TUNNEL_QP) {
                struct mlx4_ib_qp_tunnel_init_attr *tnl_init =
                        container_of(init_attr,
                                     struct mlx4_ib_qp_tunnel_init_attr, init_attr);
                if ((tnl_init->proxy_qp_type != IB_QPT_SMI &&
                     tnl_init->proxy_qp_type != IB_QPT_GSI)   ||
                    !mlx4_is_master(dev->dev))
                        return -EINVAL;
                if (tnl_init->proxy_qp_type == IB_QPT_GSI)
                        qp_type = MLX4_IB_QPT_TUN_GSI;
                else if (tnl_init->slave == mlx4_master_func_num(dev->dev) ||
                         mlx4_vf_smi_enabled(dev->dev, tnl_init->slave,
                                             tnl_init->port))
                        qp_type = MLX4_IB_QPT_TUN_SMI_OWNER;
                else
                        qp_type = MLX4_IB_QPT_TUN_SMI;
                /* we are definitely in the PPF here, since we are creating
                 * tunnel QPs. base_tunnel_sqpn is therefore valid. */

                qpn = dev->dev->phys_caps.base_tunnel_sqpn + 8 * tnl_init->slave
                        + tnl_init->proxy_qp_type * 2 + tnl_init->port - 1;
                sqpn = qpn;
        }

        if (init_attr->qp_type == IB_QPT_SMI ||
            init_attr->qp_type == IB_QPT_GSI || qp_type == MLX4_IB_QPT_SMI ||
            qp_type == MLX4_IB_QPT_GSI ||
            (qp_type & (MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_SMI_OWNER |
                        MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER))) {
                qp->sqp = kzalloc(sizeof(struct mlx4_ib_sqp), GFP_KERNEL);
                if (!qp->sqp)
                        return -ENOMEM;
        }

        qp->mlx4_ib_qp_type = qp_type;

        spin_lock_init(&qp->sq.lock);
        spin_lock_init(&qp->rq.lock);
        INIT_LIST_HEAD(&qp->gid_list);
        INIT_LIST_HEAD(&qp->steering_rules);

        qp->state = IB_QPS_RESET;
        if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
                qp->sq_signal_bits = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE);

        if (udata) {
                struct mlx4_ib_create_qp ucmd;
                size_t copy_len;
                int shift;
                int n;

                copy_len = sizeof(struct mlx4_ib_create_qp);

                if (ib_copy_from_udata(&ucmd, udata, copy_len)) {
                        err = -EFAULT;
                        goto err;
                }

                qp->inl_recv_sz = ucmd.inl_recv_sz;

                if (init_attr->create_flags & IB_QP_CREATE_SCATTER_FCS) {
                        if (!(dev->dev->caps.flags &
                              MLX4_DEV_CAP_FLAG_FCS_KEEP)) {
                                pr_debug("scatter FCS is unsupported\n");
                                err = -EOPNOTSUPP;
                                goto err;
                        }

                        qp->flags |= MLX4_IB_QP_SCATTER_FCS;
                }

                err = set_rq_size(dev, &init_attr->cap, udata,
                                  qp_has_rq(init_attr), qp, qp->inl_recv_sz);
                if (err)
                        goto err;

                qp->sq_no_prefetch = ucmd.sq_no_prefetch;

                err = set_user_sq_size(dev, qp, &ucmd);
                if (err)
                        goto err;

                qp->umem =
                        ib_umem_get(pd->device, ucmd.buf_addr, qp->buf_size, 0);
                if (IS_ERR(qp->umem)) {
                        err = PTR_ERR(qp->umem);
                        goto err;
                }

                shift = mlx4_ib_umem_calc_optimal_mtt_size(qp->umem, 0, &n);
                err = mlx4_mtt_init(dev->dev, n, shift, &qp->mtt);

                if (err)
                        goto err_buf;

                err = mlx4_ib_umem_write_mtt(dev, &qp->mtt, qp->umem);
                if (err)
                        goto err_mtt;

                if (qp_has_rq(init_attr)) {
                        err = mlx4_ib_db_map_user(udata, ucmd.db_addr, &qp->db);
                        if (err)
                                goto err_mtt;
                }
                qp->mqp.usage = MLX4_RES_USAGE_USER_VERBS;
        } else {
                err = set_rq_size(dev, &init_attr->cap, udata,
                                  qp_has_rq(init_attr), qp, 0);
                if (err)
                        goto err;

                qp->sq_no_prefetch = 0;

                if (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO)
                        qp->flags |= MLX4_IB_QP_LSO;

                if (init_attr->create_flags & IB_QP_CREATE_NETIF_QP) {
                        if (dev->steering_support ==
                            MLX4_STEERING_MODE_DEVICE_MANAGED)
                                qp->flags |= MLX4_IB_QP_NETIF;
                        else
                                goto err;
                }

                err = set_kernel_sq_size(dev, &init_attr->cap, qp_type, qp);
                if (err)
                        goto err;

                if (qp_has_rq(init_attr)) {
                        err = mlx4_db_alloc(dev->dev, &qp->db, 0);
                        if (err)
                                goto err;

                        *qp->db.db = 0;
                }

                if (mlx4_buf_alloc(dev->dev, qp->buf_size,  PAGE_SIZE * 2,
                                   &qp->buf)) {
                        err = -ENOMEM;
                        goto err_db;
                }

                err = mlx4_mtt_init(dev->dev, qp->buf.npages, qp->buf.page_shift,
                                    &qp->mtt);
                if (err)
                        goto err_buf;

                err = mlx4_buf_write_mtt(dev->dev, &qp->mtt, &qp->buf);
                if (err)
                        goto err_mtt;

                qp->sq.wrid = kvmalloc_array(qp->sq.wqe_cnt,
                                             sizeof(u64), GFP_KERNEL);
                qp->rq.wrid = kvmalloc_array(qp->rq.wqe_cnt,
                                             sizeof(u64), GFP_KERNEL);
                if (!qp->sq.wrid || !qp->rq.wrid) {
                        err = -ENOMEM;
                        goto err_wrid;
                }
                qp->mqp.usage = MLX4_RES_USAGE_DRIVER;
        }

        if (sqpn) {
                if (qp->mlx4_ib_qp_type & (MLX4_IB_QPT_PROXY_SMI_OWNER |
                    MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI)) {
                        if (alloc_proxy_bufs(pd->device, qp)) {
                                err = -ENOMEM;
                                goto err_wrid;
                        }
                }
        } else {
                /* Raw packet QPNs may not have bits 6,7 set in their qp_num;
                 * otherwise, the WQE BlueFlame setup flow wrongly causes
                 * VLAN insertion. */
                if (init_attr->qp_type == IB_QPT_RAW_PACKET)
                        err = mlx4_qp_reserve_range(dev->dev, 1, 1, &qpn,
                                                    (init_attr->cap.max_send_wr ?
                                                     MLX4_RESERVE_ETH_BF_QP : 0) |
                                                    (init_attr->cap.max_recv_wr ?
                                                     MLX4_RESERVE_A0_QP : 0),
                                                    qp->mqp.usage);
                else
                        if (qp->flags & MLX4_IB_QP_NETIF)
                                err = mlx4_ib_steer_qp_alloc(dev, 1, &qpn);
                        else
                                err = mlx4_qp_reserve_range(dev->dev, 1, 1,
                                                            &qpn, 0, qp->mqp.usage);
                if (err)
                        goto err_proxy;
        }

        if (init_attr->create_flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK)
                qp->flags |= MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK;

        err = mlx4_qp_alloc(dev->dev, qpn, &qp->mqp);
        if (err)
                goto err_qpn;

        if (init_attr->qp_type == IB_QPT_XRC_TGT)
                qp->mqp.qpn |= (1 << 23);

        /*
         * Hardware wants QPN written in big-endian order (after
         * shifting) for send doorbell.  Precompute this value to save
         * a little bit when posting sends.
         */
        qp->doorbell_qpn = swab32(qp->mqp.qpn << 8);

        qp->mqp.event = mlx4_ib_qp_event;

        spin_lock_irqsave(&dev->reset_flow_resource_lock, flags);
        mlx4_ib_lock_cqs(to_mcq(init_attr->send_cq),
                         to_mcq(init_attr->recv_cq));
        /* Maintain device to QPs access, needed for further handling
         * via reset flow
         */
        list_add_tail(&qp->qps_list, &dev->qp_list);
        /* Maintain CQ to QPs access, needed for further handling
         * via reset flow
         */
        mcq = to_mcq(init_attr->send_cq);
        list_add_tail(&qp->cq_send_list, &mcq->send_qp_list);
        mcq = to_mcq(init_attr->recv_cq);
        list_add_tail(&qp->cq_recv_list, &mcq->recv_qp_list);
        mlx4_ib_unlock_cqs(to_mcq(init_attr->send_cq),
                           to_mcq(init_attr->recv_cq));
        spin_unlock_irqrestore(&dev->reset_flow_resource_lock, flags);
        return 0;

err_qpn:
        if (!sqpn) {
                if (qp->flags & MLX4_IB_QP_NETIF)
                        mlx4_ib_steer_qp_free(dev, qpn, 1);
                else
                        mlx4_qp_release_range(dev->dev, qpn, 1);
        }
err_proxy:
        if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_GSI)
                free_proxy_bufs(pd->device, qp);
err_wrid:
        if (udata) {
                if (qp_has_rq(init_attr))
                        mlx4_ib_db_unmap_user(context, &qp->db);
        } else {
                kvfree(qp->sq.wrid);
                kvfree(qp->rq.wrid);
        }

err_mtt:
        mlx4_mtt_cleanup(dev->dev, &qp->mtt);

err_buf:
        if (!qp->umem)
                mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf);
        ib_umem_release(qp->umem);

err_db:
        if (!udata && qp_has_rq(init_attr))
                mlx4_db_free(dev->dev, &qp->db);

