G1 calculates survivor capacity based on eden capacity and SurvivorRatio (literally current_eden_capacity / SurvivorRatio).

If eden capacity is very large, e.g. set by limiting min young gen size by G1NewSizePercent to something > 50%, and SurvivorRatio=1, the total of eden, survivor and old gen capacities will exceed total heap size.

This is just an issue with logging, e.g. on a 256M heap it is possible that the following log output is given:

[2.335s][info ][gc,heap ] GC(0) Eden regions: 153->0(133)
[2.335s][info ][gc,heap ] GC(0) Survivor regions: 0->20(133)

I.e. eden size before gc + survivor capacity = 286 regions, which are more regions than available in the heap.

The problem is that in the calculation of the next survivor capacity in G1CollectorPolicy::update_survivors_policy() does not guarantee consistency.

This seems to only be a printing issue.

This is a day one G1 bug.