[PATCH V3] clk: imx: Fix reparenting of UARTs not associated with sdout

[Sascha Hauer]

On Wed, Jan 20, 2021 at 06:14:21PM +0200, Abel Vesa wrote:
> On 21-01-20 16:50:01, Sascha Hauer wrote:
> > On Wed, Jan 20, 2021 at 05:28:13PM +0200, Abel Vesa wrote:
> > > On 21-01-20 16:13:05, Sascha Hauer wrote:
> > > > Hi Abel,
> > > > 
> > > > On Wed, Jan 20, 2021 at 04:44:54PM +0200, Abel Vesa wrote:
> > > > > On 21-01-18 08:00:43, Adam Ford wrote:
> > > > > > On Mon, Jan 18, 2021 at 6:52 AM Abel Vesa <abel.vesa at nxp.com> wrote:
> > > 
> > > ...
> > > 
> > > > > > 
> > > > > > >
> > > > > > > TBH, I'm against the idea of having to call consumer API from a clock provider driver.
> > > > > > > I'm still investigating a way of moving the uart clock control calls in drivers/serial/imx,
> > > > > > > where they belong.
> > > > > > 
> > > > > > That makes sense.
> > > > > > 
> > > > > 
> > > > > Just a thought. The uart clock used for console remains on from u-boot,
> > > > > so maybe it's enough to just add the CLK_IGNORE_UNUSED flag to all the
> > > > > uart root clocks and remove the prepare/enable calls for uart clocks 
> > > > > for good. I don't really have a way to test it right now, but maybe
> > > > > you could give it a try.
> > > > 
> > > > That would mean that UART clocks will never be disabled, regardless of
> > > > whether they are used for console or not. That doesn't sound very
> > > > appealing.
> > > 
> > > AFAIK, the only uart clock that is enabled by u-boot is the one used for
> > > the console. Later on, when the serial driver probes, it will enable it itself.
> > > 
> > > Unless I'm missing something, this is exactly what we need.
> > 
> > It might enable it, but with CLK_IGNORE_UNUSED the clock won't be
> > disabled again when a port is closed after usage
> 
> OK, tell me what I'm getting wrong in the following scenario:
> 
> U-boot leaves the console uart clock enabled. All the other ones are disabled.
> 
> Kernel i.MX clk driver registers the uart clocks with flag CLK_IGNORE_UNUSED.

I was wrong at that point. I originally thought the kernel will never
disable these clocks, but in fact it only leaves them enabled during the
clk_disable_unused call.

However, when CLK_IGNORE_UNUSED becomes relevant it's too late already.
I just chatted with Lucas and he told me what the original problem was
that his patch solved.

The problem comes when an unrelated device and the earlycon UART have
the same parent clocks. The parent clock is enabled, but it's reference
count is zero. Now when the unrelated device probes and toggles its
clocks then the shared parent clock will be disabled due to the
reference count being zero. Next time earlycon prints a character the
system hangs because the UART gates are still enabled, but their parent
clocks no longer are.

Overall I think Lucas' patches are still valid and relevant and with
Adams patches we even no longer have to enable all UART clocks, but
only the ones which are actually needed.

Sascha


-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |