RFR: Enable remembered set verification during global collections

[William Kemper]

On Wed, 8 Sep 2021 21:02:33 GMT, Kelvin Nilsen <kdnilsen at openjdk.org> wrote:

>> src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp line 2657:
>> 
>>> 2655:             HeapWord* tams = ctx->top_at_mark_start(r);
>>> 2656:             if (obj_addr >= tams) {
>>> 2657:               obj_addr += obj->size();
>> 
>> Were you actually seeing obj_addr >= tams here?  I'd be inclined to replace your new code with an assert(obj_addr < tams).  Or provide comments to explain.
>
> If we verify remset while old marking is "ongoing", ctx is not "reliable".  In that case, I had been arranging for ctx to equal NULL with the understanding that obj->size() would get me to the next (possibly coalesced and filled) object within old-gen.  I don't yet follow the logic for how you can always rely on ctx being not NULL here.

If `ctx` is NULL, we'll have gone into the other path and used `obj->size()` to advance `obj_addr`. I'll add an assert that `obj_addr` is below TAMS during the init-mark verification.  We only verify the remembered set during a mixed evacuation or global collect (as of this change). In both cases there should be no concurrent marking of old (I'll add an assert for this also).

-------------

PR: https://git.openjdk.java.net/shenandoah/pull/64