Hotplugging support in the Linux kernel

July 26, 2000
J. Corbet

Due to a lunch meeting which went a bit overlong, your correspondent missed much of Greg Kroah-Hartman's session on hotplugging support. It was not one of the better sessions to miss; hotplugging is a, well, hot topic. The nature of hardware is increasingly dynamic, and the Linux kernel must learn to adapt to a world that can change at any time.

Greg gave what must have been an interesting talk on how the /sbin/hotplug tool works, automatic determination of the proper driver modules to load, etc. Toward the end, however, he was talking about the work that still needs to be done, with an eye toward the 2.5 kernel. Future work includes:

SCSI hotplug support. The SCSI layer is likely to get thrashed up, again, in 2.5. Prediction: everybody will still complain about it.
Better PCMCIA support. PCMCIA currently has its own, completely different way of dealing with dynamic devices; much has been learned from the PCMCIA approach, but there seems to be a desire to get rid of it and move PCMCIA into the new hotplug mechanism.
Hotplug IDE devices. A patch is apparently in circulation.
PCI and CompactPCI devices; patches exist for these as well.
Laptop docking stations.
Hotplug CPUs; the CPU patch has been out there for a while.
Better distribution support. The hotplug stuff is very new, of course, and uptake by the distributions remains somewhat spotty. SuSE was named as being a special goal for the hotplug people.

The most interesting thing, perhaps, came right at the end, when Greg raised the point that ACPI may eventually obsolete the entire Linux hotplug effort. As ACPI gets the basic configuration and power management issues taken care of, hotplug management will increasingly come within their sites. Of course, as LWN has covered elsewhere, not everybody is happy with the design of ACPI. There could be an interesting fight in the future.

More information on Linux and hotplugging at linux-hotplug.sf.net.

Back to OLS 2001 coverage