Request for review: 6761744 Hotspot crashes if process size limit is exceeded

[Tao Mao]

I think alignment is handled slightly differently in the two/three classes.

Tao

On 4/22/13 2:56 AM, Thomas Schatzl wrote:
> Hi,
>
> On Mon, 2013-04-22 at 09:58 +0200, Bengt Rutisson wrote:
>> Hi Tao,
>>
>> On 4/19/13 7:27 PM, Tao Mao wrote:
>>
>>> 1. check overflow routine is wrapped up.
>> I'm not sure we need round_up_total_reserved_and_check_overflow() since
>> it is only being used in one place.
> Actually it is an interesting question why only the total of
> GenCollectedHeaps is rounded up to large page size and the others are
> not.
>
> It seems that in parallel old, at these heap sizes, some code that
> calculates internal alignment would result in an alignment>  large page
> size, so it would be okay, i.e. overflow already when adding together
> the components.
>
> At least with G1 with a 1 MB region size there might be an issue as e.g.
> the perm gen size is rounded to heap region size. Or there is not an
> issue at all, but why then do the rounding only for GenCollectedHeaps?
>
> (I think the rounding can at most make the total overflow into a
> resulting zero. I am not sure if the heap reserve functions give an
> error if you pass a requested virtual space size of zero anyway...)
>
> Hth,
> Thomas
>
>