On Fri, 21 Dec 2018 12:23:32 +0100 Cornelia Huck <cohuck@xxxxxxxxxx> wrote: [..] > > You've really lost me here :( I fear you're criticizing something I > don't want to implement; I'll write some code, that should make things > much easier to discuss. > Nod. > TBH, I have no idea how this will scale to many vfio-ccw devices. > > > [..] > > As long as you return -EAGAIN we are fine. But AFAIU you proposed to > > do that until the I/O is submitted to the HW subchannel via ssch(). But > > that is not the case I'm talking about here. We have already translated > > the channel program for the first request, submitted it via ssch() and > > are awaiting an interrupt that tells us the I/O is done. While waiting > > for this interrupt we get a new ssch request. I understood, you don't > > want to give -EAGAIN for this one, but make the ssch decide. The problem > > is you still need the old translated channel program for the interrupt > > handling, and at the same time you need the new channel program > > translated as well, before doing the ssch for it in the host. > > Why? You're not doing anything with that second ssch at all, it returns > before translation is started. > OK apparently I misunderstood something -- it was a long and twisty discussion. Looking forward to the code ;). [..] > > I'll try to post something in January. Have a nice holiday break :) > Don't feel pressured. :) Have nice holidays as well! Regards, Halil
