RFR (S) 8182554: Code for os::random() assumes long is 32 bits

coleen.phillimore at oracle.com coleen.phillimore at oracle.com
Wed Jun 28 18:44:03 UTC 2017

Hi Thomas,   Thanks for reviewing this.

On 6/28/17 1:35 PM, Thomas Stüfe wrote:
> Hi Coleen,
> long->int: this makes sense.
> thread safety: So, if I understand this correctly, before it could 
> happen that two threads calling at the same time would return the same 
> value from the random sequence? Your patch seems fine and solves this.

Yes, that would be the case.  It's highly unlikely now since most 
os::random calls are at the beginning and the Symbols call during Symbol 
creation while holding the SymbolTable_lock.


> Kind Regards, Thomas
> On Wed, Jun 28, 2017 at 5:46 PM, <coleen.phillimore at oracle.com 
> <mailto:coleen.phillimore at oracle.com>> wrote:
>     Summary: And make updating the _rand_seed thread safe.
>     See bug for more details.
>     open webrev at
>     http://cr.openjdk.java.net/~coleenp/8182554.01/webrev
>     <http://cr.openjdk.java.net/%7Ecoleenp/8182554.01/webrev>
>     bug link https://bugs.openjdk.java.net/browse/JDK-8182554
>     <https://bugs.openjdk.java.net/browse/JDK-8182554>
>     Tested with JPRT and performance tested addition of cas in
>     os::random call (no regressions).   The only thing that uses
>     os::random more than like once is Symbol creation.
>     Thanks,
>     Coleen

More information about the hotspot-dev mailing list