[pypy-commit] stmgc use-gcc: This write_fence() was really needed, I think

[arigo]

Author: Armin Rigo <arigo at tunes.org>
Branch: use-gcc
Changeset: r1951:bba995bf44bc
Date: 2015-09-01 18:10 +0200
http://bitbucket.org/pypy/stmgc/changeset/bba995bf44bc/

Log:	This write_fence() was really needed, I think

diff --git a/c8/stm/hashtable.c b/c8/stm/hashtable.c
--- a/c8/stm/hashtable.c
+++ b/c8/stm/hashtable.c
@@ -326,6 +326,12 @@
                 stm_allocate_preexisting(sizeof(stm_hashtable_entry_t),
                                          (char *)&initial.header);
             hashtable->additions++;
+            /* make sure .object is NULL in all segments before
+               "publishing" the entry in the hashtable.  In other words,
+               the following write_fence() prevents a partially
+               initialized 'entry' from showing up in table->items[i],
+               where it could be read from other threads. */
+            write_fence();
         }
         table->items[i] = entry;
         write_fence();     /* make sure 'table->items' is written here */