err:
        kfree(qp->sqp);
        return err;
}

static enum mlx4_qp_state to_mlx4_state(enum ib_qp_state state)
{
        switch (state) {
        case IB_QPS_RESET:      return MLX4_QP_STATE_RST;
        case IB_QPS_INIT:       return MLX4_QP_STATE_INIT;
        case IB_QPS_RTR:        return MLX4_QP_STATE_RTR;
        case IB_QPS_RTS:        return MLX4_QP_STATE_RTS;
        case IB_QPS_SQD:        return MLX4_QP_STATE_SQD;
        case IB_QPS_SQE:        return MLX4_QP_STATE_SQER;
        case IB_QPS_ERR:        return MLX4_QP_STATE_ERR;
        default:                return -1;
        }
}

static void mlx4_ib_lock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *recv_cq)
        __acquires(&send_cq->lock) __acquires(&recv_cq->lock)
{
        if (send_cq == recv_cq) {
                spin_lock(&send_cq->lock);
                __acquire(&recv_cq->lock);
        } else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
                spin_lock(&send_cq->lock);
                spin_lock_nested(&recv_cq->lock, SINGLE_DEPTH_NESTING);
        } else {
                spin_lock(&recv_cq->lock);
                spin_lock_nested(&send_cq->lock, SINGLE_DEPTH_NESTING);
        }
}

static void mlx4_ib_unlock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *recv_cq)
        __releases(&send_cq->lock) __releases(&recv_cq->lock)
{
        if (send_cq == recv_cq) {
                __release(&recv_cq->lock);
                spin_unlock(&send_cq->lock);
        } else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
                spin_unlock(&recv_cq->lock);
                spin_unlock(&send_cq->lock);
        } else {
                spin_unlock(&send_cq->lock);
                spin_unlock(&recv_cq->lock);
        }
}

static void del_gid_entries(struct mlx4_ib_qp *qp)
{
        struct mlx4_ib_gid_entry *ge, *tmp;

        list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
                list_del(&ge->list);
                kfree(ge);
        }
}

static struct mlx4_ib_pd *get_pd(struct mlx4_ib_qp *qp)
{
        if (qp->ibqp.qp_type == IB_QPT_XRC_TGT)
                return to_mpd(to_mxrcd(qp->ibqp.xrcd)->pd);
        else
                return to_mpd(qp->ibqp.pd);
}

static void get_cqs(struct mlx4_ib_qp *qp, enum mlx4_ib_source_type src,
                    struct mlx4_ib_cq **send_cq, struct mlx4_ib_cq **recv_cq)
{
        switch (qp->ibqp.qp_type) {
        case IB_QPT_XRC_TGT:
                *send_cq = to_mcq(to_mxrcd(qp->ibqp.xrcd)->cq);
                *recv_cq = *send_cq;
                break;
        case IB_QPT_XRC_INI:
                *send_cq = to_mcq(qp->ibqp.send_cq);
                *recv_cq = *send_cq;
                break;
        default:
                *recv_cq = (src == MLX4_IB_QP_SRC) ? to_mcq(qp->ibqp.recv_cq) :
                                                     to_mcq(qp->ibwq.cq);
                *send_cq = (src == MLX4_IB_QP_SRC) ? to_mcq(qp->ibqp.send_cq) :
                                                     *recv_cq;
                break;
        }
}

static void destroy_qp_rss(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
{
        if (qp->state != IB_QPS_RESET) {
                int i;

                for (i = 0; i < (1 << qp->ibqp.rwq_ind_tbl->log_ind_tbl_size);
                     i++) {
                        struct ib_wq *ibwq = qp->ibqp.rwq_ind_tbl->ind_tbl[i];
                        struct mlx4_ib_qp *wq = to_mqp((struct ib_qp *)ibwq);

                        mutex_lock(&wq->mutex);

                        wq->rss_usecnt--;

                        mutex_unlock(&wq->mutex);
                }

                if (mlx4_qp_modify(dev->dev, NULL, to_mlx4_state(qp->state),
                                   MLX4_QP_STATE_RST, NULL, 0, 0, &qp->mqp))
                        pr_warn("modify QP %06x to RESET failed.\n",
                                qp->mqp.qpn);
        }

        mlx4_qp_remove(dev->dev, &qp->mqp);
        mlx4_qp_free(dev->dev, &qp->mqp);
        mlx4_qp_release_range(dev->dev, qp->mqp.qpn, 1);
        del_gid_entries(qp);
}

static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
                              enum mlx4_ib_source_type src,
                              struct ib_udata *udata)
{
        struct mlx4_ib_cq *send_cq, *recv_cq;
        unsigned long flags;

        if (qp->state != IB_QPS_RESET) {
                if (mlx4_qp_modify(dev->dev, NULL, to_mlx4_state(qp->state),
                                   MLX4_QP_STATE_RST, NULL, 0, 0, &qp->mqp))
                        pr_warn("modify QP %06x to RESET failed.\n",
                               qp->mqp.qpn);
                if (qp->pri.smac || (!qp->pri.smac && qp->pri.smac_port)) {
                        mlx4_unregister_mac(dev->dev, qp->pri.smac_port, qp->pri.smac);
                        qp->pri.smac = 0;
                        qp->pri.smac_port = 0;
                }
                if (qp->alt.smac) {
                        mlx4_unregister_mac(dev->dev, qp->alt.smac_port, qp->alt.smac);
                        qp->alt.smac = 0;
                }
                if (qp->pri.vid < 0x1000) {
                        mlx4_unregister_vlan(dev->dev, qp->pri.vlan_port, qp->pri.vid);
                        qp->pri.vid = 0xFFFF;
                        qp->pri.candidate_vid = 0xFFFF;
                        qp->pri.update_vid = 0;
                }
                if (qp->alt.vid < 0x1000) {
                        mlx4_unregister_vlan(dev->dev, qp->alt.vlan_port, qp->alt.vid);
                        qp->alt.vid = 0xFFFF;
                        qp->alt.candidate_vid = 0xFFFF;
                        qp->alt.update_vid = 0;
                }
        }

        get_cqs(qp, src, &send_cq, &recv_cq);

        spin_lock_irqsave(&dev->reset_flow_resource_lock, flags);
        mlx4_ib_lock_cqs(send_cq, recv_cq);

        /* del from lists under both locks above to protect reset flow paths */
        list_del(&qp->qps_list);
        list_del(&qp->cq_send_list);
        list_del(&qp->cq_recv_list);
        if (!udata) {
                __mlx4_ib_cq_clean(recv_cq, qp->mqp.qpn,
                                 qp->ibqp.srq ? to_msrq(qp->ibqp.srq): NULL);
                if (send_cq != recv_cq)
                        __mlx4_ib_cq_clean(send_cq, qp->mqp.qpn, NULL);
        }

        mlx4_qp_remove(dev->dev, &qp->mqp);

        mlx4_ib_unlock_cqs(send_cq, recv_cq);
        spin_unlock_irqrestore(&dev->reset_flow_resource_lock, flags);

        mlx4_qp_free(dev->dev, &qp->mqp);

        if (!is_sqp(dev, qp) && !is_tunnel_qp(dev, qp)) {
                if (qp->flags & MLX4_IB_QP_NETIF)
                        mlx4_ib_steer_qp_free(dev, qp->mqp.qpn, 1);
                else if (src == MLX4_IB_RWQ_SRC)
                        mlx4_ib_release_wqn(
                                rdma_udata_to_drv_context(
                                        udata,
                                        struct mlx4_ib_ucontext,
                                        ibucontext),
                                qp, 1);
                else
                        mlx4_qp_release_range(dev->dev, qp->mqp.qpn, 1);
        }

        mlx4_mtt_cleanup(dev->dev, &qp->mtt);

        if (udata) {
                if (qp->rq.wqe_cnt) {
                        struct mlx4_ib_ucontext *mcontext =
                                rdma_udata_to_drv_context(
                                        udata,
                                        struct mlx4_ib_ucontext,
                                        ibucontext);

                        mlx4_ib_db_unmap_user(mcontext, &qp->db);
                }
        } else {
                kvfree(qp->sq.wrid);
                kvfree(qp->rq.wrid);
                if (qp->mlx4_ib_qp_type & (MLX4_IB_QPT_PROXY_SMI_OWNER |
                    MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI))
                        free_proxy_bufs(&dev->ib_dev, qp);
                mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf);
                if (qp->rq.wqe_cnt)
                        mlx4_db_free(dev->dev, &qp->db);
        }
        ib_umem_release(qp->umem);

        del_gid_entries(qp);
}

static u32 get_sqp_num(struct mlx4_ib_dev *dev, struct ib_qp_init_attr *attr)
{
        /* Native or PPF */
        if (!mlx4_is_mfunc(dev->dev) ||
            (mlx4_is_master(dev->dev) &&
             attr->create_flags & MLX4_IB_SRIOV_SQP)) {
                return  dev->dev->phys_caps.base_sqpn +
                        (attr->qp_type == IB_QPT_SMI ? 0 : 2) +
                        attr->port_num - 1;
        }
        /* PF or VF -- creating proxies */
        if (attr->qp_type == IB_QPT_SMI)
                return dev->dev->caps.spec_qps[attr->port_num - 1].qp0_proxy;
        else
                return dev->dev->caps.spec_qps[attr->port_num - 1].qp1_proxy;
}

static int _mlx4_ib_create_qp(struct ib_pd *pd, struct mlx4_ib_qp *qp,
                              struct ib_qp_init_attr *init_attr,
                              struct ib_udata *udata)
{
        int err;
        int sup_u_create_flags = MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK;
        u16 xrcdn = 0;

        if (init_attr->rwq_ind_tbl)
                return _mlx4_ib_create_qp_rss(pd, qp, init_attr, udata);

