On Mon, Feb 08, 2021 at 08:08:17PM +0100, Klaus Jensen wrote: > On Feb 9 03:59, Keith Busch wrote: > > This whole implementation would be much simpler with the synchronous > > blk_flush() routine instead of the AIO equivalent. This is not really a > > performant feature, so I don't think it's critical to get these > > operations happening in parallel. What do you think? > > It would definitely be simpler, but I believe that if there is a lot to > flush, then we won't just block the nvme device. We are holding the Big > QEMU Lock and will block most other devices as well.
Hm, I feel like you may have told me this same explanation for a different patch. :) Okay, I'm convinced: this is the way.