Re: current linus bk, error mounting root

[Jon Smirl]

On Mon, 21 Mar 2005 16:38:08 -0800, Greg KH <greg@xxxxxxxxx> wrote:
> On Mon, Mar 21, 2005 at 04:19:25PM -0800, Andrew Morton wrote:
> >
> > (Adds lots of cc's.  I trust that's OK).
> >
> > Jon Smirl <jonsmirl@xxxxxxxxx> wrote:
> > >
> > > No, I think Jens wants all of the distributions to fix it. I have
> > > filed a bug with Fedora on it.
> > >
> > > Something changed in the timing for loading drivers during boot. You
> > > used to be able to do:
> > > modprobe ata_piix
> > > mount /dev/sda1
> 
> You can't do that on any udev based system reliably.  That has _never_
> been true.  You might just have been getting lucky in the past.
> 

mkinitrd in Fedora needs to be modified to wait for the device. 
Maybe something like a nash command like wait /dev/sda1.

> > > Now you have to do this:
> > > modprobe ata_piix
> > > sleep 1
> > > mount /dev/sda1
> 
> That's still racy.  Rely on the /etc/dev.d/ notifier to be able to tell
> you when you can mount your device, that is what it is there for.
> 
> This is a udev issue, not a kernel issue, there's nothing we can do in
> the kernel about it (well, except for the obvious thing of giving udev
> lots of hints and making it easier for it to work properly and faster,
> which we have been doing over the past months.)
> 
> > > I suspect the problem is that udev doesn't get a chance to run anymore.
> > > The sleep 1 allows it to run and it creates /dev/sda1.
> > > Build ata_piix in and the problem goes away too.
> > >
> > > Jens is right that this is a user space issue, but how many people are
> > > going to find this out the hard way when their root drives stop
> > > mounting. Since no one is complaining I have to assume that most
> > > kernel developers have their root device drivers built into the
> > > kernel. I was loading mine as a module since for a long time Redhat
> > > was not shipping kernels with SATA built in.
> >
> > I don't agree that this is a userspace issue.  It's just not sane for a
> > driver to be in an unusable state for an arbitrary length of time after
> > modprobe returns.
> 
> It is a userspace issue.  If you have a static /dev there are no
> problems, right?  If you use udev, you need to wait for the device node
> to show up, it will not be there right after modprobe returns.
> 
> thanks,
> 
> greg k-h
> 


-- 
Jon Smirl
jonsmirl@xxxxxxxxx
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/