Read only forced on jffs2 by mtd layer?

Read only forced on jffs2 by mtd layer?

[Adam Ward]

Hi,

I'm working on a Sandgate2-P board on 2.6.15-4 kernel.
I was a little confused by the existence of mtd
initialisation code in /arch/arm/mach-pax/mainstone.c
which when duplicated for SG2-P failed to work.

So I realised that there was also
/devices/mtd/maps/mainstone-flash.c (and I had played
with this under 2.6.9 when it was pxa27x-flash.c)
which I again duplicated and have had some success
with.

But only to read!

My jffs2 image will mount as ro.
If I do 'mount / -o remount,rw' it appears to be
happy, but I soon get told 'Read-only file system' if
I try anything too presumptuous!

As running mount still declares it as read/write, I am
assuming the error is with the mtd block driver
refusing to co-operate.

Is there anything you guys can suggest that I should
pay closer attention to?

--
Adam



		
___________________________________________________________ 
NEW Yahoo! Cars - sell your car and browse thousands of new and used cars online! http://uk.cars.yahoo.com/

Re: Read only forced on jffs2 by mtd layer?

[Nicolas Pitre]

On Fri, 10 Mar 2006, Adam Ward wrote:

> My jffs2 image will mount as ro.
> If I do 'mount / -o remount,rw' it appears to be
> happy, but I soon get told 'Read-only file system' if
> I try anything too presumptuous!
> 
> As running mount still declares it as read/write, I am
> assuming the error is with the mtd block driver
> refusing to co-operate.
> 
> Is there anything you guys can suggest that I should
> pay closer attention to?

Did you unlock the flash partition with the flash_unlock tool?
Some NOR flash parts initialize themselves with all sectors locked by 
default.


Nicolas

Re: Read only forced on jffs2 by mtd layer?

[Josh Boyer]

On 3/10/06, Nicolas Pitre <nico@cam.org> wrote:
> On Fri, 10 Mar 2006, Adam Ward wrote:
>
> > My jffs2 image will mount as ro.
> > If I do 'mount / -o remount,rw' it appears to be
> > happy, but I soon get told 'Read-only file system' if
> > I try anything too presumptuous!
> >
> > As running mount still declares it as read/write, I am
> > assuming the error is with the mtd block driver
> > refusing to co-operate.
> >
> > Is there anything you guys can suggest that I should
> > pay closer attention to?
>
> Did you unlock the flash partition with the flash_unlock tool?
> Some NOR flash parts initialize themselves with all sectors locked by
> default.

Also, does your JFFS2 filesystem have free space available?  If a
minimum amount of free eraseblocks are not present, JFFS2 will mark
itself as read-only no matter what you do.

josh

Re: Read only forced on jffs2 by mtd layer?

[Artem B. Bityutskiy]

Josh Boyer wrote:
> Also, does your JFFS2 filesystem have free space available?  If a
> minimum amount of free eraseblocks are not present, JFFS2 will mark
> itself as read-only no matter what you do.
> 
Hmm, is this sane behaviour?

-- 
Best Regards,
Artem B. Bityutskiy,
St.-Petersburg, Russia.

Re: Read only forced on jffs2 by mtd layer?

[Josh Boyer]

On 3/10/06, Artem B. Bityutskiy <dedekind@yandex.ru> wrote:
> Josh Boyer wrote:
> > Also, does your JFFS2 filesystem have free space available?  If a
> > minimum amount of free eraseblocks are not present, JFFS2 will mark
> > itself as read-only no matter what you do.
> >
> Hmm, is this sane behaviour?

Sorta.  If there isn't enough physical space to write nodes out, then
you can't really have a writeable filesystem.  It'll actually just
return -ENOSPC for everything instead of -EROFS, but the effect is the
same.

Take a look at jffs2_calc_trigger_levels in build.c. 
c->resv_blocks_deletion is the number of blocks needed to delete
something.  c->resv_blocks_write is the number of blocks needed to
write something.

josh

Re: Read only forced on jffs2 by mtd layer?

[Artem B. Bityutskiy]

Josh Boyer wrote:
> Sorta.  If there isn't enough physical space to write nodes out, then
> you can't really have a writeable filesystem.  It'll actually just
> return -ENOSPC for everything instead of -EROFS, but the effect is the
> same.
I mean, if I ask to mount read-write, but end-up with RO, this is bad. 
Shouldn't I explicitely request RO mount?


> Take a look at jffs2_calc_trigger_levels in build.c. 
> c->resv_blocks_deletion is the number of blocks needed to delete
> something.  c->resv_blocks_write is the number of blocks needed to
> write something.
Ok, but where is the code that switches it to RO mode in case of lack of 
eraseblocks?

-- 
Best Regards,
Artem B. Bityutskiy,
St.-Petersburg, Russia.

Re: Read only forced on jffs2 by mtd layer?

[Josh Boyer]

On 3/10/06, Artem B. Bityutskiy <dedekind@yandex.ru> wrote:
> Josh Boyer wrote:
> > Sorta.  If there isn't enough physical space to write nodes out, then
> > you can't really have a writeable filesystem.  It'll actually just
> > return -ENOSPC for everything instead of -EROFS, but the effect is the
> > same.
> I mean, if I ask to mount read-write, but end-up with RO, this is bad.
> Shouldn't I explicitely request RO mount?