        /*
         * We only support LSO, vendor flag1, and multicast loopback blocking,
         * and only for kernel UD QPs.
         */
        if (init_attr->create_flags & ~(MLX4_IB_QP_LSO |
                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK |
                                        MLX4_IB_SRIOV_TUNNEL_QP |
                                        MLX4_IB_SRIOV_SQP |
                                        MLX4_IB_QP_NETIF |
                                        MLX4_IB_QP_CREATE_ROCE_V2_GSI))
                return -EOPNOTSUPP;

        if (init_attr->create_flags & IB_QP_CREATE_NETIF_QP) {
                if (init_attr->qp_type != IB_QPT_UD)
                        return -EINVAL;
        }

        if (init_attr->create_flags) {
                if (udata && init_attr->create_flags & ~(sup_u_create_flags))
                        return -EINVAL;

                if ((init_attr->create_flags & ~(MLX4_IB_SRIOV_SQP |
                                                 MLX4_IB_QP_CREATE_ROCE_V2_GSI  |
                                                 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK) &&
                     init_attr->qp_type != IB_QPT_UD) ||
                    (init_attr->create_flags & MLX4_IB_SRIOV_SQP &&
                     init_attr->qp_type > IB_QPT_GSI) ||
                    (init_attr->create_flags & MLX4_IB_QP_CREATE_ROCE_V2_GSI &&
                     init_attr->qp_type != IB_QPT_GSI))
                        return -EINVAL;
        }

        switch (init_attr->qp_type) {
        case IB_QPT_XRC_TGT:
                pd = to_mxrcd(init_attr->xrcd)->pd;
                xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
                init_attr->send_cq = to_mxrcd(init_attr->xrcd)->cq;
                fallthrough;
        case IB_QPT_XRC_INI:
                if (!(to_mdev(pd->device)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
                        return -ENOSYS;
                init_attr->recv_cq = init_attr->send_cq;
                fallthrough;
        case IB_QPT_RC:
        case IB_QPT_UC:
        case IB_QPT_RAW_PACKET:
        case IB_QPT_UD:
                qp->pri.vid = 0xFFFF;
                qp->alt.vid = 0xFFFF;
                err = create_qp_common(pd, init_attr, udata, 0, qp);
                if (err)
                        return err;

                qp->ibqp.qp_num = qp->mqp.qpn;
                qp->xrcdn = xrcdn;
                break;
        case IB_QPT_SMI:
        case IB_QPT_GSI:
        {
                int sqpn;

                if (init_attr->create_flags & MLX4_IB_QP_CREATE_ROCE_V2_GSI) {
                        int res = mlx4_qp_reserve_range(to_mdev(pd->device)->dev,
                                                        1, 1, &sqpn, 0,
                                                        MLX4_RES_USAGE_DRIVER);

                        if (res)
                                return res;
                } else {
                        sqpn = get_sqp_num(to_mdev(pd->device), init_attr);
                }

                qp->pri.vid = 0xFFFF;
                qp->alt.vid = 0xFFFF;
                err = create_qp_common(pd, init_attr, udata, sqpn, qp);
                if (err)
                        return err;

                if (init_attr->create_flags &
                    (MLX4_IB_SRIOV_SQP | MLX4_IB_SRIOV_TUNNEL_QP))
                        /* Internal QP created with ib_create_qp */
                        rdma_restrack_no_track(&qp->ibqp.res);

                qp->port        = init_attr->port_num;
                qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 :
                        init_attr->create_flags & MLX4_IB_QP_CREATE_ROCE_V2_GSI ? sqpn : 1;
                break;
        }
        default:
                /* Don't support raw QPs */
                return -EOPNOTSUPP;
        }
        return 0;
}

int mlx4_ib_create_qp(struct ib_qp *ibqp, struct ib_qp_init_attr *init_attr,
                      struct ib_udata *udata)
{
        struct ib_device *device = ibqp->device;
        struct mlx4_ib_dev *dev = to_mdev(device);
        struct mlx4_ib_qp *qp = to_mqp(ibqp);
        struct ib_pd *pd = ibqp->pd;
        int ret;

        mutex_init(&qp->mutex);
        ret = _mlx4_ib_create_qp(pd, qp, init_attr, udata);
        if (ret)
                return ret;

        if (init_attr->qp_type == IB_QPT_GSI &&
            !(init_attr->create_flags & MLX4_IB_QP_CREATE_ROCE_V2_GSI)) {
                struct mlx4_ib_sqp *sqp = qp->sqp;
                int is_eth = rdma_cap_eth_ah(&dev->ib_dev, init_attr->port_num);

                if (is_eth &&
                    dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
                        init_attr->create_flags |= MLX4_IB_QP_CREATE_ROCE_V2_GSI;
                        sqp->roce_v2_gsi = ib_create_qp(pd, init_attr);

                        if (IS_ERR(sqp->roce_v2_gsi)) {
                                pr_err("Failed to create GSI QP for RoCEv2 (%ld)\n", PTR_ERR(sqp->roce_v2_gsi));
                                sqp->roce_v2_gsi = NULL;
                        } else {
                                to_mqp(sqp->roce_v2_gsi)->flags |=
                                        MLX4_IB_ROCE_V2_GSI_QP;
                        }

                        init_attr->create_flags &= ~MLX4_IB_QP_CREATE_ROCE_V2_GSI;
                }
        }
        return 0;
}

static int _mlx4_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
{
        struct mlx4_ib_dev *dev = to_mdev(qp->device);
        struct mlx4_ib_qp *mqp = to_mqp(qp);

        if (is_qp0(dev, mqp))
                mlx4_CLOSE_PORT(dev->dev, mqp->port);

        if (mqp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_GSI &&
            dev->qp1_proxy[mqp->port - 1] == mqp) {
                mutex_lock(&dev->qp1_proxy_lock[mqp->port - 1]);
                dev->qp1_proxy[mqp->port - 1] = NULL;
                mutex_unlock(&dev->qp1_proxy_lock[mqp->port - 1]);
        }

        if (mqp->counter_index)
                mlx4_ib_free_qp_counter(dev, mqp);

        if (qp->rwq_ind_tbl) {
                destroy_qp_rss(dev, mqp);
        } else {
                destroy_qp_common(dev, mqp, MLX4_IB_QP_SRC, udata);
        }

        kfree(mqp->sqp);
        return 0;
}

int mlx4_ib_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
{
        struct mlx4_ib_qp *mqp = to_mqp(qp);

        if (mqp->mlx4_ib_qp_type == MLX4_IB_QPT_GSI) {
                struct mlx4_ib_sqp *sqp = mqp->sqp;

                if (sqp->roce_v2_gsi)
                        ib_destroy_qp(sqp->roce_v2_gsi);
        }

        return _mlx4_ib_destroy_qp(qp, udata);
}

static int to_mlx4_st(struct mlx4_ib_dev *dev, enum mlx4_ib_qp_type type)
{
        switch (type) {
        case MLX4_IB_QPT_RC:            return MLX4_QP_ST_RC;
        case MLX4_IB_QPT_UC:            return MLX4_QP_ST_UC;
        case MLX4_IB_QPT_UD:            return MLX4_QP_ST_UD;
        case MLX4_IB_QPT_XRC_INI:
        case MLX4_IB_QPT_XRC_TGT:       return MLX4_QP_ST_XRC;
        case MLX4_IB_QPT_SMI:
        case MLX4_IB_QPT_GSI:
        case MLX4_IB_QPT_RAW_PACKET:    return MLX4_QP_ST_MLX;

        case MLX4_IB_QPT_PROXY_SMI_OWNER:
        case MLX4_IB_QPT_TUN_SMI_OWNER: return (mlx4_is_mfunc(dev->dev) ?
                                                MLX4_QP_ST_MLX : -1);
        case MLX4_IB_QPT_PROXY_SMI:
        case MLX4_IB_QPT_TUN_SMI:
        case MLX4_IB_QPT_PROXY_GSI:
        case MLX4_IB_QPT_TUN_GSI:       return (mlx4_is_mfunc(dev->dev) ?
                                                MLX4_QP_ST_UD : -1);
        default:                        return -1;
        }
}

static __be32 to_mlx4_access_flags(struct mlx4_ib_qp *qp, const struct ib_qp_attr *attr,
                                   int attr_mask)
{
        u8 dest_rd_atomic;
        u32 access_flags;
        u32 hw_access_flags = 0;

        if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
                dest_rd_atomic = attr->max_dest_rd_atomic;
        else
                dest_rd_atomic = qp->resp_depth;

        if (attr_mask & IB_QP_ACCESS_FLAGS)
                access_flags = attr->qp_access_flags;
        else
                access_flags = qp->atomic_rd_en;

        if (!dest_rd_atomic)
                access_flags &= IB_ACCESS_REMOTE_WRITE;

        if (access_flags & IB_ACCESS_REMOTE_READ)
                hw_access_flags |= MLX4_QP_BIT_RRE;
        if (access_flags & IB_ACCESS_REMOTE_ATOMIC)
                hw_access_flags |= MLX4_QP_BIT_RAE;
        if (access_flags & IB_ACCESS_REMOTE_WRITE)
                hw_access_flags |= MLX4_QP_BIT_RWE;

        return cpu_to_be32(hw_access_flags);
}

static void store_sqp_attrs(struct mlx4_ib_sqp *sqp, const struct ib_qp_attr *attr,
                            int attr_mask)
{
        if (attr_mask & IB_QP_PKEY_INDEX)
                sqp->pkey_index = attr->pkey_index;
        if (attr_mask & IB_QP_QKEY)
                sqp->qkey = attr->qkey;
        if (attr_mask & IB_QP_SQ_PSN)
                sqp->send_psn = attr->sq_psn;
}

static void mlx4_set_sched(struct mlx4_qp_path *path, u8 port)
{
        path->sched_queue = (path->sched_queue & 0xbf) | ((port - 1) << 6);
}

static int _mlx4_set_path(struct mlx4_ib_dev *dev,
                          const struct rdma_ah_attr *ah,
                          u64 smac, u16 vlan_tag, struct mlx4_qp_path *path,
                          struct mlx4_roce_smac_vlan_info *smac_info, u8 port)
{
        int vidx;
        int smac_index;
        int err;

        path->grh_mylmc = rdma_ah_get_path_bits(ah) & 0x7f;
        path->rlid = cpu_to_be16(rdma_ah_get_dlid(ah));
        if (rdma_ah_get_static_rate(ah)) {
                path->static_rate = rdma_ah_get_static_rate(ah) +
                                    MLX4_STAT_RATE_OFFSET;
                while (path->static_rate > IB_RATE_2_5_GBPS + MLX4_STAT_RATE_OFFSET &&
                       !(1 << path->static_rate & dev->dev->caps.stat_rate_support))
                        --path->static_rate;
        } else
                path->static_rate = 0;

        if (rdma_ah_get_ah_flags(ah) & IB_AH_GRH) {
                const struct ib_global_route *grh = rdma_ah_read_grh(ah);
                int real_sgid_index =
                        mlx4_ib_gid_index_to_real_index(dev, grh->sgid_attr);

