Re: [PATCH 18/24] iommu/amd: Introduce helper function for updating device ID mapping table

["Suthikulpanit, Suravee" <suravee.suthikulpanit@amd.com>]

On 8/19/2026 5:26 PM, guanghuifeng@linux.alibaba.com wrote:
>> +static void clear_device_mapping(struct amd_iommu *iommu, u16 
>> guestId, u16 gDevId)
>> +{
>> +    u64 val;
>> +    u8 __iomem *vfctrl;
>> +
>> +    /*
>> +     * Clear the DevID in VFCTRL registers
>> +     */
>> +    val = FIELD_PREP(DEVID_ENTRY_GDEVID_MASK, gDevId) |
>> +          FIELD_PREP(DEVID_ENTRY_HDEVID_MASK, 0) |
>> +          DEVID_ENTRY_WRITE | DEVID_ENTRY_VALID;
>> +
>> +    vfctrl = VIOMMU_VFCTRL_MMIO_BASE(iommu, guestId);
>> +    writeq(val, vfctrl + VIOMMU_VFCTRL_GUEST_DID_MAP_CONTROL0_OFFSET);
>> +}
> 
> 
> The clear path sets Vld=1 with hDevID=0. This semantics is ambiguous:
> Vld=1 marks the entry as a *valid* mapping, and hDevID=0 is a legal PCI
> BDF (Bus 0, Dev 0, Func 0), instead of treating the entry as invalid. To 
> actually invalidate the mapping,
> please use Vld=0
> 
>      val = FIELD_PREP(DEVID_ENTRY_GDEVID_MASK, gDevId) |
>            FIELD_PREP(DEVID_ENTRY_HDEVID_MASK, 0) |
>            DEVID_ENTRY_WRITE;
> 
> Alternatively, during the clear dev mapping process, you might directly 
> specify the original hdevid being used.

Based on the IOMMU implementation, it requires the valid bit to be set 
to 1. Otherwise, guest command INVALIDATE_IOMMU_ALL would trigger error 
ILLEGAL_COMMAND_ERROR event.

Generally, host device id BDF 0:00.0 is the host bridge, which is not 
part of the list of devices managed by IOMMU. So, it is safe.

Thanks,
Suravee