Learn the fantastic automatic storage frame (ASF)-operating MI instruction and the implementation details behind it.

From the point of view of a programmer, the ability to change the automatic storage allocation at run time would be fabulous. Automatic storage is thread-private storage; therefore, in a multi-threaded environment you don't need to worry about thread synchronization when you store your program variables in automatic storage. Automatic storage is managed by the system for each individual invocation stack (call stack) entry, so you can simply forget the automatic storage that you have allocated after the end of the invocation. No free operation of the allocated automatic storage is needed.

As an IBM i developer, you can be proud of the fact that you've had the ability to manage automatic storage dynamically (at run time) since System/38. And what makes this ability available for us is the Modify Automatic Storage Allocation (MODASA) MI instruction.

Manage Automatic Storage Dynamically via the MODASA MI Instruction

The Modify Automatic Storage Allocation (MODASA) MI instruction is documented in the IBM i 7.1 Information Center as follows:

Description

The automatic storage frame (ASF) of the current invocation is extended or truncated by the modification size specified by operand 2. A positive value indicates that the frame is to be extended; a negative value indicates that the frame is to be truncated; a zero value does not change the ASF. If operand 1 is not null, it will be treated as follows:

1. ASF extension: receives the address of the first byte of the extension. The ASF extension might not be contiguous with the remainder of the ASF allocation.
2. ASF truncation: operand 1 should be null for truncation. If operand 1 is not null, then addressability to the first byte of the deallocated space is returned. This value should not be used as a space pointer since it locates space that has been deallocated.
3. If a value of zero is specified for operand 2: the value returned is unpredictable.

When the ASF is extended, the extension is aligned on a 16-byte boundary. An extension is not initialized.

When the ASF is extended by this instruction, the type of storage, either single level store or teraspace, always matches the existing ASF.

A scalar value invalid (hex 3203) exception is signaled if the truncation amount would include the storage for all automatic data objects for the current invocation, including the initial allocation.

A space pointer machine object cannot be specified for operand 1.

Note that, as mentioned above, when being invoked from a bound program (ILE program), the modification size operand must be a positive value. If it's not, a scalar value invalid (hex 3203) exception (aka MCH5003) is signaled. In other words, the MODASA MI instruction cannot be used for automatic storage truncation in an ILE program.

Another exception to be aware of is the automatic storage overflow (hex 2C1D) (aka MCH4429) exception. An attempt to allocate automatic storage above the limit of the automatic stack will raise the hex 2C1D exception. The limit on the size of the automatic stack of a thread in an MI process is documented in the Machine Managed Storage Limits subsection of the Storage Limitations section of Machine Interface Architecture Introduction. Note that these limitation values have changed in the past releases of IBM i. For example, the maximum teraspace automatic stack size of the initial thread of an MI process is 64 MB and 1G-8M in VRM540 and VRM710, respectively.

Consideration should also be given to the storage model selected for your program that uses the MODASA MI instruction, since the storage model—single-level store (SLS) or teraspace—of your program determines the storage of the automatic stack—SLS or teraspace—used by your program. As shown above, when the ASF is extended by this instruction, the type of storage, either SLS or teraspace, always matches the existing ASF. You determine the storage model of a program by the Storage model (STGMDL) parameter of the Create command (CRTPGM, CRTBNDRPG, CRTBNDCL, etc.) you used to create the program object.

And finally, a 16 MB SLS segment is allocated for each thread in an MI process as its automatic stack. Thus, program variables allocated in the SLS automatic stack are thread-safe by nature. By contrast, it is possible for one thread to see or change the automatic variables allocated in the teraspace automatic stack of another thread, since the teraspace is visible to all threads within an MI process.

Examples of Using the MODASA MI Instruction

The following code samples are the source of an OMI program (asa01.emi), an ILE RPG program (asa02.rpgle), and an ILE C program (asa03.c). They demonstrate how to extend the ASF (or in other words, allocate automatic storage) of the current invocation at run time. OMI program ASA01 also invokes MODASA to truncate the ASF.

asa01.emi

asa02.rpgle

asa03.c

More examples of using the MODASA MI instruction can be found in the IBM i Information Center (e.g., Example: Common MI programming techniques) or here (e.g., t133.rpgle, which retrieves UPS information via the MATMATR MI instruction).

