From 02d89b87081f516ad3993637f9b75db0d9786554 Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Wed, 23 May 2007 15:16:08 -0700 Subject: IB/mlx4: Don't allocate RQ doorbell if using SRQ If a QP is attached to a shared receive queue (SRQ), then it doesn't have a receive queue (RQ). So don't allocate an RQ doorbell (or map a doorbell from userspace for userspace QPs) for that QP. Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/qp.c | 37 ++++++++++++++++++++++--------------- 1 file changed, 22 insertions(+), 15 deletions(-) (limited to 'drivers') diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c index a824bc5f79f..88a994d8a13 100644 --- a/drivers/infiniband/hw/mlx4/qp.c +++ b/drivers/infiniband/hw/mlx4/qp.c @@ -319,20 +319,24 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, if (err) goto err_mtt; - err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context), - ucmd.db_addr, &qp->db); - if (err) - goto err_mtt; + if (!init_attr->srq) { + err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context), + ucmd.db_addr, &qp->db); + if (err) + goto err_mtt; + } } else { err = set_kernel_sq_size(dev, &init_attr->cap, init_attr->qp_type, qp); if (err) goto err; - err = mlx4_ib_db_alloc(dev, &qp->db, 0); - if (err) - goto err; + if (!init_attr->srq) { + err = mlx4_ib_db_alloc(dev, &qp->db, 0); + if (err) + goto err; - *qp->db.db = 0; + *qp->db.db = 0; + } if (mlx4_buf_alloc(dev->dev, qp->buf_size, PAGE_SIZE * 2, &qp->buf)) { err = -ENOMEM; @@ -386,7 +390,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, return 0; err_wrid: - if (pd->uobject) + if (pd->uobject && !init_attr->srq) mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &qp->db); else { kfree(qp->sq.wrid); @@ -403,7 +407,7 @@ err_buf: mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf); err_db: - if (!pd->uobject) + if (!pd->uobject && !init_attr->srq) mlx4_ib_db_free(dev, &qp->db); err: @@ -481,14 +485,16 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp, mlx4_mtt_cleanup(dev->dev, &qp->mtt); if (is_user) { - mlx4_ib_db_unmap_user(to_mucontext(qp->ibqp.uobject->context), - &qp->db); + if (!qp->ibqp.srq) + mlx4_ib_db_unmap_user(to_mucontext(qp->ibqp.uobject->context), + &qp->db); ib_umem_release(qp->umem); } else { kfree(qp->sq.wrid); kfree(qp->rq.wrid); mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf); - mlx4_ib_db_free(dev, &qp->db); + if (!qp->ibqp.srq) + mlx4_ib_db_free(dev, &qp->db); } } @@ -852,7 +858,7 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp, if (ibqp->srq) context->srqn = cpu_to_be32(1 << 24 | to_msrq(ibqp->srq)->msrq.srqn); - if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) + if (!ibqp->srq && 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 && @@ -919,7 +925,8 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp, qp->rq.tail = 0; qp->sq.head = 0; qp->sq.tail = 0; - *qp->db.db = 0; + if (!ibqp->srq) + *qp->db.db = 0; } out: -- cgit From c0be5fb5f835110652911ea8b88ad78f841e5b45 Mon Sep 17 00:00:00 2001 From: Eli Cohen Date: Thu, 24 May 2007 16:05:01 +0300 Subject: IB/mlx4: Initialize send queue entry ownership bits We need to initialize the owner bit of send queue WQEs to hardware ownership whenever the QP is modified from reset to init, not just when the QP is first allocated. This avoids having the hardware process stale WQEs when the QP is moved to reset but not destroyed