|
From: Jeff D. <jd...@ka...> - 2001-05-31 16:51:58
|
lm...@su... said: > Is it possible to make UML not run its own scheduler, but rely > entirely on the host ? I mean, many of those tasks just aren't > necessary since the host already provides them, or am I missing > something? I think so, see if this is close enough. You configure it as an SMP box with more virtual processors than processes (and with a bit of hacking, you can fire up a new virtual processor for each new process). Then the UML scheduler will have nothing do to, since there's no contention for processors (it will just bounce between processes and idle threads). Every process will always be runnable from the point of view of UML, so it will be the job of the host scheduler to do the actual scheduling. The catch is that UML doesn't support SMP at the moment. Making that work shouldn't be a big deal. There are some compilation problems when you configure SMP, plus someone needs to go through the UML arch layer looking for missing locks and adding them. I did it in a few days last year, and it pretty much worked. > but _unfortunately_ I will spend the entire weekend partying ;-) Let's all feel sorry for Lars and his rough life... > Oh - and if I resize the db0 file on the host, is that noticed by UML > so I could do resizereiserfs and grow the filesystem online? No. Currently, you'd have to unmount the filesystem and remount it. It would be possible to say on the command line "don't do I/O bounds checking on this device". That's the only place in the driver that cares about the device size. It's handy for stopping a dd on the raw device, but it doesn't matter for a filesystem. Jeff |