Review request: 8013160: NPG: Remove unnecessary mark stack draining after CodeCache::do_unloading

[Coleen Phillimore]

Stefan,

Is there a common place for this code?  It's 4 copies of the similar 
code, which takes me a while to find when I set out looking for it.  In 
the future, I think StringTable::unlink() could be moved after 
CLDG::purge() which I thought should be around here too (but it's 
not).   The only reason StringTable::unlink() used to be here was 
because it used to have Oops in it.   Can you unite this duplicated code?

Thanks,
Coleen

On 4/24/2013 4:18 PM, Stefan Karlsson wrote:
> http://cr.openjdk.java.net/~stefank/8013160/webrev.00/
>
> There's some remnants from the PermGen that's confusing in the class 
> unloading code. This patch removes the unnecessary code, cleans up 
> some comments and unifies the code structure over different GCs.
>
> From the Bug report:
> ---
> The PermGen Removal changed CompiledICHolders from being Java objects 
> to C++ objects.
>
> CodeCache::do_unloading(...) used to take a keep_alive closure that 
> was applied to CompiledICHolders deep down in nmethod::do_unloading. 
> Since CompileICHolders don't move anymore, and we have other ways to 
> keep them alive, we don't use the keep_alive closures in these 
> functions anymore.
>
> Because of this, CodeCache::do_unloading will not populate the marking 
> stacks, and we can remove the call to drain the stacks after the calls 
> to CodeCache::do_unloading.
>
> This also has the neat effect that we now can verify that the marking 
> has really completed before we start unloading classes, which makes 
> the code easier to maintain/understand.
> ---
>
> thanks,
> StefanK

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/hotspot-gc-dev/attachments/20130424/92ffbfdb/attachment.htm>