diff --git a/ompi/mca/btl/usnic/btl_usnic_ack.c b/ompi/mca/btl/usnic/btl_usnic_ack.c
index 43b81c38cb..307c76ca32 100644
--- a/ompi/mca/btl/usnic/btl_usnic_ack.c
+++ b/ompi/mca/btl/usnic/btl_usnic_ack.c
@@ -134,20 +134,22 @@ ompi_btl_usnic_handle_ack(
                 (void*)sseg, (void*)frag, bytes_acked, frag->sf_ack_bytes_left);
 #endif
 
-        /* perform completion callback for PUT here */
+        /* If all ACKs received, and this is a put or a regular send
+         * that needs a callback, perform the callback now
+         */
         if (frag->sf_ack_bytes_left == 0 &&
-            frag->sf_base.uf_dst_seg[0].seg_addr.pval != NULL) {
-#if MSGDEBUG1
-            opal_output(0, "Calling back %p for PUT completion, frag=%p\n", 
-                    (void*)(uintptr_t)frag->sf_base.uf_base.des_cbfunc, (void*)frag);
+            ((frag->sf_base.uf_dst_seg[0].seg_addr.pval != NULL) ||
+             (frag->sf_base.uf_base.des_flags &
+              MCA_BTL_DES_SEND_ALWAYS_CALLBACK))) {
+#if MSGDEBUG2
+            opal_output(0, "completion callback for frag=%p, dest=%p\n",
+                    (void*)frag, frag->sf_base.uf_dst_seg[0].seg_addr.pval);
 #endif
             frag->sf_base.uf_base.des_cbfunc(&module->super, frag->sf_endpoint,
                     &frag->sf_base.uf_base, OMPI_SUCCESS);
+            frag->sf_base.uf_base.des_flags &= ~MCA_BTL_DES_SEND_ALWAYS_CALLBACK;
         }
 
-        /* OK to return this fragment? */
-        ompi_btl_usnic_send_frag_return_cond(module, frag);
-
         /* free this segment */
         sseg->ss_ack_pending = false;
         if (sseg->ss_base.us_type == OMPI_BTL_USNIC_SEG_CHUNK &&
@@ -155,6 +157,9 @@ ompi_btl_usnic_handle_ack(
             ompi_btl_usnic_chunk_segment_return(module, sseg);
         }
 
+        /* OK to return this fragment? */
+        ompi_btl_usnic_send_frag_return_cond(module, frag);
+
         /* indicate this segment has been ACKed */
         endpoint->endpoint_sent_segs[WINDOW_SIZE_MOD(is)] = NULL;
     }
diff --git a/ompi/mca/btl/usnic/btl_usnic_frag.h b/ompi/mca/btl/usnic/btl_usnic_frag.h
index 91d08e40b0..b8f39f15b4 100644
--- a/ompi/mca/btl/usnic/btl_usnic_frag.h
+++ b/ompi/mca/btl/usnic/btl_usnic_frag.h
@@ -57,6 +57,19 @@ typedef enum {
     OMPI_BTL_USNIC_FRAG_PUT_DEST
 } ompi_btl_usnic_frag_type_t;
 
+#if MSGDEBUG2
+static inline char *
+usnic_frag_type(ompi_btl_usnic_frag_type_t t)
+{
+    switch (t) {
+    case OMPI_BTL_USNIC_FRAG_LARGE_SEND: return "large";
+    case OMPI_BTL_USNIC_FRAG_SMALL_SEND: return "small";
+    case OMPI_BTL_USNIC_FRAG_PUT_DEST: return "put dest";
+    default: return "unknown";
+    }
+}
+#endif
+
 typedef enum {
     OMPI_BTL_USNIC_SEG_ACK,
     OMPI_BTL_USNIC_SEG_FRAG,
@@ -64,6 +77,20 @@ typedef enum {
     OMPI_BTL_USNIC_SEG_RECV
 } ompi_btl_usnic_seg_type_t;
 
