net/mlx5: FPGA, Add high-speed connection routines
authorIlan Tayari <ilant@mellanox.com>
Mon, 27 Mar 2017 11:48:38 +0000 (14:48 +0300)
committerSaeed Mahameed <saeedm@mellanox.com>
Tue, 27 Jun 2017 13:36:47 +0000 (16:36 +0300)
An FPGA high-speed connection has two endpoints, an FPGA QP and a
ConnectX QP.
Add library routines to create and connect the endpoints of an
FPGA high-speed connection.

These routines allow creating and interacting with both types of
connections: Shell and Sandbox Unit (SBU).

Shell connection provides an interface to the FPGA's address space,
which includes the configuration space and the DDR.
Use of the shell connection will be introduced in a later patchset.

SBU connection provides a command and/or data interface to the
application-specific logic within the FPGA.
Use of the SBU connection will be introduced in a later patch in
this patchset.

Some struct definitions are added to a new header file sdk.h, which
will be extended in later patches in the patchset.
This header file will contain the in-kernel FPGA client driver API.

Signed-off-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
drivers/net/ethernet/mellanox/mlx5/core/Makefile
drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c [new file with mode: 0644]
drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.h [new file with mode: 0644]
drivers/net/ethernet/mellanox/mlx5/core/fpga/core.c
drivers/net/ethernet/mellanox/mlx5/core/fpga/core.h
drivers/net/ethernet/mellanox/mlx5/core/fpga/sdk.h [new file with mode: 0644]

index 738867bab21f6e5e57d16fea191bae2270d56de6..5221b1235c47eba5b99f89253f90848636518612 100644 (file)
@@ -6,7 +6,7 @@ mlx5_core-y :=  main.o cmd.o debugfs.o fw.o eq.o uar.o pagealloc.o \
                mad.o transobj.o vport.o sriov.o fs_cmd.o fs_core.o \
                fs_counters.o rl.o lag.o dev.o lib/gid.o
 
-mlx5_core-$(CONFIG_MLX5_FPGA) += fpga/cmd.o fpga/core.o
+mlx5_core-$(CONFIG_MLX5_FPGA) += fpga/cmd.o fpga/core.o fpga/conn.o
 
 mlx5_core-$(CONFIG_MLX5_CORE_EN) += wq.o eswitch.o eswitch_offloads.o \
                en_main.o en_common.o en_fs.o en_ethtool.o en_tx.o \
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c b/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.c
new file mode 100644 (file)
index 0000000..c4392f7
--- /dev/null
@@ -0,0 +1,1042 @@
+/*
+ * Copyright (c) 2017 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 <net/addrconf.h>
+#include <linux/etherdevice.h>
+#include <linux/mlx5/vport.h>
+
+#include "mlx5_core.h"
+#include "lib/mlx5.h"
+#include "fpga/conn.h"
+
+#define MLX5_FPGA_PKEY 0xFFFF
+#define MLX5_FPGA_PKEY_INDEX 0 /* RoCE PKEY 0xFFFF is always at index 0 */
+#define MLX5_FPGA_RECV_SIZE 2048
+#define MLX5_FPGA_PORT_NUM 1
+#define MLX5_FPGA_CQ_BUDGET 64
+
+static int mlx5_fpga_conn_map_buf(struct mlx5_fpga_conn *conn,
+                                 struct mlx5_fpga_dma_buf *buf)
+{
+       struct device *dma_device;
+       int err = 0;
+
+       if (unlikely(!buf->sg[0].data))
+               goto out;
+
+       dma_device = &conn->fdev->mdev->pdev->dev;
+       buf->sg[0].dma_addr = dma_map_single(dma_device, buf->sg[0].data,
+                                            buf->sg[0].size, buf->dma_dir);
+       err = dma_mapping_error(dma_device, buf->sg[0].dma_addr);
+       if (unlikely(err)) {
+               mlx5_fpga_warn(conn->fdev, "DMA error on sg 0: %d\n", err);
+               err = -ENOMEM;
+               goto out;
+       }
+
+       if (!buf->sg[1].data)
+               goto out;
+
+       buf->sg[1].dma_addr = dma_map_single(dma_device, buf->sg[1].data,
+                                            buf->sg[1].size, buf->dma_dir);
+       err = dma_mapping_error(dma_device, buf->sg[1].dma_addr);
+       if (unlikely(err)) {
+               mlx5_fpga_warn(conn->fdev, "DMA error on sg 1: %d\n", err);
+               dma_unmap_single(dma_device, buf->sg[0].dma_addr,
+                                buf->sg[0].size, buf->dma_dir);
+               err = -ENOMEM;
+       }
+
+out:
+       return err;
+}
+
+static void mlx5_fpga_conn_unmap_buf(struct mlx5_fpga_conn *conn,
+                                    struct mlx5_fpga_dma_buf *buf)
+{
+       struct device *dma_device;
+
+       dma_device = &conn->fdev->mdev->pdev->dev;
+       if (buf->sg[1].data)
+               dma_unmap_single(dma_device, buf->sg[1].dma_addr,
+                                buf->sg[1].size, buf->dma_dir);
+
+       if (likely(buf->sg[0].data))
+               dma_unmap_single(dma_device, buf->sg[0].dma_addr,
+                                buf->sg[0].size, buf->dma_dir);
+}
+
+static int mlx5_fpga_conn_post_recv(struct mlx5_fpga_conn *conn,
+                                   struct mlx5_fpga_dma_buf *buf)
+{
+       struct mlx5_wqe_data_seg *data;
+       unsigned int ix;
+       int err = 0;
+
+       err = mlx5_fpga_conn_map_buf(conn, buf);
+       if (unlikely(err))
+               goto out;
+
+       if (unlikely(conn->qp.rq.pc - conn->qp.rq.cc >= conn->qp.rq.size)) {
+               mlx5_fpga_conn_unmap_buf(conn, buf);
+               return -EBUSY;
+       }
+
+       ix = conn->qp.rq.pc & (conn->qp.rq.size - 1);
+       data = mlx5_wq_cyc_get_wqe(&conn->qp.wq.rq, ix);
+       data->byte_count = cpu_to_be32(buf->sg[0].size);
+       data->lkey = cpu_to_be32(conn->fdev->conn_res.mkey.key);
+       data->addr = cpu_to_be64(buf->sg[0].dma_addr);
