Hi Paul, Zhou,
On Fri, Nov 29, 2019 at 12:23:42PM +0100, Paul Cercueil wrote:
> > > > @@ -93,8 +93,17 @@ ingenic_pll_recalc_rate(struct clk_hw *hw,
> > > > unsigned long parent_rate)
> > > > n += pll_info->n_offset;
> > > > od_enc = ctl >> pll_info->od_shift;
> > > > od_enc &= GENMASK(pll_info->od_bits - 1, 0);
> > > > - bypass = !pll_info->no_bypass_bit &&
> > > > - !!(ctl & BIT(pll_info->bypass_bit));
> > > > +
> > > > + if (pll_info->version >= CGU_X1830) {
> > > > + spin_lock_irqsave(&cgu->lock, flags);
> > > > + ctl = readl(cgu->base + pll_info->reg[0]);
> > > > + spin_unlock_irqrestore(&cgu->lock, flags);
> > >
> > > Why the spinlock?
> > >
> >
> > The original code used spinlock when reading the control register,
> > so when reading this new control register, I think it should also
> > be added with spinlock.
>
> Well, the original code looks wrong to me. There's nothing to protect here.
>
> Maybe @Paul Burton can shed some light?
I wish I could remember, but I agree it seems pointless here. The only
way I can think it could be of any use is if writes to the CGU register
we're accessing aren't atomic (ie. if we could observe a partially
completed write), but I don't believe that's the case.
So Zhou, if you want to drop the spinlock here from your X1830 path &
ideally also add a patch to remove it in the non-X1830 path that would
be great.
Thanks,
Paul