+#if MSGDEBUG2
+static inline char *
+usnic_seg_type(ompi_btl_usnic_seg_type_t t)
+{
+    switch (t) {
+    case OMPI_BTL_USNIC_SEG_ACK: return "ACK";
+    case OMPI_BTL_USNIC_SEG_FRAG: return "FRAG";
+    case OMPI_BTL_USNIC_SEG_CHUNK: return "CHUNK";
+    case OMPI_BTL_USNIC_SEG_RECV: return "RECV";
+    default: return "unknown";
+    }
+}
+#endif
+
 
 typedef struct ompi_btl_usnic_reg_t {
     mca_mpool_base_registration_t base;
@@ -94,6 +121,7 @@ typedef enum {
  * holes.
  */
 typedef struct {
+
     /* Hashed RTE process name of the sender */
     uint64_t sender;
 
@@ -112,8 +140,9 @@ typedef struct {
 
     /* Type of BTL header (see enum, above) */
     uint8_t payload_type;
-    /* Yuck */
-    uint8_t padding;
+
+    /* tag for PML, etc */
+    mca_btl_base_tag_t tag;
 } ompi_btl_usnic_btl_header_t; 
 
 /**
@@ -146,15 +175,6 @@ typedef enum {
     FRAG_MAX = 0xff
 } ompi_btl_usnic_frag_state_flags_t;
 
-
-/* 
- * Convenience macros for states
- */
-#define FRAG_STATE_SET(frag, state) (frag)->state_flags |= (state)
-#define FRAG_STATE_CLR(frag, state) (frag)->state_flags &= ~(state)
-#define FRAG_STATE_GET(frag, state) ((frag)->state_flags & (state))
-#define FRAG_STATE_ISSET(frag, state) (((frag)->state_flags & (state)) != 0)
-
 /**
  * Descriptor for a common segment.  This is exactly one packet and may
  * be send or receive
@@ -177,7 +197,7 @@ typedef struct ompi_btl_usnic_segment_t {
 
     union {
         uint8_t *raw;
-        mca_btl_base_header_t *pml_header;
+        void *pml_header;
     } us_payload;
 } ompi_btl_usnic_segment_t;
 
@@ -270,6 +290,7 @@ typedef struct ompi_btl_usnic_large_send_frag_t {
     ompi_btl_usnic_send_frag_t lsf_base;
 
     char lsf_pml_header[64];    /* space for PML header */
+    mca_btl_base_tag_t lsf_tag; /* save tag */
 
     uint32_t lsf_frag_id;       /* fragment ID for reassembly */
     size_t lsf_cur_offset;      /* current offset into message */
@@ -424,6 +445,10 @@ ompi_btl_usnic_frag_return(
     struct ompi_btl_usnic_module_t *module,
     ompi_btl_usnic_frag_t *frag)
 {
+#if MSGDEBUG2
+    opal_output(0, "freeing frag %p, type %s\n", (void *)frag,
+            usnic_frag_type(frag->uf_type));
+#endif
     OMPI_FREE_LIST_RETURN_MT(frag->uf_freelist, &(frag->uf_base.super));
 }
 
diff --git a/ompi/mca/btl/usnic/btl_usnic_module.c b/ompi/mca/btl/usnic/btl_usnic_module.c
index 9f9ca75a34..ffe9e8a19a 100644
--- a/ompi/mca/btl/usnic/btl_usnic_module.c
+++ b/ompi/mca/btl/usnic/btl_usnic_module.c
@@ -219,35 +219,23 @@ usnic_alloc(struct mca_btl_base_module_t* btl,
 {
     ompi_btl_usnic_send_frag_t *frag;
     ompi_btl_usnic_module_t *module = (ompi_btl_usnic_module_t*) btl;
+    ompi_btl_usnic_small_send_frag_t *sfrag;
     mca_btl_base_descriptor_t *desc;
 
-    /* will this fit into a small send? */
-    if (size <= module->max_frag_payload) {
-        ompi_btl_usnic_small_send_frag_t *sfrag;
+    if (size > module->max_frag_payload)
+        size = module->max_frag_payload;
 
-        sfrag = ompi_btl_usnic_small_send_frag_alloc(module);
-        if (NULL == sfrag) {
-            return NULL;
-        }
-        frag = &sfrag->ssf_base;
 
-    } else {
-        ompi_btl_usnic_large_send_frag_t *lfrag;
-
-        lfrag = ompi_btl_usnic_large_send_frag_alloc(module);
-        if (NULL == lfrag) {
-            return NULL;
-        }
-        frag = &lfrag->lsf_base;
-
-        BTL_ERROR(("large frag in usnic_alloc()\n"));
-        abort();    /* XXX - we don't ever want to see this... */
+    sfrag = ompi_btl_usnic_small_send_frag_alloc(module);
+    if (NULL == sfrag) {
+        return NULL;
     }
+    frag = &sfrag->ssf_base;
 
