On 16/03/2017 01:46, John Snow wrote: > > - Does the presence of blk->quiesce_counter relieve the burden of needing > blk->public.io_limits_disabled? I could probably eliminate this counter > entirely and just spy on the root node's quiescent state at key moments > instead. I am confident I'm traipsing on delicate drain semantics.
Tricky question. I believe io_limits_disabled is a bit of a hack. Certainly you could get rid of disable_external now that we have drained_begin/drained_end, but it would be a separate patch. > - Should I treat the separation of a quisced BDS/BB as a drained_end request > as an analogue to how blk_insert_bs (in this patch) handles this? I think so. Paolo
