Re: [PATCH v3 2/2] iommupt: Avoid rewalking during map

[Josua Mayer <josua@solid-run.com>]

Hi Jason,

Thank you for the quick response!

Am 09.05.26 um 21:40 schrieb Jason Gunthorpe:
> On Sat, May 09, 2026 at 05:41:02PM +0000, Josua Mayer wrote:
>> This patch has landed in v7.1-rc1 and introduced a regression on the
>> arm64 SolidRun LX2160A Clearfog-CX board.
>>
>> Since d6c65b0fd621 ("iommupt: Avoid rewalking during map"),
>> boot now stalls - printing many smmu errors.
>>
>> Please find attached full boot logs for reference:
>>
>> - good.txt @ 99fb8afa16ad
>> - bad.txt @ d6c65b0fd621
>> - config.txt: .config from the build
>>
>> Is there any advice how to deal with this?
> Does this help?
>
> --- a/drivers/iommu/iommu.c
> +++ b/drivers/iommu/iommu.c
> @@ -2669,7 +2669,7 @@ int iommu_map_nosync(struct iommu_domain *domain, unsigned long iova,
>                 return 0;
>         }
>         ret = __iommu_map_domain_pgtbl(domain, iova, paddr, size, prot, gfp);
> -       if (!ret)
> +       if (ret)
>                 return ret;
>  
>         trace_map(iova, paddr, size);

I applied this on top of the problematic commit, and my system boots fine now!

When I compare the bad boot log, the good one, and the current one with above change,
only the bad one is missing the following line from boot log:

iommusmmu.0x0000000005000000: IOMMU driver was not able to establish FW requested direct mapping.

>
> Although it would be surprising your system triggers a ret!=0 while
> booting successfully, that could by why it wasn't noticed in any
> testing so far..
I can't say I understand IOMMU, but note that this system has a 
networking co-processor running its own software.

Linux communicates with it through RPC, and about 2GB of system RAM
are reserved and assined by the bootloader.


sincerely
Josua Mayer