Withdrawn: 8272807: Permit use of memory concurrent with pretouch

[Kim Barrett]

On Thu, 2 Sep 2021 18:33:56 GMT, Kim Barrett <kbarrett at openjdk.org> wrote:

> Note that this PR replaces the withdrawn https://github.com/openjdk/jdk/pull/5215.
> 
> Please review this change which adds os::touch_memory, which is similar to
> os::pretouch_memory but allows concurrent access to the memory while it is
> being touched.  This is accomplished by using an atomic add of zero as the
> operation for touching the memory, ensuring the virtual location is backed
> by physical memory while not changing any values being read or written by
> other threads.
> 
> While I was there, fixed some other lurking issues in os::pretouch_memory.
> There was a potential overflow in the iteration that has been fixed.  And if
> the range arguments weren't page aligned then the last page might not get
> touched.  The latter was even mentioned in the function's description.  Both
> of those have been fixed by careful alignment and some extra checks.  The
> resulting code is a little more complicated, but more robust and complete.
> 
> Similarly added TouchTask, which is similar to PretouchTask.  Again here,
> there is some cleaning up to avoid potential overflows and such.
> 
> - The chunk size is computed using the page size after possible adjustment
> for UseTransparentHugePages.  We want a chunk size that reflects the actual
> number of touches that will be performed.
> 
> - The chunk claim is now done using a CAS that won't exceed the range end.
> The old atomic-fetch-and-add and check the result, which is performed by
> each worker thread, could lead to overflow.  The old code has a test for
> overflow, but since pointer-arithmetic overflow is UB that's not reliable.
> 
> - The old calculation of num_chunks for parallel touching could also
> potentially overflow.
> 
> Testing:
> mach5 tier1-3

This pull request has been closed without being integrated.

-------------

PR: https://git.openjdk.java.net/jdk/pull/5353