X-Git-Url: http://pilppa.org/gitweb/?a=blobdiff_plain;f=drivers%2Finfiniband%2Fhw%2Fmthca%2Fmthca_qp.c;h=760c418d5bc9b3b77ab52f054035fa2782dfe130;hb=78b9c0f91cf908616b8f9f356e1d1220e727ea88;hp=0164b84d4ec643062892122cf6ebf567d92ec5f1;hpb=ef88b7dba2b47c70037a34a599d383462bb74bd3;p=linux-2.6-omap-h63xx.git diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c index 0164b84d4ec..760c418d5bc 100644 --- a/drivers/infiniband/hw/mthca/mthca_qp.c +++ b/drivers/infiniband/hw/mthca/mthca_qp.c @@ -36,6 +36,8 @@ */ #include +#include +#include #include #include @@ -220,6 +222,15 @@ static void *get_send_wqe(struct mthca_qp *qp, int n) (PAGE_SIZE - 1)); } +static void mthca_wq_init(struct mthca_wq *wq) +{ + spin_lock_init(&wq->lock); + wq->next_ind = 0; + wq->last_comp = wq->max - 1; + wq->head = 0; + wq->tail = 0; +} + void mthca_qp_event(struct mthca_dev *dev, u32 qpn, enum ib_event_type event_type) { @@ -329,8 +340,7 @@ static const struct { [UC] = (IB_QP_AV | IB_QP_PATH_MTU | IB_QP_DEST_QPN | - IB_QP_RQ_PSN | - IB_QP_MAX_DEST_RD_ATOMIC), + IB_QP_RQ_PSN), [RC] = (IB_QP_AV | IB_QP_PATH_MTU | IB_QP_DEST_QPN | @@ -359,8 +369,7 @@ static const struct { .trans = MTHCA_TRANS_RTR2RTS, .req_param = { [UD] = IB_QP_SQ_PSN, - [UC] = (IB_QP_SQ_PSN | - IB_QP_MAX_QP_RD_ATOMIC), + [UC] = IB_QP_SQ_PSN, [RC] = (IB_QP_TIMEOUT | IB_QP_RETRY_CNT | IB_QP_RNR_RETRY | @@ -437,8 +446,6 @@ static const struct { [UD] = (IB_QP_PKEY_INDEX | IB_QP_QKEY), [UC] = (IB_QP_AV | - IB_QP_MAX_QP_RD_ATOMIC | - IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_CUR_STATE | IB_QP_ALT_PATH | IB_QP_ACCESS_FLAGS | @@ -469,7 +476,7 @@ static const struct { .opt_param = { [UD] = (IB_QP_CUR_STATE | IB_QP_QKEY), - [UC] = (IB_QP_CUR_STATE), + [UC] = IB_QP_CUR_STATE, [RC] = (IB_QP_CUR_STATE | IB_QP_MIN_RNR_TIMER), [MLX] = (IB_QP_CUR_STATE | @@ -577,6 +584,13 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask) return -EINVAL; } + if ((attr_mask & IB_QP_PKEY_INDEX) && + attr->pkey_index >= dev->limits.pkey_table_len) { + mthca_dbg(dev, "PKey index (%u) too large. max is %d\n", + attr->pkey_index,dev->limits.pkey_table_len-1); + return -EINVAL; + } + mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL); if (IS_ERR(mailbox)) return PTR_ERR(mailbox); @@ -677,7 +691,7 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask) } if (attr_mask & IB_QP_TIMEOUT) { - qp_context->pri_path.ackto = attr->timeout; + qp_context->pri_path.ackto = attr->timeout << 3; qp_param->opt_param_mask |= cpu_to_be32(MTHCA_QP_OPTPAR_ACK_TIMEOUT); } @@ -833,8 +847,8 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask) store_attrs(to_msqp(qp), attr, attr_mask); /* - * If we are moving QP0 to RTR, bring the IB link up; if we - * are moving QP0 to RESET or ERROR, bring the link back down. + * 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 && @@ -848,9 +862,71 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask) mthca_CLOSE_IB(dev, to_msqp(qp)->port, &status); } + /* + * If