[11u] RFR (S) 8218879: Keep track of memory accesses originated from Unsafe

[Severin Gehwolf]

On Fri, 2019-08-16 at 14:41 +0200, Aleksey Shipilev wrote:
> On 8/7/19 12:16 PM, Aleksey Shipilev wrote:
> > > 11u webrev:
> > >   https://cr.openjdk.java.net/~shade/8218879/webrev.11u.01/
> 
> Ha, no wait, there is the odd conflict here:
> 
>   37 // Use the C1_UNSAFE_ACCESS decorator to mark unsafe accesses.
>   38 const DecoratorSet C1_UNSAFE_ACCESS  = DECORATOR_LAST << 3;
>   39 // The C1_WRITE_ACCESS decorator is used to mark writing accesses.
>   40 const DecoratorSet C1_WRITE_ACCESS   = DECORATOR_LAST << 3;
> 
> Notice C1_UNSAFE_ACCESS and C1_WRITE_ACCESS would be mapped to the same constant! This does not
> happen in jdk/jdk, because JDK-8209801 have removed (READ|WRITE) accessors. JDK-8209801 does not
> look backportable, because it drags in significant part of Access API.

Nice catch!

> So, the alternative is to push C1_UNSAFE_ACCESS to other index:
>   http://cr.openjdk.java.net/~shade/8218879/webrev.11u.02/

  41 // Use the C1_UNSAFE_ACCESS decorator to mark unsafe accesses.
  42 const DecoratorSet C1_UNSAFE_ACCESS  = DECORATOR_LAST << 5;

Consider adding a comment saying that "Should JDK-8209801 ever get
backported this will need adjustment".

Up to you. I don't need to see another webrev for this.

Thanks,
Severin