[PATCH] VFIO: platform: AMD xgbe reset module

[Christoffer Dall]

On Thu, Oct 15, 2015 at 5:49 PM, Arnd Bergmann <arnd at arndb.de> wrote:
> On Thursday 15 October 2015 17:03:21 Christoffer Dall wrote:
>> On Thu, Oct 15, 2015 at 04:55:13PM +0200, Arnd Bergmann wrote:
>> > On Thursday 15 October 2015 16:46:09 Eric Auger wrote:
>> > > >
>> > > > This is where we'd need a little more changes for this approach. Instead
>> > > > of unbinding the device from its driver, the idea would be that the
>> > > > driver remains bound as far as the driver model is concerned, but
>> > > > it would be in a quiescent state where no other subsystem interacts with
>> > > > it (i.e. it gets unregistered from networking core or whichever it uses).
>> > >
>> > > Currently we use the same mechanism as for PCI, ie. unbind the native
>> > > driver and then bind VFIO platform driver in its place. Don't you think
>> > > changing this may be a pain for user-space tools that are designed to
>> > > work that way for PCI?
>> > >
>> > > My personal preference would be to start with your first proposal since
>> > > it looks (to me) less complex and "unknown" that the 2d approach.
>> >
>> > We certainly can't easily change from one approach to the other without
>> > breaking user expectations, so the decision needs to be made carefully.
>> >
>> > The main observation here is that platform devices are unlike PCI in this
>> > regard because they need extra per-device code. I have argued in the
>> > past that we should not reuse the "VFIO" name here because it's actually
>> > something else.
>>
>> I've adjusted to consider VFIO a general purpose framework for mapping
>> device resources into userspace/VMs, and there are certainly a lot of
>> commonality with both PCI, platform, and potentially other devices for
>> that to make sense.
>>
>>
>> > On the other hand, there are a lot of commonalities,
>> > we just have to make sure we don't try to force the code into one model
>> > that doesn't really work just to make it look more like PCI VFIO.
>> >
>>
>> But given that we now have code for platform device passthrough that
>> works in both QEMU and the kernel side and is actually useful for
>> people, is there a clear technical advantage to go back and rework thaat
>> at this point?
>>
>> Don't get me wrong, I like the idea of having a single driver bound to a
>> platform device, and then that's it, but it just feels like that
>> discussion doesn't necessarily belong in the context of a patch that
>> 'just' seeks to add reset functionality for a specific device for VFIO?
>
> Ah, this is for qemu/kvm? If there is already upstream qemu code that
> finds it easier to use this approach, it's certainly more logical to
> deal with it in my first approach than the second.

yes, indeed.

>
> I was thinking of ODP as the primary user, and that wouldn't need
> the interface to be consistent with PCI as much, because the code
> is inherently device specific there anyway.
>

I didn't think about how this applies to ODP (haven't had time to look
at ODP in any technical detail TBH), but it sounds like ODP could be
adapted to use either approach.

In any case we can always transition towards your 2nd approach in a
userspace compatible way later, if we find good reasons to.

Thanks,
-Christoffer