ACPI and power management [LWN.net]

This article is part of LWN's 2003 Kernel Developers' Summit coverage.

The first part of this session was run by Andrew Grover, who talked about ACPI development. According to Andrew, ACPI is "going pretty well," though a couple of hecklers in the crowd expressed a different opinion. It is now used to enumerate hardware and route interrupts (though, as Andrew concedes, it "doesn't always work"). Battery and power management also generally work. However, ACPI will always be a bit of a mess because it is the dumping ground for a number of difficult issues. If there is hardware that doesn't fit into a standard scheme, it is left up to ACPI to make it work.

Pat Mochel talked about power management and sysfs, asking the developers what they would like to see happen in that area in 2.7. Matthew Wilcox stated that the best thing that could happen would be the complete removal of the device model and sysfs; his preference would be a "forest of filesystems" where each driver creates it own, private filesystem. Linus pointed out a problem with this approach, however: it ignores the issue of linkages between the filesystems. Sysfs, by virtue of being a single filesystem, can handle links between devices (and other objects) without race conditions and other problems.

Rusty Russell asked about the representation of module parameters in sysfs. Many of the hooks are there now; the new style of module parameter declaration includes a sysfs permissions field. It turns out that Rusty and Pat have each been waiting for the other to implement this functionality. The session concluded with neither having volunteered to actually finish the task.