Re: lockdep dump on devtree_lock (involving esdhc)

[Grant Likely <grant.likely@secretlab.ca>]

Thanks

On Tue, Jun 3, 2014 at 4:00 PM, Thomas Gleixner <tglx@linutronix.de> wrote:
> On Tue, 3 Jun 2014, Grant Likely wrote:
>> On Wed, Jun 12, 2013 at 8:49 AM, Thomas Gleixner <tglx@linutronix.de> wrote:
>> > On Tue, 11 Jun 2013, Scott Wood wrote:
>> >
>> >> I get the following lockdump output on p2020rdb using
>> >> v3.10-rc5-43-g34376a5.  While it's not particularly polite for the
>> >> esdhc driver to be calling OF functions while holding another lock which
>> >> can be acquired from interrupt context, why is devtree_lock usually
>> >> acquired in an irqsafe manner but sometimes not?
>> >>
>> >> Both types of usage were added by the same commit:
>> >>
>> >> commit d6d3c4e656513dcea61ce900f0ecb9ca820ee7cd
>> >> Author: Thomas Gleixner <tglx@linutronix.de>
>> >> Date:   Wed Feb 6 15:30:56 2013 -0500
>> >>
>> >>     OF: convert devtree lock from rw_lock to raw spinlock
>> >>
>> >> Stephen, you asked about this here:
>> >> http://lkml.indiana.edu/hypermail/linux/kernel/1302.1/01383.html
>> >>
>> >> Did you ever get an answer?
>> >
>> > https://patchwork.kernel.org/patch/2470731/
>> >
>> >> I'm also curious why devtree_lock was made raw to begin with...
>> >> Iterating over a device tree doesn't seem like something you'd want to
>> >> trust to be low-latency.
>> >
>> > The reason is that it's taken in low level cpu bringup code and I did
>> > not find a different solution. :(
>>
>> Hey, Thomas,
>>
>> Do you remember where the offending code is? I'm looking at completely
>> refactoring FDT locking so I'd like to take a crack and solving the
>> cpu bringup problem.
>
> Hmm, can't remember. All I found is where it got introduced:
>
>      https://lkml.org/lkml/2007/11/9/114
>
> without comments of course :)
>
> Ask someone with a pseries to grab RT and make the lock non
> raw. Enable debugging (mightsleep, lockdep ...)
>
> Boot with maxcpus=1 and then online cores from user space. Splat
> should come right away. I remember that I saw some of them in a
> pastebin.
>
> Thanks,
>
>         tglx
>