we moved a kernel QP to RESET, clean up all old CQ + * entries and reinitialize the QP. + */ + if (!err && new_state == IB_QPS_RESET && !qp->ibqp.uobject) { + mthca_cq_clean(dev, to_mcq(qp->ibqp.send_cq)->cqn, qp->qpn, + qp->ibqp.srq ? to_msrq(qp->ibqp.srq) : NULL); + if (qp->ibqp.send_cq != qp->ibqp.recv_cq) + mthca_cq_clean(dev, to_mcq(qp->ibqp.recv_cq)->cqn, qp->qpn, + qp->ibqp.srq ? to_msrq(qp->ibqp.srq) : NULL); + + mthca_wq_init(&qp->sq); + mthca_wq_init(&qp->rq); + + if (mthca_is_memfree(dev)) { + *qp->sq.db = 0; + *qp->rq.db = 0; + } + } + return err; } +static void mthca_adjust_qp_caps(struct mthca_dev *dev, + struct mthca_pd *pd, + struct mthca_qp *qp) +{ + int max_data_size; + + /* + * Calculate the maximum size of WQE s/g segments, excluding + * the next segment and other non-data segments. + */ + max_data_size = min(dev->limits.max_desc_sz, 1 << qp->sq.wqe_shift) - + sizeof (struct mthca_next_seg); + + switch (qp->transport) { + case MLX: + max_data_size -= 2 * sizeof (struct mthca_data_seg); + break; + + case UD: + if (mthca_is_memfree(dev)) + max_data_size -= sizeof (struct mthca_arbel_ud_seg); + else + max_data_size -= sizeof (struct mthca_tavor_ud_seg); + break; + + default: + max_data_size -= sizeof (struct mthca_raddr_seg); + break; + } + + /* We don't support inline data for kernel QPs (yet). */ + if (!pd->ibpd.uobject) + qp->max_inline_data = 0; + else + qp->max_inline_data = max_data_size - MTHCA_INLINE_HEADER_SIZE; + + qp->sq.max_gs = max_data_size / sizeof (struct mthca_data_seg); + qp->rq.max_gs = (min(dev->limits.max_desc_sz, 1 << qp->rq.wqe_shift) - + sizeof (struct mthca_next_seg)) / + sizeof (struct mthca_data_seg); +} + /* * Allocate and register buffer for WQEs. qp->rq.max, sq.max, * rq.max_gs and sq.max_gs must all be assigned. @@ -868,27 +944,53 @@ static int mthca_alloc_wqe_buf(struct mthca_dev *dev, size = sizeof (struct mthca_next_seg) + qp->rq.max_gs * sizeof (struct mthca_data_seg); + if (size > dev->limits.max_desc_sz) + return -EINVAL; + for (qp->rq.wqe_shift = 6; 1 << qp->rq.wqe_shift < size; qp->rq.wqe_shift++) ; /* nothing */ - size = sizeof (struct mthca_next_seg) + - qp->sq.max_gs * sizeof (struct mthca_data_seg); + size = qp->sq.max_gs * sizeof (struct mthca_data_seg); switch (qp->transport) { case MLX: size += 2 * sizeof (struct mthca_data_seg); break; + case UD: - if (mthca_is_memfree(dev)) - size += sizeof (struct mthca_arbel_ud_seg); - else - size += sizeof (struct mthca_tavor_ud_seg); + size += mthca_is_memfree(dev) ? + sizeof (struct mthca_arbel_ud_seg) : + sizeof (struct mthca_tavor_ud_seg); + break; + + case UC: + size += sizeof (struct mthca_raddr_seg); + break; + + case RC: + size += sizeof (struct mthca_raddr_seg); + /* + * An atomic op will require an atomic segment, a + * remote address segment and one scatter entry. + */ + size = max_t(int, size, + sizeof (struct mthca_atomic_seg) + + sizeof (struct mthca_raddr_seg) + + sizeof (struct