[v2,0/6] iommu/sun50i: Allwinner D1 support

Message

Samuel Holland Jan. 3, 2023, 1:08 a.m. UTC

D1 is a RISC-V SoC from Allwinner's sunxi family. This series adds IOMMU
binding and driver support. The RISC-V architecture code still needs
some small updates to use an IOMMU for DMA[1][2]. I will send those
separately.

[1]: https://lore.kernel.org/linux-riscv/[e-mail]/
[2]: https://lore.kernel.org/linux-riscv/[e-mail]/

Changes in v2:
 - Disallow the 'resets' property for the D1 variant
 - Set bypass based on attached devices instead of using a fixed value

Samuel Holland (6):
  dt-bindings: iommu: sun50i: Add compatible for Allwinner D1
  iommu/sun50i: Track masters attached to the domain
  iommu/sun50i: Keep the bypass register up to date
  iommu/sun50i: Support variants without an external reset
  iommu/sun50i: Add support for the D1 variant
  riscv: dts: allwinner: d1: Add the IOMMU node

 .../iommu/allwinner,sun50i-h6-iommu.yaml      | 20 +++++-
 .../boot/dts/allwinner/sunxi-d1s-t113.dtsi    | 10 +++
 drivers/iommu/sun50i-iommu.c                  | 68 ++++++++++++++-----
 3 files changed, 79 insertions(+), 19 deletions(-)

Comments

Jörg Rödel Jan. 20, 2023, 3:11 p.m. UTC | #1

Hi Samuel,

On Mon, Jan 02, 2023 at 07:08:57PM -0600, Samuel Holland wrote:
> Samuel Holland (6):
>   dt-bindings: iommu: sun50i: Add compatible for Allwinner D1
>   iommu/sun50i: Track masters attached to the domain
>   iommu/sun50i: Keep the bypass register up to date
>   iommu/sun50i: Support variants without an external reset
>   iommu/sun50i: Add support for the D1 variant
>   riscv: dts: allwinner: d1: Add the IOMMU node

There is a conflict between these patches and changes in the IOMMU core
branch. With those merged in there is a compile warning about
sun50i_iommu_detach_domain() being unused. Fixing that requires removing
of 3-4 functions, which I am not sure is the right solution.

I pushed the arm/allwinner branch out to the iommu tree (with the core
branch merged in) and exluded it from next for now. Can you please have
a look and fix this up in a way that does not break the driver?

Once this is fixed I will include the arm/allwinner branch into my next
branch again.

Thanks,

	Joerg

Jörg Rödel Feb. 3, 2023, 10:21 a.m. UTC | #2

Hi Samuel,

On Fri, Jan 20, 2023 at 04:11:30PM +0100, Joerg Roedel wrote:
> Once this is fixed I will include the arm/allwinner branch into my next
> branch again.

Since there was no reply to this I nuked the patches from the IOMMU
tree. If this is still relevant please resubmit them after the next
merge window.

Regards,

	Joerg

Samuel Holland Feb. 4, 2023, 2:49 p.m. UTC | #3

Hi Joerg,

On 2/3/23 04:21, Joerg Roedel wrote:
> On Fri, Jan 20, 2023 at 04:11:30PM +0100, Joerg Roedel wrote:
>> There is a conflict between these patches and changes in the IOMMU
>> core branch. With those merged in there is a compile warning about 
>> sun50i_iommu_detach_domain() being unused. Fixing that requires
>> removing of 3-4 functions, which I am not sure is the right
>> solution.
>>
>> Once this is fixed I will include the arm/allwinner branch into my
>> next branch again.
> 
> Since there was no reply to this I nuked the patches from the IOMMU 
> tree. If this is still relevant please resubmit them after the next 
> merge window.

I am not sure what the right solution is here either, and I have not had
the chance to look at it again. With my limited understanding of how the
default domain logic works, and the fact that the IOMMU driver only
supports one domain, it seems the driver should keep that domain enabled
permanently, regardless of which devices are attached. So my patch 2
would be wrong. I will investigate and send a v3 after the merge window.

Regards,
Samuel