 #if MSGDEBUG2
-        opal_output(0, "usnic_alloc: %s frag=%p, size=%d\n",
+        opal_output(0, "usnic_alloc: %s frag=%p, size=%d, flags=0x%x\n",
                 (size <= module->max_frag_payload)?"small":"large",
-                (void *)frag, (int)size);
+                (void *)frag, (int)size, flags);
 #endif
 
     /* set # of bytes remaining to be ACKed */
@@ -262,28 +250,34 @@ usnic_alloc(struct mca_btl_base_module_t* btl,
 
     /* set up descriptor */
     desc = &frag->sf_base.uf_base;
+    desc->des_flags = flags;
     desc->des_src[0].seg_len = size;
     desc->des_src_cnt = 1;
-    desc->des_flags = flags;
 
     return desc;
 }
 
 
 /**
- * Return a small send fragment
+ * Return an allocated fragment
  *
  * Return the send fragment to the appropriate list
  */
 static int usnic_free(struct mca_btl_base_module_t* btl,
                         mca_btl_base_descriptor_t* des)
 {
-    ompi_btl_usnic_frag_t* frag = (ompi_btl_usnic_frag_t*)des;
+    ompi_btl_usnic_send_frag_t* frag = (ompi_btl_usnic_send_frag_t*)des;
 
-#if MSGDEBUG1
+#if MSGDEBUG2
     opal_output(0, "usnic_free: %p\n", (void*)frag);
 #endif
+
+#if 1 /* seperate commit for seperate bug */
     OMPI_FREE_LIST_RETURN_MT(frag->uf_freelist, &(frag->uf_base.super));
+#else
+    ompi_btl_usnic_send_frag_return_cond((struct ompi_btl_usnic_module_t *)btl,
+            frag);
+#endif
 
     return OMPI_SUCCESS;
 }
@@ -338,6 +332,7 @@ usnic_prepare_src(
     uint32_t flags)
 {
     ompi_btl_usnic_module_t *module = (ompi_btl_usnic_module_t*) base_module;
+    mca_btl_base_descriptor_t *desc;
     ompi_btl_usnic_send_frag_t *frag;
     uint32_t payload_len;
     struct iovec iov;
@@ -498,18 +493,22 @@ usnic_prepare_src(
 
     /* fragment accounting */
     frag->sf_ack_bytes_left = payload_len;
+    desc = &frag->sf_base.uf_base;
 
 #if MSGDEBUG2
-        opal_output(0, "prep_src: %s %s frag %p, size=%d+%d, src=%p\n",
+        opal_output(0, "prep_src: %s %s frag %p, size=%d+%d\n",
             module->device->name,
             payload_len <= module->max_frag_payload?"small":"large",
-            (void *)frag, (int)reserve, (int)*size,
-            frag->sf_base.uf_base.des_src[0].seg_addr.pval);
-        opal_output(0, "    data_ptr = %p, conv=%p\n",
-                data_ptr, (void *)frag->sf_convertor);
+            (void *)frag, (int)reserve, (int)*size);
+        { int i;
+            for (i=0; i<desc->des_src_cnt; ++i)
+                opal_output(0, "  %d: ptr:%p len:%d\n", i,
+                        desc->des_src[i].seg_addr.pval,
+                        desc->des_src[i].seg_len);
+        }
 #endif
 
-    return &frag->sf_base.uf_base;
+    return desc;
 }
 
 static mca_btl_base_descriptor_t*
@@ -571,12 +570,20 @@ usnic_put(
     frag = (ompi_btl_usnic_send_frag_t *)des;
 
 #if MSGDEBUG2
-    opal_output(0, "usnic_put: %"PRIu64" bytes to %p\n",
-            des->des_dst->seg_len,
-            des->des_dst->seg_addr.pval);
-    opal_output(0, "  des_dst=%p, frag->uf_dst_seg=%p\n",
-            (void *)des->des_dst,
-            (void *)frag->sf_base.uf_dst_seg);
+    opal_output(0, "usnic_put, frag=%p, source=\n", frag);
+    { int i;
+        for (i=0; i<des->des_src_cnt; ++i)
+            opal_output(0, "  %d: ptr:%p len:%d\n", i,
+                    des->des_src[i].seg_addr.pval,
+                    des->des_src[i].seg_len);
+    }
+    opal_output(0, "dest:\n");
+    { int i;
+        for (i=0; i<des->des_dst_cnt; ++i)
+            opal_output(0, "  %d: ptr:%p len:%d\n", i,
+                    des->des_dst[i].seg_addr.pval,
+                    des->des_dst[i].seg_len);
+    }
 #endif
 
