Integrating OS power management with the X server

[Benjamin Herrenschmidt]

On Fri, 2004-10-29 at 13:38 +0200, Egbert Eich wrote:

> Effectively it does the same as when doing a VT switch. Saving state is
> actually not so bad as we don't save static register states as we have
> calculated them when calculating the video mode.

AFAIK, It does more on suspend/resume than on VT switches, no ? At
least, I have cases where suspend resume work when X is notified
via /dev/apm_bios (in linux) but locks up on resume when just switching
to a text console during suspend/resume instead.

I think this accounts for some of the problems Linux users are having
when trying to use ACPI for suspend instead of APM. For PowerBooks,
where I have my own core suspend mecanism, I ended up writing an
emulation driver for /dev/apm_bios so X can be properly notified.

It would be useful to create a mecanism to notify X of suspend/resume
independently of the kernel, maybe via signals causing X to call helpers
or such thing ? The idea is to isolate the X server proper from whatever
notification mecanism is provided by the kernel. Moving the
kernel-specific logic to some separate userland thingies that can be
dealt with by the distros (though X could bundle default ones for Linux
APM interface, ACPI, etc...)

On the other hand, we might add a standard mecanism in the linux kernel
to provide that service regardless of the actual suspend/resume mecanism
used too. It's still beeing discussed at this point.

Ben.