Re: Remove pmdisk from kernel

[Pavel Machek <pavel@ucw.cz>]

On Út 16-03-04 12:27:36, Nigel Cunningham wrote:
> On Tue, 2004-03-16 at 13:56, Pavel Machek wrote:
> > 
> > > Most of those changes are hooks to make the freezer for more reliable.
> > > That part of the functionality could be isolated from the bulk of
> > > suspend2. Would that make you happy?
> > 
> > Yes, that would be very good. It would make it easy to see actual
> > changes..
> > 
> > [I still do not understand why those hooks are neccessary... kill
> > -SIGSTOP works, right?]
> 
> Not always. Take for example the case where you have an NFS mount and
> happen to be doing an ls when the suspend cycle is started. If you
> signal the NFSd threads before the ls thread, the NFS threads will
> refrigerate okay, but the ls thread will fail to stop because it's
> waiting for data from the nfsd threads.

Hmm, you are right that with dead nfs server, kill -SIGSTOP will fail
on ls, and similary current refrigerator will fail. I think we can
live with that.

I agree that two-stage suspend is probably neccessary (userland first,
kernel than); but that should be possible without that big changes,
right?

> The best way to test the reliability of the current freezer
> implementation is to grab Michael's test patches. They can load the
> system down with NFS access, kernel compiles, benchmarks and so on.
> You'll quickly see the freezer fail. My implementation handles those
> loads flawlessly, and where problems are found, they're easily fixed.

Your solution is more reliable, thats right.
								Pavel
-- 
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]