Usage Considerations

Although there are a number of validity checks to help avoid misconfiguration of partitions and segments, ultimately you are responsible for ensuring correctness. This section will outline some of the best practices and limitations that should be taken into consideration when configuring a system

• All server and non-server processes associated with the same partition should use the same hardware module. Exceptions and/or limitations in this regard will be documented in the "use" information of the applicable hardware module.
• There should be a separate hardware configuration file for each partition that will be used.
• A different hardware configuration file can be used for different pci-server instances associated with the same partition, however, the following guidelines must still be observed.
• A HW module specific identifier should only appear in one hardware configuration file.
• The server ID, if not a wild card (*), must be unique for each entry in a hardware configuration file.
• Any wild carded server ID will be assigned to the first pci-server instance to parse and then register the entries in the "pci_hw" shared memory object.
• A server ID can be re-used for pci-server instances associated with different partitions.
• When multiple pci-server instances associated with the same partition are started, valid records within the [segments] section will be registered in the "pci_hw" shared memory object by only one of the pci-server instances.
• Once the records in the [segments] section of a hardware configuration file are registered in the "pci_hw" shared memory object, they will persist until the "pci_hw" is explicitly removed and the pci-server instances restarted.
  Note: While experimenting with changes on how to segment a partition, it is strongly advised that you stop all running pci-server instances and remove all of the shared memory objects associated with the partition (i.e., rm /dev/shmem/pci/<partition_name>/*) before restarting the pci-server instances with the changes. This will help avoid unexpected behavior.
• If possible, all of the pci-server instances associated with the same partition should be started before any clients are started so that all hardware resources are initialized before any clients attempt to make use of them. Any exceptions and/or limitations in this regard will be documented in the "use" information of the applicable hardware module.
• The segment first bus and segment last bus fields of a record in the [segments] section of a hardware configuration file must be unique across all records in all hardware configuration files being used by pci-server instances associated with the same partition. These records can overlap or be reused for pci-server instances associated with different partitions because the BDF space is independent for each partition.