     /* copy out address - why does he not use ours?  silly PML */
@@ -1014,6 +1021,7 @@ usnic_handle_large_send(
     chp->ch_frag_id = lfrag->lsf_frag_id;
     chp->ch_frag_size = lfrag->lsf_base.sf_size;
     chp->ch_frag_offset = lfrag->lsf_cur_offset;
+    chp->ch_hdr.tag = lfrag->lsf_tag;
 
     /* set actual packet length for verbs */
     sseg->ss_base.us_sg_entry[0].length =
@@ -1036,24 +1044,29 @@ usnic_handle_large_send(
     /* done with fragment? */
     if (lfrag->lsf_bytes_left == 0) {
 
-        /* only callback now if this was not a PUT, otherwise
-         * we need to wait until last byte is ACKed
+        /* remove this frag from sending list now because PML may
+         * decide to put it on some other list in the callback
          */
-        if (frag->sf_base.uf_dst_seg[0].seg_addr.pval == NULL) {
+        opal_list_remove_item(&endpoint->endpoint_frag_send_queue,
+                &frag->sf_base.uf_base.super.super);
 
-#if MSGDEBUG1
-            opal_output(0, "    calling back %p, len=%zd\n",
-                    (void*)(uintptr_t)frag->sf_base.uf_base.des_cbfunc,
-                    frag->sf_size);
+        /* only callback now if this was not a PUT and we own the fragment,
+         * otherwise we need to wait until last byte is ACKed
+         */
+        if (frag->sf_base.uf_dst_seg[0].seg_addr.pval == NULL &&
+                (frag->sf_base.uf_base.des_flags &
+                 MCA_BTL_DES_FLAGS_BTL_OWNERSHIP)) {
+
+#if MSGDEBUG2
+            opal_output(0, "callback for large frag %p, len=%zd\n",
+                    (void *)frag->sf_base.uf_base.des_cbfunc, frag->sf_size);
 #endif
             frag->sf_base.uf_base.des_cbfunc(&module->super,
                           frag->sf_endpoint, &frag->sf_base.uf_base,
                           OMPI_SUCCESS);
             ++module->pml_send_callbacks;
+            frag->sf_base.uf_base.des_flags &= ~MCA_BTL_DES_SEND_ALWAYS_CALLBACK;
         }
-
-        opal_list_remove_item(&endpoint->endpoint_frag_send_queue,
-                &frag->sf_base.uf_base.super.super);
     }
 }
 
@@ -1113,6 +1126,13 @@ ompi_btl_usnic_module_progress_sends(
          * Send ptr and length will be in uf_src_seg[0]
          */
         if (frag->sf_base.uf_type == OMPI_BTL_USNIC_FRAG_SMALL_SEND) {
+
+            /* remove this frag from sending list now because PML may
+             * decide to put it on some other list in the callback
+             */
+            opal_list_remove_item(&endpoint->endpoint_frag_send_queue,
+                    &frag->sf_base.uf_base.super.super);
+
             sfrag = (ompi_btl_usnic_small_send_frag_t *)frag;
             sseg = &sfrag->ssf_segment;
 
@@ -1121,9 +1141,12 @@ ompi_btl_usnic_module_progress_sends(
             sseg->ss_base.us_btl_header->payload_len = payload_len;
 
 #if MSGDEBUG1
-            opal_output(0, "send small, ptr=%"PRIu64", payload=%zd, len=%"PRIu32"\n",
-                    sseg->ss_base.us_sg_entry[0].addr, payload_len,
-                    sseg->ss_base.us_sg_entry[0].length);
+            opal_output(0, "progress send small, frag=%p, ptr=%p, payload=%zd, len=%"PRIu32", ep=%p, tag=%d\n",
+                    (void *)frag,
+                    (void *)sseg->ss_base.us_sg_entry[0].addr, payload_len,
+                    sseg->ss_base.us_sg_entry[0].length,
+                    (void *)frag->sf_endpoint,
+                    sseg->ss_base.us_btl_header->tag);
 #endif
 