+
+       conn->qp.rq.pc++;
+       conn->qp.rq.bufs[ix] = buf;
+
+       /* Make sure that descriptors are written before doorbell record. */
+       dma_wmb();
+       *conn->qp.wq.rq.db = cpu_to_be32(conn->qp.rq.pc & 0xffff);
+out:
+       return err;
+}
+
+static void mlx5_fpga_conn_notify_hw(struct mlx5_fpga_conn *conn, void *wqe)
+{
+       /* ensure wqe is visible to device before updating doorbell record */
+       dma_wmb();
+       *conn->qp.wq.sq.db = cpu_to_be32(conn->qp.sq.pc);
+       /* Make sure that doorbell record is visible before ringing */
+       wmb();
+       mlx5_write64(wqe, conn->fdev->conn_res.uar->map + MLX5_BF_OFFSET, NULL);
+}
+
+static void mlx5_fpga_conn_post_send(struct mlx5_fpga_conn *conn,
+                                    struct mlx5_fpga_dma_buf *buf)
+{
+       struct mlx5_wqe_ctrl_seg *ctrl;
+       struct mlx5_wqe_data_seg *data;
+       unsigned int ix, sgi;
+       int size = 1;
+
+       ix = conn->qp.sq.pc & (conn->qp.sq.size - 1);
+
+       ctrl = mlx5_wq_cyc_get_wqe(&conn->qp.wq.sq, ix);
+       data = (void *)(ctrl + 1);
+
+       for (sgi = 0; sgi < ARRAY_SIZE(buf->sg); sgi++) {
+               if (!buf->sg[sgi].data)
+                       break;
+               data->byte_count = cpu_to_be32(buf->sg[sgi].size);
+               data->lkey = cpu_to_be32(conn->fdev->conn_res.mkey.key);
+               data->addr = cpu_to_be64(buf->sg[sgi].dma_addr);
+               data++;
+               size++;
+       }
+
+       ctrl->imm = 0;
+       ctrl->fm_ce_se = MLX5_WQE_CTRL_CQ_UPDATE;
+       ctrl->opmod_idx_opcode = cpu_to_be32(((conn->qp.sq.pc & 0xffff) << 8) |
+                                            MLX5_OPCODE_SEND);
+       ctrl->qpn_ds = cpu_to_be32(size | (conn->qp.mqp.qpn << 8));
+
+       conn->qp.sq.pc++;
+       conn->qp.sq.bufs[ix] = buf;
+       mlx5_fpga_conn_notify_hw(conn, ctrl);
+}
+
+int mlx5_fpga_conn_send(struct mlx5_fpga_conn *conn,
+                       struct mlx5_fpga_dma_buf *buf)
+{
+       unsigned long flags;
+       int err;
+
+       if (!conn->qp.active)
+               return -ENOTCONN;
+
+       err = mlx5_fpga_conn_map_buf(conn, buf);
+       if (err)
+               return err;
+
+       spin_lock_irqsave(&conn->qp.sq.lock, flags);
+
+       if (conn->qp.sq.pc - conn->qp.sq.cc >= conn->qp.sq.size) {
+               list_add_tail(&buf->list, &conn->qp.sq.backlog);
+               goto out_unlock;
+       }
+
+       mlx5_fpga_conn_post_send(conn, buf);
+
+out_unlock:
+       spin_unlock_irqrestore(&conn->qp.sq.lock, flags);
+       return err;
+}
+
+static int mlx5_fpga_conn_post_recv_buf(struct mlx5_fpga_conn *conn)
+{
+       struct mlx5_fpga_dma_buf *buf;
+       int err;
+
+       buf = kzalloc(sizeof(*buf) + MLX5_FPGA_RECV_SIZE, 0);
+       if (!buf)
+               return -ENOMEM;
+
+       buf->sg[0].data = (void *)(buf + 1);
+       buf->sg[0].size = MLX5_FPGA_RECV_SIZE;
+       buf->dma_dir = DMA_FROM_DEVICE;
+
+       err = mlx5_fpga_conn_post_recv(conn, buf);
+       if (err)
+               kfree(buf);
+
+       return err;
+}
+
+static int mlx5_fpga_conn_create_mkey(struct mlx5_core_dev *mdev, u32 pdn,
+                                     struct mlx5_core_mkey *mkey)
+{
+       int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
+       void *mkc;
+       u32 *in;
+       int err;
+
+       in = kvzalloc(inlen, GFP_KERNEL);
+       if (!in)
+               return -ENOMEM;
+
+       mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
+       MLX5_SET(mkc, mkc, access_mode, MLX5_MKC_ACCESS_MODE_PA);
+       MLX5_SET(mkc, mkc, lw, 1);
+       MLX5_SET(mkc, mkc, lr, 1);
+
+       MLX5_SET(mkc, mkc, pd, pdn);
+       MLX5_SET(mkc, mkc, length64, 1);
+       MLX5_SET(mkc, mkc, qpn, 0xffffff);
+
+       err = mlx5_core_create_mkey(mdev, mkey, in, inlen);
+
+       kvfree(in);
+       return err;
+}
+
+static void mlx5_fpga_conn_rq_cqe(struct mlx5_fpga_conn *conn,
+                                 struct mlx5_cqe64 *cqe, u8 status)
+{
+       struct mlx5_fpga_dma_buf *buf;
+       int ix, err;
+
+       ix = be16_to_cpu(cqe->wqe_counter) & (conn->qp.rq.size - 1);
+       buf = conn->qp.rq.bufs[ix];
+       conn->qp.rq.bufs[ix] = NULL;
+       if (!status)
+               buf->sg[0].size = be32_to_cpu(cqe->byte_cnt);
+       conn->qp.rq.cc++;
+
+       if (unlikely(status && (status != MLX5_CQE_SYNDROME_WR_FLUSH_ERR)))
+               mlx5_fpga_warn(conn->fdev, "RQ buf %p on FPGA QP %u completion status %d\n",
+                              buf, conn->fpga_qpn, status);
+       else
+               mlx5_fpga_dbg(conn->fdev, "RQ buf %p on FPGA QP %u completion status %d\n",
+                             buf, conn->fpga_qpn, status);
+
+       mlx5_fpga_conn_unmap_buf(conn, buf);
+
+       if (unlikely(status || !conn->qp.active)) {
+               conn->qp.active = false;
+               kfree(buf);
+               return;
+       }
+
+       mlx5_fpga_dbg(conn->fdev, "Message with %u bytes received successfully\n",
+                     buf->sg[0].size);
+       conn->recv_cb(conn->cb_arg, buf);
+
+       buf->sg[0].size = MLX5_FPGA_RECV_SIZE;
+       err = mlx5_fpga_conn_post_recv(conn, buf);
+       if (unlikely(err)) {
+               mlx5_fpga_warn(conn->fdev,
+                              "Failed to re-post recv buf: %d\n", err);
+               kfree(buf);
+       }
+}
+
