On Tue, 21 May 2019 16:47:45 -0400 Eric Farman <far...@linux.ibm.com> wrote:
> On 5/21/19 12:32 PM, Cornelia Huck wrote: > > Why mostly? I'm not sure yet whether we handling multiple requests for > > passthrough devices correctly yet (virtual should be fine.) > > > > Start vs. (start|halt|clear) is fine, as the code checks whether > > something is already pending before poking the kernel interface. > > Likewise, halt vs. (start|halt|clear) is fine, as the code checks for > > halt or clear and start and halt use different regions. The problematic > > one is clear, as that's something that's always supposed to go through. > > Probably fine if clear should always "win", but I need to think some > > more about that. > > I suspect you are right, because of the check on the halt side, and > considering that the clear is the biggest recovery action we have. So > this does seem like things are okay. I'll ponder this overnight and > finish my review tomorrow. Ok, what's the verdict here? :) (I'm trying to clean up my pending stuff :)