Re: Regression with a0747a859ef6 ("bdi: add error handle for bdi_debug_register")

[Shaohua Li]

On Tue, Dec 19, 2017 at 10:17:43AM -0600, Bruno Wolff III wrote:
> On Sun, Dec 17, 2017 at 21:43:50 +0800,
>  weiping zhang <zwp10758@xxxxxxxxx> wrote:
> > Hi, thanks for testing, I think you first reproduce this issue(got WARNING
> > at device_add_disk) by your own build, then add my debug patch.
> 
> The problem is still in rc4. Reverting the commit still fixes the problem. I
> tested that warning level messages should appear using lkdtm. While there
> could be something weird relating to the WARN_ON macro, more likely there is
> something different about the boots with the kernels I build (the exact way
> initramfs is built is probably different) and probably that (WARN_ON) code
> is not getting executed.

Not sure if this is MD related, but could you please check if this debug patch
changes anything?

diff --git a/drivers/md/md.c b/drivers/md/md.c
index 4e4dee0..c365179 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -518,7 +518,6 @@ static void mddev_put(struct mddev *mddev)
 	    mddev->ctime == 0 && !mddev->hold_active) {
 		/* Array is not configured at all, and not held active,
 		 * so destroy it */
-		list_del_init(&mddev->all_mddevs);
 		bs = mddev->bio_set;
 		sync_bs = mddev->sync_set;
 		mddev->bio_set = NULL;
@@ -5210,6 +5209,10 @@ static void md_free(struct kobject *ko)
 	}
 	percpu_ref_exit(&mddev->writes_pending);
 
+	spin_lock(&all_mddevs_lock);
+	list_del_init(&mddev->all_mddevs);
+	spin_unlock(&all_mddevs_lock);
+
 	kfree(mddev);
 }