+static void mlx5_fpga_conn_sq_cqe(struct mlx5_fpga_conn *conn,
+                                 struct mlx5_cqe64 *cqe, u8 status)
+{
+       struct mlx5_fpga_dma_buf *buf, *nextbuf;
+       unsigned long flags;
+       int ix;
+
+       spin_lock_irqsave(&conn->qp.sq.lock, flags);
+
+       ix = be16_to_cpu(cqe->wqe_counter) & (conn->qp.sq.size - 1);
+       buf = conn->qp.sq.bufs[ix];
+       conn->qp.sq.bufs[ix] = NULL;
+       conn->qp.sq.cc++;
+
+       /* Handle backlog still under the spinlock to ensure message post order */
+       if (unlikely(!list_empty(&conn->qp.sq.backlog))) {
+               if (likely(conn->qp.active)) {
+                       nextbuf = list_first_entry(&conn->qp.sq.backlog,
+                                                  struct mlx5_fpga_dma_buf, list);
+                       list_del(&nextbuf->list);
+                       mlx5_fpga_conn_post_send(conn, nextbuf);
+               }
+       }
+
+       spin_unlock_irqrestore(&conn->qp.sq.lock, flags);
+
+       if (unlikely(status && (status != MLX5_CQE_SYNDROME_WR_FLUSH_ERR)))
+               mlx5_fpga_warn(conn->fdev, "SQ buf %p on FPGA QP %u completion status %d\n",
+                              buf, conn->fpga_qpn, status);
+       else
+               mlx5_fpga_dbg(conn->fdev, "SQ buf %p on FPGA QP %u completion status %d\n",
+                             buf, conn->fpga_qpn, status);
+
+       mlx5_fpga_conn_unmap_buf(conn, buf);
+
+       if (likely(buf->complete))
+               buf->complete(conn, conn->fdev, buf, status);
+
+       if (unlikely(status))
+               conn->qp.active = false;
+}
+
+static void mlx5_fpga_conn_handle_cqe(struct mlx5_fpga_conn *conn,
+                                     struct mlx5_cqe64 *cqe)
+{
+       u8 opcode, status = 0;
+
+       opcode = cqe->op_own >> 4;
+
+       switch (opcode) {
+       case MLX5_CQE_REQ_ERR:
+               status = ((struct mlx5_err_cqe *)cqe)->syndrome;
+               /* Fall through */
+       case MLX5_CQE_REQ:
+               mlx5_fpga_conn_sq_cqe(conn, cqe, status);
+               break;
+
+       case MLX5_CQE_RESP_ERR:
+               status = ((struct mlx5_err_cqe *)cqe)->syndrome;
+               /* Fall through */
+       case MLX5_CQE_RESP_SEND:
+               mlx5_fpga_conn_rq_cqe(conn, cqe, status);
+               break;
+       default:
+               mlx5_fpga_warn(conn->fdev, "Unexpected cqe opcode %u\n",
+                              opcode);
+       }
+}
+
+static void mlx5_fpga_conn_arm_cq(struct mlx5_fpga_conn *conn)
+{
+       mlx5_cq_arm(&conn->cq.mcq, MLX5_CQ_DB_REQ_NOT,
+                   conn->fdev->conn_res.uar->map, conn->cq.wq.cc);
+}
+
+static void mlx5_fpga_conn_cq_event(struct mlx5_core_cq *mcq,
+                                   enum mlx5_event event)
+{
+       struct mlx5_fpga_conn *conn;
+
+       conn = container_of(mcq, struct mlx5_fpga_conn, cq.mcq);
+       mlx5_fpga_warn(conn->fdev, "CQ event %u on CQ #%u\n", event, mcq->cqn);
+}
+
+static void mlx5_fpga_conn_event(struct mlx5_core_qp *mqp, int event)
+{
+       struct mlx5_fpga_conn *conn;
+
+       conn = container_of(mqp, struct mlx5_fpga_conn, qp.mqp);
+       mlx5_fpga_warn(conn->fdev, "QP event %u on QP #%u\n", event, mqp->qpn);
+}
+
+static inline void mlx5_fpga_conn_cqes(struct mlx5_fpga_conn *conn,
+                                      unsigned int budget)
+{
+       struct mlx5_cqe64 *cqe;
+
+       while (budget) {
+               cqe = mlx5_cqwq_get_cqe(&conn->cq.wq);
+               if (!cqe)
+                       break;
+
+               budget--;
+               mlx5_cqwq_pop(&conn->cq.wq);
+               mlx5_fpga_conn_handle_cqe(conn, cqe);
+               mlx5_cqwq_update_db_record(&conn->cq.wq);
+       }
+       if (!budget) {
+               tasklet_schedule(&conn->cq.tasklet);
+               return;
+       }
+
+       mlx5_fpga_dbg(conn->fdev, "Re-arming CQ with cc# %u\n", conn->cq.wq.cc);
+       /* ensure cq space is freed before enabling more cqes */
+       wmb();
+       mlx5_fpga_conn_arm_cq(conn);
+}
+
+static void mlx5_fpga_conn_cq_tasklet(unsigned long data)
+{
+       struct mlx5_fpga_conn *conn = (void *)data;
+
+       if (unlikely(!conn->qp.active))
+               return;
+       mlx5_fpga_conn_cqes(conn, MLX5_FPGA_CQ_BUDGET);
+}
+
+static void mlx5_fpga_conn_cq_complete(struct mlx5_core_cq *mcq)
+{
+       struct mlx5_fpga_conn *conn;
+
+       conn = container_of(mcq, struct mlx5_fpga_conn, cq.mcq);
+       if (unlikely(!conn->qp.active))
+               return;
+       mlx5_fpga_conn_cqes(conn, MLX5_FPGA_CQ_BUDGET);
+}
+
+static int mlx5_fpga_conn_create_cq(struct mlx5_fpga_conn *conn, int cq_size)
+{
+       struct mlx5_fpga_device *fdev = conn->fdev;
+       struct mlx5_core_dev *mdev = fdev->mdev;
+       u32 temp_cqc[MLX5_ST_SZ_DW(cqc)] = {0};
+       struct mlx5_wq_param wqp;
+       struct mlx5_cqe64 *cqe;
+       int inlen, err, eqn;
+       unsigned int irqn;
+       void *cqc, *in;
+       __be64 *pas;
+       u32 i;
+
+       cq_size = roundup_pow_of_two(cq_size);
+       MLX5_SET(cqc, temp_cqc, log_cq_size, ilog2(cq_size));
+
+       wqp.buf_numa_node = mdev->priv.numa_node;
+       wqp.db_numa_node  = mdev->priv.numa_node;
+
+       err = mlx5_cqwq_create(mdev, &wqp, temp_cqc, &conn->cq.wq,
+                              &conn->cq.wq_ctrl);
+       if (err)
+               return err;
+
