* jianchuan.wang@xxxxxxxxxxxxx | 2015-11-23 15:19:38 [+0800]:
>From: Jianchuan Wang <jianchuan.wang@xxxxxxxxxxxxx>
>
>System will hang when enabling the kernel option
>CONFIG_DEBUG_LOCKING_API_SELFTESTS and CONFIG_SCHED_CONFIG
>in the preempt-rt kernel.
>
>In the preempt-rt kernel, migrate_enable()/migrage_disable() are called
>in the spinlock and read/write lock; The lock()/unlock() aren't used
>in pairs in the selftest processing changes the migrate_disable_atomic
>so that the system will hang in the dotest(), the calltrace is :
>printk() ..-> vprintk_emit() -> migrage_disable()
>-> WARN_ON_ONCE() ..-> warn_slowpath_common() ..-> printk()
>
>For fixing it, we need save the migrate_disable_atomic before self-testing
>and restore migrate_disable_atomic after self-testing.
So this is similar to saved_preempt_count. This needs to compile with
!CONFIG_PREEMPT_RT_FULL which I believe is not the case.
>Signed-off-by: Jianchuan Wang <jianchuan.wang@xxxxxxxxxxxxx>
>---
> lib/locking-selftest.c | 13 +++++++++++++
> 1 file changed, 13 insertions(+)
>
>diff --git a/lib/locking-selftest.c b/lib/locking-selftest.c
>index b93a610..61798da 100644
>--- a/lib/locking-selftest.c
>+++ b/lib/locking-selftest.c
>@@ -997,10 +997,23 @@ static int unexpected_testcase_failures;
> static void dotest(void (*testcase_fn)(void), int expected, int lockclass_mask)
> {
> unsigned long saved_preempt_count = preempt_count();
>+#ifdef CONFIG_SCHED_DEBUG
>+ struct task_struct *p = current;
>+ int save_migrate_atomic;
>+#endif
>
> WARN_ON(irqs_disabled());
>
>+#ifdef CONFIG_SCHED_DEBUG
>+ save_migrate_atomic = p->migrate_disable_atomic;
>+#endif
Shouldn't we save p->migrate_disable as well?
Sebastian
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
