Code Review Request: 6850846: G1: extend G1 marking verification (S)

Tony Printezis Antonios.Printezis at sun.com
Fri Jun 12 20:37:12 UTC 2009


Many thanks for John Cuthbertson for spotting that I missed to make the 
VerifyObjsInRegionClosure closure "use_prev_marking"-aware. I updated 
the webrev:

http://cr.openjdk.java.net/~tonyp/6850846/webrev.1/

Tony

Tony Printezis wrote:
> Small change to extend marking verification in G1:
>
> http://cr.openjdk.java.net/~tonyp/6850846/webrev.0/
>
> Here's the blurb from the CR:
>
> G1 uses two different versions of the marking information:
>
> - "prev" is the last one completed (which is guaranteed to be 
> consistent and complete, i.e., no live object should be pointing to a 
> dead object, but maybe no up-to-date) and
> - "next" is the marking information that is being computed next (and 
> which might be incomplete, as we might be half-way through completing it)
>
> Heap verification in G1 also checks the consistency of the "prev" 
> marking information (given that it's the only marking information that 
> we can trust to be complete). However, we would also like to check the 
> consistency of the "next" marking information at the only place where 
> we can, which is the end of the remark phase.
>
> This CR proposes to change the G1 verification code to either use the 
> "prev" or "next" marking info, depending on what the caller needs.
>
> Additionally, it adds verification of the "next" marking info at the 
> end of the remark (conditional on the VerifyDuringGC flag) and removes 
> the #ifndef PRODUCT from the verification at the end of cleanup (it's 
> good to be able to do heap / marking verification in a product build).
>

-- 
---------------------------------------------------------------------
| Tony Printezis, Staff Engineer   | Sun Microsystems Inc.          |
|                                  | MS UBUR02-311                  |
| e-mail: tony.printezis at sun.com   | 35 Network Drive               |
| office: +1 781 442 0998 (x20998) | Burlington, MA 01803-2756, USA |
---------------------------------------------------------------------
e-mail client: Thunderbird (Linux)





More information about the hotspot-gc-dev mailing list