+       for (i = 0; i < mlx5_cqwq_get_size(&conn->cq.wq); i++) {
+               cqe = mlx5_cqwq_get_wqe(&conn->cq.wq, i);
+               cqe->op_own = MLX5_CQE_INVALID << 4 | MLX5_CQE_OWNER_MASK;
+       }
+
+       inlen = MLX5_ST_SZ_BYTES(create_cq_in) +
+               sizeof(u64) * conn->cq.wq_ctrl.frag_buf.npages;
+       in = kvzalloc(inlen, GFP_KERNEL);
+       if (!in) {
+               err = -ENOMEM;
+               goto err_cqwq;
+       }
+
+       err = mlx5_vector2eqn(mdev, smp_processor_id(), &eqn, &irqn);
+       if (err)
+               goto err_cqwq;
+
+       cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
+       MLX5_SET(cqc, cqc, log_cq_size, ilog2(cq_size));
+       MLX5_SET(cqc, cqc, c_eqn, eqn);
+       MLX5_SET(cqc, cqc, uar_page, fdev->conn_res.uar->index);
+       MLX5_SET(cqc, cqc, log_page_size, conn->cq.wq_ctrl.frag_buf.page_shift -
+                          MLX5_ADAPTER_PAGE_SHIFT);
+       MLX5_SET64(cqc, cqc, dbr_addr, conn->cq.wq_ctrl.db.dma);
+
+       pas = (__be64 *)MLX5_ADDR_OF(create_cq_in, in, pas);
+       mlx5_fill_page_frag_array(&conn->cq.wq_ctrl.frag_buf, pas);
+
+       err = mlx5_core_create_cq(mdev, &conn->cq.mcq, in, inlen);
+       kvfree(in);
+
+       if (err)
+               goto err_cqwq;
+
+       conn->cq.mcq.cqe_sz     = 64;
+       conn->cq.mcq.set_ci_db  = conn->cq.wq_ctrl.db.db;
+       conn->cq.mcq.arm_db     = conn->cq.wq_ctrl.db.db + 1;
+       *conn->cq.mcq.set_ci_db = 0;
+       *conn->cq.mcq.arm_db    = 0;
+       conn->cq.mcq.vector     = 0;
+       conn->cq.mcq.comp       = mlx5_fpga_conn_cq_complete;
+       conn->cq.mcq.event      = mlx5_fpga_conn_cq_event;
+       conn->cq.mcq.irqn       = irqn;
+       conn->cq.mcq.uar        = fdev->conn_res.uar;
+       tasklet_init(&conn->cq.tasklet, mlx5_fpga_conn_cq_tasklet,
+                    (unsigned long)conn);
+
+       mlx5_fpga_dbg(fdev, "Created CQ #0x%x\n", conn->cq.mcq.cqn);
+
+       goto out;
+
+err_cqwq:
+       mlx5_cqwq_destroy(&conn->cq.wq_ctrl);
+out:
+       return err;
+}
+
+static void mlx5_fpga_conn_destroy_cq(struct mlx5_fpga_conn *conn)
+{
+       tasklet_disable(&conn->cq.tasklet);
+       tasklet_kill(&conn->cq.tasklet);
+       mlx5_core_destroy_cq(conn->fdev->mdev, &conn->cq.mcq);
+       mlx5_cqwq_destroy(&conn->cq.wq_ctrl);
+}
+
+static int mlx5_fpga_conn_create_wq(struct mlx5_fpga_conn *conn, void *qpc)
+{
+       struct mlx5_fpga_device *fdev = conn->fdev;
+       struct mlx5_core_dev *mdev = fdev->mdev;
+       struct mlx5_wq_param wqp;
+
+       wqp.buf_numa_node = mdev->priv.numa_node;
+       wqp.db_numa_node  = mdev->priv.numa_node;
+
+       return mlx5_wq_qp_create(mdev, &wqp, qpc, &conn->qp.wq,
+                                &conn->qp.wq_ctrl);
+}
+
+static int mlx5_fpga_conn_create_qp(struct mlx5_fpga_conn *conn,
+                                   unsigned int tx_size, unsigned int rx_size)
+{
+       struct mlx5_fpga_device *fdev = conn->fdev;
+       struct mlx5_core_dev *mdev = fdev->mdev;
+       u32 temp_qpc[MLX5_ST_SZ_DW(qpc)] = {0};
+       void *in = NULL, *qpc;
+       int err, inlen;
+
+       conn->qp.rq.pc = 0;
+       conn->qp.rq.cc = 0;
+       conn->qp.rq.size = roundup_pow_of_two(rx_size);
+       conn->qp.sq.pc = 0;
+       conn->qp.sq.cc = 0;
+       conn->qp.sq.size = roundup_pow_of_two(tx_size);
+
+       MLX5_SET(qpc, temp_qpc, log_rq_stride, ilog2(MLX5_SEND_WQE_DS) - 4);
+       MLX5_SET(qpc, temp_qpc, log_rq_size, ilog2(conn->qp.rq.size));
+       MLX5_SET(qpc, temp_qpc, log_sq_size, ilog2(conn->qp.sq.size));
+       err = mlx5_fpga_conn_create_wq(conn, temp_qpc);
+       if (err)
+               goto out;
+
+       conn->qp.rq.bufs = kvzalloc(sizeof(conn->qp.rq.bufs[0]) *
+                                   conn->qp.rq.size, GFP_KERNEL);
+       if (!conn->qp.rq.bufs) {
+               err = -ENOMEM;
+               goto err_wq;
+       }
+
+       conn->qp.sq.bufs = kvzalloc(sizeof(conn->qp.sq.bufs[0]) *
+                                   conn->qp.sq.size, GFP_KERNEL);
+       if (!conn->qp.sq.bufs) {
+               err = -ENOMEM;
+               goto err_rq_bufs;
+       }
+
+       inlen = MLX5_ST_SZ_BYTES(create_qp_in) +
+               MLX5_FLD_SZ_BYTES(create_qp_in, pas[0]) *
+               conn->qp.wq_ctrl.buf.npages;
+       in = kvzalloc(inlen, GFP_KERNEL);
+       if (!in) {
+               err = -ENOMEM;
+               goto err_sq_bufs;
+       }
+
+       qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
+       MLX5_SET(qpc, qpc, uar_page, fdev->conn_res.uar->index);
+       MLX5_SET(qpc, qpc, log_page_size,
+                conn->qp.wq_ctrl.buf.page_shift - MLX5_ADAPTER_PAGE_SHIFT);
+       MLX5_SET(qpc, qpc, fre, 1);
+       MLX5_SET(qpc, qpc, rlky, 1);
+       MLX5_SET(qpc, qpc, st, MLX5_QP_ST_RC);
+       MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
+       MLX5_SET(qpc, qpc, pd, fdev->conn_res.pdn);
+       MLX5_SET(qpc, qpc, log_rq_stride, ilog2(MLX5_SEND_WQE_DS) - 4);
+       MLX5_SET(qpc, qpc, log_rq_size, ilog2(conn->qp.rq.size));
+       MLX5_SET(qpc, qpc, rq_type, MLX5_NON_ZERO_RQ);
+       MLX5_SET(qpc, qpc, log_sq_size, ilog2(conn->qp.sq.size));
+       MLX5_SET(qpc, qpc, cqn_snd, conn->cq.mcq.cqn);
+       MLX5_SET(qpc, qpc, cqn_rcv, conn->cq.mcq.cqn);
+       MLX5_SET64(qpc, qpc, dbr_addr, conn->qp.wq_ctrl.db.dma);
+       if (MLX5_CAP_GEN(mdev, cqe_version) == 1)
