Re: [PATCH] OMAP: Fix McBSP spin_lock deadlock.

["stanley.miao" <stanley.miao@windriver.com>]

On Wed, 2008-11-05 at 17:16 -0800, Tony Lindgren wrote:
> * David Brownell <david-b@pacbell.net> [081105 15:53]:
> > On Wednesday 05 November 2008, Stanley.Miao wrote:
> > > @@ -333,7 +333,8 @@ struct omap_mcbsp_platform_data {
> > >         u8 dma_rx_sync, dma_tx_sync;
> > >         u16 rx_irq, tx_irq;
> > >         struct omap_mcbsp_ops *ops;
> > > -       char const *clk_name;
> > > +       char const *ick_name;
> > > +       char const *fck_name;
> > >  };
> > >  
> > >  struct omap_mcbsp {
> > 
> > I thought the idea was to fix this using the clock aliasing
> > facility ... so that the mcbsp driver would just
> > 
> > 	ick = clk_get(dev, "ick");
> > 	fck = clk_get(dev, "fck");
> > 
> > (with IS_ERR checks of course) and the platform data would
> > no longer need to hold those names.
> > 
> > That approach can work on OMAP1 too ... I think this patch
> > breaks OMAP1 boards.
> 
> Russell also has some ideas, so despite the nasty bug, let's put
> the clock aliasing and virtual clocks stuff on hold for a little
> while. Meanwhile, maybe we can just add some function for the
> virtual clock that does not cause the recursion?

At the start, I tried to add a function __clk_enable().

+int __clk_enable(struct clk *clk)
+{
+       if(arch_clock->clk_enable)
+               return arch_clock->clk_enable(clk);
+       return 0;
+}
+EXPORT_SYMBOL(__clk_enable);
+
 int clk_enable(struct clk *clk)
 {
        unsigned long flags;
@@ -84,33 +92,36 @@ int clk_enable(struct clk *clk)
                return -EINVAL;
 
        spin_lock_irqsave(&clockfw_lock, flags);
-       if (arch_clock->clk_enable)
-               ret = arch_clock->clk_enable(clk);
+       ret = __clk_enable(clk);
        spin_unlock_irqrestore(&clockfw_lock, flags);
 
        return ret;
 }
 EXPORT_SYMBOL(clk_enable);

But in this way I have to add __clk_enable() to include/linux/clk.h, so
I give up this method.

Stanley.

> 
> Tony