Re: Regression in suspend to ram in 2.6.31-rc kernels

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

On Friday 18 September 2009, OGAWA Hirofumi wrote:
> "Rafael J. Wysocki" <rjw@sisk.pl> writes:
> 
> > On Saturday 12 September 2009, Chris Ball wrote:
> >> Hi,
> >> 
> >>    > Well system could check basic card ids if they match after resume
> >> 
> >> No.  That (arguably) guarantees that it's the same card, but not that
> >> it wasn't modified in another machine during the suspend.
> >
> > Generally speaking, we'd also need to check superblocks for this to work.
> >  
> >>    > if some users wants to crash his card by randomly swapping it
> >>    > during suspend/resume - I'd have no problem with that....
> >> 
> >> You should have a problem with it.  Taking a card from a suspended
> >> machine and working on it with a different machine is not a bizarre
> >> thing to want to do.
> >
> > Agreed.
> 
> Um...
> 
> What happen if we moved remove event to resume sequence?  I.e. The
> resume generates remove and insert event (or such revalidate).  With
> this, I hope the suspend is not bothered by complex one, and the resume
> just ignores (if needed) previous state and notify it to userland by
> remove/insert event.
> 
> And, userland process should unmount for removal devices before suspend
> process (as part of userland preparation)?
> 
> If we assumed the removable device can be changed before resume, fs
> would need to recover process, to make sure in-core and on-disk state
> has consistent.
> 
> Um..., for now, I feel the umount before suspend is only safe way.

Yes, with the current design it's the only really safe way.

Thanks,
Rafael