+               MLX5_SET(qpc, qpc, user_index, 0xFFFFFF);
+
+       mlx5_fill_page_array(&conn->qp.wq_ctrl.buf,
+                            (__be64 *)MLX5_ADDR_OF(create_qp_in, in, pas));
+
+       err = mlx5_core_create_qp(mdev, &conn->qp.mqp, in, inlen);
+       if (err)
+               goto err_sq_bufs;
+
+       conn->qp.mqp.event = mlx5_fpga_conn_event;
+       mlx5_fpga_dbg(fdev, "Created QP #0x%x\n", conn->qp.mqp.qpn);
+
+       goto out;
+
+err_sq_bufs:
+       kvfree(conn->qp.sq.bufs);
+err_rq_bufs:
+       kvfree(conn->qp.rq.bufs);
+err_wq:
+       mlx5_wq_destroy(&conn->qp.wq_ctrl);
+out:
+       kvfree(in);
+       return err;
+}
+
+static void mlx5_fpga_conn_free_recv_bufs(struct mlx5_fpga_conn *conn)
+{
+       int ix;
+
+       for (ix = 0; ix < conn->qp.rq.size; ix++) {
+               if (!conn->qp.rq.bufs[ix])
+                       continue;
+               mlx5_fpga_conn_unmap_buf(conn, conn->qp.rq.bufs[ix]);
+               kfree(conn->qp.rq.bufs[ix]);
+               conn->qp.rq.bufs[ix] = NULL;
+       }
+}
+
+static void mlx5_fpga_conn_flush_send_bufs(struct mlx5_fpga_conn *conn)
+{
+       struct mlx5_fpga_dma_buf *buf, *temp;
+       int ix;
+
+       for (ix = 0; ix < conn->qp.sq.size; ix++) {
+               buf = conn->qp.sq.bufs[ix];
+               if (!buf)
+                       continue;
+               conn->qp.sq.bufs[ix] = NULL;
+               mlx5_fpga_conn_unmap_buf(conn, buf);
+               if (!buf->complete)
+                       continue;
+               buf->complete(conn, conn->fdev, buf, MLX5_CQE_SYNDROME_WR_FLUSH_ERR);
+       }
+       list_for_each_entry_safe(buf, temp, &conn->qp.sq.backlog, list) {
+               mlx5_fpga_conn_unmap_buf(conn, buf);
+               if (!buf->complete)
+                       continue;
+               buf->complete(conn, conn->fdev, buf, MLX5_CQE_SYNDROME_WR_FLUSH_ERR);
+       }
+}
+
+static void mlx5_fpga_conn_destroy_qp(struct mlx5_fpga_conn *conn)
+{
+       mlx5_core_destroy_qp(conn->fdev->mdev, &conn->qp.mqp);
+       mlx5_fpga_conn_free_recv_bufs(conn);
+       mlx5_fpga_conn_flush_send_bufs(conn);
+       kvfree(conn->qp.sq.bufs);
+       kvfree(conn->qp.rq.bufs);
+       mlx5_wq_destroy(&conn->qp.wq_ctrl);
+}
+
+static inline int mlx5_fpga_conn_reset_qp(struct mlx5_fpga_conn *conn)
+{
+       struct mlx5_core_dev *mdev = conn->fdev->mdev;
+
+       mlx5_fpga_dbg(conn->fdev, "Modifying QP %u to RST\n", conn->qp.mqp.qpn);
+
+       return mlx5_core_qp_modify(mdev, MLX5_CMD_OP_2RST_QP, 0, NULL,
+                                  &conn->qp.mqp);
+}
+
+static inline int mlx5_fpga_conn_init_qp(struct mlx5_fpga_conn *conn)
+{
+       struct mlx5_fpga_device *fdev = conn->fdev;
+       struct mlx5_core_dev *mdev = fdev->mdev;
+       u32 *qpc = NULL;
+       int err;
+
+       mlx5_fpga_dbg(conn->fdev, "Modifying QP %u to INIT\n", conn->qp.mqp.qpn);
+
+       qpc = kzalloc(MLX5_ST_SZ_BYTES(qpc), GFP_KERNEL);
+       if (!qpc) {
+               err = -ENOMEM;
+               goto out;
+       }
+
+       MLX5_SET(qpc, qpc, st, MLX5_QP_ST_RC);
+       MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
+       MLX5_SET(qpc, qpc, primary_address_path.pkey_index, MLX5_FPGA_PKEY_INDEX);
+       MLX5_SET(qpc, qpc, primary_address_path.port, MLX5_FPGA_PORT_NUM);
+       MLX5_SET(qpc, qpc, pd, conn->fdev->conn_res.pdn);
+       MLX5_SET(qpc, qpc, cqn_snd, conn->cq.mcq.cqn);
+       MLX5_SET(qpc, qpc, cqn_rcv, conn->cq.mcq.cqn);
+       MLX5_SET64(qpc, qpc, dbr_addr, conn->qp.wq_ctrl.db.dma);
+
+       err = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_RST2INIT_QP, 0, qpc,
+                                 &conn->qp.mqp);
+       if (err) {
+               mlx5_fpga_warn(fdev, "qp_modify RST2INIT failed: %d\n", err);
+               goto out;
+       }
+
+out:
+       kfree(qpc);
+       return err;
+}
+
+static inline int mlx5_fpga_conn_rtr_qp(struct mlx5_fpga_conn *conn)
+{
+       struct mlx5_fpga_device *fdev = conn->fdev;
+       struct mlx5_core_dev *mdev = fdev->mdev;
+       u32 *qpc = NULL;
+       int err;
+
+       mlx5_fpga_dbg(conn->fdev, "QP RTR\n");
+
+       qpc = kzalloc(MLX5_ST_SZ_BYTES(qpc), GFP_KERNEL);
+       if (!qpc) {
+               err = -ENOMEM;
+               goto out;
+       }
+
+       MLX5_SET(qpc, qpc, mtu, MLX5_QPC_MTU_1K_BYTES);
+       MLX5_SET(qpc, qpc, log_msg_max, (u8)MLX5_CAP_GEN(mdev, log_max_msg));
+       MLX5_SET(qpc, qpc, remote_qpn, conn->fpga_qpn);
+       MLX5_SET(qpc, qpc, next_rcv_psn,
+                MLX5_GET(fpga_qpc, conn->fpga_qpc, next_send_psn));
+       MLX5_SET(qpc, qpc, primary_address_path.pkey_index, MLX5_FPGA_PKEY_INDEX);
+       MLX5_SET(qpc, qpc, primary_address_path.port, MLX5_FPGA_PORT_NUM);
+       ether_addr_copy(MLX5_ADDR_OF(qpc, qpc, primary_address_path.rmac_47_32),
+                       MLX5_ADDR_OF(fpga_qpc, conn->fpga_qpc, fpga_mac_47_32));
+       MLX5_SET(qpc, qpc, primary_address_path.udp_sport,
+                MLX5_CAP_ROCE(mdev, r_roce_min_src_udp_port));
+       MLX5_SET(qpc, qpc, primary_address_path.src_addr_index,
+                conn->qp.sgid_index);
+       MLX5_SET(qpc, qpc, primary_address_path.hop_limit, 0);
+       memcpy(MLX5_ADDR_OF(qpc, qpc, primary_address_path.rgid_rip),
+              MLX5_ADDR_OF(fpga_qpc, conn->fpga_qpc, fpga_ip),
+              MLX5_FLD_SZ_BYTES(qpc, primary_address_path.rgid_rip));
+