                if (real_sgid_index < 0)
                        return real_sgid_index;
                if (real_sgid_index >= dev->dev->caps.gid_table_len[port]) {
                        pr_err("sgid_index (%u) too large. max is %d\n",
                               real_sgid_index, dev->dev->caps.gid_table_len[port] - 1);
                        return -1;
                }

                path->grh_mylmc |= 1 << 7;
                path->mgid_index = real_sgid_index;
                path->hop_limit  = grh->hop_limit;
                path->tclass_flowlabel =
                        cpu_to_be32((grh->traffic_class << 20) |
                                    (grh->flow_label));
                memcpy(path->rgid, grh->dgid.raw, 16);
        }

        if (ah->type == RDMA_AH_ATTR_TYPE_ROCE) {
                if (!(rdma_ah_get_ah_flags(ah) & IB_AH_GRH))
                        return -1;

                path->sched_queue = MLX4_IB_DEFAULT_SCHED_QUEUE |
                        ((port - 1) << 6) | ((rdma_ah_get_sl(ah) & 7) << 3);

                path->feup |= MLX4_FEUP_FORCE_ETH_UP;
                if (vlan_tag < 0x1000) {
                        if (smac_info->vid < 0x1000) {
                                /* both valid vlan ids */
                                if (smac_info->vid != vlan_tag) {
                                        /* different VIDs.  unreg old and reg new */
                                        err = mlx4_register_vlan(dev->dev, port, vlan_tag, &vidx);
                                        if (err)
                                                return err;
                                        smac_info->candidate_vid = vlan_tag;
                                        smac_info->candidate_vlan_index = vidx;
                                        smac_info->candidate_vlan_port = port;
                                        smac_info->update_vid = 1;
                                        path->vlan_index = vidx;
                                } else {
                                        path->vlan_index = smac_info->vlan_index;
                                }
                        } else {
                                /* no current vlan tag in qp */
                                err = mlx4_register_vlan(dev->dev, port, vlan_tag, &vidx);
                                if (err)
                                        return err;
                                smac_info->candidate_vid = vlan_tag;
                                smac_info->candidate_vlan_index = vidx;
                                smac_info->candidate_vlan_port = port;
                                smac_info->update_vid = 1;
                                path->vlan_index = vidx;
                        }
                        path->feup |= MLX4_FVL_FORCE_ETH_VLAN;
                        path->fl = 1 << 6;
                } else {
                        /* have current vlan tag. unregister it at modify-qp success */
                        if (smac_info->vid < 0x1000) {
                                smac_info->candidate_vid = 0xFFFF;
                                smac_info->update_vid = 1;
                        }
                }

                /* get smac_index for RoCE use.
                 * If no smac was yet assigned, register one.
                 * If one was already assigned, but the new mac differs,
                 * unregister the old one and register the new one.
                */
                if ((!smac_info->smac && !smac_info->smac_port) ||
                    smac_info->smac != smac) {
                        /* register candidate now, unreg if needed, after success */
                        smac_index = mlx4_register_mac(dev->dev, port, smac);
                        if (smac_index >= 0) {
                                smac_info->candidate_smac_index = smac_index;
                                smac_info->candidate_smac = smac;
                                smac_info->candidate_smac_port = port;
                        } else {
                                return -EINVAL;
                        }
                } else {
                        smac_index = smac_info->smac_index;
                }
                memcpy(path->dmac, ah->roce.dmac, 6);
                path->ackto = MLX4_IB_LINK_TYPE_ETH;
                /* put MAC table smac index for IBoE */
                path->grh_mylmc = (u8) (smac_index) | 0x80;
        } else {
                path->sched_queue = MLX4_IB_DEFAULT_SCHED_QUEUE |
                        ((port - 1) << 6) | ((rdma_ah_get_sl(ah) & 0xf) << 2);
        }

        return 0;
}

static int mlx4_set_path(struct mlx4_ib_dev *dev, const struct ib_qp_attr *qp,
                         enum ib_qp_attr_mask qp_attr_mask,
                         struct mlx4_ib_qp *mqp,
                         struct mlx4_qp_path *path, u8 port,
                         u16 vlan_id, u8 *smac)
{
        return _mlx4_set_path(dev, &qp->ah_attr,
                              mlx4_mac_to_u64(smac),
                              vlan_id,
                              path, &mqp->pri, port);
}

static int mlx4_set_alt_path(struct mlx4_ib_dev *dev,
                             const struct ib_qp_attr *qp,
                             enum ib_qp_attr_mask qp_attr_mask,
                             struct mlx4_ib_qp *mqp,
                             struct mlx4_qp_path *path, u8 port)
{
        return _mlx4_set_path(dev, &qp->alt_ah_attr,
                              0,
                              0xffff,
                              path, &mqp->alt, port);
}

static void update_mcg_macs(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
{
        struct mlx4_ib_gid_entry *ge, *tmp;

        list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
                if (!ge->added && mlx4_ib_add_mc(dev, qp, &ge->gid)) {
                        ge->added = 1;
                        ge->port = qp->port;
                }
        }
}

static int handle_eth_ud_smac_index(struct mlx4_ib_dev *dev,
                                    struct mlx4_ib_qp *qp,
                                    struct mlx4_qp_context *context)
{
        u64 u64_mac;
        int smac_index;

        u64_mac = atomic64_read(&dev->iboe.mac[qp->port - 1]);

        context->pri_path.sched_queue = MLX4_IB_DEFAULT_SCHED_QUEUE | ((qp->port - 1) << 6);
        if (!qp->pri.smac && !qp->pri.smac_port) {
                smac_index = mlx4_register_mac(dev->dev, qp->port, u64_mac);
                if (smac_index >= 0) {
                        qp->pri.candidate_smac_index = smac_index;
                        qp->pri.candidate_smac = u64_mac;
                        qp->pri.candidate_smac_port = qp->port;
                        context->pri_path.grh_mylmc = 0x80 | (u8) smac_index;
                } else {
                        return -ENOENT;
                }
        }
        return 0;
}

static int create_qp_lb_counter(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
{
        struct counter_index *new_counter_index;
        int err;
        u32 tmp_idx;

        if (rdma_port_get_link_layer(&dev->ib_dev, qp->port) !=
            IB_LINK_LAYER_ETHERNET ||
            !(qp->flags & MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK) ||
            !(dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_LB_SRC_CHK))
                return 0;

        err = mlx4_counter_alloc(dev->dev, &tmp_idx, MLX4_RES_USAGE_DRIVER);
        if (err)
                return err;

        new_counter_index = kmalloc(sizeof(*new_counter_index), GFP_KERNEL);
        if (!new_counter_index) {
                mlx4_counter_free(dev->dev, tmp_idx);
                return -ENOMEM;
        }

        new_counter_index->index = tmp_idx;
        new_counter_index->allocated = 1;
        qp->counter_index = new_counter_index;

        mutex_lock(&dev->counters_table[qp->port - 1].mutex);
        list_add_tail(&new_counter_index->list,
                      &dev->counters_table[qp->port - 1].counters_list);
        mutex_unlock(&dev->counters_table[qp->port - 1].mutex);

        return 0;
}

enum {
        MLX4_QPC_ROCE_MODE_1 = 0,
        MLX4_QPC_ROCE_MODE_2 = 2,
        MLX4_QPC_ROCE_MODE_UNDEFINED = 0xff
};

static u8 gid_type_to_qpc(enum ib_gid_type gid_type)
{
        switch (gid_type) {
        case IB_GID_TYPE_ROCE:
                return MLX4_QPC_ROCE_MODE_1;
        case IB_GID_TYPE_ROCE_UDP_ENCAP:
                return MLX4_QPC_ROCE_MODE_2;
        default:
                return MLX4_QPC_ROCE_MODE_UNDEFINED;
        }
}

/*
 * Go over all RSS QP's childes (WQs) and apply their HW state according to
 * their logic state if the RSS QP is the first RSS QP associated for the WQ.
 */
static int bringup_rss_rwqs(struct ib_rwq_ind_table *ind_tbl, u8 port_num,
                            struct ib_udata *udata)
{
        int err = 0;
        int i;

        for (i = 0; i < (1 << ind_tbl->log_ind_tbl_size); i++) {
                struct ib_wq *ibwq = ind_tbl->ind_tbl[i];
                struct mlx4_ib_qp *wq = to_mqp((struct ib_qp *)ibwq);

                mutex_lock(&wq->mutex);

                /* Mlx4_ib restrictions:
                 * WQ's is associated to a port according to the RSS QP it is
                 * associates to.
                 * In case the WQ is associated to a different port by another
                 * RSS QP, return a failure.
                 */
                if ((wq->rss_usecnt > 0) && (wq->port != port_num)) {
                        err = -EINVAL;
                        mutex_unlock(&wq->mutex);
                        break;
                }
                wq->port = port_num;
                if ((wq->rss_usecnt == 0) && (ibwq->state == IB_WQS_RDY)) {
                        err = _mlx4_ib_modify_wq(ibwq, IB_WQS_RDY, udata);
                        if (err) {
                                mutex_unlock(&wq->mutex);
                                break;
                        }
                }
                wq->rss_usecnt++;

                mutex_unlock(&wq->mutex);
        }

        if (i && err) {
                int j;

                for (j = (i - 1); j >= 0; j--) {
                        struct ib_wq *ibwq = ind_tbl->ind_tbl[j];
                        struct mlx4_ib_qp *wq = to_mqp((struct ib_qp *)ibwq);

                        mutex_lock(&wq->mutex);

                        if ((wq->rss_usecnt == 1) &&
                            (ibwq->state == IB_WQS_RDY))
                                if (_mlx4_ib_modify_wq(ibwq, IB_WQS_RESET,
                                                       udata))
                                        pr_warn("failed to reverse WQN=0x%06x\n",
                                                ibwq->wq_num);
                        wq->rss_usecnt--;

