Instead of turning all irq_work requests into lazy ones on -rt, just
move their execution from hard into soft-irq context.

This resolves deadlocks of ftrace which will queue work from arbitrary
contexts, including those that have locks held that are needed for
raising a soft-irq.

Signed-off-by: Jan Kiszka <jan.kis...@siemens.com>
---

Second try, looks much better so far. And it also removes my concerns
regarding other potential cases besides ftrace.

 kernel/irq_work.c | 26 ++++++++++----------------
 1 file changed, 10 insertions(+), 16 deletions(-)

diff --git a/kernel/irq_work.c b/kernel/irq_work.c
index 9dda38a..3f6ffcd 100644
--- a/kernel/irq_work.c
+++ b/kernel/irq_work.c
@@ -85,12 +85,9 @@ bool irq_work_queue_on(struct irq_work *work, int cpu)
                raise_irqwork = llist_add(&work->llnode,
                                          &per_cpu(hirq_work_list, cpu));
        else
-               raise_irqwork = llist_add(&work->llnode,
-                                         &per_cpu(lazy_list, cpu));
-#else
+#endif
                raise_irqwork = llist_add(&work->llnode,
                                          &per_cpu(raised_list, cpu));
-#endif
 
        if (raise_irqwork)
                arch_send_call_function_single_ipi(cpu);
@@ -114,21 +111,20 @@ bool irq_work_queue(struct irq_work *work)
        if (work->flags & IRQ_WORK_HARD_IRQ) {
                if (llist_add(&work->llnode, this_cpu_ptr(&hirq_work_list)))
                        arch_irq_work_raise();
-       } else {
+       } else
+#endif
+       if (work->flags & IRQ_WORK_LAZY) {
                if (llist_add(&work->llnode, this_cpu_ptr(&lazy_list)) &&
                    tick_nohz_tick_stopped())
+#ifdef CONFIG_PREEMPT_RT_FULL
                        raise_softirq(TIMER_SOFTIRQ);
-       }
 #else
-       if (work->flags & IRQ_WORK_LAZY) {
-               if (llist_add(&work->llnode, this_cpu_ptr(&lazy_list)) &&
-                   tick_nohz_tick_stopped())
                        arch_irq_work_raise();
+#endif
        } else {
                if (llist_add(&work->llnode, this_cpu_ptr(&raised_list)))
                        arch_irq_work_raise();
        }
-#endif
 
        preempt_enable();
 
@@ -202,6 +198,8 @@ void irq_work_run(void)
 {
 #ifdef CONFIG_PREEMPT_RT_FULL
        irq_work_run_list(this_cpu_ptr(&hirq_work_list));
+       if (!llist_empty(this_cpu_ptr(&raised_list)))
+               raise_softirq(TIMER_SOFTIRQ);
 #else
        irq_work_run_list(this_cpu_ptr(&raised_list));
        irq_work_run_list(this_cpu_ptr(&lazy_list));
@@ -211,15 +209,11 @@ EXPORT_SYMBOL_GPL(irq_work_run);
 
 void irq_work_tick(void)
 {
-#ifdef CONFIG_PREEMPT_RT_FULL
-       irq_work_run_list(this_cpu_ptr(&lazy_list));
-#else
-       struct llist_head *raised = &__get_cpu_var(raised_list);
+       struct llist_head *raised = this_cpu_ptr(&raised_list);
 
        if (!llist_empty(raised) && !arch_irq_work_has_interrupt())
                irq_work_run_list(raised);
-       irq_work_run_list(&__get_cpu_var(lazy_list));
-#endif
+       irq_work_run_list(this_cpu_ptr(&lazy_list));
 }
 
 /*
-- 
2.1.4
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Reply via email to