Use EXTERNAL_REFERENCE instead of RUNTIME_ENTRY for Runtime::PerformGC and

Use EXTERNAL_REFERENCE instead of RUNTIME_ENTRY for Runtime::PerformGC and
CallApiFunctionAndReturn in X64

Committed: http://code.google.com/p/v8/source/detail?r=13287

[haitao.feng, 2012-12-28 07:04:18]

From reading the codes, it seems that RUNTIME_ENTRY is added for IA32 for
relative direct jump/conditional branch/call instructions. RUNTIME_ENTRY is
encoded and decoded as an EXTERNAL_REFERENCE and they are both fixed (not moved
by GC). The difference is that for RUNTIME_ENTRY, when the IA32 generated codes
are moved by GC, the offset in the instruction stream need to be adjusted by
RelocInfo::apply. So in X64 for the RUNTIME_ENTRY usages in this CL, they could
be replaced by EXTERNAL_REFERENCE. 

After this change, the RUNTIME_ENTRY could be freed for the deoptimization table
entries (they will be generated in the code range and fixed) in X64 and will be
used for relative direct jump/conditional branch/call in the same way for IA32. 


There is another RUNTIME_ENTRY usage "masm->call(x64_coverage_function,
RelocInfo::RUNTIME_ENTRY)" which is guarded by "#ifdef GENERATED_CODE_COVERAGE"
in macro-assembler-x64.h. I have tried to enable GENERATED_CODE_COVERAGE in the
Makefile and common.gypi, and this functionality is broken for all the
platforms.

[danno, 2012-12-28 12:16:34]

This is a good idea, and thanks for all your work trying to untangle this
RelocInfo business, I think you're almost there.

I'd ask you to please alter the code coverage macro even though it might appear
to be broken, i.e.

masm->movq(rax, reinterpret_cast<int64_t>(x64_coverage_function),
RelocInfo::EXTERNAL_REFERENCE)
masm->call(eax);

instead of

masm->call(x64_coverage_function, RelocInfo::RUNTIME_ENTRY);

That way it's at least not more broken than it already is.

[haitao.feng, 2012-12-28 14:00:33]

danno, You are welcome :) Thanks a lot for pointing out this solution. I really
enjoy reading the source codes by following your suggestion on using
RUNTIME_ENTRY.

The code is updated for the GENERATED_CODE_COVERAGE macro.

[danno, 2012-12-28 14:36:24]

lgtm, I'll land this for you