Re: [PATCH/RFC] Add support to resume swsusp from initrd

[Pavel Machek <pavel@suse.cz>]

Hi!

> > > echo -n "set 03:02" >/sys/power/resume
> > 
> > I'd prefer not to have this one. Is it actually usefull? Then resume
> > could be triggered by echo -n "03:02" > /sys/power/resume...
> 
> resume_device is set by swsusp_read, which requires name_to_dev_t to be
> working. At the point where that's called, the device driver hasn't been
> loaded and we don't have the information to get the dev_t. Once the
> driver has been loaded, name_to_dev_t has already been discarded (it's
> marked __init). So we need to set resume_device somehow.

What about move of resume_device setup somewhere sooner?

> Heh. Yes, that's no problem. A new bigdiff for -rc3 would be
> helpful.

Hmm, I'm still on 2.6.9, but this code did not change much. I'll
generate it.

> > You really need to make sure that userland processes are stopped
> > before swsusp-resume is started. You should do freeze_process(). Then
> > resume process depends on having enough memory available, so you
> > probably want to free_some_memory() and warn in documentation about
> > the fact.
> 
> Ok. I'll look into that. The main reason I want code like this is that
> Debian use modular IDE drivers that are stored in the initrd. The disks
> won't be touched until the root file system is mounted, and we'll
> trigger the resume before then, so there shouldn't be any risk of data
> loss. At this point, there shouldn't be any userspace running other than
> a single shell script - do you think it's still a problem?

Single shell script would probably do no harm, but then, you want this
to go into mainline, not into Debian kernel, right? ;-).

Actually freezing processes is good thing to do even for normal
resume. We pretty much know there are no harmfull processes running
there, but better safe than sorry.
								Pavel
-- 
People were complaining that M$ turns users into beta-testers...
...jr ghea gurz vagb qrirybcref, naq gurl frrz gb yvxr vg gung jnl!