Re: [patch-4.16] arm/smmuv1,v2: Protect smmu master list with a lock

[Ian Jackson <iwj@xxxxxxxxxxxxxx>]

Stefano Stabellini writes ("Re: [patch-4.16] arm/smmuv1,v2: Protect smmu master 
list with a lock"):
> In regards to this specific patch and also the conversation about 4.16
> or 4.17: I think it would be fine to take this patch in 4.16 in its
> current form. Although it is not required because PCI passthrough is
> not going to be complete in 4.16 anyway, I like that this patch makes
> the code consistent in terms of protection of rbtree accesses.  With
> this patch the arm_smmu_master rbtree is consistently protected from
> concurrent accesses. Without this patch, it is sometimes protected and
> sometimes not, which is not great.

It sounds like this is a possible latent bug, or at least a bad state
of the code that might lead to the introduction of bad bugs later.

So I think I understand the upside.

> So I think that is something that could be good to have in 4.16. But
> like you said, the patch is not strictly required so it is fine either
> way.

Can you set out the downside for me too ?  What are the risks ?  How
are the affected code paths used in 4.16 ?

A good way to think about this is: if taking this patch for 4.16
causes problems, what would that look like ?

Thanks,
Ian.