                        mutex_unlock(&wq->mutex);
                }
        }

        return err;
}

static void bring_down_rss_rwqs(struct ib_rwq_ind_table *ind_tbl,
                                struct ib_udata *udata)
{
        int i;

        for (i = 0; i < (1 << ind_tbl->log_ind_tbl_size); i++) {
                struct ib_wq *ibwq = ind_tbl->ind_tbl[i];
                struct mlx4_ib_qp *wq = to_mqp((struct ib_qp *)ibwq);

                mutex_lock(&wq->mutex);

                if ((wq->rss_usecnt == 1) && (ibwq->state == IB_WQS_RDY))
                        if (_mlx4_ib_modify_wq(ibwq, IB_WQS_RESET, udata))
                                pr_warn("failed to reverse WQN=%x\n",
                                        ibwq->wq_num);
                wq->rss_usecnt--;

                mutex_unlock(&wq->mutex);
        }
}

static void fill_qp_rss_context(struct mlx4_qp_context *context,
                                struct mlx4_ib_qp *qp)
{
        struct mlx4_rss_context *rss_context;

        rss_context = (void *)context + offsetof(struct mlx4_qp_context,
                        pri_path) + MLX4_RSS_OFFSET_IN_QPC_PRI_PATH;

        rss_context->base_qpn = cpu_to_be32(qp->rss_ctx->base_qpn_tbl_sz);
        rss_context->default_qpn =
                cpu_to_be32(qp->rss_ctx->base_qpn_tbl_sz & 0xffffff);
        if (qp->rss_ctx->flags & (MLX4_RSS_UDP_IPV4 | MLX4_RSS_UDP_IPV6))
                rss_context->base_qpn_udp = rss_context->default_qpn;
        rss_context->flags = qp->rss_ctx->flags;
        /* Currently support just toeplitz */
        rss_context->hash_fn = MLX4_RSS_HASH_TOP;

        memcpy(rss_context->rss_key, qp->rss_ctx->rss_key,
               MLX4_EN_RSS_KEY_SIZE);
}

static int __mlx4_ib_modify_qp(void *src, enum mlx4_ib_source_type src_type,
                               const struct ib_qp_attr *attr, int attr_mask,
                               enum ib_qp_state cur_state,
                               enum ib_qp_state new_state,
                               struct ib_udata *udata)
{
        struct ib_srq  *ibsrq;
        const struct ib_gid_attr *gid_attr = NULL;
        struct ib_rwq_ind_table *rwq_ind_tbl;
        enum ib_qp_type qp_type;
        struct mlx4_ib_dev *dev;
        struct mlx4_ib_qp *qp;
        struct mlx4_ib_pd *pd;
        struct mlx4_ib_cq *send_cq, *recv_cq;
        struct mlx4_ib_ucontext *ucontext = rdma_udata_to_drv_context(
                udata, struct mlx4_ib_ucontext, ibucontext);
        struct mlx4_qp_context *context;
        enum mlx4_qp_optpar optpar = 0;
        int sqd_event;
        int steer_qp = 0;
        int err = -EINVAL;
        int counter_index;

        if (src_type == MLX4_IB_RWQ_SRC) {
                struct ib_wq *ibwq;

                ibwq        = (struct ib_wq *)src;
                ibsrq       = NULL;
                rwq_ind_tbl = NULL;
                qp_type     = IB_QPT_RAW_PACKET;
                qp          = to_mqp((struct ib_qp *)ibwq);
                dev         = to_mdev(ibwq->device);
                pd          = to_mpd(ibwq->pd);
        } else {
                struct ib_qp *ibqp;

                ibqp        = (struct ib_qp *)src;
                ibsrq       = ibqp->srq;
                rwq_ind_tbl = ibqp->rwq_ind_tbl;
                qp_type     = ibqp->qp_type;
                qp          = to_mqp(ibqp);
                dev         = to_mdev(ibqp->device);
                pd          = get_pd(qp);
        }

        /* APM is not supported under RoCE */
        if (attr_mask & IB_QP_ALT_PATH &&
            rdma_port_get_link_layer(&dev->ib_dev, qp->port) ==
            IB_LINK_LAYER_ETHERNET)
                return -ENOTSUPP;

        context = kzalloc(sizeof *context, GFP_KERNEL);
        if (!context)
                return -ENOMEM;

        context->flags = cpu_to_be32((to_mlx4_state(new_state) << 28) |
                                     (to_mlx4_st(dev, qp->mlx4_ib_qp_type) << 16));

        if (!(attr_mask & IB_QP_PATH_MIG_STATE))
                context->flags |= cpu_to_be32(MLX4_QP_PM_MIGRATED << 11);
        else {
                optpar |= MLX4_QP_OPTPAR_PM_STATE;
                switch (attr->path_mig_state) {
                case IB_MIG_MIGRATED:
                        context->flags |= cpu_to_be32(MLX4_QP_PM_MIGRATED << 11);
                        break;
                case IB_MIG_REARM:
                        context->flags |= cpu_to_be32(MLX4_QP_PM_REARM << 11);
                        break;
                case IB_MIG_ARMED:
                        context->flags |= cpu_to_be32(MLX4_QP_PM_ARMED << 11);
                        break;
                }
        }

        if (qp->inl_recv_sz)
                context->param3 |= cpu_to_be32(1 << 25);

        if (qp->flags & MLX4_IB_QP_SCATTER_FCS)
                context->param3 |= cpu_to_be32(1 << 29);

        if (qp_type == IB_QPT_GSI || qp_type == IB_QPT_SMI)
                context->mtu_msgmax = (IB_MTU_4096 << 5) | 11;
        else if (qp_type == IB_QPT_RAW_PACKET)
                context->mtu_msgmax = (MLX4_RAW_QP_MTU << 5) | MLX4_RAW_QP_MSGMAX;
        else if (qp_type == IB_QPT_UD) {
                if (qp->flags & MLX4_IB_QP_LSO)
                        context->mtu_msgmax = (IB_MTU_4096 << 5) |
                                              ilog2(dev->dev->caps.max_gso_sz);
                else
                        context->mtu_msgmax = (IB_MTU_4096 << 5) | 13;
        } else if (attr_mask & IB_QP_PATH_MTU) {
                if (attr->path_mtu < IB_MTU_256 || attr->path_mtu > IB_MTU_4096) {
                        pr_err("path MTU (%u) is invalid\n",
                               attr->path_mtu);
                        goto out;
                }
                context->mtu_msgmax = (attr->path_mtu << 5) |
                        ilog2(dev->dev->caps.max_msg_sz);
        }

        if (!rwq_ind_tbl) { /* PRM RSS receive side should be left zeros */
                if (qp->rq.wqe_cnt)
                        context->rq_size_stride = ilog2(qp->rq.wqe_cnt) << 3;
                context->rq_size_stride |= qp->rq.wqe_shift - 4;
        }

        if (qp->sq.wqe_cnt)
                context->sq_size_stride = ilog2(qp->sq.wqe_cnt) << 3;
        context->sq_size_stride |= qp->sq.wqe_shift - 4;

        if (new_state == IB_QPS_RESET && qp->counter_index)
                mlx4_ib_free_qp_counter(dev, qp);

        if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) {
                context->sq_size_stride |= !!qp->sq_no_prefetch << 7;
                context->xrcd = cpu_to_be32((u32) qp->xrcdn);
                if (qp_type == IB_QPT_RAW_PACKET)
                        context->param3 |= cpu_to_be32(1 << 30);
        }

        if (ucontext)
                context->usr_page = cpu_to_be32(
                        mlx4_to_hw_uar_index(dev->dev, ucontext->uar.index));
        else
                context->usr_page = cpu_to_be32(
                        mlx4_to_hw_uar_index(dev->dev, dev->priv_uar.index));

        if (attr_mask & IB_QP_DEST_QPN)
                context->remote_qpn = cpu_to_be32(attr->dest_qp_num);

        if (attr_mask & IB_QP_PORT) {
                if (cur_state == IB_QPS_SQD && new_state == IB_QPS_SQD &&
                    !(attr_mask & IB_QP_AV)) {
                        mlx4_set_sched(&context->pri_path, attr->port_num);
                        optpar |= MLX4_QP_OPTPAR_SCHED_QUEUE;
                }
        }

        if (cur_state == IB_QPS_INIT && new_state == IB_QPS_RTR) {
                err = create_qp_lb_counter(dev, qp);
                if (err)
                        goto out;

                counter_index =
                        dev->counters_table[qp->port - 1].default_counter;
                if (qp->counter_index)
                        counter_index = qp->counter_index->index;

                if (counter_index != -1) {
                        context->pri_path.counter_index = counter_index;
                        optpar |= MLX4_QP_OPTPAR_COUNTER_INDEX;
                        if (qp->counter_index) {
                                context->pri_path.fl |=
                                        MLX4_FL_ETH_SRC_CHECK_MC_LB;
                                context->pri_path.vlan_control |=
                                        MLX4_CTRL_ETH_SRC_CHECK_IF_COUNTER;
                        }
                } else
                        context->pri_path.counter_index =
                                MLX4_SINK_COUNTER_INDEX(dev->dev);

                if (qp->flags & MLX4_IB_QP_NETIF) {
                        mlx4_ib_steer_qp_reg(dev, qp, 1);
                        steer_qp = 1;
                }

                if (qp_type == IB_QPT_GSI) {
                        enum ib_gid_type gid_type = qp->flags & MLX4_IB_ROCE_V2_GSI_QP ?
                                IB_GID_TYPE_ROCE_UDP_ENCAP : IB_GID_TYPE_ROCE;
                        u8 qpc_roce_mode = gid_type_to_qpc(gid_type);

                        context->rlkey_roce_mode |= (qpc_roce_mode << 6);
                }
        }

        if (attr_mask & IB_QP_PKEY_INDEX) {
                if (qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV)
                        context->pri_path.disable_pkey_check = 0x40;
                context->pri_path.pkey_index = attr->pkey_index;
                optpar |= MLX4_QP_OPTPAR_PKEY_INDEX;
        }

        if (attr_mask & IB_QP_AV) {
                u8 port_num = mlx4_is_bonded(dev->dev) ? 1 :
                        attr_mask & IB_QP_PORT ? attr->port_num : qp->port;
                u16 vlan = 0xffff;
                u8 smac[ETH_ALEN];
                int is_eth =
                        rdma_cap_eth_ah(&dev->ib_dev, port_num) &&
                        rdma_ah_get_ah_flags(&attr->ah_attr) & IB_AH_GRH;

                if (is_eth) {
                        gid_attr = attr->ah_attr.grh.sgid_attr;
                        err = rdma_read_gid_l2_fields(gid_attr, &vlan,
                                                      &smac[0]);
                        if (err)
                                goto out;
                }

                if (mlx4_set_path(dev, attr, attr_mask, qp, &context->pri_path,
                                  port_num, vlan, smac))
                        goto out;

                optpar |= (MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH |
                           MLX4_QP_OPTPAR_SCHED_QUEUE);

                if (is_eth &&
                    (cur_state == IB_QPS_INIT && new_state == IB_QPS_RTR)) {
                        u8 qpc_roce_mode = gid_type_to_qpc(gid_attr->gid_type);

