Re: [Xen-devel] [PATCH V2] x86/AMD-Vi: Add additional check for invalid special->handle

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 12.09.13 at 00:31, Suravee Suthikulpanit <suravee.suthikulpanit@xxxxxxx> 
>>> wrote:
> On 9/5/2013 2:14 AM, Jan Beulich wrote:
>>>>> On 05.09.13 at 00:48, Suravee Suthikulpanit 
>>>>> <suravee.suthikulpanit@xxxxxxx> wrote:
>>> After looking through the examples above, I still think we only need the
>>> "ivrs_ioapic[<sbdf>]=<id>" version.
>> Perhaps, but then could you talk to your Linux side colleagues to
>> find out why they picked (only) the other alternative? I'd really
>> like to keep such workarounds largely in sync (a superset is fine,
>> but a subset, not to speak of a disjoint set, are not) with Linux...
>>
> At this point, I think it's mainly syntax.  Since the implementation is 
> different.  If you want to keep the syntax the same (i.e. 
> ivrs_ioapic[id]=<sdbf>), we can do that.  I don't think Linux tries to 
> interpret as "pin id" or "pin sdbf".  All it does is just ties both 
> values together.

I continue to disagree: add_special_device() matches on only ->id,
i.e. there's no replacing of IVRS entries with a bad ID. The
difference is that they simply add a device with the provided
attributes, whereas we truly use the command line setting as an
override for an existing entry. By adding the inverse syntax, we'd
provide more fine grained control to the use than Linux does.

I guess I should try to get to send you a prototype for this.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel