On Wednesday 10 November 2004 22:40, Benjamin Herrenschmidt wrote: > On Wed, 2004-11-10 at 14:00 -0800, David Brownell wrote: > > I'm not sure I like "on == freeze", but certainly it > > seems plausible to model "freeze" as a power state > > with minimal semantics. If the state is really needed, > > that is ... I suspect the kexec() model would be just > > as well achieved by unbinding all drivers from the > > devices. Ditto halt(). > > > > You've claimed that APM needs a FREEZE state. Why? > > If the issue is "BIOS handles hardware power states" > > that could be addressed more directly in for example > > pci_set_power_state(): "if (system_uses_APM()) return". > > Ugh... hidden magic inside of the PCI layer ? nope please ;) Call it firmware integration, then explain why you'd worry about a one-liner ... ;) ACPI needs some hooks there too it seems, if Linux is to have the option of using certain advice the BIOS may have for us ... like what PCI D-state is appropriate for a given system sleep state and device. Hmm, that pci_choose_state() patch might be a good place to hide those things. There could be a function hook, set by ACPI or APM, that provides the policy. That'd even be a reasonable hook for the board-specific code intelligence you asked for a while back. > Besides, freeze is absolutely necessary for STD right ? As I was saying: I'm unconvinced it's needed anywhere. (If for no other reason than wanting solid reasons to adopt this particular solution over others that don't need to integrate/debug new mechanisms...) But there are also a few dozen emails in my queue, which could enlighten me about such reasons. - Dave > Ben. > >
