mcm: support incompatable verbs definitions inter-node within the platform

OFA verbs 3.5 and 1.5.4 are incompatable so there can be no direct
mappings to verbs within any MIC to Host communications. Remove
all direct verbs mappings in MIX and create inline construct fuctions
to convert verbs to new dat_mix_wr and dat_mix_wc types for both work requests
and work completions.

Signed-off-by: Arlin Davis <arlin.r.davis@intel.com>

diff --git a/dapl/openib_mcm/mix.c b/dapl/openib_mcm/mix.c
index 6fae0b25885e4965531e4532670486df23e72b2c..f80a79a023658f50d1e16466812676226921e044 100644 (file)
--- a/dapl/openib_mcm/mix.c
+++ b/dapl/openib_mcm/mix.c
@@ -34,7 +34,27 @@
 #include "dapl_ep_util.h"
 #include "dapl_osd.h"
 
-
+static inline void const_ib_wc(struct ibv_wc *iwc, struct dat_mix_wc *mwc, int entries)
+{
+       int i;
+
+       for (i=0;i<entries;i++) {
+               memset((void*)&iwc[i].wr_id, 0, sizeof(*iwc));
+               iwc[i].wr_id = mwc[i].wr_id;
+               iwc[i].status = mwc[i].status;
+               iwc[i].opcode = mwc[i].opcode;
+               iwc[i].vendor_err = mwc[i].vendor_err;
+               iwc[i].byte_len = mwc[i].byte_len;
+               iwc[i].imm_data = mwc[i].imm_data;
+               iwc[i].qp_num = mwc[i].qp_num;
+               iwc[i].src_qp = mwc[i].src_qp;
+               iwc[i].wc_flags = mwc[i].wc_flags;
+               iwc[i].pkey_index = mwc[i].pkey_index;
+               iwc[i].slid = mwc[i].slid;
+               iwc[i].sl = mwc[i].sl;
+               iwc[i].dlid_path_bits = mwc[i].dlid_path_bits;
+       }
+}
 
 /*
  * CM proxy services, MCM on MIC to MPXYD via SCIF
@@ -994,12 +1014,14 @@ int dapli_mix_dto_event_in(ib_hca_transport_t *tp, scif_epd_t scif_ep, dat_mix_d
        m_cq = (void*)pmsg->cq_ctx;
 
        for (i=0; i<pmsg->wc_cnt; i++) {
+               struct ibv_wc ib_wc;
                /* possible segmentation on mpxyd side, update length if success */
                if (pmsg->wc[i].status == 0) {
                        cookie = (DAPL_COOKIE *) (uintptr_t) pmsg->wc[i].wr_id;
                        pmsg->wc[i].byte_len = cookie->val.dto.size;
                }
-               dapls_evd_cqe_to_event(m_cq->evd, &pmsg->wc[i]);
+               const_ib_wc(&ib_wc, &pmsg->wc[i], 1);
+               dapls_evd_cqe_to_event(m_cq->evd, &ib_wc);
        }
 
        return 0;

diff --git a/dapl/svc/mpxyd.c b/dapl/svc/mpxyd.c
index 1c1b1cded0a78a0dd8ca51f8c2700b9bc3bcf1d6..3709a19e1eae46cbd7d177bced3adf0631036306 100644 (file)
--- a/dapl/svc/mpxyd.c
+++ b/dapl/svc/mpxyd.c
@@ -976,6 +976,38 @@ static inline int scif_send_msg(scif_epd_t ep, void *msg, int len)
        return 0;
 }
 
