Updated: 11 December 1998

OpenVMS Programming Concepts Manual

This chapter describes the use of memory management system services and run-time routines on Alpha systems. Although the operating system's memory management concepts are much the same on VAX systems and Alpha systems, details of the memory management system are different. These details may be critical to certain uses of the operating system's memory management system services and routines on an Alpha system. This chapter highlights those differences by using the Alpha icon.

This chapter contains the following sections:

Section 22.1 describes the page sizes of Alpha systems.

Section 22.2 describes the layout of virtual address space.

Section 22.3 describes the three levels of the operating system's memory allocation routines.

Section 22.4 discusses how to use system services to add virtual address space, adjust working sets, control process swapping, and create and manage sections.

22.1 Virtual Page Sizes (Alpha Only)

On Alpha systems, in order to facilitate memory protection and mapping, the virtual address space is subdivided into segments of 8 KB, 16 KB, 32 KB, or 64 KB sizes called CPU-specific pages. On VAX systems, the page sizes are 512 bytes.

Wherever possible, the Alpha system's versions of the system services and run-time library routines that manipulate memory attempt to preserve compatibility with the VAX system's services and routines. The Alpha system's versions of the routines that accept page count values as arguments still interpret these arguments in 512-byte quantities, which are called pagelets to distinguish them from CPU-specific page sizes. The routines convert pagelet values into CPU-specific pages. The routines that return page count values convert from CPU-specific pages to pagelets, so that return values expected by applications are still measured in the same 512-byte units.

This difference in page size does not affect memory allocation using higher-level routines, such as run-time library routines that manipulate virtual memory zones or language-specific memory allocation routines such as the malloc and free routines in C.

To determine system page size, you make a call to the SYS$GETSYI system service, specifying the SYI$_PAGE_SIZE item code. See the description of SYS$GETSYI and SYI$_PAGE_SIZE in the OpenVMS System Services Reference Manual for details.

22.2 Virtual Address Space

The Alpha system defines the same virtual address space layout as the VAX system. The Alpha system virtual address space allows for growth of the P0 and P1 regions in the same directions as on VAX systems.

The initial size of a process's virtual address space depends on the size of the image being executed. The virtual address space of an executing program consists of the following three regions:

• Process program region (P0)
  The process program region is also referred to as P0 space. P0 space contains the instructions and data for the current image.
  Your program can dynamically allocate storage in the process program region by calling run-time library (RTL) dynamic memory allocation routines or the operating system's system services.
• Process control region (P1)
  The process control region is also referred to as P1 space. P1 space contains system control information and the user-mode process stack. The user mode stack expands as necessary toward the lower-addressed end of P1 space.
• Common system region (S0)
  The common system region is also referred to as S0 space, or system space. S0 space contains the operating system. Your program cannot allocate or free memory within the common system region from the user access mode.

The operating system's memory management routines map and control the relationship between physical memory and the virtual address space of a process. These activities are, for the most part, transparent to you and your programs. In some cases, however, you can make a program more efficient by explicitly controlling its virtual memory usage.

The maximum size to which a process can increase its address space is controlled by the system parameter VIRTUALPAGECNT.

Using memory management system services, a process can add a specified number of pages to the end of either the program region or the control region. Adding pages to the program region provides the process with additional space for image execution, for example, for the dynamic creation of tables or data areas. Adding pages to the control region increases the size of the user stack. As new pages are referenced, the stack is automatically expanded (see Figure 22-1). (By using the STACK= option in a linker options file, you can also expand the user stack when you link the image.)

Figure 22-1 illustrates the layout of a process's virtual memory. The initial size of a process's virtual address space depends on the size of the image being executed.

Figure 22-1 Layout of Alpha Process Virtual Address Space

Sophisticated software systems must often create and manage complex data structures. In these systems, the size and number of elements are not always known in advance. You can tailor the memory allocation for these elements by using dynamic memory allocation. By managing the memory allocation, you can avoid allocating fixed tables that may be too large or too small for your program. Managing memory directly can improve program efficiency. By allowing you to allocate specific amounts of memory, the operating system provides a hierarchy of routines and services for memory management. Memory allocation and deallocation routines allow you to allocate and free storage within the virtual address space available to your process.

