RFR: Fix reclamation of humongous objects in mark-compact

[Aleksey Shipilev]

On 11/29/2016 01:06 PM, Roman Kennke wrote:
> Aleksey's latest tests revealed a bug that deserves a bug DUH. Mark-
> compact would never reclaim any humongous regions.
> 
> 
> - mark-compact reclaims humongous regions *after* compaction, i.e. it
> cannot utilize the new free space for compaction
> 
> this is unfortunate, but the real bug is this:
> 
> - it reclaims humongous regions after having cleared the bitmap. note:
> we need marking bits for deciding whether or not to reclaim a humongous
> region.
> 
> I hear you saying: shuoldn't this reclaim *all* humongous regions,
> instead of none? Yes, but:
> 
> - before doing so, it resets the top-at-mark-start ptr to bottom,
> treating all objects as implicitely marked.
> 
> Triple-strike!! :-D

/me facepalms
/me facepalms again


> http://cr.openjdk.java.net/~rkennke/mark-compact-humongous/webrev.00/

Yes, looks good to me.

Thanks,
-Aleksey