RFR: 8207851 JEP Draft: Support ByteBuffer mapped over non-volatile memory

[Andrew Dinn]

Round 2
-------
I have updated the JEP and uploaded a revised webrev in the light of
existing feedback

  JEP JIRA: https://bugs.openjdk.java.net/browse/JDK-8207851

  Formatted JEP: http://openjdk.java.net/jeps/8207851

  New webrev: http://cr.openjdk.java.net/~adinn/pmem/webrev.02/

I would welcome any further comments (I guess I'd be happy to push this
as is but I find it hard to believe it does not require more work.
Someone must have something to add :-)

List of Changes
---------------

I have followed Alan's request:

 - extending FileChannel.MapMode to add two new modes
READ_ONLY_PERISTENT and READ_WRITE_PERSISTENT

  - retaining the single method FileChannel.map to handle the new modes
(i.e. dumping the idea of a separate map_peristent API)

  - FileChannelImpl.map now handles the relevant XXX_PERSISTENT cases
(rather than delegating to an internal method)

I have followed Florian's advice:

  corrected spelling and format of the JEP, incuding made a clearer
separation of description of API changes from implementation details

  modified the conditional compilation conditions and associated
comments in FileChannelImpl.c to define MAP_SYNC and MAP_PRIVATE when
the OS does not do so

I have also brought the implementation of the pre-sync and post-sync
operations in line with the libpmem code:

  On x86

  pre-sync is a no-op
  post-sync is an no-op if clflush is used for writeback
  post-sync is an sfence if clflushopt or clwb is used for writeback

  On AArch64

  pre-sync is a no-op
  post-sync is a dmb(ISH)

Finally, I have updated the JEP text to reflect all the above changes.


regards,


Andrew Dinn
-----------
Senior Principal Software Engineer
Red Hat UK Ltd
Registered in England and Wales under Company Registration No. 03798903
Directors: Michael Cunningham, Michael ("Mike") O'Neill, Eric Shander