Command Retry Considerations

Command retry is a function of many IBM® 3990 and newer storage controllers. When the channel subsystem receives a retry request, it repeats the execution of the CCW, without requiring any additional input/output interrupts. For example, a storage control might initiate a retry procedure to recover from a transient error.

A command retry during the execution of a channel program can cause the following conditions to be detected by the initiating program:

Modifying CCWs: For EXCPVR and EXCP V=R requests, a CCW used in a channel program must not be modified before the CCW operation has been successfully completed. Without the command retry function, a command is fetched only once from storage by a channel. This allowed a program to determine through condition codes or program controlled interruptions (PCI) that a CCW had been fetched and accepted by the channel. The CCW could be modified before execution. With the command retry function, this procedure cannot be repeated because the channel fetches the CCW from storage again on a command retry sequence. In the case of data chaining, the channel retries commands starting with the first CCW in the data chain.
Program Controlled Interrupts (PCI): A CCW containing a PCI flag can cause multiple program-controlled interrupts. This will happen if the PCI-flagged CCW was retried during a command retry procedure and a PCI could be generated each time the CCW is executed.
Residual Count: If a channel program is prematurely terminated during the retry of a command, the residual count in the channel status word (CSW) will not necessarily indicate how much storage was used. For example, if the storage control detects a wrong-length record error condition, an erroneous residual count is stored in the CSW until the command retry is successful. When the retry is successful, the residual in the CSW reflects the correct length of the data transfer.
Command Address: When data chaining with command retry, the CSW might not indicate how many CCWs have been executed at the time of a PCI. For example:

CCW# Channel Program
1 Read, data chain
2 Read, data chain
3 Read, data chain, PCI
4 Read, command chain

In this example, assume that the storage control signals command retry on Read #3 and the processor accepts the PCI after the channel resets the command address to Read #1 because of command retry. The CSW stored for the PCI will contain the command address of Read #1 when the channel has actually progressed to Read #3.
Testing Buffer Contents on Data Read: Any program that tests a buffer to determine when a CCW has been executed and continues to execute based on this data can get incorrect results if an error is detected and the CCW is retried.