Re: [Qemu-devel] address order of virtio-mmio devices

[Laszlo Ersek]

On 01/30/15 11:48, Daniel P. Berrange wrote:
> On Fri, Jan 30, 2015 at 10:29:46AM +0000, Peter Maydell wrote:
>> On 30 January 2015 at 09:54, Daniel P. Berrange <address@hidden> wrote:
>>> While it is clear there is no solution that works correctly with all
>>> kernels, I hate to think that we're going to stick with an ordering
>>> that is clearly wrong for modern kernels, forever going forward. The
>>> aarch64 world is only just starting out, so on balance I think we
>>> should optimize for the future rather than the past, since that gives
>>> right behaviour for orders of magnitude more people in the long term.
>>
>> Yeah, I agree it's awkward. But I hate breaking people's
>> working setups, and we have no guarantee the kernel won't
>> change again in the future.
>>
>> You could try asking the kernel folk to revert that patch on
>> the basis that it breaks things...
> 
> Might be worth a shot - the patch is only a month old. Or at least do a
> followup patch to put the ordering back the way it was, rather than plain
> revert

Please note that (as far as I understand) the patch that I referenced is
indeed very new, it's not part of v3.18, but the reversal can easily be
seen with v3.18. In other words, the kernel patch I referenced
introduces no functional change, it just reorganizes stuff in the kernel
(AIUI), with the benefit of killing a superfluous field.

The reason I referenced it because its *commit message* gives good
background. If we really wanted to find the kernel change that reversed
the traversal, we'd have to talk to Grant and/or bisect the kernel.

Thanks
Laszlo

> Long term though it will be much better of AArch64 would just do PCI
> instead of MMIO bus. Then we have proper device addressing which we
> can control in a predictable manner that will be stable across hotplug
> and unplug and migration.  I hear there's work on PCI for AArch64 but
> is there a near term ETA yet ?
> 
> Regards,
> Daniel
>