On Mon, Jul 1, 2019 at 11:14 AM Poul-Henning Kamp <phk_at_phk.freebsd.dk> wrote: > -------- > In message < > CANCZdfoFBvmxPtnEL4GOqXTvp6Zd-xrtja4rmUO1rAcy0JdeSw_at_mail.gmail.com>, > Warner Losh writes: > > >The only issue, really, is that this timeout is a busy loop and there may > >be I/O bus contention introduced on these systems. > > Does it have to be a busy loop for the entire duration ? > > Spin for the median, timeout+poll for the rest of the time ? > That's a good suggestion. I'd be inclined to spin for 1 tick or so, then do a timeout per tick after that (eg, shift from DELAY to pause(1)). It won't be super accurate or high performance, but when the devices are slow, that would add only a little extra time. Ideally, that's what we'd do. In the short term, bumping the timeout wouldn't be horrible. WarnerReceived on Mon Jul 01 2019 - 15:25:24 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:41:21 UTC