There are three levels of memory allocation routines:

1. Memory management system services
   The memory management system services comprise the lowest level of memory allocation routines. These services include, but are not limited to, the following:
   • SYS$EXPREG (Expand Region)
   • SYS$CRETVA (Create Virtual Address Space)
   • SYS$DELTVA (Delete Virtual Address Space)
   • SYS$CRMPSC (Create and Map Section)
   • SYS$MGBLSC (Map Global Section)
   • SYS$DGBLSC (Delete Global Section)
   
   For most cases in which a system service is used for memory allocation, the Expand Region (SYS$EXPREG) system service is used to create pages of virtual memory.
   Because system services provide more control over allocation procedures than RTL routines, you must manage the allocation precisely. System services provide extensive control over address space allocation by allowing you to do the following types of tasks:
   • Add or delete virtual address space to the process's program region (P0) or control region (P1)
   • Add or delete virtual address space at a specific range of addresses
   • Increase or decrease the number of pages in a program's working set
   • Lock or delete pages of a program's working set in memory
   • Lock the entire program's working set in memory (by disabling process swapping)
   • Define disk files containing data or shareable images and map the files into the virtual address space of a process
2. RTL page management routines
   The RTL routines exist for creating, deleting, and accessing information about virtual address space. You can either allocate a specified number of contiguous pages or create a zone of virtual address space. A zone is a logical unit of the memory pool or subheap that you can control as an independent area. It can be any size required by your program. Refer to Chapter 23 for more information about zones.
   The RTL page management routines LIB$GET_VM_PAGE and LIB$FREE_VM_PAGE provide a convenient mechanism for allocating and freeing pages of memory.
   These routines maintain a processwide pool of free pages. If unallocated pages are not available when LIB$GET_VM_PAGE is called, it calls SYS$EXPREG to create the required pages in the program region (P0 space).
3. RTL heap management routines
   The RTL heap management routines LIB$GET_VM and LIB$FREE_VM provide a mechanism for allocating and freeing blocks of memory of arbitrary size.
   The following are heap management routines based on the concept of zones:
   • LIB$CREATE_VM_ZONE
   • LIB$CREATE_USER_VM_ZONE
   • LIB$DELETE_VM_ZONE
   • LIB$FIND_VM_ZONE
   • LIB$RESET_VM_ZONE
   • LIB$SHOW_VM_ZONE
   • LIB$VERIFY_VM_ZONE
   
   If an unallocated block is not available to satisfy a call to LIB$GET_VM, LIB$GET_VM calls LIB$GET_VM_PAGE to allocate additional pages.

Modular application programs can call routines in any or all levels of the hierarchy, depending on the kinds of services the application program needs. You must observe the following basic rule when using multiple levels of the hierarchy:

• Memory that is allocated by an allocation routine at one level of the hierarchy must be freed by calling a deallocation routine at the same level of the hierarchy. For example, if you allocated a page of memory by calling LIB$GET_VM_PAGE, you can free it only by calling LIB$FREE_VM_PAGE.

Figure 22-2 shows the three levels of memory allocation routines.

Figure 22-2 Hierarchy of Alpha Memory Management Routines

For information about using memory management RTLs, see Chapter 23.

22.4 Using System Services for Memory Allocation

This section describes how to use system services to perform the following tasks:

• Increase and decrease virtual address space
• Input and return address arrays
• Control page ownership and protection
• Control working set paging
• Control process swapping

The system services allow you to add address space anywhere within the process's program region (P0) or control region (P1). To add address space at the end of P0 or P1, use the Expand Program/Control Region (SYS$EXPREG) system service. SYS$EXPREG optionally returns the range of virtual addresses for the new pages. To add address space in other portions of P0 or P1, use SYS$CRETVA.

The format for SYS$EXPREG is as follows:

SYS$EXPREG (pagcnt ,[retadr] ,[acmode] ,[region])

Specifying the Number of Pages

On Alpha systems, use the pagcnt argument to specify the number of pagelets to add to the end of the region. The Alpha system rounds the specified pagelet value to the next integral number of Alpha pages for the system where it is executing. To check the exact boundaries of the memory allocated by the system, specify the optional retadr argument. The retadr argument contains the start address and the end address of the memory allocated by the system service.

Specifying the Access Mode

Use the acmode argument to specify the access to be assigned to the newly created pages.