+       err = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_INIT2RTR_QP, 0, qpc,
+                                 &conn->qp.mqp);
+       if (err) {
+               mlx5_fpga_warn(fdev, "qp_modify RST2INIT failed: %d\n", err);
+               goto out;
+       }
+
+out:
+       kfree(qpc);
+       return err;
+}
+
+static inline int mlx5_fpga_conn_rts_qp(struct mlx5_fpga_conn *conn)
+{
+       struct mlx5_fpga_device *fdev = conn->fdev;
+       struct mlx5_core_dev *mdev = fdev->mdev;
+       u32 *qpc = NULL;
+       u32 opt_mask;
+       int err;
+
+       mlx5_fpga_dbg(conn->fdev, "QP RTS\n");
+
+       qpc = kzalloc(MLX5_ST_SZ_BYTES(qpc), GFP_KERNEL);
+       if (!qpc) {
+               err = -ENOMEM;
+               goto out;
+       }
+
+       MLX5_SET(qpc, qpc, log_ack_req_freq, 8);
+       MLX5_SET(qpc, qpc, min_rnr_nak, 0x12);
+       MLX5_SET(qpc, qpc, primary_address_path.ack_timeout, 0x12); /* ~1.07s */
+       MLX5_SET(qpc, qpc, next_send_psn,
+                MLX5_GET(fpga_qpc, conn->fpga_qpc, next_rcv_psn));
+       MLX5_SET(qpc, qpc, retry_count, 7);
+       MLX5_SET(qpc, qpc, rnr_retry, 7); /* Infinite retry if RNR NACK */
+
+       opt_mask = MLX5_QP_OPTPAR_RNR_TIMEOUT;
+       err = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_RTR2RTS_QP, opt_mask, qpc,
+                                 &conn->qp.mqp);
+       if (err) {
+               mlx5_fpga_warn(fdev, "qp_modify RST2INIT failed: %d\n", err);
+               goto out;
+       }
+
+out:
+       kfree(qpc);
+       return err;
+}
+
+static int mlx5_fpga_conn_connect(struct mlx5_fpga_conn *conn)
+{
+       struct mlx5_fpga_device *fdev = conn->fdev;
+       int err;
+
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, state, MLX5_FPGA_QPC_STATE_ACTIVE);
+       err = mlx5_fpga_modify_qp(conn->fdev->mdev, conn->fpga_qpn,
+                                 MLX5_FPGA_QPC_STATE, &conn->fpga_qpc);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to activate FPGA RC QP: %d\n", err);
+               goto out;
+       }
+
+       err = mlx5_fpga_conn_reset_qp(conn);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to change QP state to reset\n");
+               goto err_fpga_qp;
+       }
+
+       err = mlx5_fpga_conn_init_qp(conn);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to modify QP from RESET to INIT\n");
+               goto err_fpga_qp;
+       }
+       conn->qp.active = true;
+
+       while (!mlx5_fpga_conn_post_recv_buf(conn))
+               ;
+
+       err = mlx5_fpga_conn_rtr_qp(conn);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to change QP state from INIT to RTR\n");
+               goto err_recv_bufs;
+       }
+
+       err = mlx5_fpga_conn_rts_qp(conn);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to change QP state from RTR to RTS\n");
+               goto err_recv_bufs;
+       }
+       goto out;
+
+err_recv_bufs:
+       mlx5_fpga_conn_free_recv_bufs(conn);
+err_fpga_qp:
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, state, MLX5_FPGA_QPC_STATE_INIT);
+       if (mlx5_fpga_modify_qp(conn->fdev->mdev, conn->fpga_qpn,
+                               MLX5_FPGA_QPC_STATE, &conn->fpga_qpc))
+               mlx5_fpga_err(fdev, "Failed to revert FPGA QP to INIT\n");
+out:
+       return err;
+}
+
+struct mlx5_fpga_conn *mlx5_fpga_conn_create(struct mlx5_fpga_device *fdev,
+                                            struct mlx5_fpga_conn_attr *attr,
+                                            enum mlx5_ifc_fpga_qp_type qp_type)
+{
+       struct mlx5_fpga_conn *ret, *conn;
+       u8 *remote_mac, *remote_ip;
+       int err;
+
+       if (!attr->recv_cb)
+               return ERR_PTR(-EINVAL);
+
+       conn = kzalloc(sizeof(*conn), GFP_KERNEL);
+       if (!conn)
+               return ERR_PTR(-ENOMEM);
+
+       conn->fdev = fdev;
+       INIT_LIST_HEAD(&conn->qp.sq.backlog);
+
+       spin_lock_init(&conn->qp.sq.lock);
+
+       conn->recv_cb = attr->recv_cb;
+       conn->cb_arg = attr->cb_arg;
+
+       remote_mac = MLX5_ADDR_OF(fpga_qpc, conn->fpga_qpc, remote_mac_47_32);
+       err = mlx5_query_nic_vport_mac_address(fdev->mdev, 0, remote_mac);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to query local MAC: %d\n", err);
+               ret = ERR_PTR(err);
+               goto err;
+       }
+
+       /* Build Modified EUI-64 IPv6 address from the MAC address */
+       remote_ip = MLX5_ADDR_OF(fpga_qpc, conn->fpga_qpc, remote_ip);
+       remote_ip[0] = 0xfe;
+       remote_ip[1] = 0x80;
+       addrconf_addr_eui48(&remote_ip[8], remote_mac);
+
+       err = mlx5_core_reserved_gid_alloc(fdev->mdev, &conn->qp.sgid_index);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to allocate SGID: %d\n", err);
+               ret = ERR_PTR(err);
+               goto err;
+       }
+
+       err = mlx5_core_roce_gid_set(fdev->mdev, conn->qp.sgid_index,
+                                    MLX5_ROCE_VERSION_2,
+                                    MLX5_ROCE_L3_TYPE_IPV6,
+                                    remote_ip, remote_mac, true, 0);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to set SGID: %d\n", err);
+               ret = ERR_PTR(err);
+               goto err_rsvd_gid;
+       }
+       mlx5_fpga_dbg(fdev, "Reserved SGID index %u\n", conn->qp.sgid_index);
+
+       /* Allow for one cqe per rx/tx wqe, plus one cqe for the next wqe,
+        * created during processing of the cqe