                        if (qpc_roce_mode == MLX4_QPC_ROCE_MODE_UNDEFINED) {
                                err = -EINVAL;
                                goto out;
                        }
                        context->rlkey_roce_mode |= (qpc_roce_mode << 6);
                }

        }

        if (attr_mask & IB_QP_TIMEOUT) {
                context->pri_path.ackto |= attr->timeout << 3;
                optpar |= MLX4_QP_OPTPAR_ACK_TIMEOUT;
        }

        if (attr_mask & IB_QP_ALT_PATH) {
                if (attr->alt_port_num == 0 ||
                    attr->alt_port_num > dev->dev->caps.num_ports)
                        goto out;

                if (attr->alt_pkey_index >=
                    dev->dev->caps.pkey_table_len[attr->alt_port_num])
                        goto out;

                if (mlx4_set_alt_path(dev, attr, attr_mask, qp,
                                      &context->alt_path,
                                      attr->alt_port_num))
                        goto out;

                context->alt_path.pkey_index = attr->alt_pkey_index;
                context->alt_path.ackto = attr->alt_timeout << 3;
                optpar |= MLX4_QP_OPTPAR_ALT_ADDR_PATH;
        }

        context->pd = cpu_to_be32(pd->pdn);

        if (!rwq_ind_tbl) {
                context->params1 = cpu_to_be32(MLX4_IB_ACK_REQ_FREQ << 28);
                get_cqs(qp, src_type, &send_cq, &recv_cq);
        } else { /* Set dummy CQs to be compatible with HV and PRM */
                send_cq = to_mcq(rwq_ind_tbl->ind_tbl[0]->cq);
                recv_cq = send_cq;
        }
        context->cqn_send = cpu_to_be32(send_cq->mcq.cqn);
        context->cqn_recv = cpu_to_be32(recv_cq->mcq.cqn);

        /* Set "fast registration enabled" for all kernel QPs */
        if (!ucontext)
                context->params1 |= cpu_to_be32(1 << 11);

        if (attr_mask & IB_QP_RNR_RETRY) {
                context->params1 |= cpu_to_be32(attr->rnr_retry << 13);
                optpar |= MLX4_QP_OPTPAR_RNR_RETRY;
        }

        if (attr_mask & IB_QP_RETRY_CNT) {
                context->params1 |= cpu_to_be32(attr->retry_cnt << 16);
                optpar |= MLX4_QP_OPTPAR_RETRY_COUNT;
        }

        if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
                if (attr->max_rd_atomic)
                        context->params1 |=
                                cpu_to_be32(fls(attr->max_rd_atomic - 1) << 21);
                optpar |= MLX4_QP_OPTPAR_SRA_MAX;
        }

        if (attr_mask & IB_QP_SQ_PSN)
                context->next_send_psn = cpu_to_be32(attr->sq_psn);

        if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
                if (attr->max_dest_rd_atomic)
                        context->params2 |=
                                cpu_to_be32(fls(attr->max_dest_rd_atomic - 1) << 21);
                optpar |= MLX4_QP_OPTPAR_RRA_MAX;
        }

        if (attr_mask & (IB_QP_ACCESS_FLAGS | IB_QP_MAX_DEST_RD_ATOMIC)) {
                context->params2 |= to_mlx4_access_flags(qp, attr, attr_mask);
                optpar |= MLX4_QP_OPTPAR_RWE | MLX4_QP_OPTPAR_RRE | MLX4_QP_OPTPAR_RAE;
        }

        if (ibsrq)
                context->params2 |= cpu_to_be32(MLX4_QP_BIT_RIC);

        if (attr_mask & IB_QP_MIN_RNR_TIMER) {
                context->rnr_nextrecvpsn |= cpu_to_be32(attr->min_rnr_timer << 24);
                optpar |= MLX4_QP_OPTPAR_RNR_TIMEOUT;
        }
        if (attr_mask & IB_QP_RQ_PSN)
                context->rnr_nextrecvpsn |= cpu_to_be32(attr->rq_psn);

        /* proxy and tunnel qp qkeys will be changed in modify-qp wrappers */
        if (attr_mask & IB_QP_QKEY) {
                if (qp->mlx4_ib_qp_type &
                    (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER))
                        context->qkey = cpu_to_be32(IB_QP_SET_QKEY);
                else {
                        if (mlx4_is_mfunc(dev->dev) &&
                            !(qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV) &&
                            (attr->qkey & MLX4_RESERVED_QKEY_MASK) ==
                            MLX4_RESERVED_QKEY_BASE) {
                                pr_err("Cannot use reserved QKEY"
                                       " 0x%x (range 0xffff0000..0xffffffff"
                                       " is reserved)\n", attr->qkey);
                                err = -EINVAL;
                                goto out;
                        }
                        context->qkey = cpu_to_be32(attr->qkey);
                }
                optpar |= MLX4_QP_OPTPAR_Q_KEY;
        }

        if (ibsrq)
                context->srqn = cpu_to_be32(1 << 24 |
                                            to_msrq(ibsrq)->msrq.srqn);

        if (qp->rq.wqe_cnt &&
            cur_state == IB_QPS_RESET &&
            new_state == IB_QPS_INIT)
                context->db_rec_addr = cpu_to_be64(qp->db.dma);

        if (cur_state == IB_QPS_INIT &&
            new_state == IB_QPS_RTR  &&
            (qp_type == IB_QPT_GSI || qp_type == IB_QPT_SMI ||
             qp_type == IB_QPT_UD || qp_type == IB_QPT_RAW_PACKET)) {
                context->pri_path.sched_queue = (qp->port - 1) << 6;
                if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_SMI ||
                    qp->mlx4_ib_qp_type &
                    (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER)) {
                        context->pri_path.sched_queue |= MLX4_IB_DEFAULT_QP0_SCHED_QUEUE;
                        if (qp->mlx4_ib_qp_type != MLX4_IB_QPT_SMI)
                                context->pri_path.fl = 0x80;
                } else {
                        if (qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV)
                                context->pri_path.fl = 0x80;
                        context->pri_path.sched_queue |= MLX4_IB_DEFAULT_SCHED_QUEUE;
                }
                if (rdma_port_get_link_layer(&dev->ib_dev, qp->port) ==
                    IB_LINK_LAYER_ETHERNET) {
                        if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_TUN_GSI ||
                            qp->mlx4_ib_qp_type == MLX4_IB_QPT_GSI)
                                context->pri_path.feup = 1 << 7; /* don't fsm */
                        /* handle smac_index */
                        if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_UD ||
                            qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_GSI ||
                            qp->mlx4_ib_qp_type == MLX4_IB_QPT_TUN_GSI) {
                                err = handle_eth_ud_smac_index(dev, qp, context);
                                if (err) {
                                        err = -EINVAL;
                                        goto out;
                                }
                                if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_GSI)
                                        dev->qp1_proxy[qp->port - 1] = qp;
                        }
                }
        }

        if (qp_type == IB_QPT_RAW_PACKET) {
                context->pri_path.ackto = (context->pri_path.ackto & 0xf8) |
                                        MLX4_IB_LINK_TYPE_ETH;
                if (dev->dev->caps.tunnel_offload_mode ==  MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
                        /* set QP to receive both tunneled & non-tunneled packets */
                        if (!rwq_ind_tbl)
                                context->srqn = cpu_to_be32(7 << 28);
                }
        }

        if (qp_type == IB_QPT_UD && (new_state == IB_QPS_RTR)) {
                int is_eth = rdma_port_get_link_layer(
                                &dev->ib_dev, qp->port) ==
                                IB_LINK_LAYER_ETHERNET;
                if (is_eth) {
                        context->pri_path.ackto = MLX4_IB_LINK_TYPE_ETH;
                        optpar |= MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH;
                }
        }

        if (cur_state == IB_QPS_RTS && new_state == IB_QPS_SQD  &&
            attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY && attr->en_sqd_async_notify)
                sqd_event = 1;
        else
                sqd_event = 0;

        if (!ucontext &&
            cur_state == IB_QPS_RESET &&
            new_state == IB_QPS_INIT)
                context->rlkey_roce_mode |= (1 << 4);

        /*
         * Before passing a kernel QP to the HW, make sure that the
         * ownership bits of the send queue are set and the SQ
         * headroom is stamped so that the hardware doesn't start
         * processing stale work requests.
         */
        if (!ucontext &&
            cur_state == IB_QPS_RESET &&
            new_state == IB_QPS_INIT) {
                struct mlx4_wqe_ctrl_seg *ctrl;
                int i;

                for (i = 0; i < qp->sq.wqe_cnt; ++i) {
                        ctrl = get_send_wqe(qp, i);
                        ctrl->owner_opcode = cpu_to_be32(1 << 31);
                        ctrl->qpn_vlan.fence_size =
                                1 << (qp->sq.wqe_shift - 4);
                        stamp_send_wqe(qp, i);
                }
        }

        if (rwq_ind_tbl &&
            cur_state == IB_QPS_RESET &&
            new_state == IB_QPS_INIT) {
                fill_qp_rss_context(context, qp);
                context->flags |= cpu_to_be32(1 << MLX4_RSS_QPC_FLAG_OFFSET);
        }

        err = mlx4_qp_modify(dev->dev, &qp->mtt, to_mlx4_state(cur_state),
                             to_mlx4_state(new_state), context, optpar,
                             sqd_event, &qp->mqp);
        if (err)
                goto out;

        qp->state = new_state;

        if (attr_mask & IB_QP_ACCESS_FLAGS)
                qp->atomic_rd_en = attr->qp_access_flags;
        if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
                qp->resp_depth = attr->max_dest_rd_atomic;
        if (attr_mask & IB_QP_PORT) {
                qp->port = attr->port_num;
                update_mcg_macs(dev, qp);
        }
        if (attr_mask & IB_QP_ALT_PATH)
                qp->alt_port = attr->alt_port_num;

        if (is_sqp(dev, qp))
                store_sqp_attrs(qp->sqp, attr, attr_mask);

