build failure due to ROOT_DEV in mtd module (was Re: linux-2.6.22-rc5-g7c8545e build #298 failed ...)

build failure due to ROOT_DEV in mtd module (was Re: linux-2.6.22-rc5-g7c8545e build #298 failed ...)

[Satyam Sharma]

Hi,

[Added Christoph, Paolo, David and linux-mtd.]

On 6/21/07, Tom Spink <tspink@gmail.com> wrote:
> On 20/06/07, Tom Spink <tspink@gmail.com> wrote:
> > On 20/06/07, Toralf Förster <toralf.foerster@gmx.de> wrote:
> > > Hello,
> > >
> > > the build with the attached .config failed, make ends with:
> > [...]
> > The build seems to fail because of:
> > ERROR: "ROOT_DEV" [drivers/mtd/maps/nettel.ko] undefined!
> >
> > After taking a quick look at the code, I can't immediately see why
> > this would be, since there is an include for linux/root_dev.h at the
> > top, there.
> >
> > There's only one occurrence of ROOT_DEV (line 425), and after a quick
> > look at the git history, it seems the include was originally missing,
> > but was put back in, in commit
> > 6cc449c7d0292cb9b993f0df84fd3225e3099492.
> [...]
> After further inspection, I *think* it's because nettel is being built
> as a module, and it's trying to reference ROOT_DEV, which is not
> exported.

You're right. Look at:
commit a2d2b5cb8e5d8787f009eeb3bbca52687bd42a77
[ http://lkml.org/lkml/2005/7/13/166 ]

You could export ROOT_DEV to modules back again, or perhaps:
http://www.infradead.org/pipermail/linux-mtd/2005-March/012290.html
to get around the build error.

[ It might've been unexported for good reasons, of course, see:
http://lkml.org/lkml/2005/3/30/179 ]

Satyam

Re: build failure due to ROOT_DEV in mtd module (was Re: linux-2.6.22-rc5-g7c8545e build #298 failed ...)

[Christoph Hellwig]

On Thu, Jun 21, 2007 at 12:22:01PM +0530, Satyam Sharma wrote:
> >> The build seems to fail because of:
> >> ERROR: "ROOT_DEV" [drivers/mtd/maps/nettel.ko] undefined!
> >>
> >> After taking a quick look at the code, I can't immediately see why
> >> this would be, since there is an include for linux/root_dev.h at the
> >> top, there.
> >>
> >> There's only one occurrence of ROOT_DEV (line 425), and after a quick
> >> look at the git history, it seems the include was originally missing,
> >> but was put back in, in commit
> >> 6cc449c7d0292cb9b993f0df84fd3225e3099492.

Please just the reference to ROOT_DEV from this driver.  Just because
someone builds this driver there should be no change in the default root
device.

Re: build failure due to ROOT_DEV in mtd module (was Re: linux-2.6.22-rc5-g7c8545e build #298 failed ...)

[Satyam Sharma]

On 6/21/07, Christoph Hellwig <hch@infradead.org> wrote:
> On Thu, Jun 21, 2007 at 12:22:01PM +0530, Satyam Sharma wrote:
> > >> The build seems to fail because of:
> > >> ERROR: "ROOT_DEV" [drivers/mtd/maps/nettel.ko] undefined!
> > >>
> > >> After taking a quick look at the code, I can't immediately see why
> > >> this would be, since there is an include for linux/root_dev.h at the
> > >> top, there.
> > >>
> > >> There's only one occurrence of ROOT_DEV (line 425), and after a quick
> > >> look at the git history, it seems the include was originally missing,
> > >> but was put back in, in commit
> > >> 6cc449c7d0292cb9b993f0df84fd3225e3099492.
>
> Please just the reference to ROOT_DEV from this driver.  Just because
> someone builds this driver there should be no change in the default root
> device.

I agree, but this (drivers/mtd/maps/nettel.c) isn't the only modular driver
referencing ROOT_DEV. We also have drivers/mtd/maps/pmcmsp-ramroot.c
using ROOT_DEV (in fact the purpose of that driver seems to be precisely
to special-case the root fs and do something with it ...) but considering
that other driver's (tristate) Kconfig option depends on another symbol that
is non-existent in the mainline tree, there is no way someone can build
pmcmsp-ramroot and so we'll never actually  hit that problem even with an
allmodconfig build.

Anyway, I'll leave this up to David / linux-mtd to sort out. [ No other
modular user of ROOT_DEV in the tree other than the two mtd drivers
mentioned here. ]