[PATCH v2 0/3] iommu: Enable non-strict DMA on QCom SD/MMC

[Robin Murphy]

On 2021-07-14 11:15, Joerg Roedel wrote:
> Hi Robin,
> 
> On Fri, Jul 09, 2021 at 02:56:47PM +0100, Robin Murphy wrote:
>> As I mentioned before, conceptually I think this very much belongs in sysfs
>> as a user decision. We essentially have 4 levels of "strictness":
>>
>> 1: DMA domain with bounce pages
>> 2: DMA domain
>> 3: DMA domain with flush queue
>> 4: Identity domain
> 
> Together with reasonable defaults (influenced by compile-time
> options) it seems to be a good thing to configure at runtime via
> sysfs.
> 
> We already have CONFIG_IOMMU_DEFAULT_PASSTHROUGH, which can probably be
> extended to be an option list:
> 
> 	- CONFIG_IOMMU_DEFAULT_PASSTHROUGH: Trusted devices are identity
> 					    mapped
> 
> 	- CONFIG_IOMMU_DEFAULT_DMA_STRICT: Trusted devices are DMA
> 					   mapped with strict flush
> 					   behavior on unmap
> 
> 	- CONFIG_IOMMU_DEFAULT_DMA_LAZY: Trusted devices are DMA mapped
> 					 with flush queues for performance

Indeed, I got focused on the sysfs angle, but rearranging the Kconfig 
default that way to match makes a lot of sense, and is another thing 
which should fall out really easily from my domain type rework, so I'll 
add that to my branch now before I forget again.

> Untrusted devices always get into the DMA domain with bounce pages by
> default.
> 
> The defaults can be changed at runtime via sysfs. We already have basic
> support for runtime switching of the default domain, so that can be
> re-used.

As mentioned yesterday, already done! I'm hoping to be able to post the 
patches next week after some testing :)

Cheers,
Robin.