Interface Card OL* Support Guide

Chapter 1
PCI Card OL* Overview and Concepts
Planning and Preparation
16
Planning and Preparation
For the most part, these tools prevent you from performing OL* procedures that would adversely affect other
areas of the server. This section provides you with important information that can help minimize errors or
problems when performing OL* procedures.
Card Compatibility
Online Addition
When adding an interface card online, the first issue that must be resolved is whether the new card is
compatible with the system. Each OL* capable PCI slot provides a set amount of power. The card to be added
cannot require more power than is available. Some systems may have only one slot per bus with sufficient
power.
The card must also operate at the slot’s bus frequency. See “Frequency Mismatch” on page 21.
When a card is added to the system, the appropriate driver for that card must be configured in the kernel
before beginning the operation. In most cases, the added card should be the same type as a card already in the
system, and the driver will be in the kernel. Kernel Configuration is accessible through kcweb. If the required
driver is not present, a reboot will be required to load the driver. The card could be added while the system is
down, or added online after rebooting. A reboot may not be needed if the OS version supports Dynamic
Loadable Kernel Modules (DLKM).
If the necessary driver is not present you can load it manually.
If the driver is static and not configured in the kernel, then the card cannot be Online Added. The card
could be physically inserted online, but no driver would claim it.
Online Replacement
When replacing an interface card online, the replacement card must be identical. This is referred to as
like-for-like replacement and should be adhered to because using a similar card, but not identical, may cause
unpredictable results. For example, a newer version of the target card which is identical in terms of hardware
may contain an updated firmware version that could potentially conflict with the current driver. If a new card
is not acceptable, pdweb will report that the card cannot be resumed, the Attention Button method will result
in the slot not being powered up, and olrad will return an error.
During the replacement process, the driver instance for each port on the target card runs in a suspended
state. I/O to the ports are either queued or failed while the drivers are suspended. When the replacement
card is brought online, the driver instances resume normal operation. Each driver instance must be
capable of resuming and controlling the corresponding port on the replacement card.
The PCI specification allows a single physical card to contain more than one port. Attempting to replace a
card with another card that has more ports than the original could result in the additional port(s) being
claimed by other drivers if an ioscan occurs while the slot power is on. The ports would be claimed properly,
and recovering from that condition would require a system reboot.
Critical Resources
Replacing a card that is still operating can have extensive ramifications. Since power to the slot must be off
when the old card is removed and the new card is inserted, the effects of shutting down the card’s functions
must be considered.