RFR(s): 8212108: SafepointSynchronizer never ending counter (big enough)

[David Holmes]

Hi Robbin,

On 21/11/2018 11:45 pm, Robbin Ehn wrote:
> Hi all, please review.
> 
> The counter can loop around, it would mainly be an issue in reading
> such flight- recorder recording. I have two upcoming features which
> uses the fact that it is always increasing >
> CR:
> https://bugs.openjdk.java.net/browse/JDK-8212108
> 
> Webrev:
> http://cr.openjdk.java.net/~rehn/8212108/webrev/

I don't understand the logic for safepoint_counter_addr(). Why is 
endian-ness an issue here and why do you need to treat it as 32-bit in 
that case??

I've been trying to see if there could be any issues with word-tearing 
on 32-bit systems ... I can't convince myself you may not see false 
indicators if the count is read at the same time as being updated and 
the 32-bit count is going to rollover from 0xFFFFFFFF to 0x00000000.

I'm not familiar with JFR types - is "ulong" guaranteed to be 64-bit?

> Passes t1-3 and the two updated FR tests.

Does any of that actually run long enough to overflow a 32-bit counter?

Thanks,
David

> 
> Thanks Robbin