+        */
+       err = mlx5_fpga_conn_create_cq(conn,
+                                      (attr->tx_size + attr->rx_size) * 2);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to create CQ: %d\n", err);
+               ret = ERR_PTR(err);
+               goto err_gid;
+       }
+
+       mlx5_fpga_conn_arm_cq(conn);
+
+       err = mlx5_fpga_conn_create_qp(conn, attr->tx_size, attr->rx_size);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to create QP: %d\n", err);
+               ret = ERR_PTR(err);
+               goto err_cq;
+       }
+
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, state, MLX5_FPGA_QPC_STATE_INIT);
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, qp_type, qp_type);
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, st, MLX5_FPGA_QPC_ST_RC);
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, ether_type, ETH_P_8021Q);
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, vid, 0);
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, next_rcv_psn, 1);
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, next_send_psn, 0);
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, pkey, MLX5_FPGA_PKEY);
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, remote_qpn, conn->qp.mqp.qpn);
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, rnr_retry, 7);
+       MLX5_SET(fpga_qpc, conn->fpga_qpc, retry_count, 7);
+
+       err = mlx5_fpga_create_qp(fdev->mdev, &conn->fpga_qpc,
+                                 &conn->fpga_qpn);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to create FPGA RC QP: %d\n", err);
+               ret = ERR_PTR(err);
+               goto err_qp;
+       }
+
+       err = mlx5_fpga_conn_connect(conn);
+       if (err) {
+               ret = ERR_PTR(err);
+               goto err_conn;
+       }
+
+       mlx5_fpga_dbg(fdev, "FPGA QPN is %u\n", conn->fpga_qpn);
+       ret = conn;
+       goto out;
+
+err_conn:
+       mlx5_fpga_destroy_qp(conn->fdev->mdev, conn->fpga_qpn);
+err_qp:
+       mlx5_fpga_conn_destroy_qp(conn);
+err_cq:
+       mlx5_fpga_conn_destroy_cq(conn);
+err_gid:
+       mlx5_core_roce_gid_set(fdev->mdev, conn->qp.sgid_index, 0, 0, NULL,
+                              NULL, false, 0);
+err_rsvd_gid:
+       mlx5_core_reserved_gid_free(fdev->mdev, conn->qp.sgid_index);
+err:
+       kfree(conn);
+out:
+       return ret;
+}
+
+void mlx5_fpga_conn_destroy(struct mlx5_fpga_conn *conn)
+{
+       struct mlx5_fpga_device *fdev = conn->fdev;
+       struct mlx5_core_dev *mdev = fdev->mdev;
+       int err = 0;
+
+       conn->qp.active = false;
+       tasklet_disable(&conn->cq.tasklet);
+       synchronize_irq(conn->cq.mcq.irqn);
+
+       mlx5_fpga_destroy_qp(conn->fdev->mdev, conn->fpga_qpn);
+       err = mlx5_core_qp_modify(mdev, MLX5_CMD_OP_2ERR_QP, 0, NULL,
+                                 &conn->qp.mqp);
+       if (err)
+               mlx5_fpga_warn(fdev, "qp_modify 2ERR failed: %d\n", err);
+       mlx5_fpga_conn_destroy_qp(conn);
+       mlx5_fpga_conn_destroy_cq(conn);
+
+       mlx5_core_roce_gid_set(conn->fdev->mdev, conn->qp.sgid_index, 0, 0,
+                              NULL, NULL, false, 0);
+       mlx5_core_reserved_gid_free(conn->fdev->mdev, conn->qp.sgid_index);
+       kfree(conn);
+}
+
+int mlx5_fpga_conn_device_init(struct mlx5_fpga_device *fdev)
+{
+       int err;
+
+       err = mlx5_nic_vport_enable_roce(fdev->mdev);
+       if (err) {
+               mlx5_fpga_err(fdev, "Failed to enable RoCE: %d\n", err);
+               goto out;
+       }
+
+       fdev->conn_res.uar = mlx5_get_uars_page(fdev->mdev);
+       if (IS_ERR(fdev->conn_res.uar)) {
+               err = PTR_ERR(fdev->conn_res.uar);
+               mlx5_fpga_err(fdev, "get_uars_page failed, %d\n", err);
+               goto err_roce;
+       }
+       mlx5_fpga_dbg(fdev, "Allocated UAR index %u\n",
+                     fdev->conn_res.uar->index);
+
+       err = mlx5_core_alloc_pd(fdev->mdev, &fdev->conn_res.pdn);
+       if (err) {
+               mlx5_fpga_err(fdev, "alloc pd failed, %d\n", err);
+               goto err_uar;
+       }
+       mlx5_fpga_dbg(fdev, "Allocated PD %u\n", fdev->conn_res.pdn);
+
+       err = mlx5_fpga_conn_create_mkey(fdev->mdev, fdev->conn_res.pdn,
+                                        &fdev->conn_res.mkey);
+       if (err) {
+               mlx5_fpga_err(fdev, "create mkey failed, %d\n", err);
+               goto err_dealloc_pd;
+       }
+       mlx5_fpga_dbg(fdev, "Created mkey 0x%x\n", fdev->conn_res.mkey.key);
+
+       return 0;
+
+err_dealloc_pd:
+       mlx5_core_dealloc_pd(fdev->mdev, fdev->conn_res.pdn);
+err_uar:
+       mlx5_put_uars_page(fdev->mdev, fdev->conn_res.uar);
+err_roce:
+       mlx5_nic_vport_disable_roce(fdev->mdev);
+out:
+       return err;
+}
+
+void mlx5_fpga_conn_device_cleanup(struct mlx5_fpga_device *fdev)
+{
+       mlx5_core_destroy_mkey(fdev->mdev, &fdev->conn_res.mkey);
+       mlx5_core_dealloc_pd(fdev->mdev, fdev->conn_res.pdn);
+       mlx5_put_uars_page(fdev->mdev, fdev->conn_res.uar);
+       mlx5_nic_vport_disable_roce(fdev->mdev);
+}
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.h b/drivers/net/ethernet/mellanox/mlx5/core/fpga/conn.h
new file mode 100644 (file)
index 0000000..44bd9ec
--- /dev/null
@@ -0,0 +1,96 @@
+/*
+ * Copyright (c) 2017 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.
