On Solaris, the process calling mlock needs the "proc_lock_memory" privilege in RBAC.

Even with that privilege, mlock on Solaris fails if the address passed is not a multiple of the page size. (i.e. it must be the start of a page.) errno is set to EINVAL in this case. This causes test 2 in s2n_ecc_test to fail.

This restriction is part of the POSIX standard that you are compiling against (_POSIX_C_SOURCE=200112L) so it looks to me like your use of mlock() is not correct. Does it make sense to make the use of mlock configurable, or should you be using posix_memalign() instead of realloc()? Something like the following allows a significant number of tests to pass:

diff --git a/utils/s2n_mem.c b/utils/s2n_mem.c
index 7591274..9a4ce1d 100644
--- a/utils/s2n_mem.c
+++ b/utils/s2n_mem.c
@@ -38,7 +38,8 @@ int s2n_realloc(struct s2n_blob *b, uint32_t size)
         return 0;
     }

-    uint8_t *data = realloc(b->data, size);
+    void *data = NULL;
+    GUARD(posix_memalign(&data, 4096, size));
     if (data == NULL) {
         S2N_ERROR(S2N_ERR_REALLOC);
     }