Specifying the Region

Use the region argument to specify whether to add the pages to the end of the P0 or P1 region.

To deallocate pages allocated with SYS$EXPREG and SYS$CRETVA, use SYS$DELTVA.

For Alpha systems, the following example illustrates the addition of 4 pagelets to the program region of a process by writing a call to the SYS$EXPREG system service.

#include <stdio.h> #include <ssdef.h> main() { unsigned int status, retadr[2],pagcnt=4, region=0; /* Add 4 pages to P0 space */ status = SYS$EXPREG( pagcnt, &retadr, 0, region); if (( status & 1) != 1) LIB$SIGNAL( status ); else printf("Starting address: %d Ending address: %d\n", retadr[0],retadr[1]; }

The value 0 is passed in the region argument to specify that the pages are to be added to the program region. To add the same number of pages to the control region, you would specify REGION=1.

Note that the region argument to the SYS$EXPREG service is optional; if it is not specified, the pages are added to or deleted from the program region by default.

On Alpha systems, the SYS$EXPREG service can add pagelets only in the direction of growth of a particular region. When you need to add pages to the middle of these regions, you can use the Create Virtual Address Space (SYS$CRETVA) system service. Likewise, when you need to delete pages created by either SYS$EXPREG or SYS$CRETVA, you can use the Delete Virtual Address Space (SYS$DELTVA) system service. For example, if you have used the SYS$EXPREG service twice to add pages to the program region and want to delete the first range of pages but not the second, you could use the SYS$DELTVA system service, as shown in the following example:

#include <stdio.h> #include <ssdef.h> struct { unsigned int lower, upper; }retadr1, retadr2, retadr3; main() { unsigned int status, pagcnt=4, region=0; /* Add 4 pages to P0 space */ status = SYS$EXPREG( pagcnt, &retadr1, 0, region); if (( status & 1) != 1) LIB$SIGNAL( status ); else printf("Starting address: %d ending address: %d\n", retadr1.lower,retadr1.upper); /* Add 3 more pages to P0 space */ pagcnt = 3; status = SYS$EXPREG( pagcnt, &retadr2, 0, region); if (( status & 1) != 1) LIB$SIGNAL( status ); else printf("Starting address: %d ending address: %d\n", retadr2.lower,retadr2.upper); /* Delete original allocation */ status = SYS$DELTVA( &retadr1, &retadr3, 0 ); if (( status & 1) != 1) LIB$SIGNAL( status ); else printf("Starting address: %d ending address: %d\n", retadr1.lower,retadr1.upper); }

In this example, the first call to SYS$EXPREG rounds up the requested pagelet count to an integral number of CPU-specific pages and adds that number of pages to the program region; the virtual addresses of the created pages are returned in the 2-longword array at retadr1. The second request converts the pagelet count to pages, adds them to the program region, and returns the addresses at retadr2. The call to SYS$DELTVA deletes the area created by the first SYS$EXPREG call.

Section 22.4.3 mentions some other possible risks in using SYS$CRETVA for allocating memory.

22.4.2 Input Address Arrays and Return Address Arrays

When the SYS$EXPREG system service adds pages to a region, it adds them in the normal direction of growth for the region. The return address array, if requested, indicates the order in which the pages were added. For example:

• If the program region is expanded, the starting virtual address is smaller than the ending virtual address.
• If the control region is expanded, the starting virtual address is larger than the ending virtual address.

The addresses returned indicate the first byte in the first page that was added or deleted and the last byte in the last page that was added or deleted, respectively.

When input address arrays are specified for the Create and Delete Virtual Address Space (SYS$CRETVA and SYS$DELTVA, respectively) system services, these services add or delete pages beginning with the address specified in the first longword and ending with the address specified in the second longword.

On Alpha systems, the order in which the pages are added or deleted does not have to be in the normal direction of growth for the region. Moreover, because these services add or delete only whole pages, they ignore the low-order bits of the specified virtual address (the low-order bits contain the byte offset within the page). Table 22-1 shows the page size and byte offset.

Table 22-1 Page and Byte Offset Within Pages on Alpha Systems

Page Size (Bytes)	Byte Within Page (Bits)
8K	13
16K	14
32K	15
64K	16

Table 22-2 shows some sample virtual addresses in hexadecimal that may be specified as input to SYS$CRETVA or SYS$DELTVA and shows the return address arrays if all pages are successfully added or deleted. Table 22-2 assumes a page size of 8 KB = 2000 hex.

