Re: [PATCH 1/2] clk: imx: do not sleep if IRQ's are still disabled

[Shawn Guo <shawnguo@kernel.org>]

On Tue, Apr 26, 2016 at 01:51:13PM +0800, Dong Aisheng wrote:
> Hi Shawn,
> 
> On Tue, Apr 26, 2016 at 9:23 AM, Shawn Guo <shawnguo@kernel.org> wrote:
> > On Thu, Apr 21, 2016 at 11:45:20AM +0800, Dong Aisheng wrote:
> >> On Fri, Jan 29, 2016 at 02:49:23PM -0800, Stefan Agner wrote:
> >> > If a clock gets enabled early during boot time, it can lead to a PLL
> >> > startup. The wait_lock function makes sure that the PLL is really
> >> > stareted up before it gets used. However, the function sleeps which
> >> > leads to scheduling and an error:
> >> > bad: scheduling from the idle thread!
> >> > ...
> >> >
> >> > Use udelay in case IRQ's are still disabled.
> >> >
> >> > Signed-off-by: Stefan Agner <stefan@agner.ch>
> >> > ---
> >> >  drivers/clk/imx/clk-pllv3.c | 5 ++++-
> >> >  1 file changed, 4 insertions(+), 1 deletion(-)
> >> >
> >> > diff --git a/drivers/clk/imx/clk-pllv3.c b/drivers/clk/imx/clk-pllv3.c
> >> > index c05c43d..b5ff561 100644
> >> > --- a/drivers/clk/imx/clk-pllv3.c
> >> > +++ b/drivers/clk/imx/clk-pllv3.c
> >> > @@ -63,7 +63,10 @@ static int clk_pllv3_wait_lock(struct clk_pllv3 *pll)
> >> >                     break;
> >> >             if (time_after(jiffies, timeout))
> >> >                     break;
> >> > -           usleep_range(50, 500);
> >> > +           if (unlikely(irqs_disabled()))
> >>
> >> This causes a bit confusion that clk_pllv3_prepare is sleepable.
> >> But this line indicates it's possible to be called in irq context.
> >> Although it's only happened during kernel boot phase where irq is
> >> still not enabled.
> >> It seems schedule_debug() somehow did not catch it during early boot
> >> phase. Maybe schedule guys can help explain.
> >>
> >> My question is if it's really worthy to introduce this confusion
> >> to fix the issue since the delay is minor?
> >
> > I do not understand why it's confusing.  The code already obviously
> > indicates this is a special handling for cases where irq is still not
> > enabled, rather than for irq context.
> >
> 
> The code itself has nothing telling it's a special handling for the
> case where irq is
> still not enabled.

I think the following if-clause is telling that.

	if (unlikely(irqs_disabled()))

> Even it tells, it may still cause confusing by adding complexity in
> clk_pllv3_prepare()
> which actually should be called in non-atomic context as it could sleep.

I agree with you on that.

> > The patch is to fix the "bad: scheduling from the idle thread!" warning
> > rather than minimize the delay.  Do you have an opinion on how to fix
> > the warning?
> >
> 
> I just wonder maybe we could simply just using udelay(50) instead of
> usleep_range(50, 500) to eliminate the confusing since it's minor cast.
> What do you think of it?

I'm fine with it.  Since I haven't sent the patch to clk maintainers, I
could replace Stefan's patch with yours, if you can send me a patch
quickly.

Shawn