RFR (S): 8026775: nsk/jvmti/RedefineClasses/StressRedefine crashes due to EXCEPTION_ACCESS_VIOLATION

Mikael Vidstedt mikael.vidstedt at oracle.com
Tue Nov 5 11:51:50 PST 2013

John - that is a very good suggestion indeed, I took the liberty to copy 
your exact wording.


I also added the closest I have gotten to a reproducer. Unfortunately it 
is not completely reliable, in that it does not always reproduce the 
bug, but I got it to the point where from empirical results it tickles 
the bug some 80% of the cases (at least on my machine).

Please review the updated comments and the added test!


On 2013-11-04 11:58, John Rose wrote:
> +   // Skip the first one because that was already touched in the above
> +   // loop - the post decrement of temp means it's now a page below the
> +   // last touch
> The comments should say 'tmp' not 'temp'.  Also, the phrases 'first one' and 'it's now a page below' are hard to understand, and (to me) slightly misleading.
> Suggest:
> +   // At this point, (tmp-0) is the last address touched, so don't touch it again.
> +   // (It was touched as (tmp-pagesize) but then tmp was post-decremented.)
> +   // Skip this address by starting at i=1, and touch a few more pages below.
> +   // N.B.  It is important to touch all the down to and including i=StackShadowPages.
> — John

More information about the hotspot-compiler-dev mailing list