XIP issues introduced between 2.6.12 and 2.6.13

XIP issues introduced between 2.6.12 and 2.6.13

[Konstantin Kletschke]

Well, as two other guys also I experience Problems getting XIP
functional on 2.6.13.
In order to debug this I set up a nfs root to play from there.

The funny thing I relize now, is the following:

With jffs2 build into the kernel image the system freezes reproducable
when touching a file.

When build and load as a module I can do anything with the mounted
jffs2 partition, everything works fine.

Now I am puzzled. The not being a module or being a module is the only
difference and I testet both situations a dozen of times.
Another sidenote: I once had an XIP Kernel today running, root over
nfs and used /dev/mtdblock4 as a swap device with no problems. How
comes that :) ?

Regards, Konsti

-- 
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF

Re: XIP issues introduced between 2.6.12 and 2.6.13

[Nicolas Pitre]

On Wed, 12 Oct 2005, Konstantin Kletschke wrote:

> 
> Well, as two other guys also I experience Problems getting XIP
> functional on 2.6.13.
> In order to debug this I set up a nfs root to play from there.
> 
> The funny thing I relize now, is the following:
> 
> With jffs2 build into the kernel image the system freezes reproducable
> when touching a file.
> 
> When build and load as a module I can do anything with the mounted
> jffs2 partition, everything works fine.
> 
> Now I am puzzled. The not being a module or being a module is the only
> difference and I testet both situations a dozen of times.
> Another sidenote: I once had an XIP Kernel today running, root over
> nfs and used /dev/mtdblock4 as a swap device with no problems. How
> comes that :) ?

OK....  Let me resume that:

1) you have an XIP kernel

2) if you use NFS everything is fine

3) if you use JFFS2 built as a module everything _appears_ fine

4) if you use JFFS2 built-in it crashes.

Is the above right?

When JFFS2 is modular, are MTD drivers modular as well?


Nicolas

Re: XIP issues introduced between 2.6.12 and 2.6.13

[Konstantin Kletschke]

At Wed, 12 Oct 2005 12:52:49 -0400 (EDT),
Nicolas Pitre <nico@cam.org> wrote:

> OK....  Let me resume that:
> 
> 1) you have an XIP kernel

Yes.

> 2) if you use NFS everything is fine

NFS used in both cases as root file system.

> 3) if you use JFFS2 built as a module everything _appears_ fine

Yes.

> 4) if you use JFFS2 built-in it crashes.

Yes.

> When JFFS2 is modular, are MTD drivers modular as well?

In both cases built in.
Only difference a "jffs2" in lsmod output or not :-)

K

-- 
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF

Re: XIP issues introduced between 2.6.12 and 2.6.13

[Nicolas Pitre]

On Wed, 12 Oct 2005, Konstantin Kletschke wrote:

> At Wed, 12 Oct 2005 12:52:49 -0400 (EDT),
> Nicolas Pitre <nico@cam.org> wrote:
> 
> > OK....  Let me resume that:
> > 
> > 1) you have an XIP kernel
> 
> Yes.
> 
> > 2) if you use NFS everything is fine
> 
> NFS used in both cases as root file system.

What both cases?

> > 3) if you use JFFS2 built as a module everything _appears_ fine
> 
> Yes.
> 
> > 4) if you use JFFS2 built-in it crashes.
> 
> Yes.

Does it crash when JFFS2 is built-in but _not_ used as root?
I mean, does it not crash up to the point you mount it and try to access 
it?

> > When JFFS2 is modular, are MTD drivers modular as well?
> 
> In both cases built in.
> Only difference a "jffs2" in lsmod output or not :-)

If so something is really weird.

CAn you try with and without CONFIG_JFFS2_FS_WRITEBUFFER?

Looking at the changelog between v2.6.12 and v2.6.13 for fs/jffs2/ I 
don't see anything that could obviously explain the above.

Can you try the 2.6.12 code for JFFS2 in your 2.6.13 kernel?


Nicolas

Re: XIP issues introduced between 2.6.12 and 2.6.13

[Konstantin Kletschke]

At Wed, 12 Oct 2005 13:47:01 -0400 (EDT),
Nicolas Pitre <nico@cam.org> wrote:

> > NFS used in both cases as root file system.
> 
> What both cases?

Yes, to clear out a concrete difference.

When I use jffs2 as the root fs I can only build it into the kernel
image and debugging is plain guessing. As a side note I will explain
the behaviour build in and / on jffs2: It crashes most often after
"Initializing random number generator" Which is one of the first
points the userspace is "stressed". Sometimes here:
http://lists.infradead.org/pipermail/linux-mtd/2005-October/014005.html

In the other half of boots at another random place.

> Does it crash when JFFS2 is built-in but _not_ used as root?

Yes. I boot with root on nfs:
flash_unlock /dev/mtd3 -> works
mount -t jffs2 /dev/mtdblock3 /mnt -> works
cd /mnt -> works
ls -laR -> works
touch a -> immediate freeze.

> > In both cases built in.
> > Only difference a "jffs2" in lsmod output or not :-)
> 
> If so something is really weird.
> 
> CAn you try with and without CONFIG_JFFS2_FS_WRITEBUFFER?

Yes. 

> Looking at the changelog between v2.6.12 and v2.6.13 for fs/jffs2/ I 
> don't see anything that could obviously explain the above.
> 
> Can you try the 2.6.12 code for JFFS2 in your 2.6.13 kernel?

Also yes.

Konsti

-- 
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF

Re: XIP issues introduced between 2.6.12 and 2.6.13

[Konstantin Kletschke]

At Wed, 12 Oct 2005 22:11:50 +0200,
Konstantin Kletschke <lists@ku-gbr.de> wrote:

> > > NFS used in both cases as root file system.
> > 
> > What both cases?
> 
> Yes, to clear out a concrete difference.

ARGH! My answer was bogus.

NFS root used in both cases, accessing a jffs2 mtdblock device with
jffs2 built into kernel and loaded as a module.

K

-- 
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF

XIP issues introduced between 2.6.12 and 2.6.13

[Konstantin Kletschke]

Well, I want to be more precise:

If I build in jffs2 into Kernel, the Kernel freezes immediatly after
starting sometimes. In seldom times it boots to prompt (with / on
nfs). If it boots to prompt, I can mount mtdblock partitions rw and ls
them. If I touch a file then on it, it breaks 100%.

With jffs2 as module, nor Problem.

How this could correlate with 
http://lists.arm.linux.org.uk/pipermail/linux-arm-kernel/2005-October/031686.html
I have really no clue!

Regards, Konsti

-- 
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF