WIP: Reuse free memory from GMP in PGC

[LinHu2016]

Step1: Recover Region tails during post-sweep of GMP and Reuse them during PGC
Recover tail point to the last freespace which connect to region top,
can be reused as survivor or tenure space during copyforward.
the first sweep after GMP does reset all of region tails.
recover region tails during post-sweep could reduce copyforward
abort cases.

- New XXgc:tarokEnableRecoverRegionTailsAfterSweep option for enabling
region tail recovery(default == false).
- New XXgc:tarokEnableAllocationPointerAssertion option for enabling
recovered tail verification(default == false).
- Maintain free list during sweep for MemoryPoolBumpPointer.
- Maintain last free entry during sweep for MemoryPoolBumpPointer.
- Rcover region tail from last free entry.
- Verify recovered region tail via reverse searching markmap
- Align recovered tail with the card and clear card for the tail.

Step2:Reuse the largest free memory from GMP in PGC
The largest free space inside region(if it is not tail) can be used during
collector allocation. the free list has been created during the sweep after GMP,
but in order to keep collector allocation simple and efficient, only region tail
and the largest free space would be used during collector allocation.

- New XXggc:tarokEnableRecoverRegionLargestFreeMemory option for
 enabling region largest free memory recovery(default == false).
 the option would be ignored if tarokEnableRecoverRegionTailsAfterSweep
 is disabled.
- Maintain largest free entryt sweep for MemoryPoolBumpPointer.
- Align the largest free memory with the card and clear card for the free spaces.
- Reuse the regions with the largest free memory first, then try
to reuse the regions with free tail only before using free regions.
- Still using region base lock for the allocation synchronization,
for the region with both largest free memory and region tail, only
can be used by one thread at the same time.

Signed-off-by: Lin Hu linhu@ca.ibm.com

depends on eclipse-omr/omr#4284
depends on #8804
depends on eclipse-omr/omr#5514

[amicic]

We need to better understand why there could still be dirty cards in the tail. Let's hold off this change till then. Then, let's hold #7033 and #7031 too, which are not that useful without this change.

[amicic]

These changes are very good from a perspective that you've identified all spots that are affected due to attempt to use free memory, not just from the tail. However, they are restrictive since you've only try to identify one additional (largest) free entry (while there could be several medium to large free entries in a region). It adds more complexity (duplication) in code since you have to separately deal with largest free entry and tail.

So, let's separately try (leave this code as-is for now, and create a new branch) to generalize things and try to completely replace MP Bump Pointer with MP Address Ordered List and try to reuse any largish free entry. Let's complete remove a notion of a free tail (candidate region), and deal with free memory (candidate region). Some things should get simpler (relative to this approach), like no 2 separate cases to deal with, but for sure some things might get more complex like overkill complexity of AOL for Mutator allocation from Eden regions that effectively always have only a tail.

Then, we'll compare them and see what's better (code wise and perf wise).

[genie-openj9]

Can one of the admins verify this patch?

[LinHu2016]

close this PR, #11862 is better replacement solution.