Re: BUG: lock held when returning to user space

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Sat, 2007-10-27 at 17:12 +0200, Jiri Kosina wrote:
> On Sat, 27 Oct 2007, Gabriel C wrote:
> 
> > I found that today in dmesg after booting current git ( 
> > ec3b67c11df42362ccda81261d62829042f223f0 ) :
> > ...
> > [  592.752777]
> > [  592.752781] ================================================
> > [  592.753478] [ BUG: lock held when returning to user space! ]
> > [  592.753880] ------------------------------------------------
> > [  592.754262] hwclock/1452 is leaving the kernel with locks still held!
> > [  592.754655] 1 lock held by hwclock/1452:
> > [  592.755007]  #0:  (&rtc->char_lock){--..}, at: [<c02a7ebb>] rtc_dev_open+0x2e/0x7e                                        
> 
> Yes, this is because rtc keeps a char_lock mutex locked as long as the 
> device is open, to avoid concurrent accessess.
> 
> It could be easily substituted by some counting -- setting and clearing 
> bit in struct rtc_device instead of using char_lock, but doing this just 
> to shut the lockdep off is questionable imho.
> 
> Peter, what is the preferred way to annotate these kinds of locking for 
> lockdep to express that it is intended?

Not sure, I'd not thought that anyone would actually want to do this.
I'm also not sure how I stand on this, I'd prefer to say: don't do this!

I think, in this case, the lock is associated with a kernel object that
is properly cleaned up if the holding tasks gets a SIGKILL. But in
general I'd like to see this kind of thing go away.

Now I could probably come up with an annotation to hide it, but what do
other people think, Ingo, Linus, Andrew, do we want to keep kernel locks
held over userspace?