RE-sending with correct subject line (should be V2).
Support for XRC RCV-only QP (requested by userspace,
but resides in kernel space).
Implements create_xrc_rcv_qp, modify_xrc_rcv_qp, query_xrc_rcv_qp,
and destroy_xrc_rcv_qp.
Since query_xrc_rcv_qp and query_qp are very similar, the common portions
have been placed in a helper function which is invoked by both query methods.
Signed-off-by: Jack Morgenstein <ja...@dev.mellanox.co.il>
---
drivers/infiniband/hw/mlx4/cq.c | 4 +-
drivers/infiniband/hw/mlx4/main.c | 60 ++++++++--
drivers/infiniband/hw/mlx4/mlx4_ib.h | 13 ++
drivers/infiniband/hw/mlx4/qp.c | 208 +++++++++++++++++++++++++++++++++-
4 files changed, 272 insertions(+), 13 deletions(-)
Index: infiniband/drivers/infiniband/hw/mlx4/cq.c
===================================================================
--- infiniband.orig/drivers/infiniband/hw/mlx4/cq.c
+++ infiniband/drivers/infiniband/hw/mlx4/cq.c
@@ -176,7 +176,7 @@ struct ib_cq *mlx4_ib_create_cq(struct i
if (entries < 1 || entries > dev->dev->caps.max_cqes)
return ERR_PTR(-EINVAL);
- cq = kmalloc(sizeof *cq, GFP_KERNEL);
+ cq = kzalloc(sizeof *cq, GFP_KERNEL);
if (!cq)
return ERR_PTR(-ENOMEM);
@@ -545,7 +545,7 @@ static int mlx4_ib_poll_one(struct mlx4_
struct mlx4_cqe *cqe;
struct mlx4_qp *mqp;
struct mlx4_ib_wq *wq;
- struct mlx4_ib_srq *srq;
+ struct mlx4_ib_srq *uninitialized_var(srq);
struct mlx4_srq *msrq;
int is_send;
int is_error;
Index: infiniband/drivers/infiniband/hw/mlx4/main.c
===================================================================
--- infiniband.orig/drivers/infiniband/hw/mlx4/main.c
+++ infiniband/drivers/infiniband/hw/mlx4/main.c
@@ -420,7 +420,7 @@ static struct ib_pd *mlx4_ib_alloc_pd(st
struct mlx4_ib_pd *pd;
int err;
- pd = kmalloc(sizeof *pd, GFP_KERNEL);
+ pd = kzalloc(sizeof *pd, GFP_KERNEL);
if (!pd)
return ERR_PTR(-ENOMEM);
@@ -462,12 +462,18 @@ static int mlx4_ib_mcg_detach(struct ib_
&to_mqp(ibqp)->mqp, gid->raw);
}
+static void mlx4_dummy_comp_handler(struct ib_cq *cq, void *cq_context)
+{
+}
+
static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
struct ib_ucontext *context,
struct ib_udata *udata)
{
struct mlx4_ib_xrcd *xrcd;
struct mlx4_ib_dev *mdev = to_mdev(ibdev);
+ struct ib_pd *pd;
+ struct ib_cq *cq;
int err;
if (!(mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
@@ -478,23 +484,51 @@ static struct ib_xrcd *mlx4_ib_alloc_xrc
return ERR_PTR(-ENOMEM);
err = mlx4_xrcd_alloc(mdev->dev, &xrcd->xrcdn);
- if (err) {
- kfree(xrcd);
- return ERR_PTR(err);
+ if (err)
+ goto err_xrcd;
+
+ pd = mlx4_ib_alloc_pd(ibdev, NULL, NULL);
+ if (IS_ERR(pd)) {
