Re: [Qemu-devel] virtio-pci msix error

[Peter Maydell]

On 20 May 2015 at 12:55, Fam Zheng <address@hidden> wrote:
> On Wed, 05/20 11:41, Peter Maydell wrote:
>> On 20 May 2015 at 10:58, Michael S. Tsirkin <address@hidden> wrote:
>> > On Tue, May 19, 2015 at 09:11:26PM +0100, Peter Maydell wrote:
>> >> On 19 May 2015 at 20:16, Michael S. Tsirkin <address@hidden> wrote:
>> >> > On Tue, May 19, 2015 at 08:54:36AM -0700, Richard Henderson wrote:
>> >> >> Re patch c7ff5482.  What's the point of this error?
>> >> >> It's going to always appear for older targets that
>> >> >> predate such new fangled things as msix.
>> >> >>
>> >> >> Obviously the patch has been there a while, and it's
>> >> >> not actually causing any problems, but today it got
>> >> >> on my nerves...
>> >>
>> >> > So don't specify nvectors > 0 for these platforms then?
>> >>
>> >> How do you do that? I did a quick 'git grep' for nvectors
>> >> and none of the hits are in platform-dependent code...
>> >>
>> >> Why can't the virtio-pci device automatically detect
>> >> whether the PCI bus it's plugged into supports MSIx
>> >> and just do the right thing?
>>
>> > I mean why does *user* specify nvectors > 0?
>>
>> The user isn't specifying nvectors at all. That's why
>> the message is annoying...
>
> So I think it's better to fix the default for old targets?

What default? No platform or PCI controller code specifies any
default value for nvectors, and the user doesn't specify a
value for nvectors.

The only thing that tries to specify a value for nvectors
is the virtio-*-pci device itself, which it then pointlessly
complains that it can't set.

-- PMM