Re: [Qemu-devel] [big lock] Discussion about the convention of device's DMA each other after breaking down biglock

[Paolo Bonzini]

Il 20/09/2012 10:09, liu ping fan ha scritto:
> >
> > No, MMIO dispatch has to discard the fine-grained lock before acquiring
> > the big lock.
>
> This will cause the device state broken, and expose device under changing 
> risk.

We do it all the time with asynchronous I/O.

It's just an invariant that fine-grain-locked devices have to obey.  You
cannot keep the same semantics you enjoyed when a single lock covered
everything.

> > If you allow
> >
> >   lock(fine)
> >   lock(coarse)
> >
> > then the (presumably higher-priority) thread that is requesting the
> > fine-grained lock must wait for the lower-priority thread that holds the
> > coarse-grained lock.  Then you get priority inversion.
>
> Which thread has higher-priority? Why does the thread with coarse lock
> have lower-priority?

Because there can be so many things happening under the coarse lock,
that just _assuming_ it has lower priority is a good thing.

On the other hand, if the device didn't need immediate servicing (which
means higher priority), you wouldn't bother taking it out of the big
QEMU lock.

Paolo