336
Unconsumed Uncorrectable Error Handling
RAS features are defined to inform software of uncorrectable errors that have been
discovered on the persistent memory media but have not yet been consumed by
software. The goal of this feature is to allow the operating system to opportunistically
offline or clear pages with known uncorrectable errors before they can be used by an
application. If the address of the uncorrectable error is already in use by an application,
the operating system may also choose to notify it of the unconsumed uncorrectable error
or wait until the application consumes the error. The operating system may choose to
wait on the chance that the application never tries to access the affected page and later
return the page to the operating system for recycling. At this time, the operating system
would clear or offline the uncorrectable error.
Figure 17-1. Linux consumed uncorrectable error-handling sequence
Chapter 17 reliability, availability, and ServiCeability (raS)