Implementation Details Behind the MODASA MI Instruction

To understand the implementation details behind the MODASA MI instruction, you should start with the user code generated for a program or procedure that invokes MODASA. The PowerPC instruction stream generated (at VRM540) for the modasa p@, 95 MI instruction (which extends the ASF) in OMI program asa01.emi might look like the following:

Notes:

1. Operand 2 (of value 95) of MODASA is set in General Purpose Register (GPR) 3 (i.e., r3) by the Add Immediate instruction ADDI 3,0,95.
2. Branch to absolute address FFFFFFFFFF 800980 with the Link Register (LR) set to the effective address (EA) of the instruction following instruction BLA 0X3800980. Address FFFFFFFFFF 800980 is within the text section of LIC module #cfgrbla and the instruction at address FFFFFFFFFF 800980, ba 0x2131520, simply branches to absolute address FFFFFFFFFE 131520, which is the start of the LIC module that actually implements the ASF extension aimodasa1. aimodasa1 implements its task, stores the address of the start of the new ASF extension in r3, and returns to the user code via the bclr 20,0 instruction (branch unconditionally to the effective address saved in LR).
3. After returning from aimodasa1, the 8-byte address (stored in r3) of the new ASF extension is tested for validity. If the higher 5 byte segment (Segment ID (SID) portion) of the returned address is hex 0000000000, it is regarded as an invalid address and the higher 8 bytes (the pointer type bytes) of the resulting 16-byte MI pointer are set to hex AF00000000000000 (meaning an invalid MI pointer). Otherwise, the pointer type bytes are set to hex 8000000000000000, which indicates an MI space pointer.
4. Finally the STQ 2,0X50(30) instruction in conjunction with the previous SETTAG instruction store the 8-byte pointer type bytes and the returned address of the new ASF extension stored in r2 and r3 to the resulting 16-byte MI space pointer in storage and make sure tag bits of the MI pointer are set properly.

As for the details of how the LIC module aimodasa1 implements an ASF extension, I decided not to explain the LIC module instruction by instruction. If you really love the system, you can investigate the PowerPC instructions of the LIC module aimodasa1 yourself via tools such as the System Service Tools (SST). Generally speaking, tasks of the LIC module aimodasa1 include rounding the input ASF extension value (stored in r3) to a 32-byte boundary, checking the extension value for validity (e.g., the extension value should be neither zero nor negative, and the extension value should not be greater than the maximum value of a single ASF extension), increasing r31 (which always addresses the end of ASF of the current invocation) by the rounded extension value to extend the ASF, and returning the old value of r31 to its caller as the address of the new ASF extension via r3.

For detailed documentation about the PowerPC instruction set, please refer to the Assembler language reference in the AIX Information Center. The Programming Environments Manual for 64-bit Microprocessors would be a nice reference for the PowerPC architecture. Also note that instructions such as SETTAG (Set Tag), SELRR, and STQ (Store Quad Word) are IBM i-specific PowerPC instructions. I've never found public documentation about these instructions. However, they have been discussed by several AS/400 gurus in the MI400 mailing list.

One additional point to mention is that in the user code generated for an ILE program that invokes the _MODASA system built-in (e.g., asa02.rpgle or asa03.c), the returned ASF address (after extension) is also checked for a teraspace address. If the higher 4 bits of the returned address is equal to hex 9, the 8-byte pointer type bytes of the result MI space pointer is set to hex 4000000000000000, which indicates a teraspace pointer.

The user code generated (at VRM540) for the modasa p@, -95 MI instruction (which truncates the ASF) in OMI program asa01.emi is the following:

The important difference in the user code generated for MI instruction modasa p@, -95 is the LIC module being invoked. The bla 0X3800988 instruction branches to the absolute address FFFFFFFFFF 800988, where a ba 0x22170a0 instruction branches to the target LIC module aimodasa2 at address FFFFFFFFFE 2170A0. The aimodasa2 module truncates the ASF by decreasing r31 with the rounded truncation value stored in r3 and returns the old value of r31 via r3. As mentioned in MODASA's document, "This value should not be used as a space pointer since it locates space that has been deallocated."