code review request for MacOS X compressed oops work around (7118648)

[Vladimir Kozlov]

Could you move the comment before #ifndef __APPLE__?

Vladimir

Daniel D. Daugherty wrote:
> Greetings,
> 
> The UseCompressedOops flag is enabled ergonomically in a 64-bit VM.
> With MacOS X, we're seeing intermittent crashes in our automated
> build/test system (JPRT) that look like this:
> 
> Internal Error 
> (/private/tmp/jprt/P1/050642.jmelvin/source/src/share/vm/runtime/virtualspace.cpp:460), 
> pid=53634, tid=4531154944
> #  assert((_noaccess_prefix != 0) == (UseCompressedOops && _base != NULL 
> && (size_t(_base + _size) > OopEncodingHeapMax) && 
> Universe::narrow_oop_use_implicit_null_checks())) failed: 
> noaccess_prefix should be used only with non zero based compressed oops
> 
> I have filed the following bug to track that failure mode:
> 
>     7118647 3/3 compressed oops crashes on MacOS X with JPRT GCBasher test
> 
> In order to make more reliable progress on getting MacOS X fully
> enabled in JPRT, I'm planning to disable compressed oops by default
> in the 64-bit MacOS X VM until 7118647 can be fixed.
> 
> The work around is trivial so I'm including it here rather than
> generate a webrev:
> 
> diff -r a5a9db0e0203 src/share/vm/runtime/arguments.cpp
> --- a/src/share/vm/runtime/arguments.cpp        Tue Dec 06 07:23:42 2011 
> -0800
> +++ b/src/share/vm/runtime/arguments.cpp        Tue Dec 06 15:29:17 2011 
> -0800
> @@ -1359,9 +1359,12 @@
>    // by ergonomics.
>    if (MaxHeapSize <= max_heap_for_compressed_oops()) {
>  #if !defined(COMPILER1) || defined(TIERED)
> +#ifndef __APPLE__
> +    // disable UseCompressedOops by default on MacOS X until 7118647 is 
> fixed
>      if (FLAG_IS_DEFAULT(UseCompressedOops)) {
>        FLAG_SET_ERGO(bool, UseCompressedOops, true);
>      }
> +#endif // !__APPLE__
>  #endif
>  #ifdef _WIN64
>      if (UseLargePages && UseCompressedOops) {
> 
> 
> What the above code does is prevent the UseCompressedOops flag
> from being set to true ergonomically on MacOS X. It can still be
> set to true with the '-XX:+UseCompressedOops' option.
> 
> I'm planning to push this fix to RT_Baseline for the HSX-23-B08
> snapshot. Jim Melvin has already reviewed the fix so I just need
> one more reviewer.
> 
> Thanks, in advance, for any additional reviews.
> 
> Dan
>