             /* post the send */
@@ -1131,23 +1154,30 @@ ompi_btl_usnic_module_progress_sends(
 
             /* don't do callback yet if this is a put */
             if (frag->sf_base.uf_dst_seg[0].seg_addr.pval == NULL) {
-#if MSGDEBUG1
-                opal_output(0, "    calling back %p, len=%"PRIu64"\n",
-                        (void*)(uintptr_t)frag->sf_base.uf_base.des_cbfunc,
-                        frag->sf_base.uf_src_seg[0].seg_len);
+                /* we have copied the data, perform a callback if
+                 * we own the fragment and callback is requested.
+                 * If we don't own the fragment, we cannot callback yet
+                 * because we are not done with the segment inside.
+                 * (ACK not received yet)
+                 */
+                if ((frag->sf_base.uf_base.des_flags &
+                        (MCA_BTL_DES_SEND_ALWAYS_CALLBACK |
+                         MCA_BTL_DES_FLAGS_BTL_OWNERSHIP)) ==
+                        (MCA_BTL_DES_SEND_ALWAYS_CALLBACK |
+                         MCA_BTL_DES_FLAGS_BTL_OWNERSHIP)) {
+#if MSGDEBUG2
+                opal_output(0, "callback frag small %p, len=%"PRIu64"\n",
+                        (void*)frag, frag->sf_base.uf_src_seg[0].seg_len);
 #endif
-                /* we have copied the data, proceed with callback */
-                /* could be done in usnic_send? XXX */
-                frag->sf_base.uf_base.des_cbfunc(&module->super,
-                              frag->sf_endpoint, &frag->sf_base.uf_base,
-                              OMPI_SUCCESS);
-                ++module->pml_send_callbacks;
+                    frag->sf_base.uf_base.des_cbfunc(&module->super,
+                                  frag->sf_endpoint, &frag->sf_base.uf_base,
+                                  OMPI_SUCCESS);
+                    ++module->pml_send_callbacks;
+                    frag->sf_base.uf_base.des_flags &= 
+                        ~MCA_BTL_DES_SEND_ALWAYS_CALLBACK;
+                }
             }
 
-            /* remove frag from sending list */
-            opal_list_remove_item(&endpoint->endpoint_frag_send_queue,
-                    &sfrag->ssf_base.sf_base.uf_base.super.super);
-
         /* Large sends... */
         } else {
             usnic_handle_large_send(module, endpoint, frag);
@@ -1189,6 +1219,21 @@ ompi_btl_usnic_module_progress_sends(
 
 /*
  *  Initiate a send.
+ *
+ *  Send completion callbacks can be done from a few different places.
+ *
+ *  If this is a send from a fragment we do not own, we always have
+ *  to wait for the last ACK of the fragment, because we cannot allow
+ *  the fragment to be re-used until we know we have no more retransmits to do.
+ *
+ *  If this is a send from a fragment we own, and we know we have copied the
+ *  data from the user's buffer, we can perform the callback immediately
+ *  (or possibly not at all, simply returning "1" to indicate completion.
+ *
+ *  If this is a send from a fragment we own and we have not yet copied out
+ *  all the data (as is the case in a large send) then we defer the callback
+ *  until the last of the data has been copied out by routines called
+ *  from ompi_btl_usnic_progress_sends()
  */
 static int usnic_send(struct mca_btl_base_module_t* base_module,
                         struct mca_btl_base_endpoint_t* base_endpoint,
@@ -1210,7 +1255,12 @@ static int usnic_send(struct mca_btl_base_module_t* base_module,
     opal_output(0, "usnic_send: frag=%p, endpoint=%p, tag=%d, sendreq=%p\n",
             (void *)frag, (void *)endpoint,
             tag, (void *)descriptor->des_cbdata);
-    opal_output(0, "   data = %p\n", descriptor->des_src[0].seg_addr.pval);
+    { int i;
+        for (i=0; i<descriptor->des_src_cnt; ++i)
+            opal_output(0, "  %d: ptr:%p len:%d\n", i,
+                    descriptor->des_src[i].seg_addr.pval,
+                    descriptor->des_src[i].seg_len);
+    }
 #endif
 
