RFR(S): 8010190: os_windows missing winsock initialization

David Holmes david.holmes at oracle.com
Wed Mar 20 03:10:01 PDT 2013


For the email record the initialization problems here were already 
identified in 7034299.

David

On 19/03/2013 10:51 AM, David Holmes wrote:
> Hi Nils,
>
> You fix is fine as far as it goes but ...
>
> ... basically if we make changes as to how this code gets used we will
> hit new failures until slowly we add an initialization check to all of
> the methods. This is horribly ad-hoc. Worse there is a race condition (a
> couple actually) in the initialization code which we are more likely to
> hit now that we have more paths to this code.
>
> Seems to me that the real fix for this code might be to ensure sockInit
> is called during VM startup instead of lazily? Otherwise, as I said,
> there's a bunch of race conditions that need fixing anyway.
>
> Cheers,
> David
>
> On 18/03/2013 9:57 PM, Nils Eliasson wrote:
>> Hi all,
>>
>> A small change that adds lazy initalization of winsock (os::initsock())
>> on some paths that needs it. WinSock has only a few uses (streaming
>> graphs to IdealGraphVisualizers a.f.a.i.k.) so the paths are not hot,
>> and the initialization will not add any overhead to regular VM use.
>>
>> Webrev:
>> http://cr.openjdk.java.net/~neliasso/8010190/webrev.01/
>>
>> Bug:
>> http://bugs.sun.com/view_bug.do?bug_id=80010190
>>
>> Thanks,
>> Nils Eliasson


More information about the hotspot-runtime-dev mailing list