+static inline void const_mix_wc(struct dat_mix_wc *mwc, struct ibv_wc *iwc, int entries)
+{
+       int i;
+
+       for (i=0;i<entries;i++) {
+               memset((void*)&mwc[i].wr_id, 0, sizeof(*mwc));
+               mwc[i].wr_id = iwc[i].wr_id;
+               mwc[i].status = iwc[i].status;
+               mwc[i].opcode = iwc[i].opcode;
+               mwc[i].vendor_err = iwc[i].vendor_err;
+               mwc[i].byte_len = iwc[i].byte_len;
+               mwc[i].imm_data = iwc[i].imm_data;
+               mwc[i].qp_num = iwc[i].qp_num;
+               mwc[i].src_qp = iwc[i].src_qp;
+               mwc[i].wc_flags = iwc[i].wc_flags;
+               mwc[i].pkey_index = iwc[i].pkey_index;
+               mwc[i].slid = iwc[i].slid;
+               mwc[i].sl = iwc[i].sl;
+               mwc[i].dlid_path_bits = iwc[i].dlid_path_bits;
+       }
+}
+
+static inline void const_ib_wr(struct ibv_send_wr *iwr, struct dat_mix_wr *mwr, struct ibv_sge *sg_ptr)
+{
+       memset((void*)iwr, 0, sizeof(*iwr));
+       iwr->sg_list = sg_ptr;
+       iwr->opcode = mwr->opcode;
+       iwr->send_flags = mwr->send_flags;
+       iwr->imm_data = mwr->imm_data;
+       iwr->wr.rdma.remote_addr = mwr->wr.rdma.remote_addr;
+       iwr->wr.rdma.rkey = mwr->wr.rdma.rkey;
+}
 
 static int init_scif()
 {
@@ -2411,6 +2443,7 @@ static int mix_cq_poll(mcm_scif_dev_t *smd, dat_mix_dto_comp_t *pmsg)
 {
        int len, ret, pcnt;
        struct mcm_cq *m_cq;
+       struct ibv_wc local_wc[DAT_MIX_WC_MAX];
 
        /* hdr already read, get operation data */
        len = sizeof(dat_mix_dto_comp_t) - sizeof(dat_mix_hdr_t);
@@ -2428,7 +2461,7 @@ static int mix_cq_poll(mcm_scif_dev_t *smd, dat_mix_dto_comp_t *pmsg)
        else
                pcnt = pmsg->wc_cnt;
 
-       pmsg->wc_cnt = ibv_poll_cq(m_cq->ib_cq, pcnt, pmsg->wc);
+       pmsg->wc_cnt = ibv_poll_cq(m_cq->ib_cq, pcnt, local_wc);
        mlog(1," completions = %d \n", pmsg->wc_cnt);
        if (pmsg->wc_cnt < 0) {
                mlog(0, " ibv_poll_cq ERR %s on cq_id 0x%x, cq_ctx %p\n", strerror(errno), pmsg->cq_id, pmsg->cq_ctx);
@@ -2436,11 +2469,14 @@ static int mix_cq_poll(mcm_scif_dev_t *smd, dat_mix_dto_comp_t *pmsg)
        } else
                pmsg->hdr.status = 0; /* success, check pmsg->wc_cnt WC's */
 
-       if (pmsg->wc_cnt)
+       const_mix_wc(pmsg->wc, local_wc, pmsg->wc_cnt);
+
+       if (pmsg->wc_cnt) {
                mlog(1, " (ep=%d) cq %p id %d ctx %p stat %d op %d ln %d wr_id %p wc's %d verr 0x%x\n",
                        m_cq->smd->scif_op_ep, m_cq, m_cq->cq_id, m_cq->cq_ctx,
                        pmsg->wc[0].status, pmsg->wc[0].opcode, pmsg->wc[0].byte_len,
                        pmsg->wc[0].wr_id, pmsg->wc_cnt, pmsg->wc[0].vendor_err);
+       }
 
        /* send back response, with client context */
        pmsg->cq_id = m_cq->cq_id;
@@ -2661,7 +2697,7 @@ resp:
        return (scif_send_msg(smd->scif_op_ep, (void*)pmsg, len));
 }
 
-static void mix_dto_event(struct mcm_cq *m_cq, struct ibv_wc *wc, int nc)
+static void mix_dto_event(struct mcm_cq *m_cq, struct dat_mix_wc *wc, int nc)
 {
        dat_mix_dto_comp_t msg;
        int i;
@@ -3219,7 +3255,7 @@ static void m_post_pending_wr(mcm_scif_dev_t *smd, int *data, int *events)
                                m_wr->wr.wr_id = (uint64_t)m_wr;
                                ret = ibv_post_send(m_qp->ib_qp, &m_wr->wr, &bad_wr);
                                if (ret || (cn_signal && mix_eager_completion)) {
-                                       struct ibv_wc wc;
+                                       struct dat_mix_wc wc;
                                        mlog(1, " dto_event: sig %d ret %d, %s - m_wr %p, wr_id %p\n",
                                                cn_signal, ret, strerror(errno),
                                                m_wr->wr.wr_id, m_wr->org_id);
@@ -3314,16 +3350,11 @@ static int m_proxy_data(mcm_scif_dev_t *smd, dat_mix_send_t *pmsg, struct mcm_qp
 
        mlog(1, " m_wr %p m_sge %p \n", m_wr, m_sge);
 