and then modified to init again. Signed-off-by: Eli Cohen Signed-off-by: Roland Dreier --- drivers/infiniband/hw/mlx4/qp.c | 22 +++++++++++++++------- 1 file changed, 15 insertions(+), 7 deletions(-) (limited to 'drivers') diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c index 88a994d8a13..dc137dec230 100644 --- a/drivers/infiniband/hw/mlx4/qp.c +++ b/drivers/infiniband/hw/mlx4/qp.c @@ -270,9 +270,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, struct ib_qp_init_attr *init_attr, struct ib_udata *udata, int sqpn, struct mlx4_ib_qp *qp) { - struct mlx4_wqe_ctrl_seg *ctrl; int err; - int i; mutex_init(&qp->mutex); spin_lock_init(&qp->sq.lock); @@ -352,11 +350,6 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd, if (err) goto err_mtt; - for (i = 0; i < qp->sq.max; ++i) { - ctrl = get_send_wqe(qp, i); - ctrl->owner_opcode = cpu_to_be32(1 << 31); - } - qp->sq.wrid = kmalloc(qp->sq.max * sizeof (u64), GFP_KERNEL); qp->rq.wrid = kmalloc(qp->rq.max * sizeof (u64), GFP_KERNEL); @@ -878,6 +871,21 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp, else sqd_event = 0; + /* + * Before passing a kernel QP to the HW, make sure that the + * ownership bits of the send queue are set so that the + * hardware doesn't start processing stale work requests. + */ + if (!ibqp->uobject && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) { + struct mlx4_wqe_ctrl_seg *ctrl; + int i; + + for (i = 0; i < qp->sq.max; ++i) { + ctrl = get_send_wqe(qp, i); + ctrl->owner_opcode = cpu_to_be32(1 << 31); + } + } + err = mlx4_qp_modify(dev->dev, &qp->mtt, to_mlx4_state(cur_state), to_mlx4_state(new_state), context, optpar, sqd_event, &qp->mqp); -- cgit From 65a2c841d68ae3402ea4cad8d00fe4b9b0a5bc80 Mon Sep 17 00:00:00 2001 From: Stefan Roscher Date: Thu, 24 May 2007 16:51:08 +0200 Subject: IB/ehca: Fix number of send WRs reported for new QP Due to a typo, the driver was reporting the wrong number of "actual send WRs" after ehca_create_qp(). Signed-off-by: Joachim Fenkes Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ehca/hcp_if.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/infiniband/hw/ehca/hcp_if.c b/drivers/infiniband/hw/ehca/hcp_if.c index 7f0beec74f7..5766ae3a202 100644 --- a/drivers/infiniband/hw/ehca/hcp_if.c +++ b/drivers/infiniband/hw/ehca/hcp_if.c @@ -331,7 +331,7 @@ u64 hipz_h_alloc_resource_qp(const struct ipz_adapter_handle adapter_handle, 0); qp->ipz_qp_handle.handle = outs[0]; qp->real_qp_num = (u32)outs[1]; - parms->act_nr_send_sges = + parms->act_nr_send_wqes = (u16)EHCA_BMASK_GET(H_ALL_RES_QP_ACT_OUTST_SEND_WR, outs[2]); parms->act_nr_recv_wqes = (u16)EHCA_BMASK_GET(H_ALL_RES_QP_ACT_OUTST_RECV_WR, outs[2]); -- cgit From 8fd357a6e3375083f7d321413eb8f6739491f342 Mon Sep 17 00:00:00 2001 From: "Michael S. Tsirkin" Date: Thu, 24 May 2007 14:02:39 -0700 Subject: IPoIB/cm: Fix timeout check in ipoib_cm_dev_stop() time_after() was used backwards, so the timeout occurred immediately. Signed-off-by: Michael S. Tsirkin Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/ipoib/ipoib_cm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/infiniband/ulp/ipoib/ipoib_cm.c b/drivers/infiniband/ulp/ipoib/ipoib_cm.c index ffec794b791..e8fcd628be7 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_cm.