+ *
+ */
+
+#ifndef __MLX5_FPGA_CONN_H__
+#define __MLX5_FPGA_CONN_H__
+
+#include <linux/mlx5/cq.h>
+#include <linux/mlx5/qp.h>
+
+#include "fpga/core.h"
+#include "fpga/sdk.h"
+#include "wq.h"
+
+struct mlx5_fpga_conn {
+       struct mlx5_fpga_device *fdev;
+
+       void (*recv_cb)(void *cb_arg, struct mlx5_fpga_dma_buf *buf);
+       void *cb_arg;
+
+       /* FPGA QP */
+       u32 fpga_qpc[MLX5_ST_SZ_DW(fpga_qpc)];
+       u32 fpga_qpn;
+
+       /* CQ */
+       struct {
+               struct mlx5_cqwq wq;
+               struct mlx5_frag_wq_ctrl wq_ctrl;
+               struct mlx5_core_cq mcq;
+               struct tasklet_struct tasklet;
+       } cq;
+
+       /* QP */
+       struct {
+               bool active;
+               int sgid_index;
+               struct mlx5_wq_qp wq;
+               struct mlx5_wq_ctrl wq_ctrl;
+               struct mlx5_core_qp mqp;
+               struct {
+                       spinlock_t lock; /* Protects all SQ state */
+                       unsigned int pc;
+                       unsigned int cc;
+                       unsigned int size;
+                       struct mlx5_fpga_dma_buf **bufs;
+                       struct list_head backlog;
+               } sq;
+               struct {
+                       unsigned int pc;
+                       unsigned int cc;
+                       unsigned int size;
+                       struct mlx5_fpga_dma_buf **bufs;
+               } rq;
+       } qp;
+};
+
+int mlx5_fpga_conn_device_init(struct mlx5_fpga_device *fdev);
+void mlx5_fpga_conn_device_cleanup(struct mlx5_fpga_device *fdev);
+struct mlx5_fpga_conn *
+mlx5_fpga_conn_create(struct mlx5_fpga_device *fdev,
+                     struct mlx5_fpga_conn_attr *attr,
+                     enum mlx5_ifc_fpga_qp_type qp_type);
+void mlx5_fpga_conn_destroy(struct mlx5_fpga_conn *conn);
+int mlx5_fpga_conn_send(struct mlx5_fpga_conn *conn,
+                       struct mlx5_fpga_dma_buf *buf);
+
+#endif /* __MLX5_FPGA_CONN_H__ */
index c3bb4b865f0160772a6814c0d4c69f818ab368c0..7f859a3ad5d212bfad642e538a7ef809c0493dc8 100644 (file)
@@ -37,6 +37,7 @@
 #include "mlx5_core.h"
 #include "lib/mlx5.h"
 #include "fpga/core.h"
+#include "fpga/conn.h"
 
 static const char *const mlx5_fpga_error_strings[] = {
        "Null Syndrome",
@@ -127,7 +128,17 @@ int mlx5_fpga_device_start(struct mlx5_core_dev *mdev)
 
        max_num_qps = MLX5_CAP_FPGA(mdev, shell_caps.max_num_qps);
        err = mlx5_core_reserve_gids(mdev, max_num_qps);
+       if (err)
+               goto out;
+
+       err = mlx5_fpga_conn_device_init(fdev);
+       if (err)
+               goto err_rsvd_gid;
+
+       goto out;
 
+err_rsvd_gid:
+       mlx5_core_unreserve_gids(mdev, max_num_qps);
 out:
        spin_lock_irqsave(&fdev->state_lock, flags);
        fdev->state = err ? MLX5_FPGA_STATUS_FAILURE : MLX5_FPGA_STATUS_SUCCESS;
@@ -173,6 +184,7 @@ void mlx5_fpga_device_stop(struct mlx5_core_dev *mdev)
        fdev->state = MLX5_FPGA_STATUS_NONE;
        spin_unlock_irqrestore(&fdev->state_lock, flags);
 
+       mlx5_fpga_conn_device_cleanup(fdev);
        max_num_qps = MLX5_CAP_FPGA(mdev, shell_caps.max_num_qps);
        mlx5_core_unreserve_gids(mdev, max_num_qps);
 }
index db1d22c356e0eda94b23cd62c47c247ba28205cc..f64fa1cdc195450000f9c4c3e2dab559ea38a129 100644 (file)
@@ -44,6 +44,13 @@ struct mlx5_fpga_device {
        enum mlx5_fpga_status state;
        enum mlx5_fpga_image last_admin_image;
        enum mlx5_fpga_image last_oper_image;
+
+       /* QP Connection resources */
+       struct {
+               u32 pdn;
+               struct mlx5_core_mkey mkey;
+               struct mlx5_uars_page *uar;
+       } conn_res;
 };
 
 #define mlx5_fpga_dbg(__adev, format, ...) \
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fpga/sdk.h b/drivers/net/ethernet/mellanox/mlx5/core/fpga/sdk.h
new file mode 100644 (file)
index 0000000..331798d
--- /dev/null
@@ -0,0 +1,106 @@
+/*
+ * Copyright (c) 2017 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.
+ *
+ */
+
+#ifndef MLX5_FPGA_SDK_H
+#define MLX5_FPGA_SDK_H
+
+#include <linux/types.h>
+#include <linux/dma-direction.h>
+
+/**
+ * DOC: Innova SDK
+ * This header defines the in-kernel API for Innova FPGA client drivers.
+ */
+
+struct mlx5_fpga_conn;
+struct mlx5_fpga_device;
+
+/**
+ * struct mlx5_fpga_dma_entry - A scatter-gather DMA entry
+ */
+struct mlx5_fpga_dma_entry {
+       /** @data: Virtual address pointer to the data */
+       void *data;
+       /** @size: Size in bytes of the data */
+       unsigned int size;
+       /** @dma_addr: Private member. Physical DMA-mapped address of the data */
+       dma_addr_t dma_addr;
+};
+
+/**
+ * struct mlx5_fpga_dma_buf - A packet buffer
+ * May contain up to 2 scatter-gather data entries
+ */
+struct mlx5_fpga_dma_buf {
+       /** @dma_dir: DMA direction */
+       enum dma_data_direction dma_dir;
+       /** @sg: Scatter-gather entries pointing to the data in memory */
+       struct mlx5_fpga_dma_entry sg[2];
+       /** @list: Item in SQ backlog, for TX packets */
+       struct list_head list;
+       /**
+        * @complete: Completion routine, for TX packets
+        * @conn: FPGA Connection this packet was sent to
+        * @fdev: FPGA device this packet was sent to
+        * @buf: The packet buffer
+        * @status: 0 if successful, or an error code otherwise
+        */
+       void (*complete)(struct mlx5_fpga_conn *conn,
+                        struct mlx5_fpga_device *fdev,
+                        struct mlx5_fpga_dma_buf *buf, u8 status);
+};
+
+/**
+ * struct mlx5_fpga_conn_attr - FPGA connection attributes
+ * Describes the attributes of a connection
+ */
+struct mlx5_fpga_conn_attr {
+       /** @tx_size: Size of connection TX queue, in packets */
+       unsigned int tx_size;
+       /** @rx_size: Size of connection RX queue, in packets */
+       unsigned int rx_size;
+       /**
+        * @recv_cb: Callback function which is called for received packets
+        * @cb_arg: The value provided in mlx5_fpga_conn_attr.cb_arg
+        * @buf: A buffer containing a received packet
+        *
+        * buf is guaranteed to only contain a single scatter-gather entry.
+        * The size of the actual packet received is specified in buf.sg[0].size
+        * When this callback returns, the packet buffer may be re-used for
+        * subsequent receives.
+        */
+       void (*recv_cb)(void *cb_arg, struct mlx5_fpga_dma_buf *buf);
+       void *cb_arg;
+};
+
+#endif /* MLX5_FPGA_SDK_H */