mthca_data_seg)); break; + default: - /* bind seg is as big as atomic + raddr segs */ - size += sizeof (struct mthca_bind_seg); + break; } + /* Make sure that we have enough space for a bind request */ + size = max_t(int, size, sizeof (struct mthca_bind_seg)); + + size += sizeof (struct mthca_next_seg); + + if (size > dev->limits.max_desc_sz) + return -EINVAL; + for (qp->sq.wqe_shift = 6; 1 << qp->sq.wqe_shift < size; qp->sq.wqe_shift++) ; /* nothing */ @@ -1003,16 +1105,6 @@ static void mthca_free_memfree(struct mthca_dev *dev, } } -static void mthca_wq_init(struct mthca_wq* wq) -{ - spin_lock_init(&wq->lock); - wq->next_ind = 0; - wq->last_comp = wq->max - 1; - wq->head = 0; - wq->tail = 0; - wq->last = NULL; -} - static int mthca_alloc_qp_common(struct mthca_dev *dev, struct mthca_pd *pd, struct mthca_cq *send_cq, @@ -1024,6 +1116,7 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev, int i; atomic_set(&qp->refcount, 1); + init_waitqueue_head(&qp->wait); qp->state = IB_QPS_RESET; qp->atomic_rd_en = 0; qp->resp_depth = 0; @@ -1041,6 +1134,8 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev, return ret; } + mthca_adjust_qp_caps(dev, pd, qp); + /* * If this is a userspace QP, we're done now. The doorbells * will be allocated and buffers will be initialized in @@ -1082,6 +1177,9 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev, } } + qp->sq.last = get_send_wqe(qp, qp->sq.max - 1); + qp->rq.last = get_recv_wqe(qp, qp->rq.max - 1); + return 0; } @@ -1089,8 +1187,10 @@ static int mthca_set_qp_size(struct mthca_dev *dev, struct ib_qp_cap *cap, struct mthca_qp *qp) { /* Sanity check QP size before proceeding */ - if (cap->max_send_wr > 65536 || cap->max_recv_wr > 65536 || - cap->max_send_sge > 64 || cap->max_recv_sge > 64) + if (cap->max_send_wr > dev->limits.max_wqes || + cap->max_recv_wr > dev->limits.max_wqes || + cap->max_send_sge > dev->limits.max_sg || + cap->max_recv_sge > dev->limits.max_sg) return -EINVAL; if (mthca_is_memfree(dev)) { @@ -1456,8 +1556,8 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, } wqe += sizeof (struct mthca_atomic_seg); - size += sizeof (struct mthca_raddr_seg) / 16 + - sizeof (struct mthca_atomic_seg); + size += (sizeof (struct mthca_raddr_seg) + + sizeof (struct mthca_atomic_seg)) / 16; break; case IB_WR_RDMA_WRITE: @@ -1562,15 +1662,13 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, goto out; } - if (prev_wqe) { - ((struct mthca_next_seg *) prev_wqe)->nda_op = - cpu_to_be32(((ind << qp->sq.wqe_shift) + - qp->send_wqe_offset) | - mthca_opcode[wr->opcode]); - wmb(); - ((struct mthca_next_seg *) prev_wqe)->ee_nds = - cpu_to_be32((size0 ? 0 : MTHCA_NEXT_DBD) | size); - } + ((struct mthca_next_seg *) prev_wqe)->nda_op = + cpu_to_be32(((ind << qp->sq.wqe_shift) + + qp->send_wqe_offset) | + mthca_opcode[wr->opcode]); + wmb(); + ((struct mthca_next_seg *) prev_wqe)->ee_nds = + cpu_to_be32((size0 ? 