RE: [PATCH] Change spec of callback IRQ forPV-on-HVMonIA64

RE: [PATCH] Change spec of callback IRQ forPV-on-HVMonIA64

[Tian, Kevin]

>From: Tian, Kevin
>Sent: 2006年11月21日 17:14
>>From: Doi.Tsunehisa@jp.fujitsu.com
>>Sent: 2006年11月21日 16:59
>>
>>  In my experience, the other OS has device driver interface to notice
>>GSI for the device. It tells the vector to the interrupt controller
>>in kenrel code, not device driver. So, we can't know the vector, I think.
>>
>
>OK, I now understand. Other OS tells xen GSI number as callback_irq,
>and then xen look up virtual interrupt controller to find appropriate
>vector by GSI number. This is the way to go. :-)
>
>Thanks a lot,
>Kevin
>

Keir mentioned in another thread that he will change qemu to support 
more GSIs (>16) soon, and this will become an issue with your patch 
since you're only looking at isa_irq_to_vector. Is there any common 
approach exported by linux kernel to find vector->GSI mapping in 
driver? If not, maybe an alternative is to provide a callback_irq_type 
which indicates whether the value is a GSI or vector?

Thanks,
Kevin

Re: [PATCH] Change spec of callback IRQ forPV-on-HVMonIA64

[Doi.Tsunehisa]

Hi Kevin,

  Thank you.

You (kevin.tian) said:
> Keir mentioned in another thread that he will change qemu to support 
> more GSIs (>16) soon, and this will become an issue with your patch 
> since you're only looking at isa_irq_to_vector. Is there any common 
> approach exported by linux kernel to find vector->GSI mapping in 
> driver? If not, maybe an alternative is to provide a callback_irq_type 
> which indicates whether the value is a GSI or vector?

  I worry about the chage, too.

  He said "This will change today.". So, I'll care the patch.

Thanks,
- Tsunehisa Doi

Re: [PATCH] Change spec of callback IRQ forPV-on-HVMonIA64

[Keir Fraser]

On 21/11/06 10:56, "Tian, Kevin" <kevin.tian@intel.com> wrote:

>> OK, I now understand. Other OS tells xen GSI number as callback_irq,
>> and then xen look up virtual interrupt controller to find appropriate
>> vector by GSI number. This is the way to go. :-)
>> 
>> Thanks a lot,
>> Kevin
>> 
> 
> Keir mentioned in another thread that he will change qemu to support
> more GSIs (>16) soon, and this will become an issue with your patch
> since you're only looking at isa_irq_to_vector. Is there any common
> approach exported by linux kernel to find vector->GSI mapping in
> driver? If not, maybe an alternative is to provide a callback_irq_type
> which indicates whether the value is a GSI or vector?

We plan to add a callback_vector type to the Xen interface for x86 HVM, as
we will deliver event-channel notifications via an MSI-style poke directly
into the IRR of the appropriate local APIC. We'll continue to support
callback_irq type for compatibility.

 -- Keir

RE: [PATCH] Change spec of callback IRQ forPV-on-HVMonIA64

[Tian, Kevin]

>From: Keir Fraser [mailto:keir@xensource.com]
>Sent: 2006年11月21日 19:33
>>
>> Keir mentioned in another thread that he will change qemu to support
>> more GSIs (>16) soon, and this will become an issue with your patch
>> since you're only looking at isa_irq_to_vector. Is there any common
>> approach exported by linux kernel to find vector->GSI mapping in
>> driver? If not, maybe an alternative is to provide a callback_irq_type
>> which indicates whether the value is a GSI or vector?
>
>We plan to add a callback_vector type to the Xen interface for x86 HVM,
>as
>we will deliver event-channel notifications via an MSI-style poke directly
>into the IRR of the appropriate local APIC. We'll continue to support
>callback_irq type for compatibility.
>
> -- Keir

Thanks for info, and this approach seems to be the one promising. :-)

Thanks,
Kevin