From 0db3dfa03c0881fc98d3ff2f88dcca2bc69c0003 Mon Sep 17 00:00:00 2001 From: Jianxin Xiong Date: Mon, 25 Jul 2016 13:38:37 -0700 Subject: [PATCH] IB/hfi1: Work request processing for fast register mr and invalidate In order to support extended memory management support, add send side processing of work requests of type IB_WR_REG_MR, IB_WR_LOCAL_INV, and IB_WR_SEND_WITH_INV. The first two are local operations and are supported for both RC and UC. Send with invalidate is only supported for RC because the corresponding IB opcodes are not defined for UC. Reviewed-by: Mike Marciniszyn Reviewed-by: Dennis Dalessandro Signed-off-by: Jianxin Xiong Signed-off-by: Dennis Dalessandro Signed-off-by: Doug Ledford --- drivers/infiniband/hw/hfi1/rc.c | 46 ++++++++++++++++++++++++++++-- drivers/infiniband/hw/hfi1/ruc.c | 26 +++++++++++++++++ drivers/infiniband/hw/hfi1/uc.c | 24 ++++++++++++++++ drivers/infiniband/hw/hfi1/verbs.c | 5 +++- 4 files changed, 98 insertions(+), 3 deletions(-) diff --git a/drivers/infiniband/hw/hfi1/rc.c b/drivers/infiniband/hw/hfi1/rc.c index cb474a703b33..0bc43b67d0b8 100644 --- a/drivers/infiniband/hw/hfi1/rc.c +++ b/drivers/infiniband/hw/hfi1/rc.c @@ -402,6 +402,7 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) char newreq; int middle = 0; int delta; + int err; ps->s_txreq = get_txreq(ps->dev, qp); if (IS_ERR(ps->s_txreq)) @@ -477,6 +478,35 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) qp->s_flags |= RVT_S_WAIT_FENCE; goto bail; } + /* + * Local operations are processed immediately + * after all prior requests have completed + */ + if (wqe->wr.opcode == IB_WR_REG_MR || + wqe->wr.opcode == IB_WR_LOCAL_INV) { + if (qp->s_last != qp->s_cur) + goto bail; + if (++qp->s_cur == qp->s_size) + qp->s_cur = 0; + if (++qp->s_tail == qp->s_size) + qp->s_tail = 0; + if (wqe->wr.opcode == IB_WR_REG_MR) + err = rvt_fast_reg_mr( + qp, wqe->reg_wr.mr, + wqe->reg_wr.key, + wqe->reg_wr.access); + else + err = rvt_invalidate_rkey( + qp, + wqe->wr.ex.invalidate_rkey); + hfi1_send_complete(qp, wqe, + err ? IB_WC_LOC_PROT_ERR + : IB_WC_SUCCESS); + atomic_dec(&qp->local_ops_pending); + qp->s_hdrwords = 0; + goto done_free_tx; + } + newreq = 1; qp->s_psn = wqe->psn; } @@ -491,6 +521,7 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) switch (wqe->wr.opcode) { case IB_WR_SEND: case IB_WR_SEND_WITH_IMM: + case IB_WR_SEND_WITH_INV: /* If no credit, return. */ if (!(qp->s_flags & RVT_S_UNLIMITED_CREDIT) && cmp_msn(wqe->ssn, qp->s_lsn + 1) > 0) { @@ -504,11 +535,17 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) } if (wqe->wr.opcode == IB_WR_SEND) { qp->s_state = OP(SEND_ONLY); - } else { + } else if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) { qp->s_state = OP(SEND_ONLY_WITH_IMMEDIATE); /* Immediate data comes after the BTH */ ohdr->u.imm_data = wqe->wr.ex.imm_data; hwords += 1; + } else { + qp->s_state = OP(SEND_ONLY_WITH_INVALIDATE); + /* Invalidate rkey comes after the BTH */ + ohdr->u.ieth = cpu_to_be32( + wqe->wr.ex.invalidate_rkey); + hwords += 1; } if (wqe->wr.send_flags & IB_SEND_SOLICITED) bth0 |= IB_BTH_SOLICITED; @@ -671,11 +708,16 @@ int hfi1_make_rc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) } if (wqe->wr.opcode == IB_WR_SEND) { qp->s_state = OP(SEND_LAST); - } else { + } else if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) { qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE); /* Immediate data comes after the BTH */ ohdr->u.imm_data = wqe->wr.ex.imm_data; hwords += 1; + } else { + qp->s_state = OP(SEND_LAST_WITH_INVALIDATE); + /* invalidate data comes after the BTH */ + ohdr->u.ieth = cpu_to_be32(wqe->wr.ex.invalidate_rkey); + hwords += 1; } if (wqe->wr.send_flags & IB_SEND_SOLICITED) bth0 |= IB_BTH_SOLICITED; diff --git a/drivers/infiniband/hw/hfi1/ruc.c b/drivers/infiniband/hw/hfi1/ruc.c index a659aec3c3c6..76b9c9e42d86 100644 --- a/drivers/infiniband/hw/hfi1/ruc.c +++ b/drivers/infiniband/hw/hfi1/ruc.c @@ -372,6 +372,7 @@ static void ruc_loopback(struct rvt_qp *sqp) int ret; int copy_last = 0; u32 to; + int local_ops = 0; rcu_read_lock(); @@ -440,11 +441,32 @@ again: sqp->s_sge.num_sge = wqe->wr.num_sge; sqp->s_len = wqe->length; switch (wqe->wr.opcode) { + case IB_WR_REG_MR: + if (rvt_fast_reg_mr(sqp, wqe->reg_wr.mr, wqe->reg_wr.key, + wqe->reg_wr.access)) + send_status = IB_WC_LOC_PROT_ERR; + local_ops = 1; + goto send_comp; + + case IB_WR_LOCAL_INV: + if (rvt_invalidate_rkey(sqp, wqe->wr.ex.invalidate_rkey)) + send_status = IB_WC_LOC_PROT_ERR; + local_ops = 1; + goto send_comp; + + case IB_WR_SEND_WITH_INV: + if (!rvt_invalidate_rkey(qp, wqe->wr.ex.invalidate_rkey)) { + wc.wc_flags = IB_WC_WITH_INVALIDATE; + wc.ex.invalidate_rkey = wqe->wr.ex.invalidate_rkey; + } + goto send; + case IB_WR_SEND_WITH_IMM: wc.wc_flags = IB_WC_WITH_IMM; wc.ex.imm_data = wqe->wr.ex.imm_data; /* FALLTHROUGH */ case IB_WR_SEND: +send: ret = hfi1_rvt_get_rwqe(qp, 0); if (ret < 0) goto op_err; @@ -583,6 +605,10 @@ send_comp: flush_send: sqp->s_rnr_retry = sqp->s_rnr_retry_cnt; hfi1_send_complete(sqp, wqe, send_status); + if (local_ops) { + atomic_dec(&sqp->local_ops_pending); + local_ops = 0; + } goto again; rnr_nak: diff --git a/drivers/infiniband/hw/hfi1/uc.c b/drivers/infiniband/hw/hfi1/uc.c index b7a25311bd7b..ef6c96cd3d68 100644 --- a/drivers/infiniband/hw/hfi1/uc.c +++ b/drivers/infiniband/hw/hfi1/uc.c @@ -77,6 +77,7 @@ int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) u32 len; u32 pmtu = qp->pmtu; int middle = 0; + int err; ps->s_txreq = get_txreq(ps->dev, qp); if (IS_ERR(ps->s_txreq)) @@ -118,6 +119,29 @@ int hfi1_make_uc_req(struct rvt_qp *qp, struct hfi1_pkt_state *ps) clear_ahg(qp); goto bail; } + /* + * Local operations are processed immediately + * after all prior requests have completed. + */ + if (wqe->wr.opcode == IB_WR_REG_MR || + wqe->wr.opcode == IB_WR_LOCAL_INV) { + if (qp->s_last != qp->s_cur) + goto bail; + if (++qp->s_cur == qp->s_size) + qp->s_cur = 0; + if (wqe->wr.opcode == IB_WR_REG_MR) + err = rvt_fast_reg_mr(qp, wqe->reg_wr.mr, + wqe->reg_wr.key, + wqe->reg_wr.access); + else + err = rvt_invalidate_rkey( + qp, wqe->wr.ex.invalidate_rkey); + hfi1_send_complete(qp, wqe, err ? IB_WC_LOC_PROT_ERR + : IB_WC_SUCCESS); + atomic_dec(&qp->local_ops_pending); + qp->s_hdrwords = 0; + goto done_free_tx; + } /* * Start a new request. */ diff --git a/drivers/infiniband/hw/hfi1/verbs.c b/drivers/infiniband/hw/hfi1/verbs.c index 84cc09a5eb67..57e0c0df073c 100644 --- a/drivers/infiniband/hw/hfi1/verbs.c +++ b/drivers/infiniband/hw/hfi1/verbs.c @@ -306,7 +306,10 @@ const enum ib_wc_opcode ib_hfi1_wc_opcode[] = { [IB_WR_SEND_WITH_IMM] = IB_WC_SEND, [IB_WR_RDMA_READ] = IB_WC_RDMA_READ, [IB_WR_ATOMIC_CMP_AND_SWP] = IB_WC_COMP_SWAP, - [IB_WR_ATOMIC_FETCH_AND_ADD] = IB_WC_FETCH_ADD + [IB_WR_ATOMIC_FETCH_AND_ADD] = IB_WC_FETCH_ADD, + [IB_WR_SEND_WITH_INV] = IB_WC_SEND, + [IB_WR_LOCAL_INV] = IB_WC_LOCAL_INV, + [IB_WR_REG_MR] = IB_WC_REG_MR }; /* -- 2.20.1