RE: iommu/dma: SATA RMRR reservation omission in shared group on HP Gen8

["Renato Machelett" <renato@machelett.com>]

On 10/19/2026, Baolu Lu wrote:
> The early return is still there. Have you tried reserving the second
> device's RMRR in the IOVA allocator as well? Something like:
>
> diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
> index 68ee57800fef..e9b39009f109 100644
> --- a/drivers/iommu/dma-iommu.c
> +++ b/drivers/iommu/dma-iommu.c
> @@ -704,7 +704,7 @@ static int iommu_dma_init_domain(struct iommu_domain *domain, struct device *dev
>                          return -EFAULT;
>                  }
>
> -               return 0;
> +               return iova_reserve_iommu_regions(dev, domain);
>          }
>
>          init_iova_domain(iovad, 1UL << order, base_pfn);
>
> This would at least confirm your analysis.

Hi Baolu,

I have not tested a patched kernel because this is a production server.

In the October 2024 thread, Jerry Snitselaar proposed the equivalent
change, and Robin Murphy pointed out why calling
iova_reserve_iommu_regions() in the already-initialized branch is unsafe
as a general fix (https://lkml.iu.edu/2411.1/00575.html):

> It's also possible for the second device to get here after the first
> device has already started using the domain, so at that point it's no
> longer safe to use reserve_iova() due to how it merges overlapping and
> adjacent rbtree nodes, which could really screw things up once there
> are normal allocations in the tree as well.

Best regards,
Renato