From the point of view of the LFS, there are three levels of PFS
support for 64-bit virtual addressing: None, 64-bit supporting, and
64-bit exploiting.
- None:
The PFS has no understanding of 64-bit addresses.
The LFS copies 64-bit addressable user data to an internal 31-bit
addressable buffer before it invokes the PFS for write-type operations,
and vice versa for reads.
- 64-bit supporting:
The PFS can handle 64-bit user virtual
addresses, or it makes use of the OSI services that can. It does not
itself use buffers above the 2-gigabyte line or run in AMODE 64, at
least not to the knowledge of the LFS.
- 64-bit exploiting:
The PFS supports 64-bit user virtual
addresses. It may run in AMODE 64 and have its own data buffers, or
even autodata, above the 2-gigabyte line. Some considerations for
these PFSs are:
- Unless otherwise specified, the OSI service routines expect to
be called in AMODE 31, with a 31-bit parameter list address and 31-bit
parameter addresses. The calling interface may have to be manually
constructed below the 2-gigabyte line.
- The SAF (RACF®) services
do not support 64-bit callers or addresses.
- MVS™ WAIT and POST services
do not support ECBs above the 2-gigabyte line.
Recommendation: A PFS should be at least 64-bit supporting,
in order to avoid the extra LFS data move that is otherwise required
for high user buffers.