usnic: fix module_recv_buffers perf regression

Cisco v1.6 git commit 913ec6c and upstream trunk r29593 (segfault fix)
introduced a performance regression by inadvertently disabling the
`module_recv_buffers` functionality.  With those changes in place, the
`btl_usnic_recv.c` logic would end up mallocing a buffer that should
have otherwise come from a `module_recv_buffers` pool.  It also resulted
in a small, bounded memory leak (128 buffers at each power-of-two size
interval).

The new version just places the buffer after the free list item with a
flexible array member.  I bumped the pool to allocate all 128 elements
up front because the deferred allocation was modestly impacting IMB
Sendrecv performance at a few sizes.

Reviewed-by: Reese Faucette <rfaucett@cisco.com>

This commit was SVN r29631.

The following SVN revision numbers were found above:
  r29593 --> open-mpi/ompi@1ed9b8ff43

ompi/mca/btl/usnic/btl_usnic_endpoint.h

@@ -100,8 +100,7 @@ typedef struct ompi_btl_usnic_rx_frag_info_t {
     char       *rfi_data;       /* pointer to assembly area */
     int         rfi_data_pool;  /* if 0, data malloced, else rx buf pool */
     ompi_free_list_item_t *rfi_fl_elt; /* free list elemement from buf pool
-                                          (rfi_fl_elt->ptr==rfi_data) when
-                                          rfi_data_pool is nonzero */
+                                          when rfi_data_pool is nonzero */
 } ompi_btl_usnic_rx_frag_info_t;
 
 /**

ompi/mca/btl/usnic/btl_usnic_frag.c

@@ -305,3 +305,8 @@ OBJ_CLASS_INSTANCE(ompi_btl_usnic_put_dest_frag_t,
                    ompi_btl_usnic_frag_t,
                    put_dest_frag_constructor,
                    put_dest_frag_destructor);
+
+OBJ_CLASS_INSTANCE(ompi_btl_usnic_rx_buf_t,
+                   ompi_free_list_item_t,
+                   NULL,
+                   NULL);

ompi/mca/btl/usnic/btl_usnic_frag.h

@@ -319,6 +319,18 @@ typedef struct ompi_btl_usnic_small_send_frag_t {
  */
 typedef ompi_btl_usnic_frag_t ompi_btl_usnic_put_dest_frag_t;
 
+/**
+ * A simple buffer that can be enqueued on an ompi_free_list_t that is intended
+ * to be used for fragment reassembly.  Nominally the free list code supports
+ * this via the rb_super.ptr field, but that field is only allocated and
+ * non-NULL if an mpool is used, and we don't need this reassembly memory to be
+ * registered.
+ */
+typedef struct ompi_btl_usnic_rx_buf_t {
+    ompi_free_list_item_t rb_super;
+    char buf[1]; /* flexible array member for frag reassembly */
+} ompi_btl_usnic_rx_buf_t;
+
 OBJ_CLASS_DECLARATION(ompi_btl_usnic_send_frag_t);
 OBJ_CLASS_DECLARATION(ompi_btl_usnic_small_send_frag_t);
 OBJ_CLASS_DECLARATION(ompi_btl_usnic_large_send_frag_t);
@@ -329,6 +341,8 @@ OBJ_CLASS_DECLARATION(ompi_btl_usnic_frag_segment_t);
 OBJ_CLASS_DECLARATION(ompi_btl_usnic_chunk_segment_t);
 OBJ_CLASS_DECLARATION(ompi_btl_usnic_recv_segment_t);
 
+OBJ_CLASS_DECLARATION(ompi_btl_usnic_rx_buf_t);
+
 typedef ompi_btl_usnic_send_segment_t ompi_btl_usnic_ack_segment_t;
 OBJ_CLASS_DECLARATION(ompi_btl_usnic_ack_segment_t);
 

ompi/mca/btl/usnic/btl_usnic_module.c

@@ -2006,16 +2006,17 @@ int ompi_btl_usnic_module_init(ompi_btl_usnic_module_t *module)
     assert(module->module_recv_buffers != NULL);
     for (i=module->first_pool; i<=module->last_pool; ++i) {
         OBJ_CONSTRUCT(&module->module_recv_buffers[i], ompi_free_list_t);
+        size_t elt_size = sizeof(ompi_btl_usnic_rx_buf_t) - 1 + (1 << i);
         rc = ompi_free_list_init_new(&module->module_recv_buffers[i],
-                                     1 << i,
+                                     elt_size,
                                      opal_cache_line_size,
-                                     OBJ_CLASS(ompi_free_list_item_t),
-                                     0,  /* payload size */
-                                     0,  /* payload align */
-                                     8,
-                                     128,
-                                     8,
-                                     NULL);
+                                     OBJ_CLASS(ompi_btl_usnic_rx_buf_t),
+                                     0,   /* payload size */
+                                     0,   /* payload align */
+                                     128,   /* init elts to alloc */
+                                     128, /* max elts to alloc */
+                                     128,   /* num elts per alloc */
+                                     NULL /* mpool */);
         assert(OMPI_SUCCESS == rc);
     }
 

ompi/mca/btl/usnic/btl_usnic_recv.c

@@ -215,11 +215,13 @@ void ompi_btl_usnic_recv_call(ompi_btl_usnic_module_t *module,
                 if (pool >= module->first_pool &&
                         pool <= module->last_pool) {
                     ompi_free_list_item_t* item;
+                    ompi_btl_usnic_rx_buf_t *rx_buf;
                     OMPI_FREE_LIST_GET_MT(&module->module_recv_buffers[pool],
                                           item);
-                    if (OPAL_LIKELY(NULL != item)) {
+                    rx_buf = (ompi_btl_usnic_rx_buf_t *)item;
+                    if (OPAL_LIKELY(NULL != rx_buf)) {
                         fip->rfi_fl_elt = item;
-                        fip->rfi_data = item->ptr;
+                        fip->rfi_data = rx_buf->buf;
                         fip->rfi_data_pool = pool;
                     }
                 }
@@ -302,7 +304,6 @@ void ompi_btl_usnic_recv_call(ompi_btl_usnic_module_t *module,
                 if (0 == fip->rfi_data_pool) {
                     free(fip->rfi_data);
                 } else {
-                    assert(fip->rfi_fl_elt->ptr == fip->rfi_data);
                     OMPI_FREE_LIST_RETURN_MT(
                             &module->module_recv_buffers[fip->rfi_data_pool],
                             fip->rfi_fl_elt);