     assert(frag->sf_endpoint == endpoint);
@@ -1257,6 +1307,7 @@ static int usnic_send(struct mca_btl_base_module_t* base_module,
 
         sseg->ss_send_desc.send_flags |= IBV_SEND_INLINE;
         sseg->ss_channel = USNIC_PRIORITY_CHANNEL;
+        sseg->ss_base.us_btl_header->tag = tag;
 #if MSGDEBUG2
         opal_output(0, "conv = %p\n", frag->sf_convertor);
         opal_output(0, "  inline frag %d segs %p(%d) + %p(%d)\n",
@@ -1292,14 +1343,30 @@ static int usnic_send(struct mca_btl_base_module_t* base_module,
                 sizeof(ompi_btl_usnic_btl_header_t) + frag->sf_size;
         }
 
-        /* If requested, callback now, else just return 1 to show completion */
-        if (descriptor->des_flags & MCA_BTL_DES_SEND_ALWAYS_CALLBACK) {
-            frag->sf_base.uf_base.des_cbfunc(&module->super,
-                    frag->sf_endpoint, &frag->sf_base.uf_base,
-                    OMPI_SUCCESS);
-            rc = 0;
+        /* If we own the frag and callback was requested, callback now,
+         * else just return 1 to show completion.
+         * If we don't own the frag, need to wait for ACK before
+         * performing callback on the frag
+         */
+        if (descriptor->des_flags & MCA_BTL_DES_FLAGS_BTL_OWNERSHIP) {
+            if (descriptor->des_flags & MCA_BTL_DES_SEND_ALWAYS_CALLBACK) {
+#if MSGDEBUG2
+                opal_output(0, "immediate callback for frag %p\n", (void *)frag);
+#endif
+                frag->sf_base.uf_base.des_cbfunc(&module->super,
+                        frag->sf_endpoint, &frag->sf_base.uf_base,
+                        OMPI_SUCCESS);
+                rc = 0;
+                descriptor->des_flags &= ~MCA_BTL_DES_SEND_ALWAYS_CALLBACK;
+            } else {
+#if MSGDEBUG2
+                opal_output(0, "skipping callback for frag %p\n", (void *)frag);
+#endif
+                rc = 1;
+            }
         } else {
-            rc = 1;
+            descriptor->des_flags |= MCA_BTL_DES_SEND_ALWAYS_CALLBACK;
+            rc = 0;
         }
         ++module->pml_module_sends;
         ++module->pml_send_callbacks;   /* returning "1" is an implicit CB */
diff --git a/ompi/mca/btl/usnic/btl_usnic_recv.c b/ompi/mca/btl/usnic/btl_usnic_recv.c
index 40a7f9b4cf..1e2d477507 100644
--- a/ompi/mca/btl/usnic/btl_usnic_recv.c
+++ b/ompi/mca/btl/usnic/btl_usnic_recv.c
@@ -56,6 +56,7 @@ void ompi_btl_usnic_recv_call(ompi_btl_usnic_module_t *module,
     mca_btl_active_message_callback_t* reg;
     ompi_btl_usnic_endpoint_t *endpoint;
     ompi_btl_usnic_btl_chunk_header_t *chunk_hdr;
+    ompi_btl_usnic_btl_header_t *hdr;
     uint32_t window_index;
     int rc;
 #if MSGDEBUG1
@@ -110,11 +111,11 @@ void ompi_btl_usnic_recv_call(ompi_btl_usnic_module_t *module,
             return;
         }
 
+        hdr = seg->rs_base.us_btl_header;
+
 #if MSGDEBUG1
         opal_output(0, "<-- Received FRAG ep %p, seq %" UDSEQ ", len=%d\n",
-                    (void*) endpoint,
-                    seg->rs_base.us_btl_header->seq,
-                    seg->rs_base.us_btl_header->payload_len);
+                    (void*) endpoint, hdr->seq, hdr->payload_len);
 #if 0
 
