Re: [RFC] devfs=only and boot

From: Richard Gooch (rgooch@ras.ucalgary.ca)
Date: Wed Dec 12 2001 - 01:00:25 EST


Alexander Viro writes:
>
>
> On Mon, 10 Dec 2001, Richard Gooch wrote:
>
> > Alexander Viro writes:
> > > Richard, just how devfs=only is supposed to work with
> > > loading ramdisk from floppies?
> >
> > IIRC, it's supposed to work just like normal reading from a
> > floppy. That should still work.
>
> How? blkdev_get() is not going to work in that case...

To clarify: are you saying that normal floppy access cannot work with
devfs=only? I assure you, it works with 2.4.17.

> > > BTW, with initrd exiting with real-root-dev set (regardless of
> > > devfs=only) your code still goes by root_device_name and ignores new
> > > ROOT_DEV. Again, what behaviour is expected?
> >
> > The intent is that root_device_name is changed, so it should just
> > work. Has something broken? AFAIK, this too used to work.
>
> What would change it? We have ROOT_DEV = new_root_dev; in
> change_root(), so your ROOT_DEVICE_NAME is non-NULL... What's more,
> where are you going to get the new name?

Sorry, scratch what I said, I've just had a chance to look at the
code. OK, the way it's supposed to work is that after the
change_root(), ROOT_DEVICE_NAME points to root_device_name, which
should contain the "final" root device.

But while loading the RD from floppy, it has to use ROOT_DEV, which is
the RD device number. I think what you're wondering is how that can
work if you pass "devfs=only", right? Well, it should work, because
rd.c calls register_blkdev() and *not* devfs_register_blkdev(). So a
call to blkdev_get() should always work.

Does this help?

> > Alexander Viro writes:
> > > BTW, here's one more devfs rmmod race: check_disk_changed() in
> > > fs/devfs/base.c. Calling ->check_media_change() with no protection
> > > whatsoever. If rmmod happens at that point...
> >
> > How about if I do this sequence:
> > lock_kernel();
> > devfs checks;
> > if (bd_op->owner)
> > __MOD_INC_USE_COUNT(bd_op->owner);
> > revalidate();
> > if (bd_op->owner)
> > __MOD_DEC_USE_COUNT(bd_op->owner);
> > unlock_kernel();
> >
> > Is there any reason why that won't work?
>
> For one thing, the situation when you are already half-way through the
> module removal. At least use try_inc_mod_count().

Fair enough. So do you see anything wrong with this sequence:
        lock_kernel();
        if (devfs_sees_it() && bd_op->owner && try_inc_mod_count(bd_op->owner))
        {
                revalidate();
                __MOD_DEC_USE_COUNT(bd_op->owner);
        }
        unlock_kernel();

Any more problems you can see?

                                Regards,

                                        Richard....
Permanent: rgooch@atnf.csiro.au
Current: rgooch@ras.ucalgary.ca
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Sat Dec 15 2001 - 21:00:22 EST