Re: [PATCH 1/3] mfd: support 88pm80x in 80x driver

Re: [PATCH 1/3] mfd: support 88pm80x in 80x driver

[Arnd Bergmann]

On Monday 02 July 2012, Qiao Zhou wrote:
> On 07/02/2012 06:12 PM, Mark Brown wrote:
> > On Mon, Jul 02, 2012 at 06:09:57PM +0800, Qiao Zhou wrote:
> >> On 07/02/2012 06:03 PM, Mark Brown wrote:
> >
> >>> What do you mean by pages?  regmap has paging support which just maps
> >>> everything into a single flat register map from the point of view of
> >>> callers.
> >
> >> Mark, let me explain: the 88pm800 chip has three i2c address
> >> internally, which we called different page instead. it confuses you
> >> with the register page_read/write operation. there are registers in
> >> each i2c address domain, and we need to use different i2c client to
> >> access reg in different domain. such as some common regs are in the
> >> page of i2c_addr = 0x30, and power related regs are in the page of
> >> i2c_addr = 0x31, and gpadc related regs are in the page of 0x32.
> >
> > These aren't what people normally call pages, those are just separate
> > I2C devices from a Linux point of view.
> >
> Mark, surely I'll pay attention to the terms used. thanks!
> due to there separate I2C devices, does it make sense to export separate 
> r/w interface for them? do you have suggestion in such case?

(adding the i2c mailing list to get more insight)

I think in case of device tree based probing, it would be straightforward
to represent 88pm800 as a single device with three addresses in the "reg"
property, while the natural linux representation would be one regular
i2c_client device with two dummies. Do we or should we have any
infrastructure to deal with this?

If this is a common scenario, we could probably let regmap handle it
entirely internally and represent the i2c client with its dummies
as a single regmap.

	Arnd

Re: [PATCH 1/3] mfd: support 88pm80x in 80x driver

[Qiao Zhou]

On 07/02/2012 11:58 PM, Arnd Bergmann wrote:
> On Monday 02 July 2012, Qiao Zhou wrote:
>> On 07/02/2012 06:12 PM, Mark Brown wrote:
>>> On Mon, Jul 02, 2012 at 06:09:57PM +0800, Qiao Zhou wrote:
>>>> On 07/02/2012 06:03 PM, Mark Brown wrote:
>>>
>>>>> What do you mean by pages?  regmap has paging support which just maps
>>>>> everything into a single flat register map from the point of view of
>>>>> callers.
>>>
>>>> Mark, let me explain: the 88pm800 chip has three i2c address
>>>> internally, which we called different page instead. it confuses you
>>>> with the register page_read/write operation. there are registers in
>>>> each i2c address domain, and we need to use different i2c client to
>>>> access reg in different domain. such as some common regs are in the
>>>> page of i2c_addr = 0x30, and power related regs are in the page of
>>>> i2c_addr = 0x31, and gpadc related regs are in the page of 0x32.
>>>
>>> These aren't what people normally call pages, those are just separate
>>> I2C devices from a Linux point of view.
>>>
>> Mark, surely I'll pay attention to the terms used. thanks!
>> due to there separate I2C devices, does it make sense to export separate
>> r/w interface for them? do you have suggestion in such case?
>
> (adding the i2c mailing list to get more insight)
>
> I think in case of device tree based probing, it would be straightforward
> to represent 88pm800 as a single device with three addresses in the "reg"
> property, while the natural linux representation would be one regular
> i2c_client device with two dummies. Do we or should we have any
> infrastructure to deal with this?
>
> If this is a common scenario, we could probably let regmap handle it
> entirely internally and represent the i2c client with its dummies
> as a single regmap.
actually there are many drivers under mfd which have this common issue, 
which has i2c dummy devices, such as max77693.c, max8925-i2c.c, 
ab3100-core.c, max8997.c, max8998.c, s5m-core.c etc. some use regmap 
handle directly as param in exported r/w api, some add extra param to 
differentiate i2c dummy. it seems to be a common scenario. how do we 
handle the API in short term and long term?
>
> 	Arnd
>


-- 

Best Regards
Qiao