         opal_output(0, "<-- Received FRAG ep %p, seq %" UDSEQ " from %s to %s: GOOD! (rel seq %d, lowest seq %" UDSEQ ", highest seq: %" UDSEQ ", rwstart %d) seg %p, module %p\n",
@@ -126,7 +127,7 @@ void ompi_btl_usnic_recv_call(ompi_btl_usnic_module_t *module,
                     endpoint->endpoint_highest_seq_rcvd,
                     endpoint->endpoint_rfstart,
                     (void*) seg, (void*) module);
-        if (seg->rs_base.us_btl_header->put_addr != NULL) {
+        if (hdr->put_addr != NULL) {
             opal_output(0, "  put_addr = %p\n",
                     seg->rs_base.us_btl_header->put_addr);
         }
@@ -139,12 +140,10 @@ void ompi_btl_usnic_recv_call(ompi_btl_usnic_module_t *module,
          * the frame length to meet minimum sizes, add protocol information,
          * etc.
          */
-        if (seg->rs_base.us_btl_header->put_addr == NULL) {
-            reg = mca_btl_base_active_message_trigger +
-                bseg->us_payload.pml_header->tag;
-            seg->rs_segment.seg_len = bseg->us_btl_header->payload_len;
-            reg->cbfunc(&module->super, bseg->us_payload.pml_header->tag, 
-                        &seg->rs_desc, reg->cbdata);
+        if (hdr->put_addr == NULL) {
+            reg = mca_btl_base_active_message_trigger + hdr->tag;
+            seg->rs_segment.seg_len = hdr->payload_len;
+            reg->cbfunc(&module->super, hdr->tag, &seg->rs_desc, reg->cbdata);
 
         /*
          * If this is a PUT, need to copy it to user buffer
@@ -269,15 +268,10 @@ opal_output(0, "Start PUT to %p\n", chunk_hdr->ch_hdr.put_addr);
 
         fip->rfi_bytes_left -= chunk_hdr->ch_hdr.payload_len;
         if (0 == fip->rfi_bytes_left) {
-            mca_btl_base_header_t *pml_header;
             mca_btl_base_descriptor_t desc;
             mca_btl_base_segment_t segment;
 
-            /* Get access to PML header in assembled fragment so we
-             * can pull out the tag
-             */
-            pml_header = (mca_btl_base_header_t *)(fip->rfi_data);
-            segment.seg_addr.pval = pml_header;
+            segment.seg_addr.pval = fip->rfi_data;
             segment.seg_len = fip->rfi_frag_size;
             desc.des_dst = &segment;
             desc.des_dst_cnt = 1;
@@ -289,12 +283,13 @@ opal_output(0, "Start PUT to %p\n", chunk_hdr->ch_hdr.put_addr);
 #if MSGDEBUG2
                 opal_output(0, "  large FRAG complete, pass up %p, %"PRIu64" bytes, tag=%d\n",
                         desc.des_dst->seg_addr.pval, desc.des_dst->seg_len,
-                        pml_header->tag);
+                        chunk_hdr->ch_hdr.tag);
 #endif
-                reg = mca_btl_base_active_message_trigger + pml_header->tag;
+                reg = mca_btl_base_active_message_trigger +
+                    chunk_hdr->ch_hdr.tag;
 
                 /* mca_pml_ob1_recv_frag_callback_frag() */
-                reg->cbfunc(&module->super, pml_header->tag,
+                reg->cbfunc(&module->super, chunk_hdr->ch_hdr.tag,
                         &desc, reg->cbdata);
 
                 /* free temp buffer for non-put */
diff --git a/ompi/mca/btl/usnic/btl_usnic_recv.h b/ompi/mca/btl/usnic/btl_usnic_recv.h
index 950fd19de6..c115955d17 100644
--- a/ompi/mca/btl/usnic/btl_usnic_recv.h
+++ b/ompi/mca/btl/usnic/btl_usnic_recv.h
@@ -283,10 +283,9 @@ ompi_btl_usnic_recv_fast(ompi_btl_usnic_module_t *module,
          * the frame length to meet minimum sizes, add protocol information,
          * etc.
          */
-        reg = mca_btl_base_active_message_trigger +
-            bseg->us_payload.pml_header->tag;
+        reg = mca_btl_base_active_message_trigger + bseg->us_btl_header->tag;
         seg->rs_segment.seg_len = bseg->us_btl_header->payload_len;
-        reg->cbfunc(&module->super, bseg->us_payload.pml_header->tag, 
+        reg->cbfunc(&module->super, bseg->us_btl_header->tag, 
                     &seg->rs_desc, reg->cbdata);
 
 drop:
@@ -384,10 +383,9 @@ ompi_btl_usnic_recv(ompi_btl_usnic_module_t *module,
          * the frame length to meet minimum sizes, add protocol information,
          * etc.
          */
-        reg = mca_btl_base_active_message_trigger +
-            bseg->us_payload.pml_header->tag;
+        reg = mca_btl_base_active_message_trigger + bseg->us_btl_header->tag;
         seg->rs_segment.seg_len = bseg->us_btl_header->payload_len;
-        reg->cbfunc(&module->super, bseg->us_payload.pml_header->tag, 
+        reg->cbfunc(&module->super, bseg->us_btl_header->tag,
                     &seg->rs_desc, reg->cbdata);
 
