Request for reviews (M): 7017124: VM statistic use 32-bit values which may overflow

[David Holmes]

Vladimir Kozlov said the following on 02/04/11 10:05:
> Added new Atomic method inc_counter() to increment long values,
> it is not precise on MP (lock is not used) but it is fine.

I strongly disagree! At least it has no place being in the Atomic:: set 
of functions as it is not atomic. It is not even monotonic! It is 
extremely deceptive to do this. By all means change to julong where 
needed, but unless they are truly going to be atomic updates (I 
understand the overhead makes that prohibitive) then leave them as 
simply  x+=i; statements.

> Fixed several output formats to use VM FORMAT macros.
> 
> I have to remove v9 check assert from Atomic_move_long() since
> it is called before VM_Version is initialized and C2 is built
> for v9 only anyway.
> 
> Removed unused Chunk::clean_chunk_pool().

Used by CI.

> I fixed only statistic which was interesting to me.

So the CR is misleading. It purports to deal with 32-bit stat overflows 
but in fact only deals with some stats, and primarily you are using this 
to define PrintMallocStat - so the CR should reflect that: "Fix some VM 
stats to avoid 32-bit overflow"


globals.hpp:

+           "print malloc/free statictics") 
       \

Typo: statictics

src/share/vm/opto/phase.cpp

This change seems unrelated to the CR.

David
-----