On Thu, 2005-03-10 at 10:28 +1100, Nigel Cunningham wrote: > That sounds good from the kernel->userspace side, but hideous from the > other perspective! But perhaps I'm just being ignorant :> Say a > userspace daemon learns that we've just lost AC and have 3 minutes of > UPS power left. I guess it would notify HAL, Actually HAL now knows about usbhid UPS'es now and export them like other batteries :-) > but wouldn't we then want > HAL to notify kernel space? Or would you imagine HAL directly saying to > drivers, "Video driver, turn the monitor off if you haven't already."? > I could imagine some policy piece doing this, yes. It wouldn't be HAL, but something using HAL to enforce policy - I suppose we could teach HAL (or just a script, or even the kernel) about exporting a method minimizePowerUsage() that would do the right thing and said policy enforcing pieces would use that? (possibly respecting the users settings e.g. the "[x] Allow system to suspend this device if inactive") What I think is interesting here is a) making sure that the kernel<->user space interface is there for making sure that user space plays nicely with the kernel, e.g. HAL should stop polling a storage device if the kernel says it's going to sleep (hence notification); and b) providing a way for user space to configure policy (but let the kernel enforce it), such as "never put this device to sleep" or "put to sleep after N seconds of inactivity", hence writing to a sysfs attribute. Does that clarify? Best, David
