Re: [Xen-devel] [PATCH 2/2] x86/iommu: avoid mapping the APIC configuration space for hwdom

[Jan Beulich <JBeulich@xxxxxxxx>]

On 25.07.2019 15:34, Roger Pau Monné  wrote:
> On Thu, Jul 25, 2019 at 12:47:01PM +0000, Jan Beulich wrote:
>> On 24.07.2019 11:43, Jan Beulich wrote:
>>> On 23.07.2019 18:45, Andrew Cooper wrote:
>>>> On 23/07/2019 17:09, Jan Beulich wrote:
>>>>> On 23.07.2019 17:48, Roger Pau Monne wrote:
>>>>>> Current code only prevents mapping the io-apic page into the guest
>>>>>> physical memory map. Expand the range to be 0xFECx_xxxx as described
>>>>>> in the Intel 3 Series Chipset Datasheet section 3.3.1 "APIC
>>>>>> Configuration Space (FEC0_0000h–FECF_FFFFh)".
>>>>>>
>>>>>> AMD also lists this address range in the AMD SR5690 Databook, section
>>>>>> 2.4.2 "Non-SB IOAPIC Support".
>>>>> But that's chipset specific. I don't think we can blindly assume
>>>>> this range.
>>>>
>>>> The IO-APIC has always lived in that region since its introduction, and
>>>> the location isn't even configurable on newer chipsets (If I've read the
>>>> SAD routing rules in Skylake correctly.  All that can be configured is
>>>> multiple IO-APICs being mapped adjacent to each other.)
>>>
>>> I'm pretty sure I've seen IO-APICs outside that range.
>>
>>   From my AMD Fam15 system:
>>
>> <7>ACPI: Local APIC address 0xfee00000
>> <6>IOAPIC[0]: apic_id 0, version 33, address 0xfec00000, GSI 0-23
>> <6>IOAPIC[1]: apic_id 1, version 33, address 0xc8000000, GSI 24-55
> 
> Hm, I guess the only option is to then blacklist the proposed range
> plus any of the pages of the io-apics on the system. I can send a new
> version without dropping the current io-apic blacklisting, but then
> I'm not sure there's much value in adding the FEC0_0000h–FECF_FFFFh
> range.

Neither am I, hence my initial reaction. I'm surprised you don't see
much value in there anymore - after all it's quite a bit larger an
area that gets guarded against getting populated, as we're unlikely
to see many systems with this space fully (or even just mostly) used
by many, many IO-APICs. As said, I'd be fine acking the patch with
the loop left in place, and with the description refined.

Jan
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel