[PATCH 05/11] omap3: Remove non-existent config option

[Marathe, Yogesh]

> -----Original Message-----
> From: Felipe Contreras [mailto:felipe.contreras at gmail.com]
> Sent: Saturday, September 25, 2010 9:13 PM
> To: Premi, Sanjeev
> Cc: Tony Lindgren; linux-arm-kernel at lists.infradead.org; linux-
> omap at vger.kernel.org; Marathe, Yogesh
> Subject: Re: [PATCH 05/11] omap3: Remove non-existent config option
> 
> On Fri, Sep 24, 2010 at 6:17 PM, Premi, Sanjeev <premi at ti.com> wrote:
> >> On Fri, Sep 24, 2010 at 4:50 AM, Tony Lindgren
> >> <tony at atomide.com> wrote:
> >> > From: Yogesh Marathe <yogesh_marathe at ti.com>
> >> >
> >> > The definition of "iva2" device in iommu_device
> >> > is wrapped inside CONFIG_MPU_BRIDGE_IOMMU, but
> >> > this option is not defined in KConfig.
> >> >
> >> > This patch removes the wrapper and makes "iva2"
> >> > available as another iommu_device.
> >>
> >> NAK.
> >>
> >> This would break tidspbridge's MMU. This was discussed before:
> >> http://thread.gmane.org/gmane.linux.ports.arm.kernel/58302/focus=58305
> >>
> >> When tidspbridge has migrated to iommu, then it should define
> >> CONFIG_MPU_BRIDGE_IOMMU.
> >
> >  Can you go through earlier discussion on this patch
> >  http://marc.info/?l=linux-omap&m=127979007623260&w=2
> >
> >  I believe these concerns were discussed. I am unable to
> >  open the gmane thread (getting 504) - so can't be sure
> >  if Hiroshi had referred to same discussion.
> 
> When you merge iommu support, then either you enable
> CONFIG_MPU_BRIDGE_IOMMU unconditionally, or you apply this patch, but
> this patch alone will only break things.
> 
> --
> Felipe Contreras

Felipe,

Any other driver which does not depend on bridge and interested in using iommu should get the handle when iommu_get("iva2") is called. It is not happening in original case. I think there should not be restrictions on other drivers to define un-related compile time define if they just want to use iommu driver. I feel the implementation that is breaking due to removal of this define should be fixed.

Regards,
Yogesh.