Greg Kroah-Hartman led an uncontentious discussion of the device model,
which, he claimed, was full of "things people love to complain about."
Several complaints, both valid and invalid were brought up. The valid
complaints include the lack of documentation, the difficulties in using the
device model interfaces correctly, the intermixing of kobject and sysfs
functionality, and the fact that sysfs uses excessive amounts of low
For the near future, plans include improved documentation, shrinking the
kref structure to cut its overhead, and adding features to support
its use in read-copy-update schemes (see the
July 14 Kernel Page). The sysfs backing store patches (February 5 Kernel Page) will probably be
merged in some form to address the low memory problems.
For the 2.7 series, work will include (unspecified) API changes to make it
harder to use the device model incorrectly. The sysfs functionality will
be split out of the kobject type, resulting in a cleaner interface
all around. This work may also aim for supporting multiple forms of sysfs;
apparently there is interest in the cluster community in creating some sort
of cluster-wide sysfs filesystem.
In 2.7, the device model will also likely be changed to support
multithreaded device probing. This work should make probing more robust,
and, perhaps more importantly for most users, speed up the bootstrap
process. This topic came up again later in the afternoon when desktop
performance was discussed.
>> Next: Video drivers.
to post comments)