RFR: Refactor heap verification

[Zhengyu Gu]

On 06/01/2017 04:57 PM, Zhengyu Gu wrote:
>
>
> On 06/01/2017 04:44 PM, Aleksey Shipilev wrote:
>> On 06/01/2017 10:41 PM, Zhengyu Gu wrote:
>>> Is this right?
>>>
>>>  239   if (ShenandoahVerify) {
>>>  240     ReservedSpace verify_bitmap(_bitmap_size, page_size);
>>>  241     os::commit_memory_or_exit(verify_bitmap.base(),
>>> verify_bitmap.size(),
>>> false,
>>>  242                               "couldn't allocate verification
>>> bitmap");
>>>  243     MemTracker::record_virtual_memory_type(verify_bitmap.base(),
>>> mtGC);
>>>  244     MemRegion verify_bitmap_region = MemRegion((HeapWord *)
>>> verify_bitmap.base(), verify_bitmap.size() / HeapWordSize);
>>>  245     _verification_bit_map.initialize(_heap_region,
>>> verify_bitmap_region);
>>>  246   }
>>>  247
>>>  248   _verifier = new ShenandoahVerifier(this, &_verification_bit_map);
>>>  249
>>>
>>> You want to construct _verifier without initialized bitmap?
>>
>> Verifier would barf when used without ShenandoahVerify enabled. And
>> this saves
>> NULL-checking on every verifier()->foo() call. Seems like the lesser
>> evil,
>> although we need to think if we just want to initialize verification
>> bitmap
>> unconditionally, because sometimes Universe asks us to verify anyway.
>
> Well, it can barf in SH::verifier() method, and save the checking inside
> verifier.
>

Okay, never mind. Not all verification need bitmap.

Good to me.

-Zhengyu

> -Zhengyu
>
>
>>
>> -Aleksey
>>