Black allocation can lead to overestimation of live bytes (see  issue 615489 ). I suspect that this was causing spike of 0ms-1ms GCs in M51 and biasing V8.GCFinalizeMC histogram.


Related CLs:

- https://codereview.chromium.org/1862063002 - [heap] Don't use black pages for map, code and, lo space. Instead color objects black.

This CL reduces number of black pages and decreases overestimation error. I suspect this is the reason of apparent jank regression in Canary and Dev. But then I do not know why we don't see the same regression in M51 when the CL was merge to M51.

- https://codereview.chromium.org/1889853002 - [heap] Turn on black allocation during marking finalization to avoid floating garbage.
Not merged in M51.

- https://codereview.chromium.org/1869493002 - [heap] Do not perform black allocation in memory reducing mode.
Not merged in M51.