Re: IOMMU vs Ryzen embedded EMMC controller

From: Suthikulpanit, Suravee
Date: Wed Oct 09 2019 - 15:37:02 EST


On 10/9/19 2:25 PM, Jiri Kosina wrote:
> On Fri, 27 Sep 2019, Shah, Nehal-bakulchandra wrote:
>>>>> Do you have BAR memory allocation failures in dmesg with IOMMU on?
>>> No. The device is *not* treated as PCI device and I still think that
>>> this is the source of the evil.
>>>>> Actually, sharing both working and non-working dmesg, as well as
>>>>> /proc/iomem contents, would be helpful.
>>>> Yes, can you please grab dmesg from a boot with iommu enabled and add
>>>> 'amd_iommu_dump' to the kernel command line? That should give some
>>>> hints on what is going on.
>>> For now I attach a dmesg and iomem from the boot with IOMMU enabled.
>>> Nothing much interesting without IOMMU, sdhci-acpi there just works --
>>> let me know if you still want me to send the kernel msg.
>>> Thanks for looking into this!
>> I have added Suravee from AMD in the mail loop. He works on IOMMU part.
>> As per my understanding, it needs a patch in IOMMU driver for adding
>> support of EMMC. Note that on Ryzen platform we have EMMC 5.0 as ACPI
>> device.
> Friendly ping ... any news here?
> Thanks,

Could you please boot the system w/ kernel option amd_iommu_dump=1,
and do "dmesg | grep AMD-Vi". Then provide the output.

I suspect that there is something missing in the IVRS table, where it needs
to provide ACPI HID for the eMMC device.

See kernel parameter:

ivrs_acpihid [HW,X86_64]
Provide an override to the ACPI-HID:UID<->DEVICE-ID
mapping provided in the IVRS ACPI table. For
example, to map UART-HID:UID AMD0020:0 to
PCI device 00:14.5 write the parameter as:

Here we might need to do.

ivrs_acpihid[00:13.1]=<emmc ACPI HID on that system>