PCI / PCIe Error Recovery Product Note, September 2010

1 PCI / PCIe Error Recovery Product Note

The PCI / PCIe Error Recovery feature provides the ability to detect, isolate, and automatically

recover from a PCI / PCIe error, avoiding a system crash. PCI Error Recovery is included with

the HP-UX 11i v3 operating system, and it is enabled by default.

NOTE: PCI / PCIe Error Recovery is not supported on all platforms. To determine if PCI / PCIe

Error Recovery is supported on your system, see the PCI Error Recovery Support Matrix, available

at http://www.hp.com/go/hpux-networking-docs in the PCI Error Recovery section.

With the PCI / PCIe Error Recovery feature enabled, if an error occurs on a PCI bus containing

an I/O card that supports PCI Error Recovery:

• The PCI bus is quarantined to isolate the system from further I/O and prevent the error from

damaging the system.

• The PCI Error Recovery feature will attempt to recover from the error and reinitialize the

bus so I/O can resume.

If an error occurs during the automated error recovery process, the bus and I/O card will remain

quiesced. If the bus contains a card that supports online addition, replacement, or deletion (OL*)

and the card is in a hotpluggable slot, you can use the olrad command (or the attention button)

to manually recover from the error by replacing the card.

For information on OL* operations, see the Interface Card OL* Support Guide, available at: http://

www.hp.com/go/hpux-core-docs

To determine if OL* is supported, see the I/O card documentation or support matrix available at

http://www.hp.com/go/hpux-iocards-docs

If the PCI Error Recovery feature is disabled and an error occurs on a PCI bus, a Machine Check

Abort (MCA) or a High Priority Machine Check (HPMC) will occur, and the system will crash.

NOTE: PCI / PCIe Error Recovery is enabled by default. If you use HP Serviceguard, HP

recommends the PCI Error Recovery feature only be enabled if your storage devices are configured

with multiple paths and you have not disabled HP-UX native multipathing. If PCI Error Recovery

is enabled, but your storage devices are configured with only a single path, HP Serviceguard

might not detect when connectivity is lost. If HP Serviceguard does not detect loss of connectivity,

it does not cause a failover. For instructions on using the pci_eh_enable tunable to disable

PCI Error Recovery, see “Tunable Kernel Parameters”.

If a PCI error occurs on an I/O card very early in the boot process or an OL* online addition

operation, the I/O card will not be claimed and the software state of the I/O card will be marked

as UNUSABLE in the ioscan(1) output. To recover I/O cards that are in the UNUSABLE state, a

system reboot is required.