        /*
         * If we moved QP0 to RTR, bring the IB link up; if we moved
         * QP0 to RESET or ERROR, bring the link back down.
         */
        if (is_qp0(dev, qp)) {
                if (cur_state != IB_QPS_RTR && new_state == IB_QPS_RTR)
                        if (mlx4_INIT_PORT(dev->dev, qp->port))
                                pr_warn("INIT_PORT failed for port %d\n",
                                       qp->port);

                if (cur_state != IB_QPS_RESET && cur_state != IB_QPS_ERR &&
                    (new_state == IB_QPS_RESET || new_state == IB_QPS_ERR))
                        mlx4_CLOSE_PORT(dev->dev, qp->port);
        }

        /*
         * If we moved a kernel QP to RESET, clean up all old CQ
         * entries and reinitialize the QP.
         */
        if (new_state == IB_QPS_RESET) {
                if (!ucontext) {
                        mlx4_ib_cq_clean(recv_cq, qp->mqp.qpn,
                                         ibsrq ? to_msrq(ibsrq) : NULL);
                        if (send_cq != recv_cq)
                                mlx4_ib_cq_clean(send_cq, qp->mqp.qpn, NULL);

                        qp->rq.head = 0;
                        qp->rq.tail = 0;
                        qp->sq.head = 0;
                        qp->sq.tail = 0;
                        qp->sq_next_wqe = 0;
                        if (qp->rq.wqe_cnt)
                                *qp->db.db  = 0;

                        if (qp->flags & MLX4_IB_QP_NETIF)
                                mlx4_ib_steer_qp_reg(dev, qp, 0);
                }
                if (qp->pri.smac || (!qp->pri.smac && qp->pri.smac_port)) {
                        mlx4_unregister_mac(dev->dev, qp->pri.smac_port, qp->pri.smac);
                        qp->pri.smac = 0;
                        qp->pri.smac_port = 0;
                }
                if (qp->alt.smac) {
                        mlx4_unregister_mac(dev->dev, qp->alt.smac_port, qp->alt.smac);
                        qp->alt.smac = 0;
                }
                if (qp->pri.vid < 0x1000) {
                        mlx4_unregister_vlan(dev->dev, qp->pri.vlan_port, qp->pri.vid);
                        qp->pri.vid = 0xFFFF;
                        qp->pri.candidate_vid = 0xFFFF;
                        qp->pri.update_vid = 0;
                }

                if (qp->alt.vid < 0x1000) {
                        mlx4_unregister_vlan(dev->dev, qp->alt.vlan_port, qp->alt.vid);
                        qp->alt.vid = 0xFFFF;
                        qp->alt.candidate_vid = 0xFFFF;
                        qp->alt.update_vid = 0;
                }
        }
out:
        if (err && qp->counter_index)
                mlx4_ib_free_qp_counter(dev, qp);
        if (err && steer_qp)
                mlx4_ib_steer_qp_reg(dev, qp, 0);
        kfree(context);
        if (qp->pri.candidate_smac ||
            (!qp->pri.candidate_smac && qp->pri.candidate_smac_port)) {
                if (err) {
                        mlx4_unregister_mac(dev->dev, qp->pri.candidate_smac_port, qp->pri.candidate_smac);
                } else {
                        if (qp->pri.smac || (!qp->pri.smac && qp->pri.smac_port))
                                mlx4_unregister_mac(dev->dev, qp->pri.smac_port, qp->pri.smac);
                        qp->pri.smac = qp->pri.candidate_smac;
                        qp->pri.smac_index = qp->pri.candidate_smac_index;
                        qp->pri.smac_port = qp->pri.candidate_smac_port;
                }
                qp->pri.candidate_smac = 0;
                qp->pri.candidate_smac_index = 0;
                qp->pri.candidate_smac_port = 0;
        }
        if (qp->alt.candidate_smac) {
                if (err) {
                        mlx4_unregister_mac(dev->dev, qp->alt.candidate_smac_port, qp->alt.candidate_smac);
                } else {
                        if (qp->alt.smac)
                                mlx4_unregister_mac(dev->dev, qp->alt.smac_port, qp->alt.smac);
                        qp->alt.smac = qp->alt.candidate_smac;
                        qp->alt.smac_index = qp->alt.candidate_smac_index;
                        qp->alt.smac_port = qp->alt.candidate_smac_port;
                }
                qp->alt.candidate_smac = 0;
                qp->alt.candidate_smac_index = 0;
                qp->alt.candidate_smac_port = 0;
        }

        if (qp->pri.update_vid) {
                if (err) {
                        if (qp->pri.candidate_vid < 0x1000)
                                mlx4_unregister_vlan(dev->dev, qp->pri.candidate_vlan_port,
                                                     qp->pri.candidate_vid);
                } else {
                        if (qp->pri.vid < 0x1000)
                                mlx4_unregister_vlan(dev->dev, qp->pri.vlan_port,
                                                     qp->pri.vid);
                        qp->pri.vid = qp->pri.candidate_vid;
                        qp->pri.vlan_port = qp->pri.candidate_vlan_port;
                        qp->pri.vlan_index =  qp->pri.candidate_vlan_index;
                }
                qp->pri.candidate_vid = 0xFFFF;
                qp->pri.update_vid = 0;
        }

        if (qp->alt.update_vid) {
                if (err) {
                        if (qp->alt.candidate_vid < 0x1000)
                                mlx4_unregister_vlan(dev->dev, qp->alt.candidate_vlan_port,
                                                     qp->alt.candidate_vid);
                } else {
                        if (qp->alt.vid < 0x1000)
                                mlx4_unregister_vlan(dev->dev, qp->alt.vlan_port,
                                                     qp->alt.vid);
                        qp->alt.vid = qp->alt.candidate_vid;
                        qp->alt.vlan_port = qp->alt.candidate_vlan_port;
                        qp->alt.vlan_index =  qp->alt.candidate_vlan_index;
                }
                qp->alt.candidate_vid = 0xFFFF;
                qp->alt.update_vid = 0;
        }

        return err;
}

enum {
        MLX4_IB_MODIFY_QP_RSS_SUP_ATTR_MSK = (IB_QP_STATE       |
                                              IB_QP_PORT),
};

static int _mlx4_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
                              int attr_mask, struct ib_udata *udata)
{
        struct mlx4_ib_dev *dev = to_mdev(ibqp->device);
        struct mlx4_ib_qp *qp = to_mqp(ibqp);
        enum ib_qp_state cur_state, new_state;
        int err = -EINVAL;
        mutex_lock(&qp->mutex);

        cur_state = attr_mask & IB_QP_CUR_STATE ? attr->cur_qp_state : qp->state;
        new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;

        if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type,
                                attr_mask)) {
                pr_debug("qpn 0x%x: invalid attribute mask specified "
                         "for transition %d to %d. qp_type %d,"
                         " attr_mask 0x%x\n",
                         ibqp->qp_num, cur_state, new_state,
                         ibqp->qp_type, attr_mask);
                goto out;
        }

        if (ibqp->rwq_ind_tbl) {
                if (!(((cur_state == IB_QPS_RESET) &&
                       (new_state == IB_QPS_INIT)) ||
                      ((cur_state == IB_QPS_INIT)  &&
                       (new_state == IB_QPS_RTR)))) {
                        pr_debug("qpn 0x%x: RSS QP unsupported transition %d to %d\n",
                                 ibqp->qp_num, cur_state, new_state);

                        err = -EOPNOTSUPP;
                        goto out;
                }

                if (attr_mask & ~MLX4_IB_MODIFY_QP_RSS_SUP_ATTR_MSK) {
                        pr_debug("qpn 0x%x: RSS QP unsupported attribute mask 0x%x for transition %d to %d\n",
                                 ibqp->qp_num, attr_mask, cur_state, new_state);

                        err = -EOPNOTSUPP;
                        goto out;
                }
        }

        if (mlx4_is_bonded(dev->dev) && (attr_mask & IB_QP_PORT)) {
                if ((cur_state == IB_QPS_RESET) && (new_state == IB_QPS_INIT)) {
                        if ((ibqp->qp_type == IB_QPT_RC) ||
                            (ibqp->qp_type == IB_QPT_UD) ||
                            (ibqp->qp_type == IB_QPT_UC) ||
                            (ibqp->qp_type == IB_QPT_RAW_PACKET) ||
                            (ibqp->qp_type == IB_QPT_XRC_INI)) {
                                attr->port_num = mlx4_ib_bond_next_port(dev);
                        }
                } else {
                        /* no sense in changing port_num
                         * when ports are bonded */
                        attr_mask &= ~IB_QP_PORT;
                }
        }

        if ((attr_mask & IB_QP_PORT) &&
            (attr->port_num == 0 || attr->port_num > dev->num_ports)) {
                pr_debug("qpn 0x%x: invalid port number (%d) specified "
                         "for transition %d to %d. qp_type %d\n",
                         ibqp->qp_num, attr->port_num, cur_state,
                         new_state, ibqp->qp_type);
                goto out;
        }

        if ((attr_mask & IB_QP_PORT) && (ibqp->qp_type == IB_QPT_RAW_PACKET) &&
            (rdma_port_get_link_layer(&dev->ib_dev, attr->port_num) !=
             IB_LINK_LAYER_ETHERNET))
                goto out;

        if (attr_mask & IB_QP_PKEY_INDEX) {
                int p = attr_mask & IB_QP_PORT ? attr->port_num : qp->port;
                if (attr->pkey_index >= dev->dev->caps.pkey_table_len[p]) {
                        pr_debug("qpn 0x%x: invalid pkey index (%d) specified "
                                 "for transition %d to %d. qp_type %d\n",
                                 ibqp->qp_num, attr->pkey_index, cur_state,
                                 new_state, ibqp->qp_type);
                        goto out;
                }
        }

