On Fri, Jun 28, 2019 at 10:30:11AM -0700, Paul E. McKenney wrote: > On Fri, Jun 28, 2019 at 12:45:59PM -0400, Joel Fernandes wrote: > > On Fri, Jun 28, 2019 at 12:40:08PM -0400, Joel Fernandes wrote: > > > On Thu, Jun 27, 2019 at 11:41:07AM -0700, Paul E. McKenney wrote: > > > [snip] > > > > > > > And we should document this somewhere for future sanity preservation > > > > > > > :-D > > > > > > > > > > > > Or adjust the code and requirements to make it more sane, if feasible. > > > > > > > > > > > > My current (probably wildly unreliable) guess that the conditions in > > > > > > rcu_read_unlock_special() need adjusting. I was assuming that in_irq() > > > > > > implies a hardirq context, in other words that in_irq() would return > > > > > > false from a threaded interrupt handler. If in_irq() instead returns > > > > > > true from within a threaded interrupt handler, then this code in > > > > > > rcu_read_unlock_special() needs fixing: > > > > > > > > > > > > if ((exp || in_irq()) && irqs_were_disabled && use_softirq && > > > > > > (in_irq() || !t->rcu_read_unlock_special.b.deferred_qs)) { > > > > > > // Using softirq, safe to awaken, and we get > > > > > > // no help from enabling irqs, unlike bh/preempt. > > > > > > raise_softirq_irqoff(RCU_SOFTIRQ); > > > > > > > > > > > > The fix would be replacing the calls to in_irq() with something that > > > > > > returns true only if called from within a hardirq context. > > > > > > Thoughts? > > > > > > > > > > I am not sure if this will fix all cases though? > > > > > > > > > > I think the crux of the problem is doing a recursive wake up. The threaded > > > > > IRQ probably just happens to be causing it here, it seems to me this problem > > > > > can also occur on a non-threaded irq system (say current_reader() in your > > > > > example executed in a scheduler path in process-context and not from an > > > > > interrupt). Is that not possible? > > > > > > > > In the non-threaded case, invoking raise_softirq*() from hardirq context > > > > just sets a bit in a per-CPU variable. Now, to Sebastian's point, we > > > > are only sort of in hardirq context in this case due to being called > > > > from irq_exit(), but the failure we are seeing might well be a ways > > > > downstream of the actual root-cause bug. > > > > > > Hi Paul, > > > I was talking about calling of rcu_read_unlock_special from a normal process > > > context from the scheduler. > > > > > > In the below traces, it shows that only the PREEMPT_MASK offset is set at the > > > time of the issue. Both HARD AND SOFT IRQ masks are not enabled, which means > > > the lock up is from a normal process context. > > > > > > I think I finally understood why the issue shows up only with threadirqs in > > > my setup. If I build x86_64_defconfig, the CONFIG_IRQ_FORCED_THREADING=y > > > option is set. And booting this with threadirqs, it always tries to > > > wakeup_ksoftirqd in invoke_softirq. > > > > > > I believe what happens is, at an in-opportune time when the .blocked field is > > > set for the preempted task, an interrupt is received. This timing is quite in > > > auspicious because t->rcu_read_unlock_special just happens to have its > > > .blocked field set even though it is not in a reader-section. > > Thank you for tracing through this! > > > I believe the .blocked field remains set even though we are not any more in a > > reader section because of deferred processing of the blocked lists that you > > mentioned yesterday. > > That can indeed happen. However, in current -rcu, that would mean > that .deferred_qs is also set, which (if in_irq()) would prevent > the raise_softirq_irqsoff() from being invoked. Which was why I was > asking the questions about whether in_irq() returns true within threaded > interrupts yesterday. If it does, I need to find if there is some way > of determining whether rcu_read_unlock_special() is being called from > a threaded interrupt in order to suppress the call to raise_softirq() > in that case. > > But which version of the kernel are you using here? Current -rcu? > v5.2-rc1? Something else? And if this turns out to be current -rcu, and if there is no reasonable way for rcu_read_unlock_special() to know if it is being invoked from within a threaded interrupt handler, then the patch below would be one way out. Thanx, Paul ------------------------------------------------------------------------ diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h index 82c925df1d92..5140e792c1c2 100644 --- a/kernel/rcu/tree_plugin.h +++ b/kernel/rcu/tree_plugin.h @@ -625,7 +625,7 @@ static void rcu_read_unlock_special(struct task_struct *t) tick_nohz_full_cpu(rdp->cpu); // Need to defer quiescent state until everything is enabled. if ((exp || in_irq()) && irqs_were_disabled && use_softirq && - (in_irq() || !t->rcu_read_unlock_special.b.deferred_qs)) { + !t->rcu_read_unlock_special.b.deferred_qs) { // Using softirq, safe to awaken, and we get // no help from enabling irqs, unlike bh/preempt. raise_softirq_irqoff(RCU_SOFTIRQ);