On Wed, Oct 16, 2024 at 05:40:48PM +0200, Sebastian Andrzej Siewior wrote: > On 2024-10-15 15:47:20 [+0800], kernel test robot wrote: > > Hello, > Hi, > > it took me a while to reproduce this because it does not trigger without > the ltp userland and this is not downloaded properly so I had to > workaround it. However… > > > [ 98.006999][ T543] busted-torture: rtc: 00000000639e821e ver: 6796 tfle: 0 rta: 6796 rtaf: 0 rtf: 6787 rtmbe: 1 rtmbkf: 0/0 rtbe: 0 rtbke: 0 rtbf: 0 rtb: 0 nt: 89 barrier: 0/0:0 read-exits: 64 nocb-toggles: 0:0 > > [ 98.017386][ T543] busted-torture: !!! > > [ 98.017662][ T543] ------------[ cut here ]------------ > > [ 98.019330][ T543] WARNING: CPU: 0 PID: 543 at kernel/rcu/rcutorture.c:2258 rcu_torture_stats_print+0x24c/0x610 [rcutorture] > > [ 98.021831][ T543] Modules linked in: rcutorture torture > … > > [ 98.237025][ T543] ------------[ cut here ]------------ > > [ 98.238052][ T543] WARNING: CPU: 0 PID: 543 at kernel/rcu/rcutorture.c:2263 rcu_torture_stats_print+0x373/0x610 [rcutorture] > > [ 98.240000][ T543] Modules linked in: rcutorture torture > … > > It took me a while to figure out that this test is using > rcu_busted_torture_deferred_free() which in turn invokes the callback > before the grace period. Buh. > So it looks like LAZY preempt triggers this more reliably than the > normal preempt version… Agreed! Please don't run any rcutorture scenario whose name contains "BUSTED" or "busted" unless you are trying to test rcutorture's ability to find broken RCU implementations. Thanx, Paul
