Re: [PATCH 2/2] drivers: clk: st: address sparse warnings

From: Nicholas Mc Guire
Date: Thu Jul 26 2018 - 01:52:16 EST


On Wed, Jul 25, 2018 at 01:40:53PM -0700, Stephen Boyd wrote:
> Quoting Nicholas Mc Guire (2018-07-15 03:18:24)
> > Refactoring of code to make it more readable and at the same time make
> > sparse happy again.
> >
> > Signed-off-by: Nicholas Mc Guire <hofrat@xxxxxxxxx>
> > ---
> >
> > sparse complained about:
> > drivers/clk/st/clkgen-pll.c:225:12: warning: context imbalance in 'clkgen_pll_enable' - different lock contexts for basic block
> > drivers/clk/st/clkgen-pll.c:267:9: warning: context imbalance in 'clkgen_pll_disable' - different lock contexts for basic block
> > drivers/clk/st/clkgen-pll.c:413:9: warning: context imbalance in 'set_rate_stm_pll3200c32' - different lock contexts for basic block
> > drivers/clk/st/clkgen-pll.c:570:9: warning: context imbalance in 'set_rate_stm_pll4600c28' - different lock contexts for basic block
> >
> > Which are technically false positives as the pll->lock which is being
> > checked is determined at configuration time, thus the locks are balanced.
> > Never the less the refactored code seems more readable and was
> > commented to make it clear.
>
> This stuff above should go into the commit text.

ok - then I´ll resend that with that moved into the commit message.

>
> >
> > Patch was compile tested with: multi_v7_defconfig (implies
> > CONFIG_ARCH_STI=y)
> >
> > Patch is against 4.18-rc4 (localversion-next is next-20180713)
> >
> > drivers/clk/st/clkgen-pll.c | 51 +++++++++++++++++++++++++--------------------
> > 1 file changed, 28 insertions(+), 23 deletions(-)
>
> I believe this driver is in stasis mode. Not sure anyone is testing this
> right now. Please Cc people who have worked on this driver, like Gabriel
> Fernandez.
>
> >
> > diff --git a/drivers/clk/st/clkgen-pll.c b/drivers/clk/st/clkgen-pll.c
> > index 7a7106dc..cbb5184 100644
> > --- a/drivers/clk/st/clkgen-pll.c
> > +++ b/drivers/clk/st/clkgen-pll.c
> > @@ -228,13 +228,13 @@ static int clkgen_pll_enable(struct clk_hw *hw)
> > unsigned long flags = 0;
> > int ret = 0;
> >
> > - if (pll->lock)
> > + if (pll->lock) {
> > + /* stih418 and stih407 */
> > spin_lock_irqsave(pll->lock, flags);
> > -
> > - ret = __clkgen_pll_enable(hw);
> > -
> > - if (pll->lock)
> > + ret = __clkgen_pll_enable(hw);
> > spin_unlock_irqrestore(pll->lock, flags);
> > + } else
> > + ret = __clkgen_pll_enable(hw);
> >
> > return ret;
> > }
> > @@ -259,13 +259,13 @@ static void clkgen_pll_disable(struct clk_hw *hw)
> > struct clkgen_pll *pll = to_clkgen_pll(hw);
> > unsigned long flags = 0;
> >
> > - if (pll->lock)
> > + if (pll->lock) {
> > + /* stih418 and stih407 */
> > spin_lock_irqsave(pll->lock, flags);
> > -
> > - __clkgen_pll_disable(hw);
> > -
> > - if (pll->lock)
> > + __clkgen_pll_disable(hw);
> > spin_unlock_irqrestore(pll->lock, flags);
> > + } else
> > + __clkgen_pll_disable(hw);
> > }
> >
> > static int clk_pll3200c32_get_params(unsigned long input, unsigned long output,
> > @@ -400,15 +400,18 @@ static int set_rate_stm_pll3200c32(struct clk_hw *hw, unsigned long rate,
> >
> > __clkgen_pll_disable(hw);
> >
> > - if (pll->lock)
> > + if (pll->lock) {
> > + /* stih407 and stih418 */
> > spin_lock_irqsave(pll->lock, flags);
> > -
> > - CLKGEN_WRITE(pll, ndiv, pll->ndiv);
> > - CLKGEN_WRITE(pll, idf, pll->idf);
> > - CLKGEN_WRITE(pll, cp, pll->cp);
> > -
> > - if (pll->lock)
> > + CLKGEN_WRITE(pll, ndiv, pll->ndiv);
> > + CLKGEN_WRITE(pll, idf, pll->idf);
> > + CLKGEN_WRITE(pll, cp, pll->cp);
> > spin_unlock_irqrestore(pll->lock, flags);
> > + } else {
> > + CLKGEN_WRITE(pll, ndiv, pll->ndiv);
> > + CLKGEN_WRITE(pll, idf, pll->idf);
> > + CLKGEN_WRITE(pll, cp, pll->cp);
> > + }
>
> Please don't duplicate this code. The sparse warnings can be fixed by
> adding a fake lock acquire to the else of the if condition. We do this
> in drivers/clk/clk.c so you should be able to copy it from there.

the duplication is not a mater of the sparse warning only - the inetnt was
to improve readability - atleast for the one-line cases it seems more
readable to have it this way than to have the two lock checks - notably
as you can then comment what the difference here really is.

I agree that for this case with the three lines in the body it is not
that reasonable - this was simply a matter of consistency as the other lock
checks were moved into a if-else construct for readability and commenting.

thx!
hofrat