Re: 2.6.19-rc5: grub is much slower resuming from suspend-to-disk than in 2.6.18

[Zan Lynx <zlynx@acm.org>]

On Mon, 2006-11-13 at 21:54 +0300, Andrey Borzenkov wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> On Monday 13 November 2006 11:15, Stefan Seyfried wrote:
> > Hi,
> >
> > On Mon, Nov 13, 2006 at 06:42:15AM +0300, Andrey Borzenkov wrote:
> > > On Sunday 12 November 2006 17:55, Pavel Machek wrote:
> > > > On Sun 12-11-06 14:36:41, Andrey Borzenkov wrote:
> > > > > -----BEGIN PGP SIGNED MESSAGE-----
> > > > > Hash: SHA1
> > > > >
> > > > > This is rather funny; in 2.6.19-rc5 grub is *really* slow loading
> > > > > kernel when I switch on the system after suspend to disk. Actually,
> > > > > after kernel has been loaded, the whole resuming (up to the point I
> > > > > have usable
> >
> > The most important question:
> > What filesystem is your /boot on? I'd bet quite some money that it is
> > reiser or some other journaling FS (not ext3).
> >
> 
> there is no /boot, I use single / which is reiser.
> 
> > > > > desktop again) takes about three time less than the process of
> > > > > loading kernel + initrd. During loading disk LED is constantly lit.
> > > > > This almost looks like kernel leaves HDD in some strange state,
> > > > > although I always assumed HDD/IDE is completely reinitialized in this
> > > > > case.
> > > >
> > > > Seems like broken hw, really. No state should survive machine
> > > > poweroff.
> >
> > No. Broken FS / crappy GRUB.
> >
> > > To recap - this never happens upon simple power off; I do not remember
> > > this to
> >
> > I am pretty sure that it will also happen if you do "updatedb &", wait a
> > minute and then do a _HARD_ power off.
> >
> > I am pretty sure that it has nothing to do with the kernel version, just
> > with the layout of your /boot partition (which of course changes with every
> > kernel update). In other words: until now, you just have been lucky.
> 
> The idea is nice; unfortunately it fails to explain the difference 
> between 'poweroff' and 'suspend disk' cases. I doubt disk layout is changed 
> between them.

I have not checked if this is true, but it is a possible explanation:

Perhaps the filesystem is not properly unmounted during a suspend?  That
would mean GRUB is reading from a incoherent filesystem on resume.
GRUB's filesystem drivers are not very fancy.  It could be it does
something silly like check the journal before returning each block.

Maybe its a journal size thing, you could try "sync" before suspend and
see if it helps.  Another thing would be to create /boot as a separate
partition.
-- 
Zan Lynx <zlynx@acm.org>

Attachment: signature.asc
Description: This is a digitally signed message part