[PATCH 6/6] iommu/arm-smmu: Support buggy implemenations where all config accesses are secure

[will.deacon@arm.com (Will Deacon)]

On Tue, Oct 01, 2013 at 07:17:09PM +0100, Andreas Herrmann wrote:
> On Tue, Oct 01, 2013 at 09:28:51AM -0400, Rob Herring wrote:
> > On 10/01/2013 07:39 AM, Andreas Herrmann wrote:
> > > In such a case we have to use secure aliases of some non-secure
> > > registers.
> > > 
> > > This behaviour is controlled via a flag in smmu->bugs. It is set
> > > based on DT information when probing an SMMU device.
> > 
> > Perhaps quirks would be a nicer word...
> 
> In fact it might be better to rename this as "options".
> For device isolation I had added a flag to indicate this in
> smmu->features.
> But it might be better to not mix hardware features with software
> features or options.

Yes, I also didn't like the mixing of hardware and software features, but I
couldn't think of a good way to separate them out. Sounds like you might be
on to something.

> So what do you (Will and Rob) think about a change like
> 
>  struct arm_smmu_device {
>    ...
> 
>  #define ARM_SMMU_FEAT_COHERENT_WALK     (1 << 0)
>  #define ARM_SMMU_FEAT_STREAM_MATCH      (1 << 1)
>  #define ARM_SMMU_FEAT_TRANS_S1          (1 << 2)
>  #define ARM_SMMU_FEAT_TRANS_S2          (1 << 3)
>  #define ARM_SMMU_FEAT_TRANS_NESTED      (1 << 4)
>          u32                             features;
> + #define ARM_SMMU_OPT_SECURE_CFG_ACCESS  (1 << 0)
> + #define ARM_SMMU_OPT_ISOLATE_DEVICES    (1 << 1)
> +        u32                             options;
>          int                             version;

I'm ok with this change, but it gets you guys off the hook with your broken
integration!

>    ...
>  }
> 
> DT property "calxeda,smmu-secure-cfg-access" would set
> ARM_SMMU_OPT_SECURE_CFG_ACCESS and
> DT property "linux,arm-smmu-isolate-devices" would set
> ARM_SMMU_OPT_ISOLATE_DEVICES.
> 
> (Maybe in future there will be more driver options. )

Yeah, and you could probably go all out with macros to generate this stuff
based on compatible string and OPT #define too.

Will