+ err = PTR_ERR(pd);
+ goto err_pd;
}
+ pd->device = ibdev;
+
+ cq = mlx4_ib_create_cq(ibdev, 1, 0, NULL, NULL);
+ if (IS_ERR(cq)) {
+ err = PTR_ERR(cq);
+ goto err_cq;
+ }
+ cq->device = ibdev;
+ cq->comp_handler = mlx4_dummy_comp_handler;
if (context)
if (ib_copy_to_udata(udata, &xrcd->xrcdn, sizeof(__u32))) {
- mlx4_xrcd_free(mdev->dev, xrcd->xrcdn);
- kfree(xrcd);
- return ERR_PTR(-EFAULT);
+ err = -EFAULT;
+ goto err_copy;
}
+ xrcd->cq = cq;
+ xrcd->pd = pd;
return &xrcd->ibxrcd;
+
+err_copy:
+ mlx4_ib_destroy_cq(cq);
+err_cq:
+ mlx4_ib_dealloc_pd(pd);
+err_pd:
+ mlx4_xrcd_free(mdev->dev, xrcd->xrcdn);
+err_xrcd:
+ kfree(xrcd);
+ return ERR_PTR(err);
}
static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
{
+ struct mlx4_ib_xrcd *mxrcd = to_mxrcd(xrcd);
+
+ mlx4_ib_destroy_cq(mxrcd->cq);
+ mlx4_ib_dealloc_pd(mxrcd->pd);
mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
kfree(xrcd);
@@ -700,18 +734,28 @@ static void *mlx4_ib_add(struct mlx4_dev
ibdev->ib_dev.create_xrc_srq = mlx4_ib_create_xrc_srq;
ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
+ ibdev->ib_dev.create_xrc_rcv_qp = mlx4_ib_create_xrc_rcv_qp;
+ ibdev->ib_dev.modify_xrc_rcv_qp = mlx4_ib_modify_xrc_rcv_qp;
+ ibdev->ib_dev.query_xrc_rcv_qp = mlx4_ib_query_xrc_rcv_qp;
+ ibdev->ib_dev.destroy_xrc_rcv_qp = mlx4_ib_destroy_xrc_rcv_qp;
ibdev->ib_dev.uverbs_cmd_mask |=
(1ull << IB_USER_VERBS_CMD_CREATE_XRC_SRQ) |
(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
- (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
+ (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD) |
+ (1ull << IB_USER_VERBS_CMD_CREATE_XRC_RCV_QP) |
+ (1ull << IB_USER_VERBS_CMD_MODIFY_XRC_RCV_QP) |
+ (1ull << IB_USER_VERBS_CMD_QUERY_XRC_RCV_QP) |
+ (1ull << IB_USER_VERBS_CMD_DESTROY_XRC_RCV_QP) |
+ (1ull << IB_USER_VERBS_CMD_REG_XRC_RCV_QP) |
+ (1ull << IB_USER_VERBS_CMD_UNREG_XRC_RCV_QP);
}
-
if (init_node_data(ibdev))
goto err_map;
spin_lock_init(&ibdev->sm_lock);
mutex_init(&ibdev->cap_mask_mutex);
+ mutex_init(&ibdev->xrc_rcv_mutex);
if (ib_register_device(&ibdev->ib_dev))
goto err_map;
Index: infiniband/drivers/infiniband/hw/mlx4/mlx4_ib.h
===================================================================
--- infiniband.orig/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ infiniband/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -59,6 +59,8 @@ struct mlx4_ib_pd {
struct mlx4_ib_xrcd {
struct ib_xrcd ibxrcd;
u32 xrcdn;
+ struct ib_pd *pd;
+ struct ib_cq *cq;
};
struct mlx4_ib_cq_buf {
@@ -115,6 +117,7 @@ struct mlx4_ib_wq {
enum mlx4_ib_qp_flags {
MLX4_IB_QP_LSO = 1 << 0,
MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK = 1 << 1,
+ MLX4_IB_XRC_RCV = 1 << 2,
};
struct mlx4_ib_qp {
@@ -181,6 +184,7 @@ struct mlx4_ib_dev {
spinlock_t sm_lock;
struct mutex cap_mask_mutex;
+ struct mutex xrc_rcv_mutex;
bool ib_active;
};
@@ -329,6 +333,15 @@ int mlx4_ib_map_phys_fmr(struct ib_fmr *
u64 iova);
int mlx4_ib_unmap_fmr(struct list_head *fmr_list);
int mlx4_ib_fmr_dealloc(struct ib_fmr *fmr);
+int mlx4_ib_create_xrc_rcv_qp(struct ib_qp_init_attr *init_attr,
+ u32 *qp_num);
+int mlx4_ib_modify_xrc_rcv_qp(struct ib_xrcd *xrcd, u32 qp_num,
+ struct ib_qp_attr *attr, int attr_mask);
+int mlx4_ib_query_xrc_rcv_qp(struct ib_xrcd *xrcd, u32 qp_num,
+ struct ib_qp_attr *attr, int attr_mask,
+ struct ib_qp_init_attr *init_attr);
+int mlx4_ib_destroy_xrc_rcv_qp(struct ib_xrcd *xrcd, u32 qp_num);
+
static inline int mlx4_ib_ah_grh_present(struct mlx4_ib_ah *ah)
{
Index: infiniband/drivers/infiniband/hw/mlx4/qp.c
===================================================================
--- infiniband.orig/drivers/infiniband/hw/mlx4/qp.c
+++ infiniband/drivers/infiniband/hw/mlx4/qp.c
@@ -59,6 +59,7 @@ enum {
MLX4_IB_LSO_HEADER_SPARE = 128,
};
+
struct mlx4_ib_sqp {
struct mlx4_ib_qp qp;
int pkey_index;
@@ -209,14 +210,14 @@ static inline unsigned pad_wraparound(st
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;
+ struct mlx4_ib_qp *mqp = to_mibqp(qp);
+ struct ib_qp *ibqp = &mqp->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;
@@ -248,6 +249,12 @@ static void mlx4_ib_qp_event(struct mlx4
return;
}
+ if (unlikely(ibqp->qp_type == IB_QPT_XRC &&
+ mqp->flags & MLX4_IB_XRC_RCV)) {
+ event.event |= IB_XRC_QP_EVENT_FLAG;
+ event.element.xrc_qp_num = ibqp->qp_num;
+ } else
+ event.element.qp = ibqp;
ibqp->event_handler(&event, ibqp->qp_context);
}
}
@@ -1886,27 +1893,15 @@ static void to_ib_ah_attr(struct mlx4_de
}
}
-int mlx4_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int
qp_attr_mask,
- struct ib_qp_init_attr *qp_init_attr)
+int mlx4_ib_query_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
+ struct ib_qp_attr *qp_attr,
+ struct ib_qp_init_attr *qp_init_attr)
{
- struct mlx4_ib_dev *dev = to_mdev(ibqp->device);
- struct mlx4_ib_qp *qp = to_mqp(ibqp);
struct mlx4_qp_context context;
int mlx4_state;
- int err = 0;
- mutex_lock(&qp->mutex);
-
- if (qp->state == IB_QPS_RESET) {
- qp_attr->qp_state = IB_QPS_RESET;
- goto done;
- }
-
- err = mlx4_qp_query(dev->dev, &qp->mqp, &context);
- if (err) {
- err = -EINVAL;
- goto out;
- }
+ if (mlx4_qp_query(dev->dev, &qp->mqp, &context))
+ return -EINVAL;
mlx4_state = be32_to_cpu(context.flags) >> 28;