Oh, the mount will still show it as being RW from what I recall.  But
you won't actually be able to write anything.  Which is also
confusing.  Especially in the case where it won't even let you
_delete_ things.

> > Take a look at jffs2_calc_trigger_levels in build.c.
> > c->resv_blocks_deletion is the number of blocks needed to delete
> > something.  c->resv_blocks_write is the number of blocks needed to
> > write something.
> Ok, but where is the code that switches it to RO mode in case of lack of
> eraseblocks?

It doesn't do that.  It just returns -ENOSPC.  Sorry if I've confused anyone.

josh

Re: Read only forced on jffs2 by mtd layer?

[Artem B. Bityutskiy]

On Fri, 2006-03-10 at 10:21 -0600, Josh Boyer wrote:
> It doesn't do that.  It just returns -ENOSPC.  Sorry if I've confused anyone.
I have a feeling that it used to behave this way, not sure. Well, it
doesn't matter now.

-- 
Best Regards,
Artem B. Bityuckiy,
St.-Petersburg, Russia.

Re: Read only forced on jffs2 by mtd layer?

[Nicolas Pitre]

On Fri, 10 Mar 2006, Josh Boyer wrote:

> On 3/10/06, Artem B. Bityutskiy <dedekind@yandex.ru> wrote:
> > Josh Boyer wrote:
> > > Sorta.  If there isn't enough physical space to write nodes out, then
> > > you can't really have a writeable filesystem.  It'll actually just
> > > return -ENOSPC for everything instead of -EROFS, but the effect is the
> > > same.
> > I mean, if I ask to mount read-write, but end-up with RO, this is bad.
> > Shouldn't I explicitely request RO mount?
> 
> Oh, the mount will still show it as being RW from what I recall.  But
> you won't actually be able to write anything.  Which is also
> confusing.  Especially in the case where it won't even let you
> _delete_ things.

Sure, but what is important here is the exact error returned, be it 
-ENOSPC or -EROFS.


Nicolas

Re: Read only forced on jffs2 by mtd layer?

[Adam Ward]

Not doing anything beyond what is there for the
Mainstone... I don't know enough about this!

Also, there is enough free space, I do not have any
problem under the 2.6.9 kernel.

There shouldn't be anything special about the hardware
, it uses a stacked SDRAM/flash part (RD38F4470LLYBH0)
but the flash is 2 * 28F256L18B - only difference
being it's 2 * 28F128L18B parts on the Mainstone. (I
think that's on C5 revision processor cards?)

Is anyone having better results with 2.6.15-4 on a
Mainstone? 

I am wondering if the movement of configuration from
drivers/mtd/maps/mainstone-flash.c into
arch/arm/mach-pxa/mainstone.c being incomplete is
indicative of deeper changes that break the former,
but are not far enough to enable the later.


--- Josh Boyer <jwboyer@gmail.com> wrote:

> On 3/10/06, Nicolas Pitre <nico@cam.org> wrote:
> > On Fri, 10 Mar 2006, Adam Ward wrote:
> >
> > > My jffs2 image will mount as ro.
> > > If I do 'mount / -o remount,rw' it appears to be
> > > happy, but I soon get told 'Read-only file
> system' if
> > > I try anything too presumptuous!
> > >
> > > As running mount still declares it as
> read/write, I am
> > > assuming the error is with the mtd block driver
> > > refusing to co-operate.
> > >
> > > Is there anything you guys can suggest that I
> should
> > > pay closer attention to?
> >
> > Did you unlock the flash partition with the
> flash_unlock tool?
> > Some NOR flash parts initialize themselves with
> all sectors locked by
> > default.
> 
> Also, does your JFFS2 filesystem have free space
> available?  If a
> minimum amount of free eraseblocks are not present,
> JFFS2 will mark
> itself as read-only no matter what you do.
> 
> josh
> 



		
___________________________________________________________ 
Yahoo! Photos – NEW, now offering a quality print service from just 8p a photo http://uk.photos.yahoo.com

Re: Read only forced on jffs2 by mtd layer?

[Todd Poynor]

Adam Ward wrote:
...
> but the flash is 2 * 28F256L18B - only difference
> being it's 2 * 28F128L18B parts on the Mainstone. (I
> think that's on C5 revision processor cards?)

The L18 StrataFlash chips featured on some Mainstone II boards do power 
up with all blocks locked.  Run flash_unlock on writeable partitions 
prior to mounting.

-- 
Todd

Re: Read only forced on jffs2 by mtd layer?

[Adam Ward]

Thanks for your help, that did the trick so I put in
some code to do the unlock, in the kernel map file.

I still had to do a remount, but a quick fiddle with
/etc/fstab and it's all how it should be.

Cheers guys,

Adam

--- Todd Poynor <tpoynor@mvista.com> wrote:

> Adam Ward wrote:
> ...
> > but the flash is 2 * 28F256L18B - only difference
> > being it's 2 * 28F128L18B parts on the Mainstone.
> (I
> > think that's on C5 revision processor cards?)
> 
> The L18 StrataFlash chips featured on some Mainstone
> II boards do power 
> up with all blocks locked.  Run flash_unlock on
> writeable partitions 
> prior to mounting.
> 
> -- 
> Todd
> 



		
___________________________________________________________ 
Yahoo! Photos – NEW, now offering a quality print service from just 8p a photo http://uk.photos.yahoo.com