Add support for exception handling using the PNaClSjLjEH LLVM pass

Add support for exception handling using the PNaClSjLjEH LLVM pass

Extend libcxxabi with support for PNaCl's SJLJ
(setjmp()+longjmp()-based) C++ exception handling.  The routines
assume that the program has been transformed using the PNaClSjLjEH
LLVM pass, which inserts setjmp() calls around function calls that
need to handle C++ exceptions.

As noted in the comments, this will coexist with non-SJLJ (zero cost)
exception handling within the same build of libcxxabi.  When SJLJ is
enabled, pnacl-ld.py will rename
__pnacl_eh_sjlj_Unwind_RaiseException() to _Unwind_RaiseException() at
link time to enable cxa_pnacl_sjlj_exception.cpp.

This is similar to the eh_pnacl.cc file I added before to libsupc++.
Some of the differences are:

 * Coding style:  4-space indentation, etc.

 * Use libcxxabi's can_catch() instead of libsupc++'s __do_catch().
   We no longer need the __is_pointer_p() check, because that's built
   into can_catch().  So get_adjusted_ptr() becomes trivial and can be
   inlined.

 * Add casting between std::type_info and __shim_type_info, which are
   equivalent.

 * Add helpers: get_object_from_ue(), get_exception_header_from_ue().
   libcxxabi tends to open code the implementation of these.

BUG=https://code.google.com/p/nativeclient/issues/detail?id=3696
TEST=run_eh_throw_tests_sjlj_test from the NaCl build

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-libcxxabi.git;a=commit;h=bf2aa59

[Derek Schuff, 2013-12-04 18:50:54]

On 2013/12/04 18:38:18, Mark Seaborn wrote:

+jfb in case he is interested: also for reference the libstdc++ analog is in
https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-gcc.git;a=blo...

[JF, 2013-12-04 20:54:09]

Does this fix all the EH tests marked as failing for:
  ./pnacl/scripts/llvm-test.py --libcxx-tests
And pass the GCC torture tests?

The implementation looks good in general, but I'll let Derek dive deeper since
he looked at the libsupc++ one.

[Derek Schuff, 2013-12-04 22:26:26]

lgtm

https://codereview.chromium.org/95613004/diff/70001/src/cxa_pnacl_sjlj_except...
File src/cxa_pnacl_sjlj_exception.cpp (right):

https://codereview.chromium.org/95613004/diff/70001/src/cxa_pnacl_sjlj_except...
src/cxa_pnacl_sjlj_exception.cpp:70: // matches none of the exception types in a
C++ exception
s/none/one... small change, big meaning :)

https://codereview.chromium.org/95613004/diff/70001/src/cxa_pnacl_sjlj_except...
src/cxa_pnacl_sjlj_exception.cpp:233: if (!find_match((__shim_type_info *)
xh->exceptionType, &obj,
does libcxxabi's can_catch handle dependent exceptions correctly such that we
can just pass xh->exceptionType here (whereas get_adjusted_ptr did not)?

[Mark Seaborn, 2013-12-04 23:03:17]

https://codereview.chromium.org/95613004/diff/70001/src/cxa_pnacl_sjlj_except...
File src/cxa_pnacl_sjlj_exception.cpp (right):

https://codereview.chromium.org/95613004/diff/70001/src/cxa_pnacl_sjlj_except...
src/cxa_pnacl_sjlj_exception.cpp:70: // matches none of the exception types in a
C++ exception
On 2013/12/04 22:26:26, Derek Schuff wrote:
> s/none/one... small change, big meaning :)

That was intentional!  I do mean "none".

The meaning of exception_spec_can_catch() is the opposite of libsupc++'s
check_exception_spec(), to match the function of the same name in libcxxabi's
cxa_personality.cpp.

https://codereview.chromium.org/95613004/diff/70001/src/cxa_pnacl_sjlj_except...
src/cxa_pnacl_sjlj_exception.cpp:233: if (!find_match((__shim_type_info *)
xh->exceptionType, &obj,
On 2013/12/04 22:26:26, Derek Schuff wrote:
> does libcxxabi's can_catch handle dependent exceptions correctly such that we
> can just pass xh->exceptionType here (whereas get_adjusted_ptr did not)?

It's not that can_catch() handles them, but __cxa_exception and
__cxa_dependent_exception have almost the same layout in libcxxabi: they both
have an exceptionType field at the same offset.

So this does work for dependent exceptions, and NaCl's eh_throw_tests.cc covers
this.

[Mark Seaborn, 2013-12-05 00:42:14]

@Derek: Can you have a look at the fix I sneaked in for patch set 6, which fixes
the case tested by https://codereview.chromium.org/105593002/?

[Mark Seaborn, 2013-12-05 00:54:21]

On 4 December 2013 12:54, <jfb@chromium.org> wrote:

> Does this fix all the EH tests marked as failing for:
>   ./pnacl/scripts/llvm-test.py --libcxx-tests
> And pass the GCC torture tests?


It passes the EH tests from the GCC torture tests, except for uncaught1.C,
which I investigated.

uncaught1.C fails because libcxxabi does "globals->uncaughtExceptions += 1"
in __cxa_throw(), but libsupc++'s comments say it should be done
in __cxa_allocate_exception() instead.

I haven't tried libc++'s EH test suite yet.

Mark

-- 
You received this message because you are subscribed to the Google Groups
"Native-Client-Reviews" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to native-client-reviews+unsubscribe@googlegroups.com.
To post to this group, send email to native-client-reviews@googlegroups.com.
Visit this group at http://groups.google.com/group/native-client-reviews.
For more options, visit https://groups.google.com/groups/opt_out.

[Derek Schuff, 2013-12-05 01:23:10]

virtual base case LGTM too

[Mark Seaborn, 2013-12-05 05:25:27]

Committed patchset #6 manually as rbf2aa59 (presubmit successful).