@@ -1950,7 +1945,26 @@ int mlx4_ib_query_qp(struct ib_qp *ibqp,
qp_attr->rnr_retry = (be32_to_cpu(context.params1) >> 13) &
0x7;
qp_attr->alt_timeout = context.alt_path.ackto >> 3;
-done:
+ return 0;
+}
+
+int mlx4_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int
qp_attr_mask,
+ struct ib_qp_init_attr *qp_init_attr)
+{
+ struct mlx4_ib_dev *dev = to_mdev(ibqp->device);
+ struct mlx4_ib_qp *qp = to_mqp(ibqp);
+ int err = 0;
+
+ mutex_lock(&qp->mutex);
+
+ if (qp->state == IB_QPS_RESET)
+ qp_attr->qp_state = IB_QPS_RESET;
+ else {
+ err = mlx4_ib_query_qp_common(dev, qp, qp_attr, qp_init_attr);
+ if (err)
+ goto out;
+ }
+
qp_attr->cur_qp_state = qp_attr->qp_state;
qp_attr->cap.max_recv_wr = qp->rq.wqe_cnt;
qp_attr->cap.max_recv_sge = qp->rq.max_gs;
@@ -1983,3 +1997,164 @@ out:
return err;
}
+int mlx4_ib_create_xrc_rcv_qp(struct ib_qp_init_attr *init_attr,
+ u32 *qp_num)
+{
+ struct mlx4_ib_dev *dev = to_mdev(init_attr->xrcd->device);
+ struct mlx4_ib_xrcd *xrcd = to_mxrcd(init_attr->xrcd);
+ struct ib_qp_init_attr lcl_init;
+ struct mlx4_ib_qp *qp;
+ struct ib_qp *ibqp;
+ int err;
+
+ if (!(dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
+ return -ENOSYS;
+
+ memcpy(&lcl_init, init_attr, sizeof(struct ib_qp_init_attr));
+ lcl_init.qp_type = IB_QPT_XRC;
+ lcl_init.sq_sig_type = 0;
+ lcl_init.cap.max_inline_data = 0;
+ lcl_init.cap.max_recv_sge = 0;
+ lcl_init.cap.max_recv_wr = 0;
+ lcl_init.cap.max_send_sge = 1;
+ lcl_init.cap.max_send_wr = 1;
+
+ qp = kzalloc(sizeof *qp, GFP_KERNEL);
+ if (!qp)
+ return -ENOMEM;
+
+ qp->flags = MLX4_IB_XRC_RCV;
+ qp->xrcdn = xrcd->xrcdn;
+ err = create_qp_common(dev, xrcd->pd, &lcl_init, NULL, 0, qp);
+ if (err) {
+ kfree(qp);
+ return err;
+ }
+
+ ibqp = &qp->ibqp;
+ /* set the ibpq attributes which will be used by the mlx4 module */
+ ibqp->qp_num = qp->mqp.qpn;
+ ibqp->device = lcl_init.xrcd->device;
+ ibqp->pd = xrcd->pd;
+ ibqp->send_cq = ibqp->recv_cq = xrcd->cq;
+ ibqp->event_handler = lcl_init.event_handler;
+ ibqp->qp_context = lcl_init.qp_context;
+ ibqp->qp_type = lcl_init.qp_type;
+ ibqp->xrcd = lcl_init.xrcd;
+
+ *qp_num = qp->mqp.qpn;
+ return 0;
+}
+
+int mlx4_ib_modify_xrc_rcv_qp(struct ib_xrcd *ibxrcd, u32 qp_num,
+ struct ib_qp_attr *attr, int attr_mask)
+{
+ struct mlx4_ib_dev *dev = to_mdev(ibxrcd->device);
+ struct mlx4_ib_xrcd *xrcd = to_mxrcd(ibxrcd);
+ struct mlx4_ib_qp *qp;
+ struct mlx4_qp *mqp;
+ int err = -EINVAL;
+
+ if (!(dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
+ return -ENOSYS;
+
+ mutex_lock(&dev->xrc_rcv_mutex);
+ mqp = __mlx4_qp_lookup(dev->dev, qp_num);
