IBM System/370 145 Manual page 74

Usually, a page-in is required also to bring in the referenced
instruction or data.
While page-ins usually are initiated as a result of a page fault,
OS/VSl and OS/VS2 provide an Assembler Language macro that can be used
to cause one or more pages to
be
brought into real storage before they
are referenced. such requests are sometimes referred to as page-ahead
requests. A page-ahead is required if, for reasons of proper system
operation, a routine must operate without incurring any page faults.
Use of this macro is restricted because unlimited use of this facility
can defeat the objective of demand paging. DOS/VS does not support a
page-ahead facility.
When a page fault occurs and the control program determines that a
page frame is not currently available for allocation, a choice must be
made as to which allocated page frame will be taken away from the page
to which i t is currently assigned. The rule governing this choice is
called the ~ replacement algorithm. If t.he page replacement
algorithm is designed to choose from among only those page frames
currently allocated to the program that caused the page fault, it is
said to operate locally. If a page frame can be chosen from among all
those available for allocation to all executing programs, the algorithm
is said to operate globally_ DOS/VS, OS/VS1, and OS/VS2 implement a
global page replacement algorithm. VM/310 implements a global page
replacement algorithm and supports a local page replacement algorithm as
an option. The algorithms used attempt to keep the most active pages of
executing programs present in real storage. Hardware is included in
System/310 models with dynamic address translation that indicates
whether or not a page has been referenced or changed. Hence, when a
page frame is required, a page determined by the algorithm to be
relatively inactive is chosen for replacement.
Prior to loading a new page into the page frame chosen, the existing
contents of the page frame must
be
saved if they were modified during
processing. If modification occurred, a page-out operation is required;
otherwise, an exact copy of the page already exists in external page
storage. Code that is not modified during its execution, therefore, has
an additional advantage in a virtual storage environment in that it need
never be paged out once it has been written in external page storage. A
program requiring a page-in is placed in the wait state until the page
it requires has been loaded, during which time CPU control is given to
another ready task, if one is available.
For various reasons, it is necessary to prevent a page-out of certain
pages that are in real storage. One reason is for better operation of
the system. This reason applies to all (in DOS/VS) or a portion (in
OS/Vsl and OS/VS2) of the control program, certain routines that operate
with the CPU in a disabled state (masked for I/O and external
interruptions), most system tables, and most system control blocks.
Integrity of system operation is another reason. Pages associated with
certain types of operations must not
be
paged out while the operation is
in progress, in order for the operation to proceed correctly. For
example, pages that contain I/O buffer areas must remain in real storage
While the buffers are being referenced during an I/O operation, after
which a page-out can take place, if necessary. Another reason is the
existence of time dependency. A page should not be written out if the
program to which the page belongs must complete a logical operation that
requires the page in less time than it takes to perform a page-in.
Programs that handle I/O device testing operations, such as online tests
(OLT's), can have such a time dependency.
A
page that is identified as one that cannot be paged out (or, that
is nonpageable) is called a
fixed~.
IBM-supplied operating systems
support both long-·term fixing and short-term fixing. Pages that should
never be paged out when they are present in real storage are marked
64 A Guide to the IBM System/310 Model 145