-       memcpy(&m_wr->wr.num_sge, &pmsg->wr.num_sge, 40);
        m_wr->org_id = pmsg->wr.wr_id;
-       m_wr->wr.sg_list = m_sge;
-       m_wr->wr.next = 0;
-       m_wr->wr.wr.rdma.remote_addr += total_offset;
-       m_wr->wr.num_sge = 0;
-       m_wr->wr.wr_id = 0;
        m_wr->w_idx = m_qp->wr_hd;
        m_wr->flags = 0;
        m_wr->context = (uint64_t)m_qp;
+       const_ib_wr(&m_wr->wr, &pmsg->wr, m_sge);
 
        for (i=0;i<pmsg->wr.num_sge;i++) {
 
@@ -3471,16 +3502,12 @@ retry_mr:
 
                                mlog(1, " next m_wr %p m_sge %p \n", m_wr, m_wr->sg);
 
-                               memcpy(&m_wr->wr.num_sge, &pmsg->wr.num_sge, 40);
                                m_wr->org_id = pmsg->wr.wr_id;
-                               m_wr->wr.sg_list = m_wr->sg;
-                               m_wr->wr.next = 0;
-                               m_wr->wr.wr.rdma.remote_addr += total_offset;
-                               m_wr->wr.num_sge = 0;
-                               m_wr->wr.wr_id = 0;
                                m_wr->w_idx = m_qp->wr_hd;
                                m_wr->flags = 0;
                                m_wr->context = (uint64_t)m_qp;
+                               const_ib_wr(&m_wr->wr, &pmsg->wr, m_sge);
+                               m_wr->wr.wr.rdma.remote_addr += total_offset;
                        }
                }
        }
@@ -3517,7 +3544,7 @@ retry_mr:
        ret = 0;
 bail:
        if (ret) {
-               struct ibv_wc wc;
+               struct dat_mix_wc wc;
 
                wc.wr_id = pmsg->wr.wr_id;
                wc.byte_len = 0;
@@ -3550,8 +3577,8 @@ static int mix_post_send(mcm_scif_dev_t *smd, dat_mix_send_t *pmsg)
        }
        m_qp = (struct mcm_qp*)pmsg->qp_ctx;
 
-       mlog(1, " q_id %d, q_num %x ln %d, wr_id %p, sge %d, op %x flgs %x pst %d,%d cmp %d, inl %d, %s\n",
-               pmsg->qp_id, m_qp->ib_qp->qp_num, pmsg->len, pmsg->wr.wr_id,
+       mlog(1, " q_id %d, q_num %x data %d pkt %d wr_id %p, sge %d, op %x flgs %x pst %d,%d cmp %d, inl %d, %s\n",
+               pmsg->qp_id, m_qp->ib_qp->qp_num, pmsg->len, sizeof(dat_mix_send_t) + pmsg->len, pmsg->wr.wr_id,
                pmsg->wr.num_sge, pmsg->wr.opcode, pmsg->wr.send_flags, m_qp->post_cnt,
                m_qp->post_sig_cnt, m_qp->comp_cnt, pmsg->hdr.flags & MIX_OP_INLINE ? 1:0,
                pmsg->wr.opcode == IBV_WR_SEND ? "SND":"WR");
@@ -3577,9 +3604,8 @@ static int mix_post_send(mcm_scif_dev_t *smd, dat_mix_send_t *pmsg)
                m_wr, m_wr_prev, m_wr->sg, len, m_qp->wr_hd, m_qp->wr_tl);
 
        /* IB_WR */
-       memcpy(&m_wr->wr.num_sge, &pmsg->wr.num_sge, 40);
+       const_ib_wr(&m_wr->wr, &pmsg->wr, m_wr->sg);
        m_wr->wr.sg_list = m_wr->sg;
-       m_wr->wr.next = 0;
        m_wr->wr.num_sge = len ? 1:0;
 
