Re: [Qemu-devel] [PATCH v5 11/11] pci: Convert msi_init() to Error and fix callers to check it

[Cao jin <caoj.fnst@cn.fujitsu.com>]

On 05/23/2016 06:06 PM, Marcel Apfelbaum wrote:
> On 05/17/2016 01:08 PM, Cao jin wrote:
>>
>>
>> On 05/15/2016 09:41 PM, Marcel Apfelbaum wrote:
>>
>>
>> That is why I am not quite sure about this device, msi has a relation
>> with shpc.
>>
>>  From its previous behaviour, it can be seen that it don`t treat 'on'
>> as 'auto'.(I am not sure why it is different with others)
>>
>> Its previous behaviour treat msi_init`s failure as fatal, and lead to
>> device creation fail. According to Markus`s comments(if I understand
>> him right), this device has no non-msi variants.
>
> Actually it has. The bridge needs msi for the SHPC controller, as you said.
> If you look into shpc_interrupt_update function (hw/pci/shpc.c) you can
> see it can fall
> back to legacy interrupts.
>
> The only complication here is that msi is needed only if shpc is present.
> So maybe having msi=on/off/auto is OK.
>
> msi=auto => if shpc not present or msi broken results in msi = off, else
> msi = on
> msi=on => fails if shpc present and msi broken
> msi=off => use legacy interrupts if shpc is present
>
>
> Basically the msi flag has no meaning if shpc not present.
>

Thanks for the hint, v6 is on the way:)

-- 
Yours Sincerely,

Cao jin