[PATCH] mtd: allow mtd and jffs2 when ARCH=um

[Artem Bityutskiy]

On Tue, 2010-12-07 at 10:20 -0800, Jason Lunz wrote:
> On Tue, Dec 07, 2010 at 10:39:41AM +0100, richard -rw- weinberger wrote:
> > On Tue, Dec 7, 2010 at 8:29 AM, Jason Lunz <lunz at acm.org> wrote:
> > >
> > > Allow parts of drivers/mtd to compile on uml by pushing the HAS_IOMEM
> > > dependencies down closer to the parts of mtd that actually need it.
> > > This allows enough of mtd to build to let jffs2 be used on uml.
> > >
> > > Signed-off-by: Jason Lunz <lunz at acm.org>
> > > ---
> > >  arch/um/Kconfig.rest        |    4 +---
> > >  drivers/mtd/Kconfig         |    1 -
> > >  drivers/mtd/Makefile        |    3 ++-
> > >  drivers/mtd/chips/Kconfig   |    4 ++++
> > >  drivers/mtd/devices/Kconfig |    8 ++++++++
> > >  drivers/mtd/maps/Kconfig    |    4 ++++
> > >  drivers/mtd/mtdchar.c       |    4 ++++
> > >  drivers/mtd/nand/Kconfig    |    3 +++
> > >  drivers/mtd/onenand/Kconfig |    3 +++
> > >  include/linux/mtd/map.h     |    2 ++
> > >  10 files changed, 31 insertions(+), 5 deletions(-)
> > >
> > > diff --git a/arch/um/Kconfig.rest b/arch/um/Kconfig.rest
> > > index 0ccad0f..e34f399 100644
> > > --- a/arch/um/Kconfig.rest
> > > +++ b/arch/um/Kconfig.rest
> > > @@ -28,9 +28,7 @@ source "drivers/scsi/Kconfig"
> > >
> > >  source "drivers/md/Kconfig"
> > >
> > > -if BROKEN
> > > -       source "drivers/mtd/Kconfig"
> > > -endif
> > > +source "drivers/mtd/Kconfig"
> > 
> > drivers/mtd was already marked as broken when we moved over to git (2.6.12-rc2).
> > 
> > Do you exactly know why it was marked and fixes this patch all issues?
> 
> I don't know exactly why it was originally marked BROKEN. I imagine it
> was because most of the low-level mtd drivers need readb/writeb and
> friends, which don't exist on uml. But now that mtd has virtual ram- and
> blockdev-backed drivers, that's no longer sufficient reason to omit the
> whole subsystem.
> 
> I can't claim that this fixes all possible issues of course, but I've
> privately been using this patch to run block2mtd under uml for years. My
> original motivation was to allow the use of jffs2 in a uml-based test
> environment and for that it's worked very well.

But I think your solution is a bit dirty, because it adds a great deal
of little 'if HAS_IOMEM' and '#ifdef CONFIG_HAS_IOMEM' to many places.
This is error-prone.

Instead, you should solve this problem in UML code. I do not know how,
but may be you can add readb/writeb there which actually do nothing or
print a scary warning, or do BUG(), and let things which use them just
fail run-time.

Anyway, please, think about a solution which needs few changes in UML
code instead of myriads of small little changes in many places.

Thanks!

-- 
Best Regards,
Artem Bityutskiy (Артём Битюцкий)