        /* M_WR */
@@ -3694,7 +3720,7 @@ bail:
        if (ret ||
            ((m_wr->flags & M_SEND_POSTED) &&
             (m_wr->flags & M_SEND_CN_EAGER_SIG))) {
-               struct ibv_wc wc;
+               struct dat_mix_wc wc;
 
                wc.wr_id = pmsg->wr.wr_id;
                wc.byte_len = len;
@@ -4649,7 +4675,7 @@ void mcm_cq_event(struct mcm_cq *m_cq)
        void *cq_ctx;
        int i, ret, num, notify = 0;
        struct ibv_wc wc[DAT_MIX_WC_MAX];
-       struct ibv_wc wc_ev[DAT_MIX_WC_MAX];
+       struct dat_mix_wc wc_ev[DAT_MIX_WC_MAX];
 
        mlog(2," m_cq(%p) \n", m_cq);
 
@@ -4707,6 +4733,7 @@ retry:
                                m_cq->prev_id = m_wr->org_id;
                                wc_ev[num].wr_id = m_wr->org_id;
                                wc_ev[num].status = wc[i].status;
+                               wc_ev[num].vendor_err = wc[i].vendor_err;
                                num++;
                        }
                }

diff --git a/dat/include/dat2/dat_mic_extensions.h b/dat/include/dat2/dat_mic_extensions.h
index fa6e2d74f9cd6b13b200637e9e4bc49158cde973..59c576673d3478e5c1d68692a9aae868f584c467 100755 (executable)
--- a/dat/include/dat2/dat_mic_extensions.h
+++ b/dat/include/dat2/dat_mic_extensions.h
@@ -447,6 +447,56 @@ typedef struct dat_mix_cm_event
 
 } dat_mix_cm_event_t;
 
+typedef struct dat_mix_wc
+{
+       uint64_t        wr_id;
+       uint32_t        status;
+       uint32_t        opcode;
+       uint32_t        vendor_err;
+       uint32_t        byte_len;
+       uint32_t        imm_data;       /* in network byte order */
+       uint32_t        qp_num;
+       uint32_t        src_qp;
+       int             wc_flags;
+       uint16_t        pkey_index;
+       uint16_t        slid;
+       uint8_t         sl;
+       uint8_t         dlid_path_bits;
+} dat_mix_wc_t;
+
+typedef struct dat_mix_sge {
+       uint64_t                addr;
+       uint32_t                length;
+       uint32_t                lkey;
+} dat_mix_sge_t;
+
+typedef struct dat_mix_wr {
+       uint64_t                wr_id;
+       struct dat_mix_send_wr  *next;
+       struct dat_mix_sge      *sg_list;
+       int                     num_sge;
+       uint32_t                opcode;
+       int                     send_flags;
+       uint32_t                imm_data;       /* in network byte order */
+       union {
+               struct {
+                       uint64_t        remote_addr;
+                       uint32_t        rkey;
+               } rdma;
+               struct {
+                       uint64_t        remote_addr;
+                       uint64_t        compare_add;
+                       uint64_t        swap;
+                       uint32_t        rkey;
+               } atomic;
+               struct {
+                       struct ibv_ah  *ah;
+                       uint32_t        remote_qpn;
+                       uint32_t        remote_qkey;
+               } ud;
+       } wr;
+} dat_mix_wr_t;
+
 #define DAT_MIX_WC_MAX 4
 typedef struct dat_mix_dto_comp
 {
@@ -454,7 +504,7 @@ typedef struct dat_mix_dto_comp
        uint64_t                cq_ctx;
        uint32_t                cq_id;
        uint32_t                wc_cnt;
-       struct ibv_wc           wc[DAT_MIX_WC_MAX];
+       struct dat_mix_wc       wc[DAT_MIX_WC_MAX];
 
 } dat_mix_dto_comp_t;
 
@@ -465,8 +515,8 @@ typedef struct dat_mix_send
        uint64_t                qp_ctx;
        uint32_t                qp_id;
        uint32_t                len;
-       struct ibv_send_wr      wr;
-       struct ibv_sge          sge[DAT_MIX_SGE_MAX];
+       struct dat_mix_wr       wr;
+       struct dat_mix_sge      sge[DAT_MIX_SGE_MAX];
 
 } dat_mix_send_t;