Re: syncing the disks when entering sleep

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Wednesday 10 February 2010, Pavel Machek wrote:
> > Hi.
> > 
> > Pavel Machek wrote:
> > >> So you're asking to give this knob "one shot behavior" (i.e. "then
> > >> next sleep won't sync")?
> > > 
> > > Yes.
> > > 
> > >> But I'm primarily interested in the behavior on embedded systems
> > >> (where you control all the processes running -- there's no "user"
> > >> involved.
> > >> 
> > > 
> > > Well, then "one shot behaviour" does not hurt you, right?
> > > 
> > >> If a user starts messing with default settings, any unwanted
> > >> behavior is the users problem (besides, this should only be
> > >> writable as root).
> > >> 
> > > 
> > > I'd rather not add traps for the user unless absolutely neccessary.
> > > Not even for root user. Pavel
> > 
> > But that's precisely what you're doing. You're advocating making the
> > behaviour inconsistent. If what you're suggesting is done, you won't be
> > able to simply cat the sysfs entry to know whether sys_syncing is going
> > to be done on the next cycle. You'll also have to have knowledge of
> > whether a cycle has been done since the last time the value is set. The
> > end result will be someone getting trapped and caught out because they
> > think '1' in /sys/power/dont_sync (or whatever it's called) means what
> > it says.
> 
> I'm simply advocating that setting from one suspend should not change
> other suspends ... because you have multiple different programs
> wanting to suspend the system, all independend.

Which is wrong.  There should be one power manager everybody else calls to
suspend the system.  Yes, even if the battery is running critical.

And I guess on the embedded systems in question the situation is exactly like
that.

> See the example about system running low on battery earlier in the thread. 

Which is irrelevant here.

Actually, doing the sync() in the kernel is a hack that we decided not to
remove just because the users space didin't do the right thing on some systems.
If the user space always synced disks before suspending, we wouldn't have to
do that in the kernel.

Same goes for the kernel VT switch, BTW.

So really, I don't see anything wrong with a knob that will turn the kernel
sync off entirely, because that basically means "my user space is not broken".

Rafael