RFR(xs): 8079449: Improve os::attempt_reserve_memory_at() fallback coding on Linux, BSD, Solaris

[Coleen Phillimore]

Hi,

I think this fallback code may be used for CDS (class data sharing) to 
get the archive address.  I also think Solaris is more likely to not 
give a requested address (solaris 10 doesn't at all).

Since we haven't observed a performance problem with this code on 
solaris, I don't think it should be fixed, unless you have such a test case.

Did you test this on 32 bit linux with CDS?  (java -Xshare:dump, java 
-Xshare:on -version) I think you might see this code in action.

The patch looks good though, and I believe it fixes an NMT double 
tracking bug (which we also haven't seen) with 
os::pd_attempt_reserve_memory_at calling os::reserve_memory.   It's 
worth fixing with your patch for that reason.   Can you also fix this 
call for solaris?

Thanks,
Coleen

On 5/6/15, 10:07 AM, Thomas Stüfe wrote:
> Hi all,
>
> please take a look at the following patch:
>
> http://cr.openjdk.java.net/~stuefe/webrevs/8079449/webrev.00/webrev/
> https://bugs.openjdk.java.net/browse/JDK-8079449
>
> More details are in my original mail:
> http://mail.openjdk.java.net/pipermail/hotspot-runtime-dev/2015-April/014600.html
>
>
> This patch adds a fix to fallback coding used in
> os::attempt_reserve_memory_at() on Linux and BSD.
>
> If this patch meets with approval, I will port the fix to Solaris, but
> Solaris implementation was just different enough for this to be not
> straightforward. Therefore I'd like to get some feedback for this issue
> first.
>
> Note - as I wrote in my original mail - that I have my doubts that this
> fallback coding is even needed at all - in tests I never saw it doing
> anything useful. I'd actually rather just remove it than fix it. What do
> you think?`
>
> Kind Regards, Thomas