Re: [PATCH 1/6] iommu: Move IOVA power-of-2 roundup into allocator

[John Garry]

On 06/04/2021 17:54, John Garry wrote:

Hi Robin,

> > Sorry if the phrasing was unclear there - the allusion to default 
> > domains is new, it just occurred to me that what we do there is in 
> > fact fairly close to what I've suggested previously for this. In that 
> > case, we have a global policy set by the command line, which *can* be 
> > overridden per-domain via sysfs at runtime, provided the user is 
> > willing to tear the whole thing down. Using a similar approach here 
> > would give a fair degree of flexibility but still mean that changes 
> > never have to be made dynamically to a live domain.
>
> So are you saying that we can handle it similar to how we now can handle 
> changing default domain for an IOMMU group via sysfs? If so, that just 
> is not practical here. Reason being that this particular DMA engine 
> provides the block device giving / mount point, so if we unbind the 
> driver, we lose / mount point.
>
> And I am not sure if the end user would even know how to set such a 
> tunable. Or, in this case, why the end user would not want the optimized 
> range configured always.
>
> I'd still rather if the device driver could provide info which can be 
> used to configure this before or during probing.

As a new solution, how about do both of these:
a. Add a per-IOMMU group sysfs file to set this tunable. Works same as 
how we change the default domain, and has all the same 
restrictions/steps. I think that this is what you are already suggesting.
b. Provide a DMA mapping API to set this value, similar to this current 
series. In the IOMMU backend for that API, we record a new range value 
and return -EPROBE_DEFER when successful. In the reprobe we reset the 
default domain for the devices' IOMMU group, with the IOVA domain rcache 
range configured as previously requested. Again, allocating the new 
default domain is similar to how we change default domain type today.
This means that we don't play with a live domain. Downside is that we 
need to defer the probe.

Thanks,
John