On Sun, Jan 04, 2026 at 03:36:42PM -0500, Joel Fernandes wrote: > During some recent expedited RCU testing with rcutorture, I noticed RCU > stalls when using ftrace_dump() to analyze traces. The issue that since > ftrace_dump() can take a long time, any accumulated issue can show up > just after the stall is unsupressed. Due to this a stall can be > detected and reported. > > The problem is something like this: > > 1. rcu_ftrace_dump_stall_suppress() sets suppression value. > > 2. ftrace_dump() runs for a long time. > > 3. rcu_ftrace_dump_stall_unsuppress() clears suppression. > > 4. Immediately later, stall is detected. > > Fix this by calling rcu_cpu_stall_reset() when clearing suppression. > This resets jiffies_stall to ULONG_MAX, preventing the issue. > > Tested with gp_exp=true enabled with rcutorture and ftrace_dump() and no > stall is reported, with only clean traces in the output. > > Signed-off-by: Joel Fernandes <[email protected]>
Good catch! Reviewed-by: Paul E. McKenney <[email protected]> > --- > This is a non-urgent fix and could go in either this or the next merge > window (Boqun's call). > > kernel/rcu/rcu.h | 5 ++++- > kernel/rcu/tree_stall.h | 1 + > 2 files changed, 5 insertions(+), 1 deletion(-) > > diff --git a/kernel/rcu/rcu.h b/kernel/rcu/rcu.h > index 9cf01832a6c3..87fcf528b9ac 100644 > --- a/kernel/rcu/rcu.h > +++ b/kernel/rcu/rcu.h > @@ -288,6 +288,7 @@ extern int rcu_cpu_stall_cputime; > extern bool rcu_exp_stall_task_details __read_mostly; > int rcu_jiffies_till_stall_check(void); > int rcu_exp_jiffies_till_stall_check(void); > +void rcu_cpu_stall_reset(void); > > static inline bool rcu_stall_is_suppressed(void) > { > @@ -302,8 +303,10 @@ do { \ > > #define rcu_ftrace_dump_stall_unsuppress() \ > do { \ > - if (rcu_cpu_stall_suppress == 3) \ > + if (rcu_cpu_stall_suppress == 3) { \ > rcu_cpu_stall_suppress = 0; \ > + rcu_cpu_stall_reset(); \ > + } \ > } while (0) > > #else /* #endif #ifdef CONFIG_RCU_STALL_COMMON */ > diff --git a/kernel/rcu/tree_stall.h b/kernel/rcu/tree_stall.h > index b67532cb8770..9a036fc27fed 100644 > --- a/kernel/rcu/tree_stall.h > +++ b/kernel/rcu/tree_stall.h > @@ -192,6 +192,7 @@ void rcu_cpu_stall_reset(void) > WRITE_ONCE(rcu_state.nr_fqs_jiffies_stall, 3); > WRITE_ONCE(rcu_state.jiffies_stall, ULONG_MAX); > } > +EXPORT_SYMBOL_GPL(rcu_cpu_stall_reset); > > > ////////////////////////////////////////////////////////////////////////////// > // > -- > 2.34.1 >