     } else {
diff --git a/ompi/mca/btl/usnic/btl_usnic_send.c b/ompi/mca/btl/usnic/btl_usnic_send.c
index 8e51e24aee..9cc8c6141c 100644
--- a/ompi/mca/btl/usnic/btl_usnic_send.c
+++ b/ompi/mca/btl/usnic/btl_usnic_send.c
@@ -152,6 +152,7 @@ ompi_btl_usnic_send_slower(
 
         /* use standard channel */
         sseg->ss_channel = USNIC_DATA_CHANNEL;
+        sseg->ss_base.us_btl_header->tag = tag;
 #if MSGDEBUG2
     opal_output(0, "  small frag %d segs %p(%d) + %p(%d)\n",
             (int)frag->sf_base.uf_base.des_src_cnt,
@@ -166,6 +167,12 @@ ompi_btl_usnic_send_slower(
             (void *)sseg->ss_send_desc.sg_list[1].addr,
             sseg->ss_send_desc.sg_list[1].length);
 #endif
+    } else {
+        ompi_btl_usnic_large_send_frag_t *lfrag;
+
+        lfrag = (ompi_btl_usnic_large_send_frag_t *)frag;
+
+        lfrag->lsf_tag = tag;
     }
 
     /* queue this fragment into the send engine */
diff --git a/ompi/mca/btl/usnic/btl_usnic_send.h b/ompi/mca/btl/usnic/btl_usnic_send.h
index 586305bf2a..a6b9fdff94 100644
--- a/ompi/mca/btl/usnic/btl_usnic_send.h
+++ b/ompi/mca/btl/usnic/btl_usnic_send.h
@@ -53,6 +53,21 @@ ompi_btl_usnic_check_rts(
     }
 }
 
+#if MSGDEBUG2
+static inline
+int sge_total(struct ibv_send_wr *wr)
+{
+    int i;
+    int len;
+    len=0;
+    for (i=0; i<wr->num_sge; ++i) {
+        len += wr->sg_list[i].length;
+    }
+
+    return len;
+}
+#endif
+
 /*
  * Common point for posting a segment to VERBS
  */
@@ -68,10 +83,11 @@ ompi_btl_usnic_post_segment(
     int ret;
 
 #if MSGDEBUG1
-    opal_output(0, "post_send: type=%d, addr=%p, len=%d\n",
-                sseg->ss_base.us_type,
+    opal_output(0, "post_send: type=%s, addr=%p, len=%d, payload=%d\n",
+                usnic_seg_type(sseg->ss_base.us_type),
                 (void*) sseg->ss_send_desc.sg_list->addr, 
-                sseg->ss_send_desc.sg_list->length);
+                sge_total(&sseg->ss_send_desc),
+                sseg->ss_base.us_btl_header->payload_len);
     /*ompi_btl_usnic_dump_hex((void *)(sseg->ss_send_desc.sg_list->addr + sizeof(ompi_btl_usnic_btl_header_t)), 16); */
 #endif
 
@@ -215,8 +231,9 @@ ompi_btl_usnic_endpoint_enqueue_frag(
     ompi_btl_usnic_send_frag_t *frag)
 {
 #if MSGDEBUG1
-    opal_output(0, "enq_frag: frag=%p, endpoint=%p, type=%d, len=%"PRIu64"\n",
-            (void*)frag, (void*)endpoint, frag->sf_base.uf_type,
+    opal_output(0, "enq_frag: frag=%p, endpoint=%p, %s, len=%"PRIu64"\n",
+            (void*)frag, (void*)endpoint,
+            usnic_frag_type(frag->sf_base.uf_type),
             frag->sf_base.uf_base.des_src->seg_len);
     if (frag->sf_base.uf_type == OMPI_BTL_USNIC_FRAG_LARGE_SEND) {
         ompi_btl_usnic_large_send_frag_t *lfrag;