[PATCH] MXS: Convert mutexes in clock.c to spinlocks

[marek.vasut@gmail.com (Marek Vasut)]

> > On Mon, Dec 19, 2011 at 10:05:25PM +0100, Marek Vasut wrote:
> > > > On Mon, Dec 19, 2011 at 09:54:25PM +0100, Marek Vasut wrote:
> > > > > > So, in summary, you have everything you require to fix it outside
> > > > > > the driver.  You just have to decide which of the two options you
> > > > > > want to proceed with, and actually (and finally) do it instead of
> > > > > > endlessly procrastinating and waiting for more and more bug
> > > > > > reports (which is exactly what has happened so far.)
> > > > > 
> > > > > What the hell, I just recently found this bug and I submitted a
> > > > > patch right away! What are you complaining about?!
> > > > 
> > > > If you want to take that attitude to my attempt to help you
> > > > understand the problem and see solutions, I'll ignore you
> > > > permanently for being an absolute twit.
> > > 
> > > Go ahead, but you accused me of procrastinating and waiting even if the
> > > first thing I did when I saw the bug was start solving it. That's just
> > > insane!
> > > 
> > > > I'm not going to spend time giving a detailed explaination
> > > > about the background and options over something to only then have it
> > > > immediately shoved back in my face with such a response.
> > > 
> > > I consider my response to the last part of your email appropriate.
> > 
> > Sorry, it wasn't directed personally at you, but to the entire MXS
> 
> > community.  The facts over this are:
> Ah! I'm sorry I was so direct and rude too. I was unaware it was discussed
> before, I started this effort on my own just recently.
> 
> > 1. This problem has been known about since October.
> 
> I was really away from the kernel community for a while so I didn't know.
> 
> > 2. It's been discussed several times - every time along the same lines.
> > 3. There is zero apparant progress on the issue.
> > 
> > Here's two of the discussions over it, where I've said exactly the same
> > thing:
> > 
> > http://lists.arm.linux.org.uk/lurker/thread/20111018.173744.46c4bd76.en.h
> > tm l
> > http://lists.arm.linux.org.uk/lurker/thread/20111123.183640.222b05cf.en.h
> > t ml
> > 
> > So now, tell me - is this _finally_ going to get fixed in the MXS code,
> > or is the previous discussion about converting stuff to spinlocks etc
> > just going to be repeated yet again?
> 
> Spinlocks are OK as far as the code within them is fast, right ? But hm ...
> actually, we might be able to toggle the clock in one instruction by using
> the bitwise set/clear registers. That way, we won't need the locks at all,
> but we'd loose the usecount ... which is useless anyway).
> 
> M

I see ... clk_set_rate() can be slow. OK, looking further into it.