Re: [PATCH v4 1/3] ARM: shmobile: wait for MSTP clock status to toggle, when enabling it

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Geert,

On Sunday 15 December 2013 14:06:31 Geert Uytterhoeven wrote:
> On Sat, Dec 14, 2013 at 4:23 PM, Laurent Pinchart wrote:
> >  static int sh_clk_mstp_enable(struct clk *clk)
> >  {
> >         sh_clk_write(sh_clk_read(clk) & ~(1 << clk->enable_bit), clk);
> > +       if (clk->status_reg) {
> > +               unsigned int (*read)(const void __iomem *addr);
> > +               int i;
> > +               void __iomem *mapped_status = (phys_addr_t)clk->status_reg
> > - +                       (phys_addr_t)clk->enable_reg + clk->mapped_reg;
> > +
> > +               if (clk->flags & CLK_ENABLE_REG_8BIT)
> > +                       read = r8;
> > +               else if (clk->flags & CLK_ENABLE_REG_16BIT)
> > +                       read = r16;
> > +               else
> > +                       read = r32;
> > +
> > +               for (i = 1000;
> > +                    (read(mapped_status) & (1 << clk->enable_bit)) && i;
> > +                    i--)
> > +                       cpu_relax();
> > +               if (!i) {
> > +                       pr_err("cpg: failed to enable %p[%d]\n",
> > +                              clk->enable_reg, clk->enable_bit);
> > +                       return -ETIMEDOUT;
> > +               }
> 
> Just wondering: what's a typical value of i when leaving the loop?

1000 :-)

I've tested the VSP1 on Lager (which sometimes crashed when opening the device 
without this patch) and the loop counter is always 1000 so far. Reading the 
status register once seems to be enough to ensure that the clock really 
starts, although in practice it might take more time for some devices.

-- 
Regards,

Laurent Pinchart