mtd device for rootfs with kernel 2.4.18

mtd device for rootfs with kernel 2.4.18

[Armin Schindler]

Hi,

I hope someone can help with the following :

- MTD Partitions
  (mtd from CVS about 2 weeks ago, no changes in the important
  parts since then)
- JFFS2 as root filesystem on one mtd partition
  (JFFS2 from CVS jffs2-2_4-branch)
- Kernel 2.4.18

The problem is, when the kernel tries to mount the rootfs, it just
hangs. This is because it waits for a semaphore s_umount of the
superblock structure.

This semaphore was introduced in kernel 2.4.4 (I believe) and is
down during the loop for fs_types in fs/super.c. It is still down,
when blkdev_put() is called, where the mtdblock's release function
is called. This release function calls invalidate_device() which also
calls other functions where the s_umount semaphore is "downed" too.

I'm not that familiar with the mount-rootfs or the mtdblock code to
propose a patch here. 
Maybe moving the up_write(&sb->s_umount); before the blkdev_put()
would be a solution, but I don't know the impact...

Or did I overlook something here ?

Thanx,
Armin

Re: mtd device for rootfs with kernel 2.4.18

[David Woodhouse]

acs@sysgo.de said:
>  Or did I overlook something here ? 

Yes. The mailing list archives.

--
dwmw2

Re: mtd device for rootfs with kernel 2.4.18

[Armin Schindler]

Okay, the call to invalidate_device(), which causes the semaphore
problem, was removed from the release function in nftlcore.c.

Is it safe to do the same from the release function in mtdblock.c ?
If yes, why was it added ?

Armin 

On Wed, 3 Apr 2002, David Woodhouse wrote:
> 
> acs@sysgo.de said:
> >  Or did I overlook something here ? 
> 
> Yes. The mailing list archives.
> 
> --
> dwmw2

Re: mtd device for rootfs with kernel 2.4.18

[David Woodhouse]

acs@sysgo.de said:
> Okay, the call to invalidate_device(), which causes the semaphore
> problem, was removed from the release function in nftlcore.c.

> Is it safe to do the same from the release function in mtdblock.c ? If
> yes, why was it added ? 

It should never have been there in the first place. It was removed from 
mtdblock.c in CVS last year, and that fix is in 2.4.18-pre5.

--
dwmw2

Re: mtd device for rootfs with kernel 2.4.18

[Armin Schindler]

On Wed, 3 Apr 2002, David Woodhouse wrote:
> acs@sysgo.de said:
> > Okay, the call to invalidate_device(), which causes the semaphore
> > problem, was removed from the release function in nftlcore.c.
> 
> > Is it safe to do the same from the release function in mtdblock.c ? If
> > yes, why was it added ? 
> 
> It should never have been there in the first place. It was removed from 
> mtdblock.c in CVS last year, and that fix is in 2.4.18-pre5.

Okay, thanx.
Just wondering, in CVS HEAD it is still there ? Or is this only
meant for 2.5 ?

Armin

Re: mtd device for rootfs with kernel 2.4.18

[David Woodhouse]

acs@sysgo.de said:
> Just wondering, in CVS HEAD it is still there ? Or is this only meant
> for 2.5 ? 

It was removed from CVS HEAD in November last year, and hasn't returned.

--
dwmw2