        if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC &&
            attr->max_rd_atomic > dev->dev->caps.max_qp_init_rdma) {
                pr_debug("qpn 0x%x: max_rd_atomic (%d) too large. "
                         "Transition %d to %d. qp_type %d\n",
                         ibqp->qp_num, attr->max_rd_atomic, cur_state,
                         new_state, ibqp->qp_type);
                goto out;
        }

        if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC &&
            attr->max_dest_rd_atomic > dev->dev->caps.max_qp_dest_rdma) {
                pr_debug("qpn 0x%x: max_dest_rd_atomic (%d) too large. "
                         "Transition %d to %d. qp_type %d\n",
                         ibqp->qp_num, attr->max_dest_rd_atomic, cur_state,
                         new_state, ibqp->qp_type);
                goto out;
        }

        if (cur_state == new_state && cur_state == IB_QPS_RESET) {
                err = 0;
                goto out;
        }

        if (ibqp->rwq_ind_tbl && (new_state == IB_QPS_INIT)) {
                err = bringup_rss_rwqs(ibqp->rwq_ind_tbl, attr->port_num,
                                       udata);
                if (err)
                        goto out;
        }

        err = __mlx4_ib_modify_qp(ibqp, MLX4_IB_QP_SRC, attr, attr_mask,
                                  cur_state, new_state, udata);

        if (ibqp->rwq_ind_tbl && err)
                bring_down_rss_rwqs(ibqp->rwq_ind_tbl, udata);

        if (mlx4_is_bonded(dev->dev) && (attr_mask & IB_QP_PORT))
                attr->port_num = 1;

out:
        mutex_unlock(&qp->mutex);
        return err;
}

int mlx4_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
                      int attr_mask, struct ib_udata *udata)
{
        struct mlx4_ib_qp *mqp = to_mqp(ibqp);
        int ret;

        if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
                return -EOPNOTSUPP;

        ret = _mlx4_ib_modify_qp(ibqp, attr, attr_mask, udata);

        if (mqp->mlx4_ib_qp_type == MLX4_IB_QPT_GSI) {
                struct mlx4_ib_sqp *sqp = mqp->sqp;
                int err = 0;

                if (sqp->roce_v2_gsi)
                        err = ib_modify_qp(sqp->roce_v2_gsi, attr, attr_mask);
                if (err)
                        pr_err("Failed to modify GSI QP for RoCEv2 (%d)\n",
                               err);
        }
        return ret;
}

static int vf_get_qp0_qkey(struct mlx4_dev *dev, int qpn, u32 *qkey)
{
        int i;
        for (i = 0; i < dev->caps.num_ports; i++) {
                if (qpn == dev->caps.spec_qps[i].qp0_proxy ||
                    qpn == dev->caps.spec_qps[i].qp0_tunnel) {
                        *qkey = dev->caps.spec_qps[i].qp0_qkey;
                        return 0;
                }
        }
        return -EINVAL;
}

static int build_sriov_qp0_header(struct mlx4_ib_qp *qp,
                                  const struct ib_ud_wr *wr,
                                  void *wqe, unsigned *mlx_seg_len)
{
        struct mlx4_ib_dev *mdev = to_mdev(qp->ibqp.device);
        struct mlx4_ib_sqp *sqp = qp->sqp;
        struct ib_device *ib_dev = qp->ibqp.device;
        struct mlx4_wqe_mlx_seg *mlx = wqe;
        struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx;
        struct mlx4_ib_ah *ah = to_mah(wr->ah);
        u16 pkey;
        u32 qkey;
        int send_size;
        int header_size;
        int spc;
        int err;
        int i;

        if (wr->wr.opcode != IB_WR_SEND)
                return -EINVAL;

        send_size = 0;

        for (i = 0; i < wr->wr.num_sge; ++i)
                send_size += wr->wr.sg_list[i].length;

        /* for proxy-qp0 sends, need to add in size of tunnel header */
        /* for tunnel-qp0 sends, tunnel header is already in s/g list */
        if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER)
                send_size += sizeof (struct mlx4_ib_tunnel_header);

        ib_ud_header_init(send_size, 1, 0, 0, 0, 0, 0, 0, &sqp->ud_header);

        if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER) {
                sqp->ud_header.lrh.service_level =
                        be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 28;
                sqp->ud_header.lrh.destination_lid =
                        cpu_to_be16(ah->av.ib.g_slid & 0x7f);
                sqp->ud_header.lrh.source_lid =
                        cpu_to_be16(ah->av.ib.g_slid & 0x7f);
        }

        mlx->flags &= cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE);

        /* force loopback */
        mlx->flags |= cpu_to_be32(MLX4_WQE_MLX_VL15 | 0x1 | MLX4_WQE_MLX_SLR);
        mlx->rlid = sqp->ud_header.lrh.destination_lid;

        sqp->ud_header.lrh.virtual_lane    = 0;
        sqp->ud_header.bth.solicited_event = !!(wr->wr.send_flags & IB_SEND_SOLICITED);
        err = ib_get_cached_pkey(ib_dev, qp->port, 0, &pkey);
        if (err)
                return err;
        sqp->ud_header.bth.pkey = cpu_to_be16(pkey);
        if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_TUN_SMI_OWNER)
                sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->remote_qpn);
        else
                sqp->ud_header.bth.destination_qpn =
                        cpu_to_be32(mdev->dev->caps.spec_qps[qp->port - 1].qp0_tunnel);

        sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1));
        if (mlx4_is_master(mdev->dev)) {
                if (mlx4_get_parav_qkey(mdev->dev, qp->mqp.qpn, &qkey))
                        return -EINVAL;
        } else {
                if (vf_get_qp0_qkey(mdev->dev, qp->mqp.qpn, &qkey))
                        return -EINVAL;
        }
        sqp->ud_header.deth.qkey = cpu_to_be32(qkey);
        sqp->ud_header.deth.source_qpn = cpu_to_be32(qp->mqp.qpn);

        sqp->ud_header.bth.opcode        = IB_OPCODE_UD_SEND_ONLY;
        sqp->ud_header.immediate_present = 0;

        header_size = ib_ud_header_pack(&sqp->ud_header, sqp->header_buf);

        /*
         * Inline data segments may not cross a 64 byte boundary.  If
         * our UD header is bigger than the space available up to the
         * next 64 byte boundary in the WQE, use two inline data
         * segments to hold the UD header.
         */
        spc = MLX4_INLINE_ALIGN -
              ((unsigned long) (inl + 1) & (MLX4_INLINE_ALIGN - 1));
        if (header_size <= spc) {
                inl->byte_count = cpu_to_be32(1 << 31 | header_size);
                memcpy(inl + 1, sqp->header_buf, header_size);
                i = 1;
        } else {
                inl->byte_count = cpu_to_be32(1 << 31 | spc);
                memcpy(inl + 1, sqp->header_buf, spc);

                inl = (void *) (inl + 1) + spc;
                memcpy(inl + 1, sqp->header_buf + spc, header_size - spc);
                /*
                 * Need a barrier here to make sure all the data is
                 * visible before the byte_count field is set.
                 * Otherwise the HCA prefetcher could grab the 64-byte
                 * chunk with this inline segment and get a valid (!=
                 * 0xffffffff) byte count but stale data, and end up
                 * generating a packet with bad headers.
                 *
                 * The first inline segment's byte_count field doesn't
                 * need a barrier, because it comes after a
                 * control/MLX segment and therefore is at an offset
                 * of 16 mod 64.
                 */
                wmb();
                inl->byte_count = cpu_to_be32(1 << 31 | (header_size - spc));
                i = 2;
        }

        *mlx_seg_len =
        ALIGN(i * sizeof (struct mlx4_wqe_inline_seg) + header_size, 16);
        return 0;
}

static u8 sl_to_vl(struct mlx4_ib_dev *dev, u8 sl, int port_num)
{
        union sl2vl_tbl_to_u64 tmp_vltab;
        u8 vl;

        if (sl > 15)
                return 0xf;
        tmp_vltab.sl64 = atomic64_read(&dev->sl2vl[port_num - 1]);
        vl = tmp_vltab.sl8[sl >> 1];
        if (sl & 1)
                vl &= 0x0f;
        else
                vl >>= 4;
        return vl;
}

static int fill_gid_by_hw_index(struct mlx4_ib_dev *ibdev, u8 port_num,
                                int index, union ib_gid *gid,
                                enum ib_gid_type *gid_type)
{
        struct mlx4_ib_iboe *iboe = &ibdev->iboe;
        struct mlx4_port_gid_table *port_gid_table;
        unsigned long flags;

        port_gid_table = &iboe->gids[port_num - 1];
        spin_lock_irqsave(&iboe->lock, flags);
        memcpy(gid, &port_gid_table->gids[index].gid, sizeof(*gid));
        *gid_type = port_gid_table->gids[index].gid_type;
        spin_unlock_irqrestore(&iboe->lock, flags);
        if (rdma_is_zero_gid(gid))
                return -ENOENT;

        return 0;
}

#define MLX4_ROCEV2_QP1_SPORT 0xC000
static int build_mlx_header(struct mlx4_ib_qp *qp, const struct ib_ud_wr *wr,
                            void *wqe, unsigned *mlx_seg_len)
{
        struct mlx4_ib_sqp *sqp = qp->sqp;
        struct ib_device *ib_dev = qp->ibqp.device;
        struct mlx4_ib_dev *ibdev = to_mdev(ib_dev);
        struct mlx4_wqe_mlx_seg *mlx = wqe;
        struct mlx4_wqe_ctrl_seg *ctrl = wqe;
        struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx;
        struct mlx4_ib_ah *ah = to_mah(wr->ah);
        union ib_gid sgid;
        u16 pkey;
        int send_size;
        int header_size;
        int spc;
        int i;
        int err = 0;
        u16 vlan = 0xffff;
        bool is_eth;
        bool is_vlan = false;
        bool is_grh;
        bool is_udp = false;
        int ip_version = 0;

        send_size = 0;
        for (i = 0; i < wr->wr.num_sge; ++i)
                send_size += wr->wr.sg_list[i].length;

        is_eth = rdma_port_get_link_layer(qp->ibqp.device, qp->port) == IB_LINK_LAYER_ETHERNET;