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_cm.c @@ -713,7 +713,7 @@ void ipoib_cm_dev_stop(struct net_device *dev) while (!list_empty(&priv->cm.rx_error_list) || !list_empty(&priv->cm.rx_flush_list) || !list_empty(&priv->cm.rx_drain_list)) { - if (!time_after(jiffies, begin + 5 * HZ)) { + if (time_after(jiffies, begin + 5 * HZ)) { ipoib_warn(priv, "RX drain timing out\n"); /* -- cgit From 2dfbfc37121d307e1f1d24c2979382cb17b19347 Mon Sep 17 00:00:00 2001 From: "Michael S. Tsirkin" Date: Thu, 24 May 2007 18:32:46 +0300 Subject: IPoIB/cm: Drain cq in ipoib_cm_dev_stop() Since NAPI polling is disabled while ipoib_cm_dev_stop() is running, ipoib_cm_dev_stop() must poll the CQ itself in order to see the packets draining. Signed-off-by: Michael S. Tsirkin Signed-off-by: Roland Dreier --- drivers/infiniband/ulp/ipoib/ipoib.h | 1 + drivers/infiniband/ulp/ipoib/ipoib_cm.c | 1 + drivers/infiniband/ulp/ipoib/ipoib_ib.c | 31 +++++++++++++++++++------------ 3 files changed, 21 insertions(+), 12 deletions(-) (limited to 'drivers') diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h index a0b3782c762..158759e28a5 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib.h +++ b/drivers/infiniband/ulp/ipoib/ipoib.h @@ -429,6 +429,7 @@ int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey); void ipoib_pkey_poll(struct work_struct *work); int ipoib_pkey_dev_delay_open(struct net_device *dev); +void ipoib_drain_cq(struct net_device *dev); #ifdef CONFIG_INFINIBAND_IPOIB_CM diff --git a/drivers/infiniband/ulp/ipoib/ipoib_cm.c b/drivers/infiniband/ulp/ipoib/ipoib_cm.c index e8fcd628be7..f133b56fd97 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_cm.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_cm.c @@ -726,6 +726,7 @@ void ipoib_cm_dev_stop(struct net_device *dev) } spin_unlock_irq(&priv->lock); msleep(1); + ipoib_drain_cq(dev); spin_lock_irq(&priv->lock); } diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c index c1aad06eb4e..8404f05b2b6 100644 --- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c +++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c @@ -550,13 +550,30 @@ static int recvs_pending(struct net_device *dev) return pending; } +void ipoib_drain_cq(struct net_device *dev) +{ + struct ipoib_dev_priv *priv = netdev_priv(dev); + int i, n; + do { + n = ib_poll_cq(priv->cq, IPOIB_NUM_WC, priv->ibwc); + for (i = 0; i < n; ++i) { + if (priv->ibwc[i].wr_id & IPOIB_CM_OP_SRQ) + ipoib_cm_handle_rx_wc(dev, priv->ibwc + i); + else if (priv->ibwc[i].wr_id & IPOIB_OP_RECV) + ipoib_ib_handle_rx_wc(dev, priv->ibwc + i); + else + ipoib_ib_handle_tx_wc(dev, priv->ibwc + i); + } + } while (n == IPOIB_NUM_WC); +} + int ipoib_ib_dev_stop(struct net_device *dev, int flush) { struct ipoib_dev_priv *priv = netdev_priv(dev); struct ib_qp_attr qp_attr; unsigned long begin; struct ipoib_tx_buf *tx_req; - int i, n; + int i; clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags); netif_poll_disable(dev); @@ -611,17 +628,7 @@ int ipoib_ib_dev_stop(struct net_device *dev, int flush) goto timeout; } - do { - n = ib_poll_cq(priv->cq, IPOIB_NUM_WC, priv->ibwc); - for (i = 0; i < n; ++i) { - if (priv->ibwc[i].wr_id & IPOIB_CM_OP_SRQ) - ipoib_cm_handle_rx_wc(dev, priv->ibwc + i); - else if (priv->ibwc[i].wr_id & IPOIB_OP_RECV) - ipoib_ib_handle_rx_wc(dev, priv->ibwc + i); - else - ipoib_ib_handle_tx_wc(dev, priv->ibwc + i); - } - } while (n == IPOIB_NUM_WC); + ipoib_drain_cq(dev); msleep(1); } -- cgit