Le Sun, Sep 10, 2023 at 11:02:09PM +0000, Joel Fernandes a écrit : > On Sun, Sep 10, 2023 at 10:25:43PM +0200, Frederic Weisbecker wrote: > > Le Sun, Sep 10, 2023 at 03:21:33AM -0700, Paul E. McKenney a écrit : > > > On Sat, Sep 09, 2023 at 10:28:36PM -0400, Joel Fernandes wrote: > > > > Hi Paul, Frederic, ever since this recent patch got merged into all > > > > recent stables: > > > > https://lore.kernel.org/all/20230818200757.1808398-1-paul.gortmaker@xxxxxxxxxxxxx/ > > > > .. > > > > > > > > .. I am seeing this happen a lot now causing rcutorture failures: > > > > "NOHZ tick-stop error: local softirq work is pending" > > > > > > > > It looks like the "tick-stop error" as local_bh_blocked always returns > > > > false on !PREEMPT_RT, so that patch itself seems correct. > > > > > > > > However, it appears the string in console-badness.sh is different: > > > > > > > > console-badness.sh:grep -v 'NOHZ tick-stop error: Non-RCU local > > > > softirq work is pending, handler' > > > > > > > > So the string in the script needs an update? > > > > > > Looks like it, though I believe that there is a fix in the offing, > > > at which point that "grep -v" should just be removed. Depending on > > > timing, of course. > > > > Yep I have a pending fix for that, will post soonish. The grep will still > > make sense afterward though. > > I was wondering why the grep -v will still make sense if the pending fix > works. Or is there another legitimate reason for the string to appear? Because my fix will only take out (hr)timer softirqs out of the report. The others will still be considered buggy if pending while the CPU goes down. Anyway, cooking the timer patchset to post. Thanks.
