Re: [PATCH] Bypass mask bit of msix entry in xen

[DuanZhenzhong <zhenzhong.duan@oracle.com>]

Jan Beulich wrote:
>>>> On 21.03.13 at 11:50, Zhenzhong Duan <zhenzhong.duan@oracle.com> wrote:
>>>>         
>> On 2013-03-21 18:12, Jan Beulich wrote:
>>     
>>> Furthermore, without explanation I also don't see how the mask
>>> bit is now being dealt with properly: From an abstract pov you'd
>>> need to merge ("or") the guest-requested mask bit state with what
>>> Xen needs for its own purposes. I don't see anything like that here
>>> or in the qemu side patch.
>>>       
>> Right, I didn't consider combine xen's masking with guest's.
>> My patch just target making irq affinity in old hvm guest work and 
>> sometimes no irq handler panic.
>> But I would like to know what's the result if guest's mask setting 
>> doesn't pass to device,
>> interrupt loss or something else? I see current implemention didn't 
>> passthrough mask bit too.
>>     
>
> The effect of this is unknown, as it depends on the particular
> driver behavior and assumptions. Loss of interrupts, however,
> doesn't appear to be problematic here; instead, interrupts
> getting delivered when the driver doesn't expect them seems
> to be the much bigger problem.
>
> Jan
>
>   
It's hard to merge guest requested mask with xen's.
Because xen lack a irq_desc.depth as kernel for nested irq disable and 
enable.

-- 
Regards
zhenzhong
--
Oracle Building, No.24 Building, Zhongguancun Software Park
Haidian District, Beijing 100193, China