RFR: 8262073: assert(allocates2(pc)) failed: not in CodeBuffer memory

[Yangfei (Felix)]

Hi,

  Please review this 8u aarch64-specific fix: 
  Bug: https://bugs.openjdk.java.net/browse/JDK-8262073 
  Webrev: http://cr.openjdk.java.net/~fyang/8262073/webrev.00/ 

  Call chain for creating itable stub: create_itable_stub -> load_klass -> decode_klass_not_null -> reinit_heapbase 
  As reported on the issue, we are emitting three instructions in reinit_heapbase for this case: mov, movk and movk. 
  Note that we also emit insns in decode_klass_not_null  moving narrow_klass_base into a register. 
  To be safe, I think we should also assume insns for moving narrow_klass_base  to be: mov, movk and movk. 
  Patch increases size for both kind of stubs: 16 bytes for itable stub (which calls load_klass twice) and 8 byte for vtable stub (which calls load_klass once).

  Previously, we reserve extra 216 bytes when DebugVtables is on.  This is enough for the itable stub creation case.
  But this won't work for the vtable stub creation in which case we will exceed the code buffer limit. 
  The reason is that we have two "if (DebugVtables)" blocks in vtable stub creation code. 
  For this case, patch adopts the same way as ppc & sparc: set size to 1000 bytes when those debug options are turned on.

  Performed full jtreg test on aarch64-linux-gnu.  OK? 

Thanks,
Felix