[Qemu-devel] Question on Qemu IOAPIC ID

[Qemu-devel] Question on Qemu IOAPIC ID

[David Kiarie]

Hello all,

I, working on AMD IOMMU interrupt remapping would like to ask some
questions on Qemu IOAPIC id.

I currently have a problem in that Linux AMD IOMMU driver expects at least
one IOAPIC on the chipset to have ID 0xa0 while Qemu IOAPIC id is always 0.
I am faced with two options
   -Change IOAPIC id so that it always matches what Linux AMD IOMMU driver
expects. In this case I am not sure which problems I could encounter with
this approach.
   -Have a way to configure IOAPIC ID based on emulated IOMMU.

I prefer the first approach since it's quite simple but would like to hear
other opinions regarding this issue.

David.

Re: [Qemu-devel] Question on Qemu IOAPIC ID

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 952 bytes --]

On 2016-07-23 10:12, David Kiarie wrote:
> Hello all,
> 
> I, working on AMD IOMMU interrupt remapping would like to ask some
> questions on Qemu IOAPIC id.
> 
> I currently have a problem in that Linux AMD IOMMU driver expects at least
> one IOAPIC on the chipset to have ID 0xa0 while Qemu IOAPIC id is always 0.
> I am faced with two options
>    -Change IOAPIC id so that it always matches what Linux AMD IOMMU driver
> expects. In this case I am not sure which problems I could encounter with
> this approach.
>    -Have a way to configure IOAPIC ID based on emulated IOMMU.
> 
> I prefer the first approach since it's quite simple but would like to hear
> other opinions regarding this issue.
> 
> David.
> 

Ping on this question by David:

If we can consolidate over AMD's IOAPIC ID now, before releasing 4.7
(later changes would be guest visible), we won't have to add a special
case for the AMD IOMMU later.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 181 bytes --]

Re: [Qemu-devel] Question on Qemu IOAPIC ID

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 1404 bytes --]

On 2016-07-30 09:41, Jan Kiszka wrote:
> On 2016-07-23 10:12, David Kiarie wrote:
>> Hello all,
>>
>> I, working on AMD IOMMU interrupt remapping would like to ask some
>> questions on Qemu IOAPIC id.
>>
>> I currently have a problem in that Linux AMD IOMMU driver expects at least
>> one IOAPIC on the chipset to have ID 0xa0 while Qemu IOAPIC id is always 0.
>> I am faced with two options
>>    -Change IOAPIC id so that it always matches what Linux AMD IOMMU driver
>> expects. In this case I am not sure which problems I could encounter with
>> this approach.
>>    -Have a way to configure IOAPIC ID based on emulated IOMMU.
>>
>> I prefer the first approach since it's quite simple but would like to hear
>> other opinions regarding this issue.
>>
>> David.
>>
> 
> Ping on this question by David:
> 
> If we can consolidate over AMD's IOAPIC ID now, before releasing 4.7
> (later changes would be guest visible), we won't have to add a special
> case for the AMD IOMMU later.
> 

OK, seems like we were on the wrong track here: We can perfectly go for
different IDs (virtual BDFs) for Intel vs. AMD IOMMU and leave things in
QEMU as they are already. There is no overlap in the code, at most in
the namespace of Q35_PSEUDO_* - but then again, an AMD IOMMU doesn't
belong to the Q35 chipset anyway, only inside QEMU until we have an AMD
chipset model.

Jan



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 181 bytes --]