Re: [PATCH 1/2] i2c: qup: Add device tree bindings information

[Ivan T. Ivanov]

Hi, 

On Tue, 2013-09-10 at 08:59 -0500, Josh Cartwright wrote: 
> [Hmm.  Fixing b0rked LKML address; that might explain why I am not
> seeing Kumar's replies.]
> 

Yes, sorry about this.

> On Tue, Sep 10, 2013 at 03:08:57PM +0300, Ivan T. Ivanov wrote:
> > Hi Kumar,
> > 
> > On Fri, 2013-08-30 at 10:21 -0500, Kumar Gala wrote: 
> > > On Aug 29, 2013, at 12:26 PM, Ivan T. Ivanov wrote:
> > > >>> +
> > > >>> +Optional properties :
> > > >>> + - Child nodes conforming to i2c bus binding
> > > >>> + - clock-frequency : Desired I2C bus clock frequency in Hz. If
> > > >>> +		not set thedefault frequency is 100kHz
> > > >>> + - qcom,src-freq : Frequency of the source clocking this bus in Hz.
> > > >>> +		Divider value is set based on soruce-frequency and
> > > >>> +		desired I2C bus frequency. If this value is not
> > > >>> +		provided, the source clock is assumed to be running
> > > >>> +		at 19.2 MHz.
> > > >> 
> > > >> I'd spell out frequency instead of 'freq' to be consistent with 'clock-frequency'
> > > > 
> > > > ok.
> > > > 
> > > >> 
> > > >> Is the frequency of the 'iface' clock?  Can we not use clk_get_rate?
> > > > 
> > > > It is for 'core' clock. I think that for higher I2C bus frequencies,
> > > > 'core' clock have to be higher, but I am not sure what is relation.
> > > 
> > > Ok, can we use clk_get_rate on the 'core' clk to get its frequency instead of needing a DT prop for it?
> > 
> > Probably I didn't explain it well. The 'core' clock have to be
> > accelerated before higher bus frequencies could be achieved.
> 
> I think what Kumar is suggesting is that the QUP driver not do an
> explicit clk_set_rate() at all (which AFAICT is what's currently being
> done to set the consuming clock to the rate specified in
> 'qcom,src-freq'), but instead assume that the consuming clock has
> already been setup properly.  Then the driver just uses clk_get_rate()
> and clock-frequency to calculate how to setup any internal dividers.

Yes, I think I got it, but the point is that clock is not already set 
and it is driver responsibility to do that.

> 
> > > >>> +Aliases: An alias may optionally be used to bind the I2C controller
> > > >>> +to bus number. Aliases are of the form i2c<n> where <n> is an integer
> > > >>> +representing the bus number to use.
> > > >>> +
> > > >>> +Example:
> > > >>> +
> > > >>> + aliases {
> > > >>> +		i2c0 = &i2c_A;
> > > >>> +		i2c1 = &i2c_B;
> > > >>> +		i2c2 = &i2c_C;
> > > >>> + };
> > > >> 
> > > >> What is the purpose here?
> > > > 
> > > > Define on which I2C bus this controller operate. I2C client 
> > > > drivers usually do i2c_get_adapter(bus_number) before its
> > > > registration. This is for drivers before invention of
> > > > of_i2c_register_devices(), I believe.
> > > 
> > > Since this is for upstream why dont we use of_i2c_register_devices() and remove this stuff related to aliases.
> > 
> > Adapter driver already is using of_i2c_register_devices(). Also OF
> > helper function will/or is already part of i2c_register_adapter().
> > Attempt here was to make it compatible with older i2c client drivers.
> 
> I agree with Kumar on removing this.  If we decide it is something worth
> keeping, logic to support it doesn't belong in the QUP driver, but in
> the i2c core.

I don't have strong objection here, will remove aliases support.

Regards,
Ivan 
> 


--
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/