btl/vader: fix fast box support for 32-bit architectures

On 32-bit architectures loads/stores of fast box headers may take
multiple instructions. This can lead to a data race between the
sender/receiver when reading/writing the sequence number. This can
lead to a situation where the receiver could process incomplete
data. To fix the issue this commit re-orders the fast box header to
put the sequence number and the tag in the same 32-bits to ensure they
are always loaded/stored together.

Fixes #473

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>

opal/mca/btl/vader/btl_vader_endpoint.h

@@ -11,7 +11,7 @@
  * Copyright (c) 2004-2005 The Regents of the University of California.
  *                         All rights reserved.
  * Copyright (c) 2006-2007 Voltaire. All rights reserved.
- * Copyright (c) 2012-2014 Los Alamos National Security, LLC. All rights
+ * Copyright (c) 2012-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
  * $COPYRIGHT$
  *
@@ -48,14 +48,16 @@ typedef struct mca_btl_base_endpoint_t {
     /* per peer buffers */
     struct {
         unsigned char *buffer; /**< starting address of peer's fast box out */
-        unsigned int start, seq;
         uint32_t *startp;
+        unsigned int start;
+        uint16_t seq;
     } fbox_in;
 
     struct {
         unsigned char *buffer; /**< starting address of peer's fast box in */
-        unsigned int start, end, seq;
         uint32_t *startp;      /**< pointer to location storing start offset */
+        unsigned int start, end;
+        uint16_t seq;
     } fbox_out;
 
     int32_t peer_smp_rank;  /**< my peer's SMP process rank.  Used for accessing

opal/mca/btl/vader/btl_vader_fbox.h

@@ -1,6 +1,6 @@
 /* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
 /*
- * Copyright (c) 2011-2014 Los Alamos National Security, LLC. All rights
+ * Copyright (c) 2011-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
  * $COPYRIGHT$
  *
@@ -18,9 +18,16 @@
 
 typedef union mca_btl_vader_fbox_hdr_t {
     struct {
+        /* NTH: on 32-bit platforms loading/unloading the header may be completed
+         * in multiple instructions. To ensure that seq is never loaded before tag
+         * and the tag is never read before seq put them in the same 32-bits of the
+         * header. */
+        /** message tag */
         uint16_t  tag;
-        uint16_t  size;
+        /** sequence number */
-        uint32_t  seq;
+        uint16_t  seq;
+        /** message size */
+        uint32_t  size;
     } data;
     uint64_t ival;
 } mca_btl_vader_fbox_hdr_t;
@@ -40,6 +47,13 @@ typedef union mca_btl_vader_fbox_hdr_t {
 
 void mca_btl_vader_poll_handle_frag (mca_btl_vader_hdr_t *hdr, mca_btl_base_endpoint_t *ep);
 
+static inline void mca_btl_vader_fbox_set_header (mca_btl_vader_fbox_hdr_t *hdr, uint16_t tag,
+                                                  uint16_t seq, uint32_t size)
+{
+    mca_btl_vader_fbox_hdr_t tmp = {.data = {.tag = tag, .seq = seq, .size = size}};
+    hdr->ival = tmp.ival;
+}
+
 /* attempt to reserve a contiguous segment from the remote ep */
 static inline unsigned char *mca_btl_vader_reserve_fbox (mca_btl_base_endpoint_t *ep, size_t size)
 {
@@ -88,12 +102,10 @@ static inline unsigned char *mca_btl_vader_reserve_fbox (mca_btl_base_endpoint_t
         /* if this is the end of the buffer and the fragment doesn't fit then mark the remaining buffer space to
          * be skipped and check if the fragment can be written at the beginning of the buffer. */
         if (OPAL_UNLIKELY(buffer_free > 0 && buffer_free < size && start <= end)) {
-            mca_btl_vader_fbox_hdr_t tmp = {.data = {.size = buffer_free - sizeof (mca_btl_vader_fbox_hdr_t),
-                                                     .seq = ep->fbox_out.seq++, .tag = 0xff}};
-
             BTL_VERBOSE(("message will not fit in remaining buffer space. skipping to beginning"));
 
-            MCA_BTL_VADER_FBOX_HDR(dst)->ival = tmp.ival;
+            mca_btl_vader_fbox_set_header (MCA_BTL_VADER_FBOX_HDR(dst), 0xff, ep->fbox_out.seq++,
+                                           buffer_free - sizeof (mca_btl_vader_fbox_hdr_t));
 
             end = MCA_BTL_VADER_FBOX_ALIGNMENT;
             /* toggle the high bit */
@@ -114,11 +126,7 @@ static inline unsigned char *mca_btl_vader_reserve_fbox (mca_btl_base_endpoint_t
                  (unsigned int) size, end, start, end, hbs, buffer_free));
 
     /* write out part of the header now. the tag will be written when the data is available */
-    {
+    mca_btl_vader_fbox_set_header (MCA_BTL_VADER_FBOX_HDR(dst), 0, ep->fbox_out.seq++, data_size);
-        mca_btl_vader_fbox_hdr_t tmp = {.data = {.size = data_size, .tag = 0, .seq = ep->fbox_out.seq++}};
-
-        MCA_BTL_VADER_FBOX_HDR(dst)->ival = tmp.ival;
-    }
 
     end += size;