Re: [libvirt] [PATCH 2/4] xen: Fix logic bug in xenDaemon*DeviceFlags
On 10/04/2010 06:01 AM, Jiri Denemark wrote:
> It seems like we need a 12-way table (in fact, this is pretty
much what
> I ended up resorting to with my vcpus stuff). Here's my shot at it,
> from reading the comments (but not actually testing it); once we fix
> this attempt to an actual table, then I can answer whether the code
> matches the table.
>
> _LIVE _CONFIG _LIVE|_CONFIG
> xen 2,running good unsupported unsupported
> xen 2,inactive error good error or silently good
> xen 3,running good good good
> xen 3,inactive error good error or silently good
Yeah, this is probably a good idea however we shouldn't forget about _CURRENT
which is an equivalent of either _CONFIG or _LIVE depending on current state
of the guest.
LI: _LIVE
CO: _CONFIG
CU: _CURRENT = active ? _LIVE : _CONFIG
g: good
i: OPERATION_INVALID (in the wrong state)
u: CONFIG_UNSUPPORTED (never possible)
LI CO LI|CO CU LI|CU CO|CU LI|CO|CU
xen 2,running g u u g g u u
xen 2,inactive i u u i i u u
xen 3,running u u g u u g g
xen 3,inactive u u i u i u i
The logic might be rendered easier by starting the function with:
if (!flags)
return -1;
if (flags & _CURRENT)
flags |= running ? _LIVE : _CONFIG;
so that the rest of the function need only operate on _LIVE and _CONFIG
(reducing 2^3==8 flag combinations down to 3).
But, as written, and with your extra context explanations, ACK to patch 2.
--
Eric Blake eblake(a)redhat.com +1-801-349-2682
Libvirt virtualization library http://libvirt.org