handle system call failure: delay vs undo, races, etc.

_From [derek.br...@gmail.com](https://code.google.com/u/117968039472581148324/) on December 10, 2010 17:57:47_

PR 408540

on syscall failure:
/\* FIXME: the shadow writes we enacted in event_pre_syscall() should be
\* considered to have NOT happened. We can detect some with checks
\* on known IN args that will cause failure (NULL, etc.).
\* How handle races though? Xref all the discussion over malloc/free
\* failure races, possibility of locking, and whether better to
\* undo or delay.
*/

_Original issue: http://code.google.com/p/drmemory/issues/detail?id=119_