Work around sigaltstack() bug on Mac OS X when unregistering a signal stack

Work around sigaltstack() bug on Mac OS X when unregistering a signal stack

This bug manifests itself when building with
"-mmacosx-version-min=10.5", which is necessary in order to build
sel_ldr for x86-64 but which causes a newer version of libc's
sigaltstack() wrapper to be used.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=1051
BUG=http://code.google.com/p/nativeclient/issues/detail?id=1053
TEST=run_hello_world_test when SConstruct is changed to use
  "-mmacosx-version-min=10.5" instead of 10.4

Committed: http://src.chromium.org/viewvc/native_client?view=rev&revision=3504

[Mark Seaborn, 2010-10-17 19:46:25]

I guess we should ignore, for now, the problem that signal handlers on Mac won't
interoperate with Breakpad (where we should use Mach exception handling
instead), and fix the code that is already in NaCl...

[Mark Seaborn, 2010-10-18 19:40:44]

Cliff wrote:
> Well, Rietveld is totally hosed for me right now (getting JavaScript
> console errors about M_handleTableDblClick not being defined), so here
> are my two questions:
> 
> 1. Have you found evidence of anyone else encountering this bug?

Nope.

> I wasn't able to find anything on OpenRadar, and of course we can't
> search the *actual* Apple bug database.  If you haven't, have we
> reported it to Apple?

No, I don't know how to.

> 2. Other POSIX platforms ignore that parameter when unregistering,
> right?  So could we take out the conditionals?

We could, but I don't like to keep it in for non-broken platforms that
don't need it.  There is a teeny risk that another bug might be to
ignore SS_DISABLE in which case we'd register a stack at the bottom of
address space; or to give an error because ss_size != 0.  Or people
might copy "ss_size = MINSIGSTKSZ" into other code uncritically and
lose the explanatory comment...

Mark

[Cliff L. Biffle, 2010-10-19 21:25:26]

LGTM