+ if (unlikely(!mqp)) {
+ printk(KERN_WARNING "mlx4_ib_modify_xrc_rcv_qp: "
+ "unknown QPN %06x\n", qp_num);
+ goto out;
+ }
+
+ qp = to_mibqp(mqp);
+ if (qp->ibqp.qp_type != IB_QPT_XRC || !(qp->flags & MLX4_IB_XRC_RCV) ||
+ !qp->ibqp.xrcd || xrcd->xrcdn != to_mxrcd(qp->ibqp.xrcd)->xrcdn)
+ goto out;
+
+ err = mlx4_ib_modify_qp(&qp->ibqp, attr, attr_mask, NULL);
+
+out:
+ mutex_unlock(&dev->xrc_rcv_mutex);
+ return err;
+}
+
+int mlx4_ib_query_xrc_rcv_qp(struct ib_xrcd *ibxrcd, u32 qp_num,
+ struct ib_qp_attr *qp_attr, int qp_attr_mask,
+ struct ib_qp_init_attr *qp_init_attr)
+{
+ struct mlx4_ib_dev *dev = to_mdev(ibxrcd->device);
+ struct mlx4_ib_xrcd *xrcd = to_mxrcd(ibxrcd);
+ struct mlx4_ib_qp *qp;
+ struct mlx4_qp *mqp;
+ int err = -EINVAL;
+
+ if (!(dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
+ return -ENOSYS;
+
+ mutex_lock(&dev->xrc_rcv_mutex);
+ mqp = __mlx4_qp_lookup(dev->dev, qp_num);
+ if (unlikely(!mqp)) {
+ printk(KERN_WARNING "mlx4_ib_query_xrc_rcv_qp: "
+ "unknown QPN %06x\n", qp_num);
+ goto err_out;
+ }
+
+ qp = to_mibqp(mqp);
+ if (qp->ibqp.qp_type != IB_QPT_XRC || !(qp->flags & MLX4_IB_XRC_RCV) ||
+ !qp->ibqp.xrcd || xrcd->xrcdn != to_mxrcd(qp->ibqp.xrcd)->xrcdn)
+ goto err_out;
+
+ if (qp->state == IB_QPS_RESET)
+ qp_attr->qp_state = IB_QPS_RESET;
+ else {
+ err = mlx4_ib_query_qp_common(dev, qp, qp_attr, qp_init_attr);
+ if (err)
+ goto err_out;
+ }
+
+ qp_attr->cur_qp_state = qp_attr->qp_state;
+ qp_attr->cap.max_recv_wr = 0;
+ qp_attr->cap.max_recv_sge = 0;
+ qp_attr->cap.max_send_wr = 0;
+ qp_attr->cap.max_send_sge = 0;
+ qp_attr->cap.max_inline_data = 0;
+ qp_init_attr->cap = qp_attr->cap;
+
+ err = 0;
+
+err_out:
+ mutex_unlock(&dev->xrc_rcv_mutex);
+ return err;
+}
+
+int mlx4_ib_destroy_xrc_rcv_qp(struct ib_xrcd *ibxrcd, u32 qp_num)
+{
+ struct mlx4_ib_dev *dev = to_mdev(ibxrcd->device);
+ struct mlx4_ib_xrcd *xrcd = to_mxrcd(ibxrcd);
+ struct mlx4_ib_qp *qp;
+ struct mlx4_qp *mqp;
+ int err = -EINVAL;
+
+ mutex_lock(&dev->xrc_rcv_mutex);
+ mqp = __mlx4_qp_lookup(dev->dev, qp_num);
+ if (unlikely(!mqp)) {
+ printk(KERN_WARNING "mlx4_ib_destroy_xrc_rcv_qp: "
+ "unknown QPN %06x\n", qp_num);
+ goto out;
+ }
+
+ qp = to_mibqp(mqp);
+
+ if (qp->ibqp.qp_type != IB_QPT_XRC || !(qp->flags & MLX4_IB_XRC_RCV) ||
+ !qp->ibqp.xrcd || xrcd->xrcdn != to_mxrcd(qp->ibqp.xrcd)->xrcdn)
+ goto out;
+
+ mlx4_ib_destroy_qp(&qp->ibqp);
+
+ err = 0;
+
+out:
+ mutex_unlock(&dev->xrc_rcv_mutex);
+ return err;
+}
+
-------------------------------------------------------
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
