From: Kimio S. <k-s...@mv...> - 2001-12-12 06:03:20
|
Hi all, > Actually, it's looking like it's not that bad. Some things need to be > integrated so everything flows a little better, but I think most of the > dirty work is done. > > Basically, you get yourself down to 1 processor at the beginning of the > suspend process. You can do that via the hotplug CPU patches: > > http://sourceforge.net/projects/lhcs/ > > They're a bit outdated, but if anyone is interested, Randy up-ported the > patches to 2.4.14(?) and one of us will probably do it for 2.5.x. I'm now making the patch for 2.5.0 and am expecting to send it in couple of days. > Based on some rumors that I heard, the x86 CPU "drivers" will be > getting a > bit of a makeover. If we could some how massage them into either the > device tree or a parallel tree, they could use a lot of the same code to > do suspend/resume. (Same thing for APICs and other board-level control > devices). > > With application CPUs, I wouldn't worry about saving state. Just schedule > everything off the processor and send the IPI to shut it down (I don't > remember the name ATM). When we come back up, bootstrap all the > application CPUs and start scheduling on them. Ummm, the hotplug CPU patch for x86 does have functions of CPU online/offline, but no CPU hot-add/remove functions. So, I suspect it can not be used just as it is. I'm now developing hot-add/remove functions for IA-64, but have no plan to implement it on x86 platforms. Regards, Kimi -- Kimio Suganuma <k-s...@mv...> |