0 : MTHCA_NEXT_DBD) | size); if (!size0) { size0 = size; @@ -1609,6 +1707,7 @@ int mthca_tavor_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr, { struct mthca_dev *dev = to_mdev(ibqp->device); struct mthca_qp *qp = to_mqp(ibqp); + __be32 doorbell[2]; unsigned long flags; int err = 0; int nreq; @@ -1626,6 +1725,22 @@ int mthca_tavor_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr, ind = qp->rq.next_ind; for (nreq = 0; wr; ++nreq, wr = wr->next) { + if (unlikely(nreq == MTHCA_TAVOR_MAX_WQES_PER_RECV_DB)) { + nreq = 0; + + doorbell[0] = cpu_to_be32((qp->rq.next_ind << qp->rq.wqe_shift) | size0); + doorbell[1] = cpu_to_be32(qp->qpn << 8); + + wmb(); + + mthca_write64(doorbell, + dev->kar + MTHCA_RECEIVE_DOORBELL, + MTHCA_GET_DOORBELL_LOCK(&dev->doorbell_lock)); + + qp->rq.head += MTHCA_TAVOR_MAX_WQES_PER_RECV_DB; + size0 = 0; + } + if (mthca_wq_overflow(&qp->rq, nreq, qp->ibqp.recv_cq)) { mthca_err(dev, "RQ %06x full (%u head, %u tail," " %d max, %d nreq)\n", qp->qpn, @@ -1667,13 +1782,11 @@ int mthca_tavor_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr, qp->wrid[ind] = wr->wr_id; - if (likely(prev_wqe)) { - ((struct mthca_next_seg *) prev_wqe)->nda_op = - cpu_to_be32((ind << qp->rq.wqe_shift) | 1); - wmb(); - ((struct mthca_next_seg *) prev_wqe)->ee_nds = - cpu_to_be32(MTHCA_NEXT_DBD | size); - } + ((struct mthca_next_seg *) prev_wqe)->nda_op = + cpu_to_be32((ind << qp->rq.wqe_shift) | 1); + wmb(); + ((struct mthca_next_seg *) prev_wqe)->ee_nds = + cpu_to_be32(MTHCA_NEXT_DBD | size); if (!size0) size0 = size; @@ -1685,8 +1798,6 @@ int mthca_tavor_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr, out: if (likely(nreq)) { - __be32 doorbell[2]; - doorbell[0] = cpu_to_be32((qp->rq.next_ind << qp->rq.wqe_shift) | size0); doorbell[1] = cpu_to_be32((qp->qpn << 8) | nreq); @@ -1780,8 +1891,8 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, } wqe += sizeof (struct mthca_atomic_seg); - size += sizeof (struct mthca_raddr_seg) / 16 + - sizeof (struct mthca_atomic_seg); + size += (sizeof (struct mthca_raddr_seg) + + sizeof (struct mthca_atomic_seg)) / 16; break; case IB_WR_RDMA_READ: @@ -1884,15 +1995,13 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, goto out; } - if (likely(prev_wqe)) { - ((struct mthca_next_seg *) prev_wqe)->nda_op = - cpu_to_be32(((ind << qp->sq.wqe_shift) + - qp->send_wqe_offset) | - mthca_opcode[wr->opcode]); - wmb(); - ((struct mthca_next_seg *) prev_wqe)->ee_nds = - cpu_to_be32(MTHCA_NEXT_DBD | size); - } + ((struct mthca_next_seg *) prev_wqe)->nda_op = + cpu_to_be32(((ind << qp->sq.wqe_shift) + + qp->send_wqe_offset) | + mthca_opcode[wr->opcode]); + wmb(); + ((struct mthca_next_seg *) prev_wqe)->ee_nds = + cpu_to_be32(MTHCA_NEXT_DBD | size); if (!size0) { size0 = size; @@ -2106,5 +2215,6 @@ void __devexit mthca_cleanup_qp_table(struct mthca_dev *dev) for (i = 0; i < 2; ++i) mthca_CONF_SPECIAL_QP(dev, i, 0, &status); + mthca_array_cleanup(&dev->qp_table.qp, dev->limits.num_qps); mthca_alloc_cleanup(&dev->qp_table.alloc); }