Table 22-2 Sample Virtual Address Arrays on Alpha Systems

Input Array			Output Array
Start	End	Region	Start	End	Number of Pages
1010	1670	P0	0	1FFF	1
2450	2451	P0	2000	3FFF	1
4200	A500	P0	4000	BFFF	5
9450	9450	P0	8000	9FFF	1
7FFEC010	7FFEC010	P1	7FFEDFFF	7FFEC000	1
7FFEC010	7FFEBCA0	P1	7FFEDFFF	7FFEA000	2

For SYS$CRETVA and SYS$DELTVA, note that if the input virtual addresses are the same, as in the fourth and fifth items in Table 22-2, a single page is added or deleted. The return address array indicates that the page was added or deleted in the normal direction of growth for the region.

Note that for SYS$CRMPSC and SYS$MGBLSC, which are discussed in Section 22.4.7, the sample virtual address arrays in Table 22-2 do not apply. The reason is that the lower address value has to be an even multiple of the machine page size; that is, it must be rounded down to an even multiple page size. In addition, the higher address value must be one less than the even multiple page size, representing the last byte on the last page. That is, it must be rounded up to an even multiple page size, minus 1.

The procedure for determining start and end virtual addresses is as follows:

1. Get the page size in bytes.
2. Subtract 1 to get the byte-with-page mask.
3. Mask the low bits of lower virtual address, which is a round-down operation to round it to the next lower page boundary.
4. Perform a logical OR operation on the higher virtual address, which is a round-up operation to round it to the highest address in the last page.

On Alpha systems, if you reallocate memory that is already in its virtual address space by using the SYS$CRETVA system service, you may need to modify the values of the following arguments to SYS$CRETVA:

• If your application explicitly rounds the lower address specified in the inadr argument to be a multiple of 512 in order to align on a page boundary, you need to modify the address. The Alpha system's version of the SYS$CRETVA system service rounds down the start address to a CPU-specific page boundary, which will vary with different implementations. It also rounds up the end address to the last byte in a CPU-specific page boundary.
• The size of the reallocation, specified by the address range in the inadr argument, may be larger on an Alpha system than on a VAX system because the request is rounded up to CPU-specific pages. This can cause the unintended destruction of neighboring data, which may also occur with single-page allocations. (When the start address and the end address specified in the inadr argument match, a single page is allocated.)

To determine whether you must modify the address as specified in inadr, specify the optional retadr argument to determine the exact boundaries of the memory allocated by the call to SYS$CRETVA.

22.4.4 Page Ownership and Protection

Each page in the virtual address space of a process is owned by the access mode that created the page. For example, pages in the program region initially provided for the execution of an image are owned by user mode. Pages that the image creates dynamically are also owned by user mode. Pages in the control region, except for the pages containing the user stack, are normally owned by more privileged access modes.

Only the owner access mode or a more privileged access mode can delete the page or otherwise affect it. The owner of a page can also indicate, by means of a protection code, the type of access that each access mode will be allowed.

The Set Protection on Pages (SYS$SETPRT) system service changes the protection assigned to a page or group of pages. The protection is expressed as a code that indicates the specific type of access (none, read-only, read/write) for each of the four access modes (kernel, executive, supervisor, user). Only the owner access mode or a more privileged access mode can change the protection for a page.

When an image attempts to access a page that is protected against the access attempted, a hardware exception called an access violation occurs. When an image calls a memory management system service, the service probes the pages to be used to determine whether an access violation would occur if the image attempts to read or write one of the pages. If an access violation occurs, the service exits with the status code SS$_ACCVIO.

Because the memory management services add, delete, or modify a single page at a time, one or more pages can be successfully changed before an access violation is detected. If the retadr argument is specified in the service call, the service returns the addresses of pages changed (added, deleted, or modified) before the error. If no pages are affected, that is, if an access violation occurs on the first page specified, the service returns a -1 in both longwords of the return address array.

If the retadr argument is not specified, no information is returned.

Copyright © Compaq Computer Corporation 1998. All rights reserved. Legal

5841PRO_058.HTML