Re: Pinmux bindings proposal V2

[Shawn Guo]

On Fri, Jan 27, 2012 at 09:05:45AM -0800, Tony Lindgren wrote:
> Hi,
> 
> * Shawn Guo <shawn.guo@xxxxxxxxxx> [120126 22:15]:
> > On Thu, Jan 26, 2012 at 06:08:33PM -0800, Tony Lindgren wrote:
> > > * Stephen Warren <swarren@xxxxxxxxxx> [120126 11:03]:
> > ...
> > > > Second, as I mentioned before, while some of the states are certainly
> > > > PM-related, I don't think all will be, e.g. the case of running an SD
> > > > controller at different clock rates to the SD card, and needing to
> > > > set different pin parameters based on the clock rate. Is runtime PM
> > > > intended cover that kind of thing? The idea here is that the common
> > > > pinctrl binding can allow the driver to require different named states
> > > > for those different clock rate cases.
> > > 
> > > For the PM related states, those should be Linux generic. For rate
> > > setting sounds like that's really something you should set up as clocks
> > > in the Tegra wrapper driver for SDHCI?
> > > 
> > That's right.
> > 
> > > Ideally the SDHCI driver would be completely arch independent, and
> > > then the SoC specific wrapper driver would know how to communicate to
> > > the pinmux/pinconf framwork or clock framework what it needs using
> > > Linux generic APIs.
> > 
> > But that wrapper driver should not be bothered to call pinmux/pinconf
> > APIs on pin basis to change the pinctrl configuration.  The elegant
> > way would be something like the following in case that it switches
> > the bus frequency from 50 MHz to 100 MHz.
> > 
> > 	pmx = pinmux_get(dev, "esdhc_50mhz");
> > 	...
> > 	pinmux_put(pmx);
> > 	pmx = pinmux_get(dev, "esdhc_100mhz");
> > 	...
> > 
> > The specific mux and config settings of states esdhc_50mhz and
> > esdhc_100mhz would be retrieved from device tree.
> 
> Yes whatever mux names can be used, same as with clock framework
> for clock names. But that means you'll have to constantly get/put
> the mux which is not efficient.
> 
The most important reason that we want to move to pinctrl subsystem
is we need its run-time configuration feature for cases like esdhc
here.  I do not think the switch here is so constant to be inefficient.

> Wouldn't it be cleaner to just clk_get esdhc_clk during init, then
> do clk_set_rate on it to toggle the rates?
>  
It's not an init-time switch but run-time one.  That said,
sdhci_ops.set_clock will be called during run-time.

> > > So I'd rather stay out of random named states for
> > > the pins coming from device tree; If we still need them, they should
> > > be common bindings rather than things like "xyz_clock_hack".
> > > 
> > The binding defines the syntax, and I do not see the necessity to
> > force the particular state name, which is really pinctrl client
> > device specific.
> 
> Do you have some other custom pin state example other than the
> clock rate change example above?
> 
I have another case PM related.  To aggressively save power, the pins
configured for particular function during active mode need to be
muxed on gpio mode and output 0 in low-power mode.

-- 
Regards,
Shawn
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/