I've done some testing on a MP CMS configuration to see if I could
ever trip any of the gutsy calls to CMS into running on other than the
IPL processor.  My tests were done on 5.2.  Ymmv.

The executive summary is that if CMS Pipelines is invoked by CMSCALL,
it is inherently thread safe, whatever your virtual machine
configuration, it always runs on the IPL processor.  If you manage to
get to the entry point by branch, the first allocation of storage will
switch it to the IPL processor.  Once on the IPL processor the thread
stays there until it yields control using the MT yield calls, not a
normal CMS wait.

If, however, you are paranoid, two conditions must be satisfied for
any possibility of exposure:  The threads must be in separate
dispatching classes and the virtual machine must have at least two
processors attached.  Even so, it is not clear whether it would be
possible to get to any of the critical regions.

The code that was fixed in 1.1.12/0007 deals with accessing a session
resource from multiple threads; that is invocations that are not
hierarchical.

   j.

On 26 August 2010 20:59, Mike Walter <[email protected]> wrote:
> Is anyone else experiencing ABENDs in PIPES when running multiple separate
> PIPE commands using CMS MultiTasking?

Reply via email to