On Sun, Oct 4, 2015 at 3:33 AM, Meelis Roos <mroos@linux.ee> wrote:
>> This is 4.3.0-rc1 on Sun E220R (dual-CPU sparc64). Sometimes it boots,
>> sometimes it fails to boot with looping errors and finally a watchdog
>> timeout. This console log from a failure. Config is below.
>
> I noticed blk-mq related changes in todays git. Retested, loop_init
> still causes the same blk-mq problem.


> [  100.063323] blk_mq_sysfs_init: init hctx_kobj fffff800ad8b95f8 0
> [  100.135151] blk_mq_sysfs_init: init ctx_kobj fffff800af80cdd8 0
> [  100.205952] blk_mq_sysfs_init: init ctx_kobj 00000000009f2dd8 1
> [  100.276785] kobject (00000000009f2dd8): tried to init an initialized object, something is seriously wrong.

From the debug log, the issue happened on ctx_kobj which is one
percpu variable, and looks it can't be initialized by blk-mq more than
one time.

I suspect it may be one memory issue, especiall in percpu allocation.
The percpu variable's address for the offline CPU1 is a bit special, not
like the other two online CPUs. And it might be helpful to enalbe
'Per cpu operations test' in' Kernel hacking  --->'/'Runtime Testing'
of Kconfig to see if it can pass the test.

Thanks,