RFR(S): 8223444: Improve CodeHeap Free Space Management

[Schmidt, Lutz]

Thank you, Tobias!
Local builds and tests are currently running. I will ask jdk/submit for its opinion before pushing.
Regards,
Lutz

On 20.05.19, 12:59, "Tobias Hartmann" <tobias.hartmann at oracle.com> wrote:

    Hi Lutz,
    
    On 20.05.19 12:49, Schmidt, Lutz wrote:
    > Hi Tobias, 
    > 
    > thank you for your comments.
    > 
    > I do not need "res". It's just personal preference. I like variable names which tell something about their contents. A reasonably capable compiler should be able to avoid stack footprint for res.
    
    Okay, fine with me.
    
    > Yes, I can convert the warning into an assert if you like. I opted for warning to get information about all segments with unexpected state, not just the first one. Now that it works, I can live with assert as well. Do you need to see a new full webrev? Here is the delta: 
    > 
    > diff -r 802d016189e0 src/hotspot/share/memory/heap.cpp
    > --- a/src/hotspot/share/memory/heap.cpp	Mon May 20 12:12:01 2019 +0200
    > +++ b/src/hotspot/share/memory/heap.cpp	Mon May 20 12:43:31 2019 +0200
    > @@ -629,14 +629,10 @@
    >        size_t segn = seg1 + b->length();
    >        for (size_t i = seg1; i < segn; i++) {
    >          nseg++;
    > -        if (is_segment_unused(seg_map[i])) {
    > -          warning("CodeHeap: unused segment. %d [%d..%d], %s block", (int)i, (int)seg1, (int)segn, b->free()? "free":"used");
    > -        }
    > +        assert(is_segment_unused(seg_map[i]), "CodeHeap: unused segment. %d [%d..%d], %s block", (int)i, (int)seg1, (int)segn, b->free()? "free":"used");
    >        }
    >      }
    > -    if (nseg != _next_segment) {
    > -      warning("CodeHeap: segment count mismatch. found %d, expected %d.", (int)nseg, (int)_next_segment);
    > -    }
    > +    assert(nseg != _next_segment, "CodeHeap: segment count mismatch. found %d, expected %d.", (int)nseg, (int)_next_segment);
    >  
    >      // Verify that the number of free blocks is not out of hand.
    >      static int free_block_threshold = 10000;
    
    Looks good. Please re-run testing before pushing.
    
    Thanks,
    Tobias