[Qemu-devel] Re: [PATCH 15/15] Implement the bus structure for PAPR virtual IO

[Blue Swirl]

On Sun, Feb 13, 2011 at 2:31 PM, Alexander Graf <address@hidden> wrote:
>
> On 13.02.2011, at 09:08, Blue Swirl wrote:
>
>> On Sun, Feb 13, 2011 at 1:15 AM, Benjamin Herrenschmidt
>> <address@hidden> wrote:
>>> On Sun, 2011-02-13 at 00:52 +0200, Blue Swirl wrote:
>>>> On Sat, Feb 12, 2011 at 11:00 PM, Benjamin Herrenschmidt
>>>> <address@hidden> wrote:
>>>>> On Sat, 2011-02-12 at 18:59 +0200, Blue Swirl wrote:
>>>>>>
>>>>>> Actually I don't quite understand the need for vty layer, why not use
>>>>>> the chardev here directly?
>>>>>
>>>>> I'm not sure what you mean here...
>>>>
>>>> Maybe it would be reasonable to leave h_put_term_char to spapr_hcall.c
>>>> instead of moving those to a separate file.
>>>
>>> Well, the VIO device instance gives the chardev instance which is all
>>> nicely encapsulated inside spapr-vty... Also VIO devices tend to have
>>> dedicated hcalls, not only VTY, so it makes a lot of sense to keep them
>>> close to the rest of the VIO driver they belong to don't you think ?
>>>
>>> (Actually veth does, vscsi uses the more "generic" CRQ stuff which we've
>>> added to the core VIO but you'll see that when we get those patches
>>> ready, hopefully soon).
>>
>> This is a bit of a special case, much like semihosting modes for m68k
>> or ARM, or like MOL hacks which were removed recently. From QEMU point
>> of view, the most natural way of handling this would be hypervisor
>> implemented in the guest side (for example BIOS). Then the hypervisor
>> would use normal IO (or virtio) to communicate with the host. If
>> inside QEMU, the interface of the hypervisor to the devices needs some
>> thought. We'd like to avoid ugly interfaces like vmmouse where a
>> device probes CPU registers directly or spaghetti interfaces like
>> APIC.
>
> In this case I disagree. While the "emulate real hardware" case would be to 
> have a full proprietary binary blob of firmware running in the guest that 
> would handle all the hypervisor stuff, this is not feasible. Implementing 
> PAPR in Qemu is hard (and slow) enough - doing it all emulated simply is 
> overkill for what we're trying to achieve here.
>
> The PAPR interfaces are well specified (in the PAPR spec - seems to be 
> power.org member restricted) and are the only thing you ever get to see on 
> recent POWER hardware. The real hardware interface is simply inaccessible to 
> you.

Hah, I'm sure that if sufficiently motivated bunch of cool hackers got
access to a truckload (or several, aren't these big machines?) of
POWER hardware, which they could open and brick at leisure, like it is
done with other firmware reverse engineering efforts, we'd have open
hypervisor at no time. ;-)

It's not impossible to imagine also IBM open sourcing theirs.

> It's basically the same as the S390 machine we have. On those machine we 
> simply don't have access to real hw, so emulating it is moot. All interfaces 
> that the OS sees are PV interfaces.
>
>>
>>> Actually, one thing I noticed is that the current patches David posted
>>> still have a single function with a switch/case statement for hcalls.
>>>
>>> I'm not 100% certain what David long term plans are here, but in our
>>> internal "WIP" tree, we've subsequently turned that into a mechanism
>>> where any module can call powerpc_register_hypercall() to add hcalls.
>>>
>>> So if David intends to move the "upstream candidate" tree in that
>>> direction, then naturally, the calls in spapr_hcall.c are going to
>>> disappear in favor of a pair of powerpc_register_hypercall() locally in
>>> the vty module.
>>
>> Is the interface new design, or are you implementing what is used also
>> on real HW?
>
> PAPR is the spec that defines the PV interface in use on POWER. Outside of 
> IBM, nobody can run Linux on POWER without going through phyp which is their 
> hypervisor.
>
> So this implements exactly what the OS sees on real HW :).

Right. As long as the resulting spaghetti is well contained, I'm OK
with this approach. But should ever the millionaire hacker team (or
IBM) appear with their open hypervisor, this shall make room for it.