On Mon, Sep 01, 2008 at 09:24:19PM -0700, Jason Thorpe wrote: > > On Aug 31, 2008, at 12:38 AM, matthew green wrote: > >> swapoff == removing a swap device from use, which means, in the >> current code, brining all those pages back into ram. > > Should not matter if all of the processes that would have cared are > gone. The trouble is the unkillable buggers, blocking on some resource that will never become available again, that have gobbled all memory and are the reason for needing the rebbot in the first place. On the whole, I like the idea - and I like that the kernel will do it as a last resort even if userland rc.shutdown will have tried first (while it's more likely that there are processes still blocked and around).. but it can't be something that blocks or even notably slows down the reboot. The 'better way' we need to find is somthing that can forcefully invalidate all the pages, rather than cause them to be paged in again. -- Dan.
Attachment:
pgpa61DV1nW2W.pgp
Description: PGP signature