Thresholding Hardware Errors

Abstract

A system is provided to protect against ill-behaved microcode by balancing between an actual occurrence of a hardware problem and a microcode bug setting a flag appearing as a hardware problem. In this method, the error recovery is performed only on a single piece of hardware and no further error recovery action is taken on other pieces of similar hardware. The approach addresses the problem by treating a hit on one card as a hardware problem, but as a bug on subsequent cards. The invention keeps track of whether or not the same event has occurred on the same type of hardware so not to take action on more than one instance of the hardware. Subsequent hits on another instance of the hardware will not trigger the hardware related recovery on additional hardware.

1 . A system of thresholding hardware errors in a storage module, said system comprising:
 multiple host adapter cards comprising storage control units; and   a first flag;   wherein a string of bits on a longitudinal track is received;   a parity bit from said string of bits on said longitudinal track is generated, to form a longitudinal-redundancy check;   said multiple host adapter cards are configured in a cluster;   in case of a problem with said storage system:   said first flag is raised in a header for said longitudinal-redundancy check;   a first card corresponding to said first flag among said multiple host adapter cards for a first job is fenced off;   said first job is reissued to a second card among said multiple host adapter cards; and   all of said multiple host adapter cards, other than said first card with a marker, are marked, so that all of said multiple host adapter cards other than said first card are not allowed to be fenced off;   in case of a new adapter card is put in service, an initial microcode load operation is started, and it is checked to see if said marker is set; and   an error recovery, after a predetermined number of error hits occurs, is prevented, based on longitudinal-redundancy checks within a predetermined period of time.