On Tue, Feb 28, 2023 at 12:33 AM Paul E. McKenney <paulmck@xxxxxxxxxx> wrote: > > On Mon, Feb 27, 2023 at 11:16:26AM -0500, Steven Rostedt wrote: > > On Tue, 28 Feb 2023 00:11:51 +0800 > > Zhouyi Zhou <zhouzhouyi@xxxxxxxxx> wrote: > > > > > > Yes certainly, the rcu_head is allocated on the caller side so it > > > > could have been trampled while the callback was still in flight. > > > Thank you all for your guidance, I learned a lot during this process > > > > > > > > > OR it could be a bug with RCU if the synchronize_rcu() ended before the > > > > > grace periods have finished. > > > Thanks again. > > > > > > By the way, the syzkaller on my local machine has been running for 8 > > > hours, only three bugs reported[1][2][3], but they don't seem to be > > > related to Sanan's original report. > > > Maybe there are some configuration mismatches between us.The test > > > continues, I will report to you once I have any new discovery. > > > > Note, the above races (either bug, the one that tramples on something in > > RCU flight, or a synchronize_sched() returning early) may be extremely hard > > to hit. It could have been the planets were lined up just right to hit the > > bug, and won't happen for another 27,000 years. > > Which turns into once per week or two across a million-system fleet. ;-) > > Not that I know of any fleets running syzkaller... My syzkaller has been running for 24 hours, the bug can't be reproduced. Yes, the above races are extremely hard to hit. I learned a lot during the process ;-) Please inform me if there are any more clues. Thank